Network Strategy (Advances in Strategic Management)

ADVANCES IN STRATEGIC MANAGEMENT Series Editor: Joel A. C. Baum Recent Volumes: Volume 15: Disciplinary Roots of Strate...

Author: Joel Baum | Tim Rowley

54 downloads 2239 Views 6MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

ADVANCES IN STRATEGIC MANAGEMENT Series Editor: Joel A. C. Baum Recent Volumes: Volume 15:

Disciplinary Roots of Strategic Management Research

Volume 16:

Population-Level Learning and Industry Change

Volume 17:

Economics Meets Sociology in Strategic Management

Volume 18:

Multiunit Organization and Multimarket Strategy

Volume 19:

The New Institutionalism in Strategic Management

Volume 20:

Geography and Strategy

Volume 21:

Business Strategy over the Industry Lifecycle

Volume 22:

Strategy Process

Volume 23:

Ecology and Strategy

Volume 24:

Real Options Theory

ADVANCES IN STRATEGIC MANAGEMENT VOLUME 25

NETWORK STRATEGY EDITED BY

JOEL A. C. BAUM Rotman School of Management, University of Toronto, Toronto, ON, Canada

TIMOTHY J. ROWLEY Rotman School of Management, University of Toronto, Toronto, ON, Canada

United Kingdom – North America – Japan India – Malaysia – China

JAI Press is an imprint of Emerald Group Publishing Limited Howard House, Wagon Lane, Bingley BD16 1WA, UK First edition 2008 Copyright r 2008 Emerald Group Publishing Limited Reprints and permission service Contact: [email protected] No part of this book may be reproduced, stored in a retrieval system, transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright Clearance Center. No responsibility is accepted for the accuracy of information contained in the text, illustrations or advertisements. The opinions expressed in these chapters are not necessarily those of the Editor or the publisher. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-7623-1442-3 ISSN: 0742-3322 (Series)

Awarded in recognition of Emerald’s production department’s adherence to quality systems and processes when preparing scholarly journals for print

LIST OF CONTRIBUTORS Barak Aharonson

Olin Business School, Washington University in St. Louis, St. Louis, MO, USA

Andreas Al-Laham

Lehrstuhl fu¨r Internationales Management, University of Kaiserslautern, Kaiserslautern

Terry L. Amburgey

Rotman School of Management, University of Toronto, Toronto, ON, Canada

Ted Baker

College of Management, North Carolina State University, Raleigh, NC, USA

Joel A. C. Baum

Rotman School of Management, University of Toronto, Toronto, ON, Canada

Ronald S. Burt

Graduate School of Business, University of Chicago, Chicago, IL, USA

Turanay Caner

College of Management, North Carolina State University, Raleigh, NC, USA

Alessandra Carlone

Bocconi University, Milan, Italy

Martin J. Conyon

ESSEC Business School, France

Robin Cowan

BETA, Universite´ Louis Pasteur, Strasbourg, France and UNU-MERIT, University of Maastricht, Maastricht, The Netherlands

Giovanni Battista Dagnino

University of Catania, Catania, Italy

ix

x

LIST OF CONTRIBUTORS

Gerald F. Davis

Ross School of Business, University of Michigan, Ann Arbor, MI, USA

Arabella Mocciaro Li Destri

University of Palermo, Viale delle Scienze, Palermo, Italy

Patrick Doreian

University of Pittsburgh, Pittsburgh, PA, USA

Hans T. W. Frankort

University of Maastricht, Maastricht, The Netherlands

Bret R. Fund

University of Colorado-Boulder, Boulder, CO, USA

John Hagedoorn

University of Maastricht, Maastricht, The Netherlands

Julie M. Hite

Brigham Young University, Provo, UT, USA

Bala Iyer

Technology, Operations and Information Management Division, Babson College, Babson Park, MA, USA

Nicolas Jonard

Universite du Luxembourg, Luxembourg

Balaji Koka

Jesse H. Jones Graduate School of Business, Rice University, Houston, TX, USA

Otto R. Koppius

Rotterdam School of Management, Erasmus University Rotterdam, Rotterdam, The Netherlands

Chi-Hyon Lee

School of Management, George Mason University, Fairfax, VA, USA

Gabriella Levanti

University of Palermo, Viale delle Scienze, Palermo, Italy

Bjørn Løva˚s

London Business School

Ravi Madhavan

Katz Graduate School of Business, University of Pittsburgh, Pittsburgh, PA, USA

xi

List of Contributors

Bill McEvily

Rotman School of Management, University of Toronto, Toronto, ON, Canada

Mark S. Mizruchi

University of Michigan, Ann Arbor, MI, USA

Mark R. Muldoon

School of Mathematics, University of Manchester, Manchester, UK

Eric J. Neuman

College of Business, University of Illinois, Champaign, IL

Timothy G. Pollock

Pennsylvania State University, University Park, PA, USA

John Prescott

Katz Graduate School of Business, University of Pittsburgh, Pittsburgh, PA, USA

Ray Reagans

Tepper School of Business, Carnegie Mellon University, Pittsburgh, PA, USA

Timothy J. Rowley

Rotman School of Management, University of Toronto, Toronto, ON, Canada

Giuseppe Soda

SDA Bocconi University School of Management, Milan, Italy

Olav Sorenson

Rotman School of Management, University of Toronto, Toronto, ON, Canada

Danny Tzabbar

University of Central Florida, Orlando, Fl, USA

Diederik W. van Liere

Rotterdam School of Management, Erasmus University Rotterdam, Rotterdam, The Netherlands

N. Venkatraman

School of Management, Boston University, Boston, MA, USA

Peter H. M. Vervest

Rotterdam School of Management, Erasmus University Rotterdam, Rotterdam, The Netherlands

xii

LIST OF CONTRIBUTORS

Gordon Walker

Cox School of Business, Southern Methodist University, Dallas, TX, USA

Adam J. Wowak

Pennsylvania State University, University Park, PA, USA

Akbar Zaheer

Carlson School of Management, University of Minnesota, Minneapolis, MN, USA

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES Knowledge and information have become increasingly important assets to contemporary ﬁrms, and as a result, network approaches have become central to their strategic organization as ﬁrms turn to cooperative arrangements to gain advantage. In industrial economies characterized by oligopolistic competition in which a few large ﬁrms dominated their industries, scale economies dominated strategic thinking. Interﬁrm collaboration was less common than competition. In knowledge and information economies, however, network economies have become vital for strategic action and success (Shapiro & Varian, 1999). Networks are increasingly important in modern-day economies because technological and competitive advantage can be rapidly eroded by knowledge and information emerging from beyond as well as within a ﬁrm’s industry. Success depends on the value and uniqueness of a ﬁrm’s knowledge to its stakeholders. To develop information and knowledge-based advantages, ﬁrms have increasingly turned to cooperative ties to access other ﬁrms’ complimentary expertise, valuable information ﬂows, and novel technological developments that reduce uncertainty and facilitate initiation of additional ties. The proliferation of such cooperative ties has created interﬁrm networks – evolving webs of linkages spanning and linking entire industries. Inside organizations, hierarchical structures have similarly given way to networks as the means of creating value for the organization and pursing individual and group goals. Made possible by modern information technologies, more ﬂexible and disaggregated forms of production have emerged, initiating a shift from hierarchical to network organizational structures characterized by horizontal linkages both within and across organizations (Nohria, 1992). Entrepreneurial ﬁrms have challenged incumbents in many industries with strategies based on knowledge advantages derived from collaboration within and beyond their boundaries. Established ﬁrms have attempted to respond by restructuring themselves along network lines, outsourcing activities to suppliers and partnering with other ﬁrms to access technological xiii

xiv

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

developments and complimentary expertise, allowing them to concentrate on reﬁning core activities in which they possess a unique advantage (Kogut, 1988; Oliver, 1991; Powell, 1990; Powell, Koput, & Smith-Doerr, 1996). In part, network models of strategic organization originated in the empirical observations of network forms operating effectively in new information-rich environments. However, the focus on networks in the study of organizations and strategy preceded the rise of network forms of strategic organization. Perhaps the sharpest motivation for adopting the network perspective has been to move away from the individualist, essentialist and atomist assumptions and explanations in economic theory that treat individual choices and exchanges as independent, toward more relational, contextual and systemic appreciations. From a network perspective, individual rationality is a variable rather than an assumption (Stinchcombe, 1985); markets are conceived not as a pricing mechanism among anonymous ﬁrms, but as meeting places and repositories of exchange histories and existing relations all of which affect future patterns of exchange (Smelser & Swedberg, 1994; Gulati & Gargiulo, 1999). Network theory has also contributed to research in strategy and organizations by affording greater precision in the conceptualization of environments in terms of the relational system surrounding individuals, groups and ﬁrms, and helping to specify sources and mechanisms underlying environmental muniﬁcence, uncertainty and change (Wasserman, 1992).

NETWORK THEORY: KEY ASSUMPTIONS AND RESEARCH Two core assumptions set network theory apart from other perspectives and direct research into speciﬁc strategic and organizational topics. From a network perspective, social actors – individuals, groups and ﬁrms – and their actions are viewed as interdependent and the relational ties among them as pipes through which inﬂuence and resources ﬂow, and prisms through which their qualities are reﬂected (Podolny, 2001). Firms’ relationships may be symbiotic, the resources accessed and status obtained through one tie making ties with other ﬁrms possible and/or valuable (Powell et al., 1996; Uzzi & Gillespie, 2002). Partnering decisions are also inﬂuenced by strategic considerations such as the ability to play one partner off against another (Burt, 1992; Simmel, 1950; Willer, 1999), the desire to create an appropriate mix of new and old ties (Levitt & March, 1988;

Introduction: Evolving Webs in Network Economies

xv

Baum & Ingram, 2002), or the need to maintain a manageable alliance portfolio size (Li & Rowley, 1999) and composition (Gomes-Casseres, 2003; Rowley, Baum, Shipilov, Greve, & Rao, 2004). Within networks, the evolving pattern of relationships – the network structure – shapes opportunities and constraints. Adopting Burt’s (this volume) language, the pattern of network ties surrounding social actors is a ‘‘causal spark’’ that can promote or hinder action. Networks take shape as actors enter into collaborative relations based on information about the quality, trustworthiness, and status of potential partners obtained through experiences in their own and their partners’ past relationships. The more the network internalizes information about potential partners, the more actors look to the network for cues about future relationships. And, in building new ties, actors contribute to the structuring of the network that shapes and constrains their future actions. Together, these assumptions suggest that networks affect actors’ behavior and performance by serving as conduits through which information, knowledge and other resources ﬂow, and reputations are signaled. And, that the topologies of networks and positions of actors within them determine which actors will have access to and control over resources and information ﬂowing through the network pipes and which will shine brightly in the network prism. In the network model, then, the social structure of organizational life is the mechanism inspiring strategic action and competitive advantage. In practical terms, these assumptions have guided research toward the area of social capital, which relates the network structure surrounding actors to their behavior and performance. The central questions in social capital research on strategy and organizations pertain to the types of network structures and positions that confer advantages on social actors. Research emphasizes to two types of advantageous network structures and positions. Burt (1992) equates social capital with the lack of ties among an actor’s partners, a structural property he terms structural holes. He argues that the spanning of structural holes provides efﬁciency and brokering advantages based on the ability to arbitrage non-redundant information and resource exchanges. In contrast to Burt, Coleman’s (1988) view of social capital, calls for a dense connections among an actor’s partners to promote trust and cooperation among them. These contrasting views of beneﬁcial network structures triggered a productive stream of research. While some studies emphasize the superior performance of actors spanning structural holes (e.g., McEvily & Zaheer, 1999; Rowley & Baum, 2004; Soda, Usai, & Zaheer, 2004) or having dense

xvi

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

ties (Ahuja, 2000; Walker, Kogut, & Shan, 1997), others adopt contingency approaches in which exogenous factors (e.g., uncertainty, opportunism) or tie function (e.g., exploration or exploitation) to clarify the conditions under which each type of structure would be most beneﬁcial (e.g., Podolny & Baron, 1997; Rowley, Behrens, & Krackhardt, 2000). These studies share a theoretical view in which actors enjoy advantages of either structural holes or dense ties among their partners – but not both. Recently, researchers have begun to examine beneﬁts of ‘‘hybrid’’ network structures that contain elements of both bridging and closure (Baldassarri & Diani, 2007; Baum, van Liere, & Rowley, 2007; Reagans & McEvily, 2003, this volume; Reagans, Zuckerman, & McEvily, 2004). Hybrid networks resemble locally clustered, sparsely connected smallworld structures (e.g., Watts, 1999), in which closure and bridging are viewed as complements that support coordinated action and create advantages. Within such structures, bridging improves vision through the circulation of diverse information and new ideas, while closure strengthens exchange relations and enforcement of cooperative norms (Schilling & Phelps, 2007). Although the small-world concept is now more than forty years old, empirical research is just now gaining traction into these network structures thanks to recent developments in physics and mathematics. While many questions remain unanswered, a vast body of research now afﬁrms the signiﬁcance of social capital to actors at individual, group and organizational levels. This afﬁrmation has spawned a closely related stream of research examining how actors choose partners and networks emerge. Partner selection research examines mechanisms of relationship formation and thus provides knowledge of how network structures emerge and change. The underlying theory, particularly at the ﬁrm level, treats partner selection as a risk-uncertainty problem. Speciﬁcally, ﬁrms’ decision makers are conceived to follow a logic of reducing uncertainty and risk in their exchanges by engaging past partners in repeated ties and forming new ties with partners’ partners based on referrals (Podolny, 1994; Gulati, Dialdin, & Wang, 2002), rather than seeking riskier and more uncertain nonlocal ties beyond these bounds (Baum, Rowley, Shipilov, & Chuang, 2005; Li & Rowley, 2002). As a result of this preference for ties with past partners and their partners, ﬁrms’ ties tend to congeal into dense, stable, and constraining local clusters (Walker et al., 1997; Gulati & Gargiulo, 1999). Thus, partner selection research tends to emphasize inertia in partner choice and network stabilizing mechanisms (Chung, Singh, & Lee, 2000; Li & Rowley, 2002).

Introduction: Evolving Webs in Network Economies

xvii

THE CHALLENGE OF NETWORK STRATEGY RESEARCH Social capital research convincingly illustrates the inﬂuence of network structures and positions on competitive advantage: Advantages are unevenly distributed across networks that confer to those in privileged positions better access to information, knowledge, resources and partners (Krackhardt, 1992; Burt, 2000). What is unclear from this work is the degree to which network structures and positions are the outcome of social actors actively seeking network advantages or by-product of other, more myopic concerns. While some theorists suggest that social actors are aware of their networks and perhaps entrepreneurial in their approach (e.g., Rowley & Baum, 2004; Burt, 1992; Obstfeld, 2005; Shipilov, Labianca, Kalnysh, & Kalnysh, 2007; Sik & Wellman, 2000; Vissa, 2008), most are silent on topics of network cognition and agency. Neither does network theory typically consider the strategic goals and self-interests of actors in shaping network positions and structures. Instead, research typically asserts a passive role of existing network structures in shaping future relationships, while discounting the proactive role individuals may play in shaping local network structures through strategic formation, maintenance and dissolution of ties. This orientation contrasts starkly with research in other areas of strategy (e.g., Porter, 1980; Teece, 1986) and organization theory (e.g., Pfeffer & Salancik, 1978) where actors’ active and strategic efforts to manipulate their positioning vis-a`-vis markets, rivals or resource holders is taken-for-granted. Relatedly, partner selection research emphasizes myopic partnering and network inertia (Podolny, 1994; Gulati, 1995). However, decision makers do not inevitably reproduce their past relationships (Baker, Faulkner, & Fisher, 1998; Palmer, 1983). As a result of its focus on the risk and uncertainty reduction partnering logic, partner selection research supplies few insights into the forces driving partner and network change. Baum et al. (2005) recently attempted to generalize the risk-avoidance model of partner selection by adopting the performance feedback model from organizational learning to specify conditions under which risk-taking is more or less likely to occur, and thus the conditions under which ﬁrms’ decision makers would select new partners and change their network. Consistent with learning theory predictions a ﬁrm was more likely to adopt the risk and uncertainty avoidance pattern of partnering with past partners and their partners when the ﬁrm met its performance aspirations, but more likely to select partners with which it had no prior direct or indirect contact when its performance fell below aspirations, or greatly exceeded them.

xviii

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

In contrast to tie formation, tie dissolution has seen far less attention. While it seems straightforward (and nearly tautological) to suggest that important ties are less likely to be severed, such claims beg the question of which relationships are considered important (Greve, Baum, Mistuhashi, & Rowley, 2008). Network change depends importantly on relationship termination because severing one or a few ties alters network positions of both ﬁrms that withdraw and their former partners. These indirect effects can be ampliﬁed as initial dissolutions trigger additional ones. Study of tie dissolution is needed to augment models of network dynamics, as the implications of partnering patterns will remain clear until termination patterns are better understood. Despite research advances identifying performance-enhancing network advantages and partner selection logics, Salancik’s (1995) lament – that network research indicates how network positions can be used to advantage, but says little about how these positions evolve or are destroyed – remains an accurate critique. The mechanisms underlying network dynamics are critical to understanding network-based advantages. Different from other types of advantage, social capital is not a property of the actors enjoying the beneﬁts because they do not control the relationships comprising their network positions. As such, social capital can ebb and ﬂow as a result of other actors’ actions. The challenge for network strategy research – taken up in this volume – is to move beyond static conceptions of networks and their effects and to promote a dynamic view that aids our understanding of how networks create (dis)advantages by exploring the origins, evolution and decay of network structures, positions and their associated advantages. To orient the volume’s contributors we identiﬁed three important network strategy themes:

Network Dynamics Although the sources of network stability are well understood, how networks emerge and change over time is not. Research suggests that interﬁrm networks commonly exhibit ‘small world’ properties, but is virtually silent regarding why they are patterned in this way (Baum, Shipilov, & Rowley, 2003). From a resource-based perspective, network theories should not only allow us to identify sources of competitive advantage, but also evaluate their emergence and sustainability over time (Barney, 1991; Dyer & Singh, 1998). Understanding both exogenous and

Introduction: Evolving Webs in Network Economies

xix

endogenous drivers of network change is vital to improving our understanding of how networks and positions within them emerge, take shape and dissolve, and to further our knowledge of competitive advantage in general and network advantage in particular. Network change cannot be studied without measuring network structures over time. A one-shot measure of the network and ongoing measures of behaviors inﬂuenced by the network is a common study design when networks are stable (or thought to be) and the behaviors caused by the network rather than the network itself are emphasized. Such designs are nearly universal in network diffusion studies. For studying the network itself, however, such a design is inadequate. In a long-overdue response to numerous calls to arms (e.g., Laumann, Galaskiewicz, & Marsden, 1978; Nohria, 1992), longitudinal analyses are now being undertaken to understand networks in terms of the processes that construct, maintain, and alter them. Many of the chapters in this volume tackle questions of network change head on.

Network Entrepreneurship We echo DiMaggio’s (1988) call for the reintroduction of ‘‘agency’’ – the capacity to ‘‘make a difference’’ in one’s situation – into institutional theorizing in the domain of social network theory. Awareness of the structural advantages available to occupants of certain network positions should inspire organizations and their managers to seek these advantages. Several perspectives, including structural hole and resource dependence theories, suggest that ﬁrms and managers take actions to inﬂuence the structure of relationships around them, constructing networks that facilitate innovation or serve as barriers to entry that new entrants must overcome. Obstfeld (2005), for example, has shown how individuals initiate ties between existing network contacts in the pursuit of innovation. However, there is limited understanding of the extent to which social actors understand their networks sufﬁciently well to manipulate the network structure and/or their position within it strategically. With an emphasis on network entrepreneurship, we seek to broaden the research focus in the strategic networks literature from a ﬁrm’s (or other actor’s) ‘partnering strategy’ to its ‘networking strategy’ by linking the ﬁrm’s partnering choices to changes in its network position over time. An understanding of actors’ potential for (re)constructing their positions within social and exchange networks is essential to network strategy research.

xx

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

Endogeneity Although the idea that structural advantages are available to occupants of certain network positions is widely accepted, this idea is largely corroborated by cross-sectional studies, and more recently by panel studies that do not typically account for the potential endogeneity of network positions. Is it possible that some advantages precede, rather than follow, network positions? Assumptions of causal order require further theoretical and empirical consideration. For example, network research applied to the area of innovation suggests well-positioned actors gain information advantages used to drive higher levels of innovation often measured through patenting rates. Reversing the causal order of this logic represents a viable alternative explanation: The most innovative actors attract many and the best partnering opportunities and as a result end up in preferred network positions. A closer examination of causality through dynamic modeling and more sophisticated empirical approaches is needed to address these issues.

ORGANIZATION AND CONTRIBUTIONS OF VOLUME 25 We distinguish the contributions to this volume on two dimensions. First, each chapter touches on one or more of the three challenges of network strategy research identiﬁed above. Second, each chapter addresses these challenges within one of ﬁve topical areas: (1) Small Worlds and Complex Systems, (2) Network Change, (3) Network Position and Performance, (4) Endogeneity and Embeddedness and (5) Network Navigation. Table 1 locates the chapters according to topic area and research challenge(s) addressed. After discussing the contributions of each chapter along these two dimensions, we conclude with a discussion of the volume’s collective contribution, highlighting several additional themes that emerge in the chapters and that we think merit attention in future network strategy research. Small Worlds and Complex Systems In ‘‘The rise of ecommerce as an epidemic in the small world of venture capital,’’ Gordon Walker examines how investment in ecommerce ﬁrms diffuses across the network of deals among venture capitalists (VCs) in the

xxi

Introduction: Evolving Webs in Network Economies

Table 1.

Chapter Topics and Themes.

Author

Small worlds and complex systems

Network change

Network position and performance

Endogeneity and embeddedness

Network navigation

Walker Conyon and Muldoon Dagnino et al. Hite Amburgey et al. Neuman et al. Doreian Reagans and McEvily Burt Løva˚s and Sorenson Venkatraman et al. Cowan and Jonard Madhavan et al. Hagedoorn and Frankort Soda et al. Fund et al. Van Liere et al. Rowley and Baum

Network Dynamics

Network Agency

X X X X X X X

X

X X

X X X

X X

X X X

X X X X X X

X X X X

X X X

X

Endogeneity

X X X X X X

1990s. Likening the diffusion of VC investment in ecommerce to the spread of a disease in an epidemic, Walker shows that ecommerce investment is stimulated when syndicated investments in the industry are located on key network paths – ‘‘shortcuts’’ in the deal network. These shortcuts integrate the network by bringing VC ﬁrms located in different regions of the syndicate network closer together. As they are brought together, they are exposed to the new information about the new industry, increasing the likelihood that they will fund industry startups. Thus, in much the same way as a disease spreads, ecommerce investments diffused through prior investments located on shortcuts in the network. Martin Conyon and Mark Muldoon’s chapter, ‘‘Ownership and control: A small-world analysis,’’ applies computational graph theory in the context of corporate interlocks to highlight some important limitations of traditional agency models of corporate ownership and control, which ignore inﬂuential links among corporations that promote knowledge diffusion. By comparing empirical data on corporate interlocks in the

xxii

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

U.K. with the results of a simulation study, they show that, relative to a set of comparable random graphs, the U.K. ownership and control network is characterized by short path-lengths and high clustering, or cliquishness. Mechanisms of tie formation and termination in the U.K. network also appear to differ from those in U.S. and German interlock networks suggesting cultural inﬂuences on the evolution of these networks. Given their ﬁndings, Conyon and Muldoon suggest that a recently developed class of small world models may offer promise in accounting for the social structure of interlock and related interﬁrm networks. ‘‘Evolutionary dynamics of interﬁrm networks: A complex system perspective,’’ by Giovanni Dagnino, Gabriella Levanti, and Arabella Li Destri, draws on in-depth case studies of STMicroelectronics and Toyota’s supplier network to explore network emergence and evolution. They develop a complex systems framework that travels across multiple levels of relational structure: the overarching network, clusters of ﬁrms and single ﬁrms. The results suggest a strong interplay of factors across levels and provide evidence that network change is triggered by changes in managers’ interpretations of the competitive domain surrounding network ties. In this study, network evolution is not the result of exogenous shocks but a recursive pattern on inﬂuence between managerial cognition and network structure.

Network Change In ‘‘The role of dyadic multidimensionality in the evolution of strategic network ties,’’ Julie Hite argues that dyadic ties are multidimensional systems that evolve according to changes within multiple contexts and levels. For example, dyadic ties are terminated or gain/lose strength as factors at the actor, dyad and network levels evolve. The implication for network research is a model of perpetual dynamics: The social structure can change a tie’s worth or purpose, which spurs tie change and in turn changes the structure. In addition, individual differences among social actors color how they perceive their networks and thus the types of ties they form and terminate. Terry Amburgey, Andreas Al-Laham, Daniel Tzabbar and Barak Aharonson develop and test an ecological model of network dynamics in their chapter ‘‘The structural evolution of multiplex organizational networks: Research and commerce in biotechnology.’’ Amburgey et al. contend that such a model should include modiﬁcation and replacement processes

Introduction: Evolving Webs in Network Economies

xxiii

and multiple levels of analysis – organizations, clusters of organizations and overarching networks. In an analysis of research and development and marketing and distribution alliances among biotechnology ﬁrms and their partners over a 30-year period, they examine the relationship between ﬁrms’ actions and the network structure. These networks show distinctive patterns of change based on preferential attachment processes that ‘jump’ between the two networks. The result is that while initially fragmented, over time the network is bound into fewer and larger components as high-status ﬁrms establish bridging ties – short cuts – across the fragments. Eric Neuman, Jerry Davis and Mark Mizruchi’s chapter, ‘‘Industry consolidation and network evolution in U.S. global banking, 1986–2004,’’ examines the interplay of network factors with other evolutionary forces. They explore how industry evolution in the form of banks’ changing role in the economic system and merger and acquisition proliferation affect networks. The study shows that networks change as a result of changes in organizations’ strategies. Notably for the study of network dynamics, Neuman et al. highlight the role of exogenous forces. They argue that banks diminished centrality in board interlock networks over time was the consequence of the changing role of the board, institutional pressure to reduce the size of boards, and geographic proximity constraints on board member commitment. In ‘‘Actor utilities, strategic action and network evolution,’’ Pat Doreian pushes beyond the network axiom that network structure is the dominant factor promoting network change. He offers a model of network dynamics based on the contention that social actors make strategic choices regarding the costs and beneﬁts of structural hole opportunities and the structure of the network in which those choices are made. Doreian’s ﬁndings suggest that structural holes may not provide the best partnering options and that direct structural competitors have conﬂicting interests. This study illustrates that strategic maneuvering to ﬁnd optimal network position may be overly complex and thus suggests that, beyond some point, actors cannot fully understand the structural and positional consequences of their partner selection decisions.

Network Position and Performance In ‘‘Contradictory or compatible? Reconsidering the tradeoff between brokerage and closure on knowledge sharing,’’ Ray Reagans and Bill McEvily challenge the widely held assumption that brokerage and closure

xxiv

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

are opposing forces in networks. They advance an integrative model of knowledge sharing that distinguishes the effects of brokerage and closure from each other, and between two distinct phases of knowledge sharing: seeking and transfer. The results of their analysis of a small contract R&D ﬁrm supports the view that brokerage and closure are mutually reinforcing, each promoting knowledge seeking and knowledge transfer, although in distinct ways and to differing extents. Patterns of collaboration are driven by both the risks and potential gains of cooperation. Reagans and McEvily’s analysis suggests that networks provide information key to knowing not only who merits cooperation, but also its value. In his chapter, ‘‘Returns to secondhand brokerage in industry networks: Spillover effects on price-cost margins in American manufacturing,’’ Ron Burt examines the boundaries of agency in networks by testing whether there are spillovers effects from partners’ network positions. Building on results from his work on second-hand brokerage effects at the individual level, Burt explores in this chapter similar mechanisms at the industry level using interindustry ﬂows in the US economy (1987–1992). Contrary to ﬁndings in earlier work at the individual level (Burt, 2007), Burt ﬁnds that industries in close proximity to other industries with low constraint (access to structural holes) beneﬁt from better price-cost margins. His chapter motivates further research to explain the differences between managers and industries and suggests that spillover beneﬁts are likely exist in manager networks but they lack the cognitive and emotional skills necessary to internalize those indirect beneﬁts. Bjørn Løva˚s and Olav Sorenson argue that successful resource mobilization depends on the nature of relationships among actors as well as the demand for the resource in question. In their chapter, ‘‘The mobilization of scarce resources,’’ they argue that because the cost of reciprocating rises as the resources in question become scarcer, actors increasingly rely on mobilization through relations embedded in a set of mutual, third-party acquaintances to reduce the risk of reneging. Survey data on senior partners at a consulting ﬁrm corroborates their idea. Resource scarcity is thus an important determinant of the value of triadic relations to exchange – one that alters the value of a given triad over time as resource scarcity varies. Notably, their ﬁndings suggest that those best positioned to assemble resources may ﬁnd it difﬁcult to identify opportunities to exploit those resources because the dense ties that facilitate the former do not aid in the latter. Løva˚s and Sorenson encourage a shift away from a focus on relationship strength to the effects of common alters on the value of ties.

Introduction: Evolving Webs in Network Economies

xxv

In ‘‘Interconnect to win: The joint effects of business strategy and network positions on the performance of software ﬁrms,’’ N. Venkat Venkatraman, Chi-Hyon Lee and Bala Iyer examine the interaction of business strategies and network position on performance. They model software ﬁrms’ alliances as links to complementary resources arguing that product scope strategy is the mechanism through which value is produced, but that the ﬁrm’s network position determines the degree to which that value can be appropriated. Moreover, and consistent with their view that network positions support execution of business strategy, the effects product scope on performance precede those of network position in time, suggesting a temporal distinction of ﬁrm and network effects on competitive advantage.

Endogeneity and Embeddedness Joint innovation networks are the focus of ‘‘If the alliance ﬁts: Innovation and network dynamics,’’ by Robin Cowan and Nicolas Jonard. Using the degree of knowledge overlap, which affects the likelihood of innovation, as the main driver of partner selection, they simulate a dynamic model of network formation, the results of which replicate the central features (e.g., clustering, short average path length, skewed degree distribution) and behavior (e.g., repeated ties) of real-world interﬁrm networks. Thus, the process of seeking partners with a high potential for joint innovation is, by itself, sufﬁcient to produce networks that share many properties of empirically observed networks. When choosing partners, ﬁrms clearly consider many factors; empirical research has, however, focused primarily on social capital as a primary antecedent. Cowan and Jonard’s ﬁndings provocatively suggest that social capital may instead be a corollary of other antecedents. In ‘‘Bringing the ﬁrm back in: Networking as an antecedent to network structure,’’ Ravi Madhavan, Turanay Caner, John Prescott and Balaji Koka explore how network factors combine with ﬁrm characteristics to inﬂuence network change and competitive advantage. They question the assumption that network advantage requires only that ﬁrms occupy preferential positions, arguing instead that network-based advantage depends on both a ﬁrm’s position and its ability and motivation to network. Building from an absorptive capacity logic, Madhavan et al. portray alliance networks as the outcome of ﬁrms’ differing network strategies and abilities to internalize innovation from their network positions. They supplement a standard structural analysis of biotechnology ﬁrms’ alliances with qualitative analysis

xxvi

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

to gain insight into ﬁrm-speciﬁc differences in networking ability and motivation to enrich their account of antecedents to and consequences of network structure. In ‘‘The gloomy side of embeddedness: The effects of over-embeddedness on partnership formation,’’ John Hagedoorn and Hans Frankort draw on detailed histories of IBM’s collaborations to develop a set of propositions suggesting a dynamic process operating at multiple, nested network levels. In particular, their case histories suggest that the partner selection processes underlying tie formation result, over time, in network overembeddedness, which diminishes the value of network ties to its members, motivating ﬁrms to forego the comfort of established embedded ties to seek out new nonembedded partners, lowering network density and opening structural holes. Hagedoorn and Frankort’s theoretical model portrays networks as dynamic entities, alternating between periods of increasing and decreasing density. In ‘‘Imitiative behavior: Network antecedents and performance consequences,’’ Giuseppe Soda, Akbar Zaheer and Alessandra Carlone examine network positions and performance in the Italian TV production industry, a network context in which competitors must collaborate to access specialized resources necessary for achieving highly creative output. Contrary to ﬁndings in other settings, Soda et al. ﬁnd that imitation declines with closure while increasing with centrality, and in addition, that imitation lowers performance. Their results suggest that exploratory efforts to uncover novel information to drive innovation need not come from far reaching brokerage ties. Under conditions of competitive interdependence among highly specialized partners, actors in this study were able to access and absorb more knowledge in a dense structure resulting in more creativity (less imitation).

Network Navigation In their chapter, ‘‘Who’s the new kid? The process of developing network centrality and embeddedness in venture capital deal networks,’’ Bret Fund, Tim Pollock, Ted Baker and Adam Wowak explore conditions under which new venture capital ﬁrms are able to move from the periphery to the core of their networks. They consider ﬁrm and network inﬂuences and, akin to several other chapters in this volume (e.g., Venkatraman et al.), argue that it is the interplay between ﬁrm and network factors that drive this dynamic process. By detailing two in-depth case studies of venture capital ﬁrms founded in 1995, they build a model of factors facilitating ﬁrms’ move

Introduction: Evolving Webs in Network Economies

xxvii

toward the network core: founders’ status, resource endowment, attractiveness as a partner, visibility of portfolio ﬁrms, and the venture capital ﬁrm’s urgency and effort. Perhaps most notable, however, is that ‘cognitive centrality’ – the perception that one is or at least belongs at the core of the network – precedes and predicts structural centrality. In ‘‘Strengthening of bridging positions: Network horizon and network horizon heterogeneity,’’ Diederik van Liere, Otto Koppius and Peter Vervest consider the question of why ﬁrms differ in their abilities to create network advantages. They contend that the answer resides in variation in actors’ ‘‘network horizons’’ – the extent of information they possess about network structure and positions. Using network experiments involving both students and managers, they examine the role of network information by varying their subjects’ ability to see their networks. Subjects with broader network horizons more often occupied networks positions rich in structural holes, particularly when experimental subjects varied in their network horizons. Their ﬁndings thus suggest that actors possessing structural information about their network can identify network-based advantages and are capable of strategically maneuvering to capture them vis-a`-vis lessinformed actors in the network. In our chapter, ‘‘The dynamics of network strategies and positions,’’ we (Tim Rowley and Joel Baum) wonder how ﬁrms come to occupy bridging ties spanning structural holes given the strong motives to form closure ties speciﬁed in partner selection research. We examine nearly 40 years of underwriting syndicates in Canada. Our ﬁndings indicate that lead investment banks, which have greater discretion in choosing syndicate partners than colead banks, are more likely to create bridging positions for themselves by selecting coleads that are not connected to one another. We also ﬁnd that bridging positions deteriorate when lead banks form syndicate ties with each other, suggesting that lead investment banks compete for bridging positions. Taken together the evidence we provide supports the idea that, like the subjects in van Liere et al.’s experiments, investment banks’ managers are aware of network structures and advantages, and act to realize them.

Emergent Themes Beyond the three challenges we emphasized in orienting the volume’s contributors, three notable additional themes emerged across multiple chapters in this volume.

xxviii

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

Network Cognition Inspired partly by a focus on questions of network agency, network cognition emerged as a theme. Actors’ ability to conceive their networks is argued to inﬂuence partner selection decisions (Cowan and Jonard; Hagedoorn and Frankort; Rowley and Baum), pursuit of network advantage (van Liere et al.), absorption of network-based advantages (Burt; Fund et al.; Madhavan et al.) and shape network evolution more generally (Doreian; Dagnino et al.; Hite). Although no single chapter provides deﬁnitive evidence of actors’ network awareness and strategic pursuit of network positions, collectively these chapters suggest, at the very least, that network cognition varies across actors and point to the need for future research designed to examine more directly actors’ mental models and cognitive abilities with respect to their networks. Situated Networks A second theme pervading the volume is that network effects and dynamics are frequently situated within or contingent on a range of non-network factors. Several chapters emphasize that ﬁrm strategies and attributes (Madhavan et al., Venkatraman et al.) as well as industry and institutional characteristics (Neuman et al., Soda et al.) inﬂuence network dynamics and the degree to which network effects are enjoyed (see also Shipilov, 2005, 2006, 2008). Cultural or country-speciﬁc factors, for example, concerning the operation of capital markets operating in different jurisdictions as observed by Conyon and Muldoon, may also result in the emergence of distinctive network structures. Indeed, the value of relationships may vary dynamically with the scarcity of resources ﬂowing through them (Løva˚s and Sorenson). Perhaps a lingering artifact of an earlier erroneously polarized view of social structure dominating rather than supplanting other social processes, network researchers have tended to overlook the situatedness of networks in their theorizing. The chapters in this volume begin a correction of this oversight that we hope continues. Multilevel Analysis Finally, several chapters highlight the need multilevel theorizing and analysis of network strategy. Network models span not only individual, group and ﬁrm actor levels of analysis, but also ‘network’ levels of analysis: dyad, ego network, clique and network. Appealing to the study of strategy and organizations is the clustering of these levels, which permit application of theory at different levels of analysis and development of multilevel models (Wellman, 1988). Amburgey et al. and Dagnino et al. explicitly model multiple

Introduction: Evolving Webs in Network Economies

xxix

levels of analysis, Hite, Hagedoorn and Frankort, and Walker argue that dyadic ties are inﬂuenced by network structure, and Burt and Rowley and Baum contend that network positions have differing effects across levels of analysis. In addition, many of these studies suggest these multiple level relationships are recursive: Network structure inﬂuences how social actors see their networks, which in turn leads to changing patterns of partner selection and thus dynamics in the network structure (e.g., Cowan and Jonard; Dagnino et al., Fund et al., van Liere et al.). These observations suggest the need to complement models of network persistence emphasized in most research on social capital and partner selection with an account of network dynamics. The work in this volume points to further development of multilevel models as holding promise for understanding the dynamics of network strategy.

CONCLUSION Of course, by necessity, our introduction offers a partial view of the scope of network strategy research. Many important questions remain open, awaiting future study. And so, it is time to turn you over to the volume’s contributors. They provide a sampling of important studies, each of which develop and extend aspects of network strategy – in areas we have touched upon as well as those we have not. By presenting their work, we hope to promote a shift from discussions of network effects to network processes and from linear to recursive models of network evolution that recognizing both top–down and bottom–up modes of inﬂuence. We hope their work inspires additional researchers to join in advancing our understanding of network strategy and aids practicing networkers to achieve network advantage. Timothy J. Rowley Joel A. C. Baum Editors

REFERENCES Ahuja, G. (2000). Collaboration networks, structural holes and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Baker, W. E., Faulkner, R. R., & Fisher, G. A. (1998). Hazards of the market: The continuity and dissolution of interorganizational market relationships. American Sociological Review, 63, 147–177. Baldassarri, D., & Diani, M. (2007). The integrative power of civic networks. American Journal of Sociology, 113, 735–780.

xxx

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

Barney, J. B. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17, 99–120. Baum, J. A. C., & Ingram, P. (2002). Interorganizational learning and network organizations: Toward a behavioral theory of the ‘interﬁrm’. In: M. Augier & J. G. March (Eds), The economics of choice, change and organization: Essays in memory of Richard M. Cyert, 191–218. Cheltenham, UK: Edward Elgar. Baum, J. A. C., Rowley, T. J., Shipilov, A. V., & Chuang, Y-T. (2005). Dancing with strangers: Aspiration performance and the search for underwriting syndicate partners. Administrative Science Quarterly, 50, 536–575. Baum, J. A. C., Shipilov, A. V., & Rowley, T. J. (2003). Where do small worlds come from?’’ Industrial and Corporate Change, 12, 597–725. Baum, J. A. C., van Liere, D., & Rowley, T. J. (2007). Between closure and holes: Hybrid network positions and ﬁrm performance. Paper presented at the Academy of Management, Philadelphia PA, August. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. (2000). The network structure of social capital. In: R. I. Sutton & B. M. Staw (Eds), Research in organizational behavior (Vol. 22, pp. 339–365). Greenwich, CT: JAI Press. Burt, R. S. (2007). Secondhand brokerage: Evidence on the importance of local structure for managers, bankers, and analysts. Academy of Management Journal, 50, 119–148. Chung, S. A., Singh, H., & Lee, K. (2000). Complementarity, status similarity and social capital as drivers of alliance formation. Strategic Management Journal, 21(1), 1–22. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. DiMaggio, P. J. (1988). Interest and agency in institutional theory. In: L. G. Zucker (Ed.), Institutional patterns and organizations: Culture and environment (pp. 3–21). Cambridge, MA: Ballinger. Dyer, J. H., & Singh, H. (1998). The relational view: Cooperative strategy and interorganizational sources of competitive advantage. Academy of Management Review, 23, 660–679. Gomes-Casseres, B. (2003). Competitive advantage in alliance constellations. Strategic Organization, 1, 327–335. Greve, H.R., Baum, J.A.C., Mistuhashi, H., & Rowley, T.J. (2008). Built to last but falling apart: Cohesion, friction and the durability of interﬁrm alliances. Academy of Management meetings, Anaheim, CA, August. Gulati, R. (1995). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R., Dialdin, D. A., & Wang, L. (2002). Organizational networks. In: J. A. C. Baum (Ed.), The Blackwell companion to organizations (pp. 281–303). Oxford: Blackwell Publishers. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1439–1494. Kogut, B. (1988). Joint ventures: Theoretical and empirical perspectives. Strategic Management Journal, 9, 319–332. Krackhardt, D. (1992). The strength of strong ties: The importance of philos in organizations. In: N. Nohria & R. C. Eccles (Eds), Networks and organizations: Structure, form and action (pp. 216–239). Boston, MA: Harvard Business School Press. Laumann, E. O., Galaskiewicz, J., & Marsden, P. V. (1978). Community structures as interorganizational linkages. Annual Review of Sociology, 4, 455–484.

Introduction: Evolving Webs in Network Economies

xxxi

Levitt, B., & March, J. G. (1988). Organization learning. Annual Review of Sociology, 14, 319–340. Li, S. X., & Rowley, T. J., (1999). Locating capable and reliable allies: The role of experiential and nonexperiential information in interorganizational partner selection. Presented at the Academy of Management Conference, Chicago, August Li, S. X., & Rowley, T. J. (2002). Inertia and evaluation mechanisms in interorganizational partner selection: Syndicate formations among U.S. investment banks. Academy of Management Journal, 45(6), 1104–1119. McEvily, B., & Zaheer, A. (1999). Bridging ties: A source of ﬁrm heterogeneity in competitive capabilities. Strategic Management Journal, 20, 1133–1156. Nohria, N. (1992). Is a network perspective a useful way of studying organizations? In: N. Nohria & R. C. Eccles (Eds), Networks and organizations: Structure, form and action (pp. 1–22). Boston, MA: Harvard Business School Press. Obstfeld, D. (2005). Social networks, the Tertius Iungens orientation and involvement in innovation. Administrative Science Quarterly, 50, 100–130. Oliver, C. (1991). Strategic responses to institutional environments. Academy of Management Review, 16, 145–179. Palmer, D. (1983). Broken ties: Interlocking directorates and intercorporate coordination. Administrative Science Quarterly, 28, 40–55. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations: A resource dependence perspective. New York, NY: Harper and Row. Podolny, J. M. (1994). Market uncertainty and the social character of economic exchange. Administrative Science Quarterly, 39, 458–470. Podolny, J. M. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107, 33–60. Podolny, J. M., & Baron, J. N. (1997). Resources and relationships: Social networks and mobility in the workplace. American Sociological Review, 62, 673–693. Porter, M. E. (1980). Competitive Strategy: Techniques for analysing industries and competitors. New York: Free Press. Powell, W. W. (1990). Neither market nor hierarchy: Network forms of organization. Research in Organizational Behavior, 12, 295–336. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in Biotechnology. Administrative Science Quarterly, 41, 116–145. Reagans, R. E., & McEvily, B. (2003). Network structure and knowledge transfer: The effect of cohesion and range. Administrative Science Quarterly, 48, 240–267. Reagans, R. E., Zuckerman, E. W., & McEvily, B. (2004). How to make the team: Social networks vs. demography as criteria for designing effective projects in a contract R&D ﬁrm. Administrative Science Quarterly, 49, 101–133. Rowley, T. J., & Baum, J. A. C. (2004). Sophistication of interﬁrm network strategies in the Canadian investment banking industry. Scandinavian Journal of Management, 20, 103–124. Rowley, T. J., Baum, J. A. C., Shipilov, A. V., Greve, H. R., & Rao, H. (2004). Competing in groups. Managerial and Decision Economics, 25, 453–471. Rowley, T. J., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21, 369–386.

xxxii

INTRODUCTION: EVOLVING WEBS IN NETWORK ECONOMIES

Salancik, G. R. (1995). WANTED: A good network theory of organization. Administrative Science Quarterly, 40, 345–349. Schilling, M. A., & Phelps, C. C. (2007). Interﬁrm collaboration networks: The impact of large-scale network structure on ﬁrm innovation. Management Science, 53, 1113–1126. Shapiro, C., & Varian, H. R. (1999). Information rules: A strategic guide to the network economy. Boston: Harvard Business School Press. Shipilov, A. (2005). Should you bank on your network? Relational and positional embeddedness in the making of ﬁnancial capital. Strategic Organization, 3, 279–309. Shipilov, A. (2006). Network strategies and performance of Canadian investment banks. Academy of Management Journal, 49, 590–604. Shipilov, A. (2008). Firm scope experience, historic multimarket contact with partners, centrality and the relationship between structural holes and performance. Organization Science, in press. Shipilov, A., Labianca, G., Kalnysh, V., & Kalnysh, Y., (2007). The constraining nature of structured foci networking: Career related network building behaviours, range social capital and promotion. Academy of Management Best Paper Proceedings. CD-ROM. Sik, E., & Wellman, B. (2000). Network capital in communist and post-communist societies. In: B. Wellman (Ed.), Networks in the global village (pp. 225–254). Boulder: Westview Press. Simmel, G. (1950). The sociology of Georg Simmel (K. H. Wolff, Trans.). Glencoe, IL: Free Press Smelser, N. J., & Swedberg, R. (1994). The handbook of economic sociology. Princeton: Princeton University Press. Soda, G., Usai, A., & Zaheer, A. (2004). Network memory: The inﬂuence of past and current networks on performance. Academy of Management Journal, 47, 893–906. Stinchcombe, A. L. (1985). Norms of Exchange. In: A. L. Stinchcombe (Ed.), Stratiﬁcation and organization: Selected papers (pp. 231–267). Cambridge, UK: Cambridge University Press. Teece, D. J. (1986). Proﬁting from technological innovation: Implications for integration, collaboration, licensing and public policy. Research Policy, 15, 285–305. Uzzi, B., & Gillespie, J. J. (2002). Knowledge spillover in corporate ﬁnancing networks: Embeddedness and the ﬁrm’s debt performance. Strategic Management Journal, 23, 595–618. Vissa, B. (2008). Entrepreneurs’ networking action patterns and referral based search for new exchange partners. Working paper. INSEAD, Singapore Campus. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8, 109–125. Wasserman, L. (1992). Invariance properties of density ratio priors. The Annals of Statistics, 20, 2177–2182. Watts, D. J. (1999). Networks, dynamics, and the small-world phenomenon. American Journal of Sociology, 105, 493–527. Wellman, B. (1988). Structural analysis: From method and metaphor to theory and substance. In: B. Wellman & S. D. Berkowitz (Eds), Social structures: A network approach (pp. 19–61). Cambridge: Cambridge University Press. Willer, D. (Ed.) (1999). Network exchange theory. New York: Praeger.

THE RISE OF ECOMMERCE AS AN EPIDEMIC IN THE SMALL WORLD OF VENTURE CAPITAL Gordon Walker ABSTRACT Syndicated investments in startups within a particular industry follow an evolutionary path consistent with models of industry growth and evolution. Some industries spend a long time gestating, while others grow and mature quickly. Entry into ecommerce industry segments has both characteristics. What spurred the sector’s slow emergence and subsequent quick rise? One obvious answer is the development of the Internet in the mid-1990s. However, a competing possibility is that the diffusion pattern resembles an epidemic among venture capital ﬁrms. This chapter examines how the existing structure of the VC syndication network in the US enabled such an epidemic. Consistent with existing theory on the spread of a disease in a small world, this study argues that the incidence of investments in ecommerce startups was a function of prior investments located on shortcuts in the network. The time frame is 1980 to just before the NASDAQ boom in 1999.

Network Strategy Advances in Strategic Management, Volume 25, 3–29 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25001-1

3

4

GORDON WALKER

OVERVIEW How do industries emerge? The usual approach to this question is to focus on new technologies and on the ﬁrms that sell the products based on them. If these ﬁrms are successful economically, the industry evolves; if not, the industry dies. From this angle, industry development is all about innovation and commercialization. Yet there is another crucial component of industry development; that is the ﬁnancial investment in startup companies typically by venture capital ﬁrms. Without investors willing to provide startups with the necessary funds for technology and product development in a nascent industry, the industry would be unable to grow. Investment in a new technology almost always requires a demonstration of its viability. Obviously, some seed funding is necessary for prototype development. Yet it is well understood that even after technological feasibility is demonstrated, and new businesses based on the technology have been formed, the viability of the industry remains highly uncertain. In the early stages of an industry’s history, it is impossible to forecast the returns to an investment in a startup ﬁrm with any reasonableness, so the standard calculus for estimating a ﬁrm’s ﬁnancial returns does not apply. Normal investors therefore will not consider them, which excludes them from a major source of capital available to more mature companies. The venture capital (VC) industry, which emerged in the U.S. in the late 1950s, deals with this problem (see Gompers & Lerner, 1999). Venture capital ﬁrms invest in businesses that are usually based on novel technologies or business ideas – e.g., biotechnology, data communications, and ecommerce software. The capital that VC ﬁrms invest comes from large institutions, such as pension funds and universities, and from wealthy individuals and families. Because the future of a startup business is highly uncertain, VC ﬁrms expect only a few of these businesses to succeed. However, as long as the amount they make on the successes exceeds what they lose on the failures, VC ﬁrms earn a positive return on their capital; and institutions will continue to give them funds to invest. Because startup performance is so unpredictable, venture capital ﬁrms develop portfolios of investments hoping that at least one startup will become successful and pay for the losses incurred by the others. To spread the risk of startup failure even further, venture capital ﬁrms very often will form a syndicate and co-invest together in a particular startup (Lerner, 1994). When a startup fails, or is sold, the VC ﬁrm’s role as an investor ends, and the syndicate, if one was formed, is dissolved.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

5

Startups that are opportunities for new investment may be in new industries that have been growing for some time (e.g., computer software) or industries experiencing their ﬁrst wave of entry (e.g., nanotechnology). Further, an industry may be composed of product segments that have different rates of entry based on the discovery of new technologies and different life cycles, some maturing sooner than others. Because startups in an existing industry have been competing longer and in some instances may have succeeded, it is likely to be seen by investors as less risky than startups in an industry that is just beginning to grow. Startups in new industries are unproven and are viewed with circumspection. How then does the funding base for a new industry develop? This chapter examines the rise of an industry as a diffusion process across VCs. Following a venerable tradition of research on diffusion processes, the percolation mechanism studied here is a network. The network is constituted of syndicated deals among VCs and is analyzed as a small world (Watts, 1999a). The spread of investments in the new industry across the small world is characterized as an epidemic (Moore & Newman, 2000). The epidemic’s growth is stimulated when syndicated investments in the new industry are located on key network paths that integrate the network by bringing the VCs closer together. As they are brought together, they are exposed to the new information about the new industry, increasing the likelihood that they will fund industry startups.

THEORY Overview The unit of analysis in this study is the syndicated investment in a startup, not the VCs that make that investment. That is, the focus is on characteristics of the relationship and not of the actors that it connects. Although the importance of both dimensions – relationship and actor – has long been recognized (McPherson, 1983; Breiger, 1974; Baker, 1986), analyzing relationships has been secondary in network studies. Nonetheless, it is intuitive that relationships between actors must be motivated. Ties between actors cannot (will not) be formed without a purpose or without at least an underlying sociological logic. A key assumption in a network analysis of actors is that the type of tie that comprises the network is sufﬁciently long lasting and homogeneous across actors (e.g., friendship, advice seeking, technology development

6

GORDON WALKER

partnerships, co-investment) to remove any suggestion that the analysis is confounded by tie heterogeneity. To achieve conceptually coherent results, the analyst must assume that he is studying a tie whose meaning maintains its integrity. Whether this is true is a question that each researcher needs to address carefully. It is apparent that it is not true for the network of VCs, which is composed of syndicated investments in startup ﬁrms. The reason is that, although the generic reason for ties between VCs – co-investment in a startup – remains the same over time, the speciﬁc investment opportunities change. Fig. 1 shows how the incidence of investments in ﬁve generic industries shifts over the 20 years examined here. Clearly, computer-related startups (hardware and software) dominated the VC industry in the early

1200 computers data communications 1000

medical products services

800

600

400

200

0 19801 19804 19813 19822 19831 19834 19843 19852 19861 19864 19873 19882 19891 19894 19903 19912 19921 19924 19933 19942 19951 19954 19963 19972 19981 19984

Number of syndications

ecommerce

Four quarter periods

Fig. 1.

Number of Syndications in an Industry/Sector.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

7

1980s but then dropped precipitously in number. At the same time, however, medical products, services, and data communications continued to rise slowly as investment opportunities. Without the continuing availability of startups in these three industries, the venture capital business would have virtually ground to a halt. Further, the growth in the number of ecommerce startups during the 1990s created a palpable new industry for VCs to invest in, even as the other industries rose at the same time. The availability of ecommerce as an investment opportunity greatly expanded the network of syndications. This study asks two questions about how ecommerce diffused as a new investment opportunity spreading through the syndication network. The ﬁrst is: does the number of investments in ecommerce increase as a function of where earlier ecommerce investments were located in the syndication network? The second question is the reverse of the ﬁrst: how do investments in ecommerce contribute to the integration of the syndication network? Both questions relate to the dynamic through which investments in ecommerce are built and sustained over time. The idea here is that these investments tend to be found more and more over time in the structural locations that induce future investment behavior, thus creating a virtuous cycle of growth through which ecommerce emerges as an industry. This approach differs from earlier studies of diffusion through networks (Coleman, Katz, & Menzel, 1966; Burt, 1987) in which an actor’s position in an existing network predicts his susceptibility to adopting an innovation. The focus here is on predicting the number of investment events, not speciﬁcally on the behavior of individual VCs (see e.g., Strang, 1991; Greve, Strang, & Tuma, 1995). The theory here then necessarily entails network rather than individual level variables.

Small Worlds The approach to network structure taken here is the small world. A small world is a network in which the actors are more highly clustered than would be expected if ties were formed randomly in a network of the same size and same average outdegree, and at the same time the actors are able to reach one another across the network in about the same number of steps as would be expected in a comparable randomly formed network (see Watts, 1999a; Watts & Strogatz, 1998). The concept of a small world emerged in academic research through the experiments of Stanley Milgram and colleagues

8

GORDON WALKER

(Milgram, 1967; Travers & Milgram, 1969) and the communication studies of Kochen (1989) and Pool and Kochen (1978). Watts (1999a) developed a set of benchmarks for calculating random clustering and path length parameters that were subsequently used in empirical studies to identify whether networks exhibited small-world traits (for early empirical applications, see Kogut & Walker, 2001; Davis, Yoo, & Baker, 2003). Several recent studies on small-world networks (Uzzi & Spiro, 2005; Schilling & Phelps, 2007) have shown that small-world attributes have an effect on ﬁrm performance and innovativeness; other studies have examined how small worlds develop (Baum, Shipilov, & Rowley, 2003; Goyal, van der Leij, & Moraga-Gonzales, 2006). Further, there is a large body of research focused on building analytical models of small-world networks beginning with White’s (1970) paper on search parameters (for recent work see Amaral, Barthelemy, Scala, & Stanley, 2000; Newman, 2000; Kleinberg, 2000; Kuperman & Abramson, 2001; Watts, Dodds, & Newman, 2002). It is safe to say that, even though the concept of a small world is roughly 40 years old, the broad research program centered on it is still in the early stages (see Watts, 2004, for a discussion of problems and issues).

Diffusion in a Small World The diffusion of a new industry in the VC network is necessarily related to network evolution, a topic that has attracted an increasing body of research following a number of approaches. One approach is to examine the role played by dominant ﬁrms as they emerge and create poles that attract other ﬁrms, both existing and entering (Gulati & Gargiulo, 1999). A second is to identify aspects of the network’s structure and analyze their role in the formation of new ties between ﬁrms (Burt, 1987; Walker, Kogut, & Shan, 1997). A third approach is to observe the persistence of network structure even as the ﬁrms in the industry or economy change (Corrado & Zollo, 2006). Last, network evolution is also directly related to studies of search within a network (see Kleinberg, 2000). Two critical premises underlie this body of research. The ﬁrst premise is that the motivations for forming relationships remain relatively stable over the course of network development. If this is not so, then the content and meaning of network ties change qualitatively over time, vitiating the power of central ﬁrms and making search based on earlier network content infeasible. Second, the opportunity structure within which ties are formed must persist sufﬁciently over time so that the network can form and evolve.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

9

For example, Uzzi and Spiro’s (2005) study of musical theater professionals in New York from 1945 to 1989 shows how the structure of relationships among them develops over time into a small world. The assumption here is that the opportunities these professionals encountered were common enough qualitatively that the small world could develop. When musical talent is drawn away by opportunities in other media, the degree of clustering erodes. In a sense, the availability of new job opportunities shifts the structure of the network, leading to a new phase of network evolution. One imagines, for example, that the search parameters of the earlier musical theater regime became less functional as jobs in new media became more available. The successful diffusion of a new investment opportunity, which is what a new industry represents for VC ﬁrms, requires both that the existing opportunity structure be altered and that the new structure that is created be sustained. If the new industry does not penetrate and change the existing opportunity structure for VC investments, the likelihood of systematic growth in funding the new industry’s startups is low. Further, if, once changed, the new opportunity structure based on the new industry does not endure, then investment in the new industry must fade. Both conditions are necessary and neither is sufﬁcient for the effective development of new industry emergence through VC ﬁnancing. The diffusion of a new industry as a new opportunity structure is analogous to the spread of a disease across a population. In the analogy, evidence of being infected by the disease is an investment in a new industry startup and the population of the potentially infected is composed of the VCs. Moore and Newman’s (2000) model of epidemics on small-world networks is directly applicable. In their model the likelihood of a disease diffusing throughout a small world of actors is determined by the proportion of shortcut positions it occupies. A shortcut is a path that joins two actors and whose removal would increase the distance between them by more than two links. Shortcuts are thus central to the integration of the network and, in the case of a small world, to the reduction of the average path length. The logic therefore is that, since a shortcut is a path that brings actors closer together, when more shortcuts are occupied by the disease, more individuals become exposed and therefore infected. Translating this argument into terms relating to the diffusion of ecommerce investments suggests that: the more shortcuts occupied by ecommerce syndications, the greater the likelihood that new investments in ecommerce will occur. But as argued above, for the funding levels in ecommerce to grow, the opportunity structure that favors the ecommerce startups must be enduring.

10

GORDON WALKER

Shortcuts occupied by ecommerce syndications must not be removed by emerging patterns of investments in other new industries or by random variation in syndications in existing industries. For the ecommerce opportunity structure to endure then: (1) the number of shortcuts that are occupied by ecommerce syndications should be autocorrelated; or (2) the greater the number of ecommerce syndications, the more shortcuts they should occupy.

Investment Riskiness Venture capital investments in startups occur in rounds (see Gompers & Lerner, 1999, Chapter 7). In the ﬁrst round, the startup is untried; and, because forecasting its performance is highly uncertain, VCs typically invest less money in it. If the startup’s prospects improve, later funding rounds may occur, with the expectation that it will be sold, either to the public or to a strategic investor. The ﬁrst round of funding therefore may entail a qualitatively higher level of risk than later ﬁnancing rounds in which investors are better informed regarding the startup’s viability (for studies differentiating between rounds, see Gompers, 1995; Podolny, 2001; Sorenson & Stuart, 2001). The difference in investment riskiness is important here since it may affect the susceptibility of VCs to the structural location of ecommerce syndications. There are two contrasting, but not mutually exclusive, possibilities. The ﬁrst possibility is that VCs may be more sensitive to structural effects when they face higher risk (ﬁrst round) decisions, as these effects are incremental to whatever local information is available about the startup. The added exposure to other ecommerce investments may tip the VC into deciding to invest. In this case, more shortcuts should lead to more ﬁrst round investments. The other possibility for ﬁrst-round investments is that VCs are careful to shield their decisions from structural inﬂuences, relying more diligently on their strategic and ﬁnancial analyses. Higher risk may produce greater wariness of exposure to other events, even if it adds information to the decision. This would be the interpretation if shortcuts have no effect on ﬁrst-round investments. As for investments after the ﬁrst round, the arguments are reversed. First, because later rounds are based on more information about the startup, the salience of the information provided by the syndication on the shortcut may be reduced, lowering its importance in the decision and the value of the number of shortcuts for the growth of the industry. Alternatively, because later rounds are less risky, VCs may be more susceptible to structural effects. If this is true, a larger number of shortcuts should lead to more later round investments.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

11

DATA The data are drawn from VentureExpert, a comprehensive database of venture capital funds, startups and deals owned and managed by Thomson Financial. VentureExpert has been used extensively in studies of the venture capital industry (Sorenson & Stuart, 2001; Gans, Hsu, & Stern, 2002; Kogut, Urso, & Walker, 2007). Data from Venture Economics, an earlier incarnation of VentureExpert, begin with deals made in 1961 and constitute the most comprehensive source of information on the venture capital industry. In an extensive check on the robustness of these data, Gompers and Lerner (1999, Chapter 16) ﬁnd that the information in the Venture Economics database from 1978 to 1989 contains 95% of the ventures they were able to ﬁnd through other sources. More recent data are assumed to be as complete. The time periods examined here run from the second quarter of 1978 to the last quarter of 1998. All deals that occur in a calendar quarter are assigned to that quarter, e.g., the ﬁrst quarter in 1998, since the data clearly show a pattern of quarterly reporting. The syndication network is constructed in rolling four-quarter windows, starting with the ﬁrst quarter of 1978 and ending with the last quarter of 1998. During this time, 22,057 startups received venture capital funding in either one round or more than one round; and 3,663 ﬁrms provided venture capital. The types of ﬁrms investing in new ventures range from investment banks to private equity ﬁrms to wealthy individuals. In the period examined here 85% of the ﬁrms making these investments were either ﬁnancial service ﬁrms (VCs, investment banks, commercial banks, or other ﬁnancial institutions), corporate subsidiaries, or small business investment corporations (SBICs). Since the theory here regarding the diffusion of ecommerce as an investment opportunity does not differentiate among the types of investing ﬁrms, the syndication network is composed of all of these types.

METHODS Identifying the Small World of the Venture Capital Syndication Network Venture capital syndications form a bipartite network, so-called because it has two modes – ﬁrst, the VC ﬁrms and second, the syndicated investments they make. The ﬁrst step in the analysis is to determine the small-world characteristics of the network. Venture capital syndications form a bipartite network,

12

GORDON WALKER

so-called because it has two modes – ﬁrst the VC ﬁrms and second the syndications that connect them. Because the network is bipartite, the standard method for assessing the presence of a small world (see Watts, 1999b) cannot be applied. The reason is that ﬁrms are tied through belonging to groups (syndicates) as well as through overlapping participation in multiple groups. If the average number of actors in a group is large, then it is likely that there will be signiﬁcant clustering in the actor network, even if the overlap of participation in the groups is small. To address this problem in bipartite networks, Newman, Strogatz, and Watts (2001) developed general expressions for the clustering coefﬁcient and average path length that take into account the distribution of group sizes in the network. These expressions are analogous to analyzing a cross-tabulation of categorical variables based on observed row and column marginals. The expressions can also be used to derive the expected clustering coefﬁcient and average path length for any statistical distribution underlying the process of tie formation in a bipartite network. As an illustration, Newman et al. derive the clustering coefﬁcient and path length for a bipartite network based on the Poisson distribution, which conveniently is the benchmark distribution for determining whether a small world is observable in a unipartite network. An interesting use of this technique is made by Conyon and Muldoon (2008 – in the present volume) who show its applicability to a network of board and UK ﬁrms traded on the London Stock Exchange in 2000. Newman et al. (2001) begin by deﬁning a clustering coefﬁcient based on the numbers of triangles and triples in the unipartite network of actors. A triangle is a closed triad of actors, and a triple is a set of three actors two of whom are connected. The clustering measure is three times the number of triangles in the network divided by the number of triples. Triangles need to be multiplied by three since each triangle constitutes three triples. The more triangles there are in the network, the higher the clustering coefﬁcient. When the network is fully connected, the ratio of (three times) the number of triangles to the number of triples equals one. When there are no triangles, the ratio is zero. Using this deﬁnition to develop a clustering coefﬁcient for a bipartite network, Newman et al. show that the numbers of triangles and triples depend on the dimensions of the network, on the average group size, and on the distribution of group sizes. The expression for the clustering coefﬁcient is: M

P

kðk 1Þðk 2Þqk P N zðz 1Þrz

k

z

(1)

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

13

where M is the number of groups, N the number of actors, k the group size, qk the distribution of group sizes, z the average number of other actors in a group, and rz the probability of having z other actors in a group. The numerator in Eq. (1) is the expected number of triangles, and the denominator the expected number of triples. Newman et al. show that the expression for the average path length does not depend on the bipartite characteristics of the network and can be used to analyze unipartite networks as well. The expression is: log ðN=z1 Þ þ1 log ðz2 =z1 Þ

(2)

where N is the number of actors, z1 the average number of actors each actor can reach directly, and z2 the average number of actors each actor can reach in two steps. Newman et al. caution that Eq. (2) should be used cautiously as a benchmark for the average path length because it does not take into account the fragmentation of the network into isolated components. The greater the percentage of ties constituted by the largest (main) component, the more the Eq. (2) is appropriate for determining the network’s expected average path length. The standard distribution for establishing a random benchmark to assess small-worlds properties is the Poisson. Newman et al. show that when the Poisson distribution is applied to Eq. (1), it becomes: 1 mþ1

(3)

where m is the average number of groups for an actor. The Poisson version of Eq. (2) is: log N log z

(4)

where N is the number of actors, as before, and z the average number of actors in a group times the average number of groups per actor. This is the same as the expected path length in a Poisson generated unipartite network. To assess the small-world properties of the venture capital syndication network, this study uses both Eqs. (1) and (3) to benchmark the empirical clustering coefﬁcient and Eqs. (2) and (4) to benchmark the average path length. Both Newman et al. and Uzzi and Spiro (2005) show that for some bipartite networks, the match of empirical coefﬁcients to Eqs. (1) and (2) can

14

GORDON WALKER

be quite weak, which suggests that there are patterns in the data they analyze that are inconsistent with the assumptions behind these expressions. The more appropriate benchmarks to determine the presence of a small world are those developed from the Poisson distribution – the expressions shown in Eqs. (3) and (4) – since these are based on the same random model used by Watts (1999b) in his original small-world analysis. Comparing the two pairs of benchmarks – Eqs. (1) and (2) vs. (3) and (4) – will therefore be interesting and instructive here. The standard procedure for testing whether a bipartite network has smallworld properties has two basic steps. First, the empirical clustering coefﬁcient – three times the actual number of triangles over the actual number of triples – is divided by a benchmark clustering coefﬁcient – expressions (1) or (3). This ratio should be substantially greater than one for the network to be characterized as a small world. Second, the empirical average path length is divided by the benchmark path length – expressions (2) or (4). This ratio in turn should be close to one. A third step, introduced by Kogut and Walker (2001), is to divide the clustering ratio by the path length ratio. This ratio of ratios, sometimes called Q (see Kogut & Walker, 2001; Davis et al., 2003; Uzzi & Spiro, 2005), should be substantially greater than one. A high Q reﬂects the key small-world property identiﬁed by Watts (1999a, b) that, compared to random benchmarks, high clustering among actors is retained in the network even as the average path length is shortened.

Identifying Shortcuts The focus in this chapter is on the incidence of investments in ecommerce startups and the locations of ecommerce investments within the structure of the venture capital syndication network. As described above, structural location means whether or not an ecommerce investment occupies a shortcut in the network. A shortcut is a syndication that, when artiﬁcially removed from the network, increases the path length between the coinvesting venture capital ﬁrms by more than two links. Identifying a shortcut therefore entails removing the syndication from the network and calculating anew the pairwise distances between the ﬁrms that participated in it. Since shortcuts are deﬁned between pairs of ﬁrms, each pair whose distance increases by more than two deﬁnes a shortcut. One syndicated investment can therefore be a shortcut between some co-investing ﬁrms but not between others. To identify the shortcuts for ecommerce as an industry,

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

15

the syndications for all investments in ecommerce startups in a data window of four quarters (e.g., ﬁrst quarter 1991 to fourth quarter 1991) were removed from the network, and the distance matrix recalculated. An ecommerce syndication whose removal produced a path length of greater than two between at least two VCs that were members of the syndicate was deemed a shortcut. This procedure was performed on quarterly data aggregated into rolling four-quarter windows starting in the second quarter of 1979, when ecommerce startups enter the market, and ending in the second quarter of 1998, when the NASDAQ boom began. In each four-quarter window, the number of shortcuts created by ecommerce syndications was calculated. These became the predictors of the number of ecommerce syndications in the subsequent time period, following Moore and Newman’s (2000) model of epidemics in small worlds. To control for the possibility that the effect of shortcuts might be due to the centrality of the VCs rather than the position of the shortcut in the network, a variable measuring how many shortcuts in a four quarter window have VCs in the top outdegree decile is included in the runs.

Hypothesis Testing The hypotheses were tested using the negative binomial variation of Poisson regression. The data are inherently counts, and the variances for each dependent variable are markedly greater than their means, indicating that Poisson regression without a correction for this over – dispersion would produce biased estimates. The predictors in each of the two equations – shortcuts predicting investment (diffusion) and investment predicting shortcuts (network integration) – are lagged one quarter. Each equation also included a lagged dependent variable, emulating a test of Granger causality. To assess how important relative riskiness of an investment is for the diffusion of ecommerce through shortcuts, the hypotheses are tested using ﬁrst rounds only (more risky) and separately the later rounds (less risky).

RESULTS Is the syndication network a small world? Table 1 shows the actual, bipartite, and Poisson clustering coefﬁcients and the ratios of the actual

16

GORDON WALKER

coefﬁcient to the other two. The two ratios show that the actual and bipartite are much closer to each other than the actual is to the Poisson. This result suggests ﬁrst that Newman et al.’s (2001) bipartite clustering coefﬁcient is a reasonable estimator of the actual clustering coefﬁcient. Moreover, the ratio between the two coefﬁcients improves over time. Second, because the actual coefﬁcient is substantively larger than the Poisson coefﬁcient, we can assert with some conﬁdence that VC ﬁrms are clustered to a greater degree than one would ﬁnd in a random network. This ﬁnding is a necessary piece of evidence that the VC network, as captured in the rolling four-quarter windows, is a small world. The results in Table 2 are also supportive of the assertion that the VC network is a small world. Here the actual path length ﬁts the Poisson very closely, and not the bipartite, which is what one would want to ﬁnd if the actual path length matched that of a network of randomly generated ties. This is the second necessary ﬁnding to support the idea that the network is Table 1. Clustering Coefﬁcients for the Venture Capital Firm Syndication Network for the Venture Capital Firm Syndication Network Measured Over Four Quarters Ending in the Last Quarter of Each Year. Year

Actual Clustering Coeff. (Triangles/ Triples)

Bipartite Clustering Coeff.

Poisson Clustering Coeff.

Actual Coeff./ Bipartite Coeff.

Actual Coeff./ Poisson Coeff.

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

0.56767 0.58676 0.59825 0.607 0.58868 0.5666 0.58985 0.56518 0.58147 0.56849 0.54893 0.56154 0.54958 0.55622 0.51821 0.50921 0.42535 0.40387 0.37014

0.33055 0.31152 0.30561 0.28468 0.27275 0.33355 0.28456 0.35461 0.35963 0.40789 0.65671 0.60136 0.46772 0.47785 0.45944 0.50598 0.5496 0.48165 0.48028

0.20817 0.18117 0.15496 0.13703 0.14422 0.15491 0.14474 0.13562 0.14222 0.14747 0.15737 0.15778 0.14132 0.15326 0.1526 0.17222 0.1481 0.14776 0.15321

1.8271 1.9189 1.9128 2.1119 2.0685 1.649 1.9779 1.5487 1.5114 1.3292 0.78043 0.8731 1.1164 1.1577 1.135 1.013 0.84841 0.92441 0.88167

2.726953932 3.238726058 3.860673722 4.42968693 4.081819443 3.657607643 4.075238358 4.167379443 4.088524821 3.85495355 3.488148948 3.559006211 3.888904614 3.629257471 3.39587156 2.956741377 2.872045915 2.733283703 2.415899745

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

17

a small world. Table 3 shows that dividing the clustering ratio by the path length ratio produces a Q value substantively greater than one, indicating again that the network has small world properties. Table 4 shows that the main component of the network generally contains a very large percentage of VCs over the 76 periods, so using Newman et al.’s (2001) expression for the path length is appropriate. Since there is evidence that the network can be characterized as a small world, Moore and Newman’s (2000) model can be applied, and we can ask how important shortcuts are for the incidence of ecommerce startups. Tables 5a and 5b show the ﬁndings using the population of syndications for all ecommerce investment rounds. Table 5a shows the results for the time periods from mid-1979 to the end of 1991, the dormant stage for ecommerce startups; and Table 5b for the periods from 1992 to the end of 1998, the growth stage. In the early period from 1979 to 1991 (Table 5a), both the number of ecommerce syndications and the number of shortcuts in a quarter are predicted by both prior syndications and by the number of shortcuts in Table 2. Path Length Coefﬁcients for the Venture Capital Firm Syndication Network for the Venture Capital Firm Syndication Network Measured Over Four Quarters Ending in the Last Quarter of Each Year. Year Actual Average Path Length 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

2.8046 2.6964 2.6219 2.5426 2.6201 2.6879 2.659 2.5823 2.707 2.7592 2.8822 2.7843 2.7753 2.7678 2.8777 3.1999 3.1755 3.2256 3.3634

Bipartite Path Poisson Path Length Length 2.0121 1.8789 1.8161 1.6427 1.6696 1.7536 1.7395 1.8271 1.9098 1.9611 1.9811 1.88 1.834 1.9362 1.9029 2.0906 2.1422 2.1431 2.2365

2.6781 2.5877 2.4315 2.2379 2.3168 2.4424 2.3853 2.4085 2.5111 2.6597 2.8205 2.6837 2.5189 2.6888 2.7211 3.2157 3.1408 3.0564 3.3633

Actual Coeff./ Bipartite Coeff.

Actual Coeff./ Poisson Coeff.

1.3939 1.4351 1.4437 1.5478 1.5693 1.5328 1.5286 1.4133 1.4174 1.407 1.4549 1.481 1.5133 1.4295 1.5123 1.5306 1.4823 1.5051 1.5038

1.0473 1.042 1.0783 1.1362 1.1309 1.1005 1.1148 1.0722 1.078 1.0374 1.0219 1.0375 1.1018 1.0294 1.0575 0.9951 1.011 1.0554 1

18

GORDON WALKER

Table 3. Q Ratios of Clustering Coefﬁcients to Path Length Coefﬁcients for the Venture Capital Firm Syndication Network Measured Over Four Quarters Ending in the Last Quarter of Each Year. Year

Bipartite Q ratio

Poisson Q ratio

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

1.2321 1.3125 1.3559 1.3776 1.3753 1.1082 1.3561 1.1277 1.1407 0.99059 0.57453 0.63049 0.7765 0.81429 0.74583 0.6575 0.52211 0.55713 0.51249

2.7703 3.1667 3.4986 3.8618 3.4592 3.2263 3.4881 3.777 3.5451 3.5438 3.187 3.2074 3.3536 3.5065 3.2313 2.9908 3.114 2.8553 2.7639

the previous year, consistent with both the hypotheses. The control variable for the outdegrees of VCs in shortcut syndications has no effect. Its removal allows the shortcut variable to become signiﬁcant as a determinant of syndications. A different pattern is found for the growth stage (Table 5b). Here only the number of shortcuts is a signiﬁcant predictor of both syndications and shortcuts in the subsequent time period, but the effect on shortcuts is weak. Again, the outdegree control is statistically irrelevant. These ﬁndings thus support the diffusion hypothesis in both stages and the integration hypothesis only in the ﬁrst stage. Tables 6a and 6b show the same regressions, now however only for ﬁrst rounds. The results show consistently that shortcuts do not predict syndications in either the dormant stage (1979–1992) or the growth stage. However, they do predict their own incidence in both stages. In the growth stage, syndications determine shortcuts and are also self-predictive. The control for the presence of high outdegree VCs on shortcuts is not signiﬁcant. The results for later rounds are shown in Tables 7a and 7b. In the dormant stage, shortcuts and syndications predict shortcuts, and

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

19

Table 4. Size of the Main Component and Its Percentage of All Ties in the Venture Capital Firm Syndication Network Measured Over Four Quarters Ending in the Last Quarter of Each Year. Year

Size of the Main Component

Main Component Percentage of All Ties in the Network

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

263 391 468 658 784 775 790 714 679 635 527 417 486 435 478 640 780 1022 1290

0.8323 0.8612 0.8556 0.9203 0.9245 0.9237 0.924 0.9072 0.8864 0.8225 0.8528 0.7958 0.8223 0.7659 0.7446 0.6995 0.6933 0.73 0.7219

Table 5a.

Negative Binomial Regressions All Rounds Dormant Stage 1979–1991.

Independent Variables

Dependent Variables No. of Synst No. of Shortcutst No. of Synst No. of Shortcutst

No. of Synst1 No. of Shortcutst1 No. of Shortcutst1 in top outdegree decile Constant Pseudo R2 df po.05; po.001.

.141 (.0423) .0222 (.0188) .0877 (.0646) .474 (.197) .104 50

.127 (.0469) .1091 (.0217) .00969 (.0727) .461 (.222) .1302 50

.159 (.0429) .0342 (.0169)

.129 (.0444) .1102 (.02)

.422 (.2021) .0966 50

.456 (.219) .1301 50

20

GORDON WALKER

Table 5b.

Negative Binomial Regressions All Rounds Growth Stage 1992–1998.

Independent Variables

Dependent Variables No. of Synst No. of Shortcutst No. of Synst No. of Shortcutst

No. of Synst1 No. of Shortcutst1 No. of Shortcutst1 in top outdegree decile Constant Pseudo R2 df

.00519 (.00574) .0111 (.00461) .01061 (.0128) 2.852 (.149) .161 25

.01017 (.00694) .0125 (.0559) .0251 (.0153) 2.8 (.184) .143 25

.00351 (.00553) .00968 (.00432)

.0062 (.0071) .00937 (.0551)

2.855

2.798 (.198) .134 25

.153 25

po.05; po.001.

Table 6a.

Negative Binomial Regressions First Round Investments Dormant Stage 1979–1991.

Independent Variables

Dependent Variables No. of Synst No. of Shortcutst No. of Synst No. of Shortcutst

No. of Synst1 No. of Shortcutst1 No. of Shortcutst1 in Top Outdegree Decile Constant Pseudo R2 df

.131 (.115) .0011 (.0293) .051 (.108) .00948 (.253) .0149 50

.142 (.0913) .125 (.024) .0273 (.0821) .62 (.218) .114 50

.147 (.1094) .00871 (.0241)

.0114 (.251) .0134 50

.153 (.0863) .129 (.0213) .6065 (.216) .114 50

po.05; po.001.

syndications are serially correlated. However, the outdegree control confounds the relationship between shortcuts and syndications. In the growth stage, shortcuts determine themselves and syndications. Syndications have no effects. Overall, the results support the hypotheses, but with interesting nuances. The analysis shows that shortcuts are a determinant of new ecommerce

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

Table 6b.

21

Negative Binomial Regressions First Round Investments Growth Stage 1992–1998.

Independent Variables

No. of Synst1 No. of Shortcutst1 No. of Shortcutst1 in top outdegree decile Constant Pseudo R2 df

Dependent Variables No. of Synst

No. of Shortcutst

No. of Synst

No. of Shortcutst

.0212 (.01) .00535 (.00445) .00148 (.013) 2.087 (.179) .152 25

.0249 (.0107) .0118 (.00463) .0148 (.0138) 2.663 (.192) .153 25

.0212 (.0101) .00495 (.00271)

.026 (.0111) .00792 (.00291)

2.086 (.179) .152 25

2.63 (.199) .149 25

po.05; po.001.

Table 7a.

Negative Binomial Regressions Investments After the First Round Dormant Stage 1979–1991.

Independent Variables

No. of Synst1 No. of Shortcutst1 No. of Shortcutst1 in top outdegree decile Constant Pseudo R2 df

Dependent Variables No. of Synst

No. of Shortcutst

.144 (.067) .0396 (.0221) .145 (.0687) .0102 (.218) .121 50

.151 (.0623) .1039 (.0225) .0454 (.071) .578 (.203) .125 50

No. of Synst .17 (.071) .059 (.021)

0.27 (.229) .1 50

No. of Shortcutst .157 (.0626) .11 (.0211) .568 (.204) .12 50

po.05; po.001.

syndications only for rounds after the ﬁrst, suggesting that shortcuts create opportunities for lower, not higher risk investments. Further, shortcuts spur subsequent round syndications only in the growth stage. Interestingly, these constitute over 60% of all investments in ecommerce startups. But shortcuts are self-predictive for all types of investment and are determined by early numbers of subsequent rounds. The stock of ecommerce shortcuts is

22

GORDON WALKER

Table 7b.

Negative Binomial Regressions Investments After the First Round Growth Stage 1992–1998.

Independent Variables

No. of Synst1 No. of Shortcutst1 No. of Shortcutst1 in top outdegree decile Constant Pseudo R2 df

Dependent Variables No. of Synst

No. of Shortcutst

No. of Synst

No. of Shortcutst

.00319 (.0073) .0149 (.00352) .0122 (.0122) 2.224 (.136) .213 25

.000979 (.0114) .018 (.00535) .0179 (.0189) 2.861 (.188) .135 25

.001 (.00626) .0139 (.00345)

.0054 (.00945) .0169 (.00526)

2.217 (.141) .209 25

2.832 (.192) .132 25

po.05; po.001.

therefore self-sustaining. It does not deteriorate as new syndications are formed. As for contagion, there is evidence of it for low risk investments in the dormant stage and high-risk investments in the growth stage, a ﬁnding that indicates a transition from risk aversion to herd behavior. Fig. 2 summarizes the results.

DISCUSSION This chapter began with the question of how industries develop and argued that it is important to attend to the factors that inﬂuence investment in startup ﬁrms, since without such investment there would be no industry. The approach taken here has been to analyze the emergence of investment in ecommerce ﬁrms as a diffusion process across the network of deals among venture capitalists. What drives this process? To answer this question the analysis focuses on the syndication network structure, speciﬁcally, the location of ecommerce syndications on shortcuts in the deal network, consistent with Moore and Newman’s (2000) model of the spread of an epidemic in a small world. But unlike an epidemic, in which members of the population are typically afﬂicted once, the diffusion of ecommerce involves repeated ‘‘infection;’’ and the later infections have by assumption less risk. This split of investments according to their riskiness has important implications for how ecommerce diffuses.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

23

First Rounds Dormant Stage

Growth Stage

Shortcuts

Shortcuts

Syndications

Syndications

Later Rounds

Fig. 2.

Dormant Stage

Growth Stage

Shortcuts

Shortcuts

Syndications

Syndications

Path Diagrams for First and Subsequent Rounds in the Dormant and Growth Stages.

First round (more risky) syndications have no predictors in the dormant stage. However, they are self-reinforcing, reﬂecting contagion, in the growth stage. The absence of an explanation for these initial investments is consistent with a view that the origins of early startup funding are not contextual but idiosyncratic and cannot be understood by analyzing interﬁrm relationships. The comparison of ﬁrst and later rounds therefore illuminates further important themes in entrepreneurship, this time from the perspective of startup ﬁnancing rather then technology and business development.

24

GORDON WALKER

For investment in the industry to grow, the structural location (in terms of shortcuts) of ecommerce investments must be sustained. So the number of ecommerce shortcuts must not be reduced as new ties are formed. In fact, the results show that they do not decrease (Fig. 3) but in fact slowly rise. They do so both because existing shortcuts endure, and because new ecommerce syndications for lower risk investments, in the dormant stage, and higher risk investments, in the growth stage, establish new shortcuts. The structural location of ecommerce investments is thus effectively preserved, promoting further diffusion. One important potential stimulator of investment in ecommerce startups is the frequency of initial public offerings of these ﬁrms. One can speculate that as the number of Initial Public Offerings (IPOs) rises, VCs will gain conﬁdence that they will receive an acceptable return on their investments. Number of shortcuts

1200

computers data communications medical products services ecommerce

800

600

400

200

0

19801 19804 19813 19822 19831 19834 19843 19852 19861 19864 19873 19882 19891 19894 19903 19912 19921 19924 19933 19942 19951 19954 19963 19972 19981 19984

Number of shortcuts

1000

Four quarter periods

Fig. 3.

Number of Shortcuts.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

25

Interestingly, a test of this assertion shows that the number of ecommerce IPOs in a calendar quarter has no effect on the number of syndications in the next quarter, controlling for the other predictors in the analysis. This possible determinant of funding can therefore be ruled out. What does the epidemic model imply for research on network evolution in general? Five aspects of this study raise questions about how future studies on this topic might proceed. These elements are: the examination of ties, not actors; the presence of a small world; the endurance of new industry syndications on shortcuts; the possibility for repeated ‘‘infection;’’ and the existence of dormant and growth stages (Fig. 2). Each of these will be discussed in turn.

Ties not Actors First, as discussed above, research on network evolution has focused on actors and not ties. Yet as is shown in this study, where ties with speciﬁc content are located in the network structure plays a critical role in network growth. This implies that network evolution should be thought of as a series of shifting opportunity structures to which actors respond in varying degrees of intensity. Those actors that can adapt repeatedly stay in the network, and those that cannot exit. This perspective is not inconsistent with research that emphasizes the rise of highly central actors (see e.g., Gulati & Gargiulo, 1999) or with research that focus on search (Baum et al., 2003). However, it seems clear, based on the results presented here, that it is necessary to be speciﬁc about the shifting opportunities that lead some actors to become central and determine what actors are searching for.

Small Worlds Since Moore and Newman’s (2000) model draws on the structural attributes of a small-world network, speciﬁcally the presence of shortcuts, it cannot apply to a network that is not a small world. When the ﬁrms in a network are either weakly clustered or far apart the value of shortcuts as a diffusion mechanism is reduced, and network evolution as a function of opportunity structures that shift but overlap in time cannot occur in the way presented in this study. The emergence of a small world is thus a key precondition for epidemic-based network development. Fortunately, the VC syndication network becomes

26

GORDON WALKER

a small world early in its history, enabling the diffusion of new industries as described here. Enduring Industry Ties on Shortcuts Unlike a disease that diffuses through a network, infecting those in the population that are structurally disadvantaged, investment opportunities in new industries do not ﬂow. Each startup is a unique opportunity that disappears once VCs invest in it (in each round). The opportunity set is thus replenished as new startups are formed. VCs are thus exposed to the new industry as a ‘‘disease’’ when a syndicate for one of the industry’s startups is formed but they cannot ‘‘contract’’ the disease from this investment. They must ﬁnd or be found by a new startup to fund. But these new investments in the emerging industry must also lie on shortcuts for the epidemic model to be applicable. Otherwise, over time the degree of exposure to investment opportunities in the new industry weakens, the probability of funding decreases, and new industry growth slows. In effect, there is no epidemic at all. More generally, it is the combination of shortcut replication and the availability of new investment opportunities that determines the applicability of the epidemic model to network evolution; neither alone is sufﬁcient. The Potential for Reinfection Is it important that investment opportunities in an evolving network be categorized into more and less risky categories? The results here suggest that this is so. The reason is that more risky investments are less sensitive to structural inﬂuence, even though they contribute to the stock of shortcuts occupied by investments in new industry startups. More generally, ﬁrms in networks with both more and less risky ties are likely to experience the epidemic effect in the latter. If all ties are risky, then the present results suggest that diffusion through the network will not occur. The Presence of Dormant and Growth Stages An interesting and important artifact of the rise of the ecommerce industry is the existence of a long period of dormancy followed by a sharp increase in entry. The results for these two stages are different, especially for higher risk investments and for the development of the stock of shortcuts occupied by

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

27

syndications for new industry startups. Without the stock of shortcuts created by lower risk investments (later rounds) in the dormant stage, it is not clear that VCs would have been sufﬁciently exposed to the ecommerce industry in the growth stage. To generalize to network evolution, the diffusion of new opportunity structures may require an early phase in which adoption decisions are less an outcome than a determinant of where opportunities are located in the network structure. In the later stage, these locations then become important sources of exposure to the industry as an opportunity set. These implications to some extent relate to the points made in the chapter by Dagnino, Levanti and Mocciaro (2008) in this volume. Their chapter distinguishes between multiple levels of network evolution, albeit in a hub as opposed to a full population of actors. Investigating how these levels participate in the diffusion of a type of tie is an important question to be pursued.

Summary This study has applied a small worlds epidemic model to the rise of ecommerce as an investment opportunity for venture capitalists. Without understanding how such opportunities diffuse, one misses a critical, but unanalyzed, part of industry development – viz., the process underlying the funding of its startup ﬁrms. The epidemic model works well statistically in the presence of key controls, especially contagion effects represented by lagged syndications. Whether ecommerce is the only industry to which the model applies is a question that can be answered in a broader analysis. Further, researchers interested in networks over time would beneﬁt from focusing on the shifting opportunities ﬁrms face to form relationships as a central underlying force in network evolution.

REFERENCES Amaral L., Scala A., Barthelemy, B., & Stanley, H. (2000). Classes of small world networks. In: Proceedings of the National Academy of Sciences, USA, pp. 11149–11152. Baker, W. (1986). Three-dimensional blockmodels. Journal of Mathematical Sociology, 12, 191–223. Baum, J., Shipilov, A., & Rowley, T. (2003). Where do small worlds come from? Industrial and Corporate Change, 12, 697–725. Breiger, R. (1974). The duality of persons and groups. Social Forces, 53, 181–190.

28

GORDON WALKER

Burt, R. (1987). Social contagion and innovation. American Journal of Sociology, 92, 1287–1335. Coleman, J., Katz, E., & Menzel, H. (1966). Medical innovation. New York: Bobbs-Merrill. Corrado, R., & Zollo, M. (2006). Small worlds evolving: Governance reforms, privatizations, and ownership networks in Italy. Industrial and Corporate Change, 15, 319–352. Davis, G., Yoo, M., & Baker, W. (2003). The small world of the American corporation elite: 1982–2001. Strategic Organization, 1, 301–326. Gans, J., Hsu, D., & Stern, S. (2002). When does start-up innovation spur the gale of creative destruction? Rand Journal of Economics, 33, 571. Gompers, P. (1995). Optimal investment, monitoring, and the staging of venture capital. Journal of Finance, 50, 1461–1490. Gompers, P., & Lerner, J. (1999). The venture capital cycle. Cambridge, MA: MIT Press. Goyal, S., van der Leij, M., & Moraga-Gonzalez, J. (2006). Economics: An emerging small world. Journal of Political Economy, 114(2), 403–412. Greve, H., Strang, D., & Tuma, N. (1995). Speciﬁcation and Estimation of Heterogeneous Diffusion Models. Sociological Methodology, 25, 377–420. Gulati, R., & Gargiulo, M. (1999). Where do networks come from? American Journal of Sociology, 104, 1439–1493. Kleinberg, J. (2000). Navigation in a small world. Nature, 406, 845. Kochen, M. (1989). The small world. Norwood, NJ: Ablex. Kogut, B., & Walker, G. (2001). The small world of Germany and the durability of national networks. American Sociological Review, 66, 317–335. Kogut, B., Urso, P., & Walker, G. (2007). Emergent properties of a new ﬁnancial market: American venture capital syndication, 1965–2005. Management Science, 53, 1181–1198. Kuperman, M., & Abramson, G. (2001). Small world effect in an epidemiological model. Physical Review Letters, 86, 2909–2912. Lerner, J. (1994). The syndication of venture capital investments. Financial Management, 23, 16–27. McPherson, M. (1983). An ecology of afﬁliation. American Sociological Review, 48, 519–542. Milgram, S. (1967). The small world problem. Psychology Today, 2, 60–67. Moore, C., & Newman, M. (2000). Epidemics and percolation in small worlds. Physical Review E., 61, 5678–5682. Newman, M. (2000). Models of the small world: A review. Journal of Statistical Physics, 101, 819–841. Newman, M., Strogatz, S., & Watts, D. (2001). Random graphs with arbitrary degree distributions and their applications. Physical Review E., 64, 0261181–02611817. Podolny, J. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107, 33. Pool, I., & Kochen, M. (1978). Contacts and inﬂuence. Social Networks, 1(1), 5–51. Schilling, M., & Phelps, C. (2007). Interﬁrm collaboration networks: The impact of large-scale network structure on ﬁrm innovation. Management Science, 53, 1113–1126. Sorenson, O., & Stuart, T. (2001). Syndication networks and the spatial distribution of venture capital investments. American Journal of Sociology, 106, 1546–1588. Strang, D. (1991). Adding social structure to diffusion models. Sociological Methods and Research, 19, 324–353. Travers, J., & Milgram, S. (1969). An experimental study of the small world problem. Sociometry, 32, 425–443.

Rise of Ecommerce as an Epidemic in the Small World of Venture Capital

29

Uzzi, B., & Spiro, J. (2005). Collaboration and creativity: The small world problem. American Journal of Sociology, 111, 447–504. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 2, 109–125. Watts, D. (1999a). Small worlds: The dynamics of networks between order and randomness. Princeton, NJ: Princeton University Press. Watts, D. (1999b). Networks, dynamics and the small world problem. American Journal of Sociology, 105, 493–527. Watts, D. (2004). The new science of networks. Annual Review of Sociology, 30, 243–270. Watts, D., Dodds, P., & Newman, M. (2002). Identity and search in social networks. Science, 296, 1302–1305. Watts, D., & Strogatz, S. (1998). Collective dynamics of small world networks. Nature, 393, 440–442. White, H. (1970). Search parameters for a small world. Social Forces, 49, 259–264.

OWNERSHIP AND CONTROL: A SMALL-WORLD ANALYSIS Martin J. Conyon and Mark R. Muldoon ABSTRACT In this chapter we investigate the ownership and control of UK ﬁrms using contemporary methods from computational graph theory. Speciﬁcally, we analyze a ‘small-world’ model of ownership and control. A small-world is a network whose actors are linked by a short chain of acquaintances (short path lengths), but at the same time have a strongly overlapping circle of friends (high clustering). We simulate a set of corporate worlds using an ensemble of random graphs introduced by Chung and Lu (2002a, 2002b). We ﬁnd that the corporate governance network structures analyzed here are more clustered (‘clubby’) than would be predicted by the random-graph model. Path lengths, though, are generally not shorter than expected. In addition, we investigate the role of ﬁnancial institutions: potentially important conduits creating connectivity in corporate networks. We ﬁnd such institutions give rise to systematically different network topologies.

Network Strategy Advances in Strategic Management, Volume 25, 31–65 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25002-3

31

32

MARTIN J. CONYON AND MARK R. MULDOON

1. INTRODUCTION Social network analysis is a critical methodology for analyzing the interrelationships between actors (e.g., Burt, 1992; Wasserman & Faust, 1994). Recent advances in computational graph theory have opened up exciting new techniques for investigating complex network structures in diverse disciplines such as sociology, economics, physics, the life-sciences and organizational studies (Newman, 2003; Jackson, 2007). In this chapter we use techniques introduced by Newman, Strogatz, and Watts (2001) and more recently Chung and Lu (2002a, 2002b); Chung, Lu, and Vu (2003) to provide a social network analysis of the ownership and control of publicly traded UK ﬁrms. This paper extends our previous network evaluation of corporate boards (Conyon & Muldoon, 2006). We contribute to the extant social network, strategy and corporate governance literature in two distinct ways. First, we argue that contemporary developments in graph theory are capable of providing deep insights into the structure of social networks in organizational settings. We illustrate this by presenting a social network analysis of the governance of publicly traded ﬁrms (i.e., the ownership and control of ﬁrms). Speciﬁcally, we discuss important concepts such as ‘bipartite graphs’, ‘small-world’ models, and ‘random graphs’ and show how they can be proﬁtably applied to the analysis of corporate governance phenomenon.1 Second, in order to illustrate the usefulness of this graph-theoretic approach to modeling corporate governance phenomena, we present empirical evidence on actual ownership and control (board) networks. Using data on a set of publicly traded UK ﬁrms for the year 2000, we test whether observed ownership and control networks can be characterized as a ‘small-world’. Our research builds upon prior small-world studies in organizational and business settings. Previous studies ﬁnd that inter-ﬁrm networks may exhibit important ‘small-world’ properties but have been largely silent as to why they exhibit such patterns and whether they deviate from random-graph alternatives. We use the methods introduced by Chung and Lu (2002b, 2002a); Chung et al. (2003) to simulate a distribution of random graphs and then compare real-world data to them. The context for our research project is the governance of publicly traded ﬁrms in ‘Anglo-Saxon’ economies. Exemplars of the Anglo-Saxon mode of corporate governance are the United States and the United Kingdom. There are two distinctive features of ownership and control in these economies. First, ownership is diffuse in the sense that there are many different shareholders each of whom own only a small fraction of the ﬁrm’s equity.

Ownership and Control

33

Second, de facto control of the ﬁrm resides with the Chief Executive Ofﬁcer (CEO) and the board of directors.2 The social network approach offered here contrasts with the canonical agency model which focuses on the ownership and control of a single ﬁrm without reference to its connections to other organizations. The separation of ownership from control gives rise to an agency problem: the CEO can potentially engage in opportunistic selfserving activities. Tirole (2005) provides an excellent account of the corporate ﬁnance approach to the classic principal-agent problem. Since Berle and Means (1932) governance questions have centered on how to resolve this agency problem. In consequence, the design of compensation contracts, monitoring by independent directors or the discipline imposed by an active takeover market have been widely studied. However, this representation of the ﬁrm is incomplete. Firms are not independent islands in a sea of economic activity. Instead, ﬁrms are frequently linked to each other in a complex network of various types of ties. For example, an example of such a tie is when a given ﬁrm shares a common owner or board member with at least one other ﬁrm. These ties are often sufﬁcient to connect ﬁrms together in a giant network. Such connectivity, we might conjecture, has signiﬁcant implications for understanding network strategies. The presence of short path lengths and high clustering in a network promotes the rapid diffusion of knowledge, information, ideas and strategic practices throughout the corporate network (see Cowan & Jonard, 2004 and Schilling & Phelps, 2005). Our chapter is motivated, in part, by a famous study by the renowned psychologist Stanley Milgram who suggested that the social world of the USA is ‘small’ in the sense that most pairs of Americans are connected by a rather short chain of acquaintance (around six intermediates) – see Milgram (1967).3 This ﬁnding turns out to be true even though (a) most Americans are acquainted with only a tiny fraction of the country’s population and (b) circles of acquaintance tend to be strongly overlapping – that is, one’s friends have a tendency to be friends in their own right. In the last decade or so scholars from diverse disciplines, including physics, sociology and economics, have become very interested in this phenomenon and have sought to understand how it arises. An important strand of this work is the exploration of various ensembles of random graphs (see Newman et al., 2001; Chung & Lu, 2002a, 2002b; Chung et al., 2003; Robins, Woolcock, & Pattison, 2005). In our study we are interested in the application of these principles to the social network of UK ownership and control. A small-world, then, is a network whose actors are linked by a short chain of acquaintances, but at the same time have a strongly overlapping circle of

34

MARTIN J. CONYON AND MARK R. MULDOON

friends. These two distinctive properties, the coexistence of short path lengths and a high degree of local clustering, constitute the small-world phenomenon. This immediately prompts the question ‘small compared to what’? In response, there has been considerable effort in the graph theory literature to develop a sensible and tractable answer to this question. Broadly, the answer is to compare the observed data to some suitably assembled ensemble (collection) of random graphs. Early work made comparisons with the so-called Poisson random graphs introduced by Erdo+ s and Re´nyi (1959). Indeed, important contributions to the small-world of corporate governance literature by Kogut and Walker (2001) and Davis, Yoo, and Baker (2003) and did precisely this. However, it is now known that social networks have particular structural properties (e.g., directors automatically know each other by virtue of sitting on the same board) that makes the comparison to the Poisson graph inappropriate. Consequently, more recent efforts have used graph ensembles designed to capture more features of the social network. In an important paper Newman et al.(2001) examined afﬁliation networks which recognize that the network of corporate governance has a bipartite structure. That is, the network has two distinct kinds of actors, for example, boards and the directors who sit on them. One of the consequences of this structure is that one should automatically expect a high degree of clustering among directors; after all, as noted above, the directors who sit on a given board know each other. The main accomplishment of Newman et al. (2001) is to provide numerical predictions of the small-world statistics that one can derive from an appropriately assembled random corporate universe. It is these expected values of the path length and clustering coefﬁcient that form a legitimate benchmark to compare actual data against. However, their methods yield only average estimates of the small-world statistics and provide little guidance as to other properties of the distribution, such as the variance of clustering and path lengths. Our chapter differs from prior social science applications of graph theory by considering a slightly different collection of random graphs introduced by Fan Chung and Linyuan Lu (see Chung & Lu, 2002a, 2002b; Chung et al., 2003). This family of random graphs have two attractive properties: ﬁrst, they are more tractable analytically and second lend themselves readily to numerical simulation. This latter property means that the researcher is able to simulate as many hundreds of corporate universes as desired and so examine the entire distribution of the small-world statistics. In our chapter, we do precisely this. The research question we ask has the ﬂavor of a hypothesis test: is the real corporate universe similar to, or different from,

35

Ownership and Control

a randomly assembled business world? Newman, Watts, and Strogatz (2001, 2002) suggested that the corporate elite was rather close, in a particular distributional sense, to a random one. If this really is the case, then the world is actually no smaller than one would expect by chance. We extend this research by focusing on the family of random graphs introduced by Chung and Lu (2002a, 2002b); Chung et al. (2003). The rest of the chapter is organized as follows. In Section 2, we show that the ownership and control of ﬁrms can be characterized as a bipartite graph. We provide concrete examples to show this. In Section 4, we describe germane aspects of small-world theory. Our focus is on the class of models introduced by Chung and Lu (2002a, 2002b); Chung et al. (2003) and how they may be applied to the corporate governance social network. In Section 4, we describe the UK data to be used in this study. We present the results and deduce that the world is, indeed, ‘small’, but only in the sense that the corporate world appears more clustered than expected. Path lengths, however, are no smaller than expected, even though they form a very small fraction of the available vertices in the network. The innovation here is that we provide an appropriate benchmark, based on Chung–Lu random graphs, to evaluate the signiﬁcance of smallness. In addition, we analyze the importance of ﬁnancial institutions in the network. It is often argued that banks are important conduits creating connectivity in corporate networks. We demonstrate that the presence of ﬁnancial institutions does indeed give rise to different network topologies. Finally in Section 5 we offer some concluding remarks.

2. SOCIAL NETWORKS OF OWNERSHIP AND CONTROL 2.1. Motivation Our concern in this chapter is to analyze the ownership and control of enterprises using the tools of contemporary social network analysis. We focus on the type of ﬁrm that is publicly traded on stock exchanges in economies like the United States and the United Kingdom. Such ﬁrms, as is well-known, are characterized by diffuse ownership in the sense that a myriad different shareholders each own only a small fraction of the ﬁrm’s equity. In addition, the ﬁrm is controlled by a CEO and the board of directors. Importantly, the ﬁrm’s owner (shareholder) does not generally

36

MARTIN J. CONYON AND MARK R. MULDOON

make strategic decisions about the ﬁrm’ assets. Instead, the task of running the ﬁrm is delegated to a separate body (the CEO and board of directors) giving rise to a de facto separation between ownership and control (Berle & Means, 1932; Tirole, 2005). The theoretical lens typically used by economists to analyze such situations is the principal–agent model. In this context the principal would be the shareholder and the agent would be the CEO.4 The divorce between ownership and control implies that the principal (shareholder) cedes the task of running the ﬁrm to an agent (the CEO). The mere fact of this delegation suggests that the agent may have access to private information that is not similarly available to the principal. For example, the shareholder may not know with certainty the true skills and quality of the CEO when he or she is initially appointed. Alternatively, the shareholder may not be able to perfectly know whether the CEO is taking the right actions to safeguard the owner’s asset. For example, the CEO may be taking unwarranted perks in the form of excessive pay or empire building. Each of these problems arises because of the private knowledge of the agent.5 A primary concern of the principal–agent literature has been how to design an optimal contract in order to resolve the tensions between the principal and agent arising from such natural asymmetries of information. There is a vast literature on this. For example, a central research theme in agency theory centers on how to design an optimal compensation contract to align the interests of CEOs with owners. Such models typically predict the use of stock options, bonus payments or proﬁt sharing as a potential solution. Alternative mechanisms to alleviate the costs associated with the separation of ownership and control include monitoring activities. The ﬁrm might appoint auditors, establish audit committees and hold regular audits to establish the veracity of corporate accounts for example (Fama & Jensen, 1983; Jensen, 1993; Tirole, 2005). Whilst the agency perspective provides important insights into the relationships between actors in an economy, it is not the only way to model such interactions. Our conjecture in this chapter is that the principal–agent representation of the nexus of ownership and control relationships is incomplete. It is incomplete in the sense that the models are generally applied to information asymmetries arising in a single context. For example, how to solve the shirking (moral hazard) problem between a manager and worker in a single ﬁrm. Or how a shareholder can motivate the CEO in a single ﬁrm. The focus of attention is not, in general, on the connections that might arise between the whole network of ﬁrms. In short, most of the important features the agency model can be articulated in a single

Ownership and Control

37

hierarchical relation between a single principal and a single agent.6 Interconnections between ﬁrms are ignored. However, ﬁrms are not independent islands in a sea of economic activity. Instead, they are connected to each other in important ways. For example, a given ﬁrm typically shares a common owner with at least one other ﬁrm. In addition, the same ﬁrm might share a common board member with another ﬁrm. Or indeed, it might share a common advisor such as a compensation consultant or audit company with another ﬁrm. These ties are sufﬁcient to connect ﬁrms together in a giant complex network. Such connectivity, we conjecture, will have signiﬁcant implications for ﬁrm strategy. For example, the presence of short path lengths in the network and/or a high degree of clustering may promote the rapid diffusion of knowledge, information, ideas and even business practices throughout the corporate network (see Cowan & Jonard, 2004; Schilling & Phelps, 2005; Jackson, 2007; Walker, 2008).

2.2. Corporate Governance Relations as Bipartite Graphs An important feature of our chapter is the representation of the interconnections between actors in the corporate governance system as bipartite graphs. To understand this we ﬁrst need to introduce some necessary terminology. A graph or network is a set of items termed vertices (sometimes also called nodes) and edges. The vertex is the fundamental unit of a network and an edge is a link connecting two vertices. Two vertices are adjacent if connected by an edge. The number of edges connected to (or incident on) a vertex is called the degree and is a local measure of the vertex’s centrality in the graph. A vertex’s connected component is that part of the graph consisting of the vertex itself and all those others that can be reached by paths running along the edges. A geodesic is a shortest path7 (in the sense of traversing the fewest edges) that connects two vertices and ﬁnally, the distance between two vertices is the number of edges in a geodesic connecting them. We are now in a position to represent many corporate governance relationships as a bipartite graph. To be concrete, a bipartite graph is one whose vertices can be divided into two distinct sets and whose edges connect only unlike kinds of vertices. Previous work has also stressed the importance of modeling boards of directors as a bipartite graph (Robins & Alexander, 2004; Conyon & Muldoon, 2006). Consider how one might represent the owners of a ﬁrm and the ﬁrm itself as a such a bipartite graph. There are two sets of vertices: the ﬁrst set are the ﬁrms themselves and the second set are

38

MARTIN J. CONYON AND MARK R. MULDOON 1

B

A

C

2

D

E

3

F

G

4

H

I

K

J

H

A 1

2

3

4

B

F

C

I D

G

E K J

Fig. 1. The Network Structure of Ownership and Control. The Top Half is a Bipartite Graph. Vertices Labeled A to K Can Represent Either Company Directors or the Firm’s Shareholders. Vertices 1 to 4 Represent Either the Board of Directors or the Firm. The Bottom Half Shows the Projections: the Board/Firm Projection is to the Left and the Director/Ownership Projection is to the Right.

the owners of the ﬁrm. Fig. 1 illustrates such a bipartite graph and its projections. In the upper-half of the ﬁgure the two sets of vertices are displayed horizontally. Vertices {1,2,3,4} represent the set of ﬁrms and vertices {A,B,y, J,K} represent shareholders. An edge exists between the two sets if a shareholder owns shares in a given ﬁrm. Notice that the edges run between the two sets of vertices. The bipartite graphs give rise to two distinct projections. These two separate graphs are drawn in the lower-half of Fig. 1 and are ‘projections’ of the bipartite graph onto one of its two sets of vertices. One might call these the ‘ownership projection’ and the ‘ﬁrm projection’. In Section 3 we calculate small-world statistics for each of these projections.8 This procedure for understanding the relation between different sets of actors in a social network is quite general. The set of vertices X ¼ {1,2,3,4} can represent a ﬁrm and vertices Y ¼ {A,B,y, J,K} represent directors. In this case an edge exists between the two vertices of a director is the member of the board. To illustrate the bipartite representation of ownership and control consider the two companies in Table 1: Marks and Spencer Plc. (a highstreet retailer) and Standard Chartered Plc. (a bank).9 Suppose we are interested in the ﬁrms’ boards and directors. The two ﬁrms can be represented like the vertices in the upper-half of Fig. 1 and the directors like

39

Ownership and Control

Table 1. Company Name Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Marks and Spencer Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc.

The Board of Directors. Name

Role

L Vandevelde R W C Colvill A McWalter D Norgrove A Ball Sir Michael Perry CBE Dame Stella Rimington DCB Sir David Sieff B F Baldock CBE Sir Ralph Robins J K Lomax Sir Patrick Gillam G S Talwar P N Kenny The Rt Hon Lord Stewartby RD C N A Castleman E M Davies K S Nargolwala C A Keljik M DeNoma D G Moir A W P Stenham R C Chan H E Norton K A V Mackrell H KwonPing Sir C Chow Sir Ralph Robins B Clare

Exec Exec Exec Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Exec Exec Exec Non-Exec Exec Exec Exec Exec Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec Non-Exec

‘Interlock’

Yes

Yes

Yes Yes

Yes

Yes Yes

Yes Yes Yes

the vertices in the lower-half. Marks and Spencer Plc. and Standard Chartered Plc. are linked since they share a common director: Sir Ralph Robins, who is a non-executive (outside) director at each ﬁrm. Table 1, though, is only a partial representation of the links between ﬁrms generated by shared board members. The ﬁnal column is a variable indicating whether the board member has at least one additional director appointment. Marks and Spencer Plc. is connected to Standard Chartered Plc. via Sir Ralph Robins, but is also connected to other listed ﬁrms (not shown) since Dame Stella Rimington, J K Lomax and I P Sedgwick also hold additional directorships at other ﬁrms.10

40

MARTIN J. CONYON AND MARK R. MULDOON

Table 2. Company Name Marks and Spencer Plc. Marks and Spencer Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc. Standard Chartered Plc.

The Ownership of Firms. Shareholder

Ownership

Clients

Franklin Resources Inc. Brandes Investment Partners CGNU Plc. Prudential Corporation Plc. Tan Sri Khoo Teck Puat Scottish Widows

5.89 4.319 3.026 4.725 13.994 3.003

40 17 184 137 1 71

In a similar fashion we could examine the ownership structure of the two ﬁrms. This is illustrated in Table 2. Under UK corporate law, ownership stakes exceeding 3% are reported and can be observed by the researcher. We therefore list all owners (who are not board members) with share stakes exceeding this critical threshold. Table 2 illustrates that Marks and Spencer has two such owners, while Standard Chartered has four. In the case of Marks and Spencer the two largest shareholders own about 10.2% of the company and the four largest shareholders in Standard Chartered own about 25%. There is not a single controlling block holder in either case – this is consistent with the Anglo-Saxon model that ownership is diffuse with each shareholder typically owning a small fraction of the ﬁrm’s common equity.11 The two example ﬁrms are not linked by a common shareholder. However, the ﬁnal column is the number of listed companies in which the named shareholder has a share stake exceeding 3%. For example, the Prudential Corporation owns about 4.7% of Standard Chartered. In addition, it owns at least 3% share stakes in 136 other publicly listed UK ﬁrms. In Section 4 we document the general pattern of ownership.

3. THE SMALL-WORLD MODEL The small-world theoretical model addressed in this chapter has been described by Newman et al. (2001), Jackson (2007), Watts (1999b), Watts and Strogatz (1998) and Robins et al. (2005) among others. We can now describe two main properties that characterize a ‘small-world’, namely the notions of mean path length and clustering. The ﬁrst property of the small-world model is short mean (average) path lengths. That is, randomly chosen pairs of vertices turn out to be unexpectedly close to each other. A graph with N vertices contains

41

Ownership and Control

N(N1)/2 unordered pairs of vertices. If these are numbered 1, 2, y, N and di,j is the distance between vertex i and j then one can calculate the typical (average) path length as: X 2 d i; j (1) L¼ NðN 1Þ ioj In our numerical work we compute L from a complete list of the di,j, found using Johnson’s All-Pairs-Shortest-Path algorithm. In the context of boards, the term L measures the number of steps in a chain of shared directors that it takes to get from a source board to a target board. The second property of the small-world model is high ‘clustering’. This is the propensity for one’s network neighbors to be neighbors in their own right. It is a measure of network density. Following Newman et al. (2001) we calculate the following clustering coefﬁcient for the graph as a whole: CD ¼

3 ðnumbe of triangles in the graphÞ ðnumber of connected triplesÞ

(2)

where a triangle is a set of three distinct vertices j, k, l in which each vertex is connected to both the other two. A connected triple is a set of three vertices j, k, l in which j is connected to k and k is connected to l (though l need not be connected to j). CD is also called the transitive closure of the graph and 0oCD o1. In our data, a large value of CD is a measure of how ‘clubby’ boards of directors are. When CD ¼ 1 everybody is connected to everyone else.

3.1. Random Graphs I: Newman, Strogatz and Watts We have identiﬁed the properties of a small-world. Next, one needs to agree on what is meant by L being ‘small’ and CD being ‘large’ in order to deduce the presence of a small-world. A now standard approach is to compare L and CD from real social networks to the numbers one would expect to measure in randomly assembled graphs that share some properties with the observed network. In their seminal work on random graphs, Erdo¨s and Re´nyi imagined ﬁxing the number of nodes N and then deciding, at random, independently and with ﬁxed probability p, whether each of the graph’s N(N1)/2 possible edges exist (see Erdo+ s & Re´nyi, 1959, 1960). In such a graph each vertex can

42

MARTIN J. CONYON AND MARK R. MULDOON

have as many (N1) edges – one connecting it to each of the remaining nodes – and in expectation a fraction p of these will exist, so such a graph has mean degree z ¼ p(N1). More generally, the degree distribution is binomial: N1 k pk ¼ p ð1 pÞNk1 k for kA{0,1,y, N1} and zero otherwise. Here pk is the probability of ﬁnding a node of degree k. Erdo¨s and Re´nyi were primarily interested in how the qualitative properties of such graphs changed as they held the mean degree z constant, but allowed the number of nodes to tend to inﬁnity. In the limit the degrees of the nodes have a Poisson distribution pk ¼ zk ez =k. As a shorthand we will refer to graphs with Poisson degree distributions as Poisson random graphs. Such graphs have studied extensively (see e.g., Bolloba´s, 2001) and have been used as the benchmark by which to compare social networks. For example, see Newman et al. (2001) as well as Davis et al. (2003), both of whom compare small-world statistics from actual social networks to those expected for random graphs with a Poisson degree distribution. More recently, an alternative and more appropriate benchmark was proposed by Newman et al. (2001, 2002). Newman, Strogatz and Watts (NSW) developed a way to calculate the expected values of both L and CD using the machinery of probability generating functions.12 Their approach takes as its input the two empirical degree sequences – the distribution of board size (number of seats on the board) and the distribution of workload among directors (the vast majority of whom sit on only a single board, though a few serve on many more) – and combines them to predict degree sequences for the board and director projections. Their predictions are large-graph limits for a certain family of randomly assembled social networks, but they are in surprisingly good agreement with empirical data, especially for the distribution of degree in the director projection. Using the machinery of generating functions one can calculate the expected mean degree, path length and clustering coefﬁcient for the graph. The degree distribution associates a probability pk with each possible value of the degree k, where k is any non-negative integer. Such distributions permit one to construct a probability generating function, G(x), which is a P k function of one variable deﬁned by the inﬁnite sum: GðxÞ ¼ 1 p k¼0 k x and it is a general property of generating functions that G(1) ¼ 1. For generating

Ownership and Control

43

functions P arising from degree distributions the mean degree is given by: 0 z ¼ hki ¼ 1 k¼0 kpk ¼ G ð1Þ. Here the angle brackets are the expectation 13 operator. A bipartite graph of either the control network (board-to-director) or the ownership network (shareholder-to-ﬁrm) has two empirical degree distributions. This gives rise to two separate generating functions. Consider the case of the control network. We denote f0(x) as the function that generates the degree distribution for the directors (i.e., the distribution of the number of boards on which a director sits). For concreteness, say that the frequency with which one P ﬁnds a director serving on j boards is pj. Then f0(x) is given by f 0 ðxÞ ¼ j pj xj . The second degree distribution for boards (i.e., the distribution of the number of directors per ﬁrm) is given as g0(x). If the empiricalP frequency for boards made up of k directors is qk then g0 ðxÞ ¼ k pk xk . Our real objects of interest are the two projections of the bipartite graph: one whose nodes are boards and whose edges represent shared directors (the graph showing board interlocks) and another whose nodes are directors and whose edges connect directors that sit on one (or more) board in common. The charm of the generating function approach is that it permits one to start with the empirical generating functions f0(x) and g0(x) and derive expressions for the generating functions for the degree distributions in the projections. We will refer to these derived distributions as the theoretical degree distributions to emphasize that they are not measured directly from the data. Instead, they describe the distribution of degrees one would ﬁnd in random corporate worlds constructed by applying the methods of the previous section to the empirical degree distributions. These theoretical degree distributions are the device by which we obtain expected values of the small-world statistics without having to generate and analyze random graphs. Suppose now that we are investigating a corporate world in which N directors sit on M boards. Suppose further that the mean number of seats on a board is n and that the mean number of directorships held is m. The bipartite graph representing this community has one edge for each seat on a board and so nM ¼ ðnumber of seats on boardsÞ ¼ mN Deﬁne the generating function for theoretical degree distribution of the projection onto directors as G0(x). We denote the expected degree as /zS, the expected path length as /LS W and the expected clustering coefﬁcient

44

MARTIN J. CONYON AND MARK R. MULDOON

as /CDS. Newman et al. (2001) show that these quantities are given by: G0 ðxÞ ¼ f 0

0 g0 ðxÞ 1 0 g ¼ f ðxÞ 0 g00 ð1Þ n 0

z ¼ G00 ð1Þ lnðN=G0 ð1ÞÞ 00 0 00 f 0 ð1Þ g0 ð1Þ ln f 00 ð1Þ g00 ð1Þ M g000 0 ð1Þ hC D i ¼ N G000 ð1Þ hLi ¼ 1 þ

(3)

The corresponding expressions for the projection whose vertices are boards may be obtained similarly and are given by the following expression: F 0 ðxÞ ¼ g0

0 f 0 ðxÞ f 00 ð1Þ

z ¼ F 00 ð1Þ lnðM=F 0 ð1ÞÞ 00 0 00 g0 ð1Þ f 0 ð1Þ ln g00 ð1Þ f 00 ð1Þ N f 000 0 ð1Þ hCD i ¼ M F 000 ð1Þ hLi ¼ 1 þ

(4)

Each quantity in Eqs. (3) and (4) (i.e., the mean degree, expected path length and clustering coefﬁcient) can be calculated from the observed data. In the empirical work below we can calculate these separately for the ownership and control networks. For the control network we use Eq. (3) for the projection whose vertices are directors and (4) for the projection whose vertices are boards. Similarly, for the ownership network we can use Eq. (3) for the projection whose vertices are shareholders and (4) for the projection whose vertices are ﬁrms. Previous empirical work on small-worlds in the context of corporate governance has used the results of the NSW theory. For example, Conyon and Muldoon (2006) evaluate whether the board of directors in the UK can be considered a small-world.

3.2. Random Graphs II: Chung and Lu More recently, Fan Chung and Linyuan Lu have introduced a very convenient and attractive ensemble of random bipartite graphs that has a

45

Ownership and Control

prescribed expected degree sequence (Chung and Lu, 2002a, 2002b; Chung et al., 2003). That is, they start with the same ingredients as Newman et al. (2001, 2002), but use them in a different way. Their approach amounts to a sort of probabilistic hiring hall for corporate leadership: the probability that a given director serves on a given board depends only on the number of directorships she wants hold and the number of seats on the board. To the extent that their approach allows one to assemble randomly a corporate network similar to the real one, Chung and Lu’s model is similar to that of Newman, Strogatz and Watts, but the Chung-Lu approach is slightly superior as it excludes certain unrealistic kinds of corporate network. For example, almost every graph in the NSW family will include a few cases in which a single director holds multiple seats on the same board: the ChungLu family excludes such examples by design. Although such anomalous directors will appear only extremely rarely in a given NSW network (far too rarely to affect such summary statistics as L and CD for any realistically sized corporate network), it is still appealing to work with a family of random graphs that includes no such cases. To be concrete, we will illustrate the construction of a board network. Ownership networks can be similarly fashioned. Imagine that in the original data director A served on dA boards (that is, she has degree dA in the afﬁliation network representing the real corporate world) and that board l consisted of d1 directors. Then, in a Chung–Lu (CL) random graph, director A would serve on board 1 with probability pA;1 ¼ d A d 1 =r where r¼

X Boards j

dj ¼

X

(5) dk

(6)

Directors k

is the total number of edges (or, equivalently, seat on boards) in the original network. This prescription ensures that, in expectation, each board has the same number of directors and each director holds the same number of directorships as in the real data. To see why, consider Fig. 2. The upper panel shows a small corporate world with three boards and eleven directors: here the total number of edges r is 15. The lower panel illustrates those steps in the construction of a CLrandom network that determine which edges are incident on director C: she serves on board 1 with probability pC;1 ¼

dCd1 2 4 8 ¼ ¼ 15 15 r

46

MARTIN J. CONYON AND MARK R. MULDOON

1

A

C

B

1

A

D

E

C

D

F

2

d1= 4

B

3

2

E

F

G

H

I

J

K

3 d3= 5

d2= 6

G

H

I

J

K

dC = 2

Fig. 2. Constructing a Chung–Lu Random Graph. The Upper Panel Shows a Small Corporate World, while the Lower Panel Illustrates that Stage in Construction of a CL-Random Network which Determines the Boards on which Director C Serves.

Similarly, she serves on board 2 with pC;2 ¼ 12=15 and on board 3 with pC;3 ¼ 10=15. Thus, her expected degree in the random network (the expected number of edges connected to vertex C) is 1 p1;C þ 0 ð1 p1;C Þ þ 1 p2;C þ0 ð1 p2;C Þ þ 1 p3;C þ 0 ð1 p3;C Þ 8 7 12 3 10 5 ¼1 þ0 þ1 þ0 þ1 þ0 15 15 15 15 15 15 30 þ0 ¼ 15 ¼2 More generally, if director C serves on dC boards in the real corporate world then her expected degree (number of directorships) in a CL-random network is X X 1 pC; j þ 0 ð1 pC; j Þ ¼ pC;j Boards j

Boards j

¼

X

Boards j

d C d j =r

47

Ownership and Control

¼ d C ð1=rÞ

X

dj

Boards j

¼ dC where the ﬁnal equality follows from Eq. (6). Of course, the same argument works for any other director and a similar one for any board, so the simple prescription of Eq. (5) allows us quickly to generate random graphs whose expected degree sequence matches that of the observed data.

3.3. Findings from Previous Research There has been a recent upsurge of interest in small-world models and their application to social networks14 (Newman, 2003; Jackson, 2007; Watts, 1999a; Rowley & Baum, 2008). A comparatively small number of papers have applied small-world network analysis to the governance and control of organizations. For example, contemporary work on small-worlds and complex systems includes Dagnino, Levanti, and Li Destri (2008), Walker (2008) and Cohen, Frazzini, and Malloy (2007). None, to our knowledge, have simultaneously examined the ownership and control of ﬁrms, or used the types of techniques proposed in this chapter (i.e., Chung–Lu graphs). However, it is important to highlight some of the salient ﬁndings from the extant empirical literature. In recent research Walker (2008) provides empirical evidence on investments in startups, speciﬁcally how the existing structure of the venture capital syndication network in the US promoted an epidemic of startups in e-commerce. He ﬁnds that ‘consistent with existing theory on the spread of a disease in a small-world y the incidence of investments in e-commerce startups was a function of prior investments located on short cuts in the network’. Dagnino et al. (2008) present evidence of ‘networks as complex dynamic systems of knowledge and capabilities’ and provide an ‘in depth analysis of the processes underlying the emergence and evolution of STMicroelectronic’s global network and of Toyota’s supplier network in the US’. Schilling and Phelps (2005) investigate alliance networks and patent performance in a set of US ﬁrms and industries. They propose and ﬁnd that a high degree of clustering and short average path lengths are positively associated with greater knowledge creation compared to ﬁrms in networks that do not have such characteristics. Nguyen-Dang (2008) investigates the small-world of the corporate elite in France. He ﬁnds that ‘socially wellconnected CEOs are less likely to be dismissed for poor performance and

48

MARTIN J. CONYON AND MARK R. MULDOON

more likely to ﬁnd new and good employment after a forced departure’. Conyon and Muldoon (2006) and Kramarz and Thesmar (2006) also investigate social network effects of boards. In an important paper Kogut and Walker (2001) examined the smallworld ownership structure of German ﬁrms in the mid-1990s. Speciﬁcally, the 500 largest non-ﬁnancial ﬁrms, and 50 ﬁnancial ﬁrms. The authors conclude that the ownership structure of German ﬁrms can be characterized as a small-world when compared with an Erdo+ s and Re´nyi (1959) Poisson random graph. In addition, their simulation results indicate the German small-world is robust to disruption. The properties of the small-world remain intact even when ownership ties are removed. In practice, however, the Poisson random graph is not a good description of social networks. Newman et al. (2001) generalize the random-graph model in a number of ways to better understand the formation of social networks. For example, they consider non-Poisson degree distributions and the representation of afﬁliation networks as bipartite graphs. As noted in Section 3 they derive exact expressions for the degree distribution, the average path length, and clustering coefﬁcient within a graph (among other results). They apply the theory to some real-world graphs including Fortune 1000 company directors. Their analysis shows remarkable agreement between actual data and the random-graph model for some statistics (the clustering coefﬁcient and the average number of directors on a board) but not others (the number of board interlocks).15 Davis et al. (2003) also examined the network structure of US corporate boards during the 1980s and 1990s. They investigated the small-world properties of directors and companies in each of the years 1982, 1990 and 1999. They ﬁnd the presence of a small-world among the corporate elite (short path lengths and high clustering) when compared with a Poisson random graph. They conclude that the aggregate connectivity of the social network of boards seems remarkably stable. Such connectivity is an intrinsic property of the network. Baum, Shiplov, and Rowley (2003) examined the small-world of Canadian capital markets, speciﬁcally investigating investment bank syndicates. Two banks are connected if they participated in an underwriting syndicate together. They document the presence and evolution of a small-world between 1956 and 1989. The Poisson distribution is used as the benchmark to compare calculated network measures. Baum, Van Liere, and Rowley (2006) further analyzed small-worlds in the context of UK investment banks. Robins and Alexander (2004) investigated the small-world of boards of directors using US and Australian data. They argue that analyzing boards

Ownership and Control

49

of directors as bipartite graphs is important. Conyon and Muldoon (2006) used techniques developed by Newman et al. (2001) to investigate the smallworld of corporate boards in the United States, Germany and the United Kingdom. They demonstrated two important ﬁndings. First, the network of boards and directors is, in fact, no smaller than would excepted by chance from a randomly assembled corporate universe. That is not to say that this is the ‘true’ mechanism by which the corporate world is constructed – only that the type of random-graph models described by Newman et al. (2001) are remarkably good at predicting the average clustering and path lengths in the network. Second, Conyon and Muldoon (2006) ﬁnd a tendency for directors who hold many directorships to sit on boards with other directors who also hold many directorships. This is consistent with a ‘social homophily’ effect described in the psychology literature (McPherson, Smith-Lovin, & Cook, 2001). Robins et al. (2005) make signiﬁcant advances in understanding small and other worlds by providing a simulation approach. They show how to simulate a distribution of Markov random graphs based on simple local process and how to examine the resulting global structure and compare it Bernoulli graphs. Like their paper, we test the predictions arising from the small-world model and contrast it to the random-graph alterative hypothesis using simulations from small-worlds too. In contrast to their chapter our analysis is based on Chung and Lu (2002a, 2002b) methods. It is also important to mention a class of research that investigates the formation and evolution of social networks by actors involved in a rational cost-beneﬁt calculus. (1) formulate a model where actors incur some costs of maintaining social ties and also beneﬁt from their location within the resulting dense network of ties. The cost and beneﬁt functions of the rational link-making actors are parameterized and the subsequent equilibrium structures analyzed. See also the related work of Dutta and Mutuswami (1997), Kim and Wong (2007) and Jackson and Watts (2002). This thematic research is also expanded in Doreian (2006) who provides an analytical model of the transitions between networks on the lattices of all graphs with a ﬁxed number of vertices through the addition and deletion of ties. See also Doreian (2008). Other theoretical research too has examined the formation of networks (see Jackson, 2007). Cowan and Jonard (2006) analyze the formation of innovation (alliance)networks using graph theory. In their model, creating an alliance increases the chances of innovation if the knowledge set of potential partners is complimentary. Since innovation changes the ﬁrms knowledge portfolio, so the network evolution changes. They show the structural properties of networks (degree and clustering)

50

MARTIN J. CONYON AND MARK R. MULDOON

arising from alliance partner choice which is driven by complementarities in ﬁrm’s knowledge portfolios. In other analytical work related to the themes of our paper Jackson and Rogers (2005) develop a model where small-worlds arise from the costs and beneﬁts of network formation. In their model the costs of maintaining a relationship between two actors (e.g., boards) depends on their proximity to each other. It is relatively easy for near acquaintances to maintain friendships and form strong bonds. These low costs help explain why one observes high local clustering in the network. It also explains the presence of short average path lengths. Consider a hypothetical situation where the network has many densely clustered groups, but the average path length is high because of the absence of links between the groups. Forming such a link across groups is, at ﬁrst sight, costly. However, the absence of links across groups means that making one such link can provide substantial beneﬁts because it opens up access to many other agents. Forming a single link can connect one agent to many other agents by reducing the distance between them and this is exactly what makes that link valuable. This type of research shows that understanding economic incentives leading to network formation is critical.

4. EMPIRICAL RESULTS 4.1. Data To examine the social network of ownership and control in Britain we used data supplied by Hemscott (http://www.hemscott.com).16 Hemscott data have been used in previous corporate governance research (e.g., Conyon & Muldoon, 2006). We treat the ownership and board networks as separate bipartite graphs. In the bipartite graph of ﬁrms and owners, the ﬁrm represents one set of vertices and the shareholders the other. In the bipartite graph of ﬁrm control, the board represents one set of vertices and the directors (i.e., the board members) the other – as illustrated in Fig. 1. The board data are snap-shot of all publicly traded ﬁrms listed on the London Stock Exchange in 2000. It consists of all executive and non-executive directors at UK public ﬁrms (other senior ofﬁcers are excluded). Likewise, the ownership data consists of ﬁrms and their owners, where there is at least an ownership stake exceeding 3%. We consider only non-board owners in year 2000. Some basic network statistics on the structure of ownership and control are contained in Table 3. The data set contains 2,055 separate ﬁrms and

51

Ownership and Control

Table 3. Descriptive Statistics – UK.

The board Board size Boards per director Ownership Shareholder per ﬁrm Firms per shareholder

N

Mean

p50

p25

p75

Min

Max

2,055 10,920

6.65 1.25

6.00 1.00

5.00 1.00

8.00 1.00

2.00 1.00

22.00 13.00

1,932 3,739

4.78 2.47

4.00 1.00

3.00 1.00

6.00 1.00

1.00 1.00

22.00 184.00

Source: Hemscott (2000).

10,920 unique board members (directors). In total there are 13,671 director seats – some directors are members of more than one board. The average board contains approximately 7 members and the average UK director is a member of 1.25 boards (including membership of his or her primary board).17 Only a minority of UK board members hold more than one directorship. Indeed, the director at the 75th percentile still has only one board position. The maximum number of directorships held by one person is 13. In the data set we can separately identify a ﬁrm and its set of non-board shareholders. The descriptive statistics are contained in Table 3. We were able to analyze 1,932 unique ﬁrms. These ﬁrms have 3,739 separate shareholders. A typical ﬁrm has, on average, about ﬁve different shareholders. Conversely, each shareholder owns shares in about 2.5 separate ﬁrms. The ﬁgure conceals a very strong skew in the data. There are some shareholders who own shares in many companies. For example, the shareholder at the 99th percentile holds shares in 31 separate ﬁrms. One shareholder (the maximum) owns at least a 3% share stake in 184 ﬁrms.

4.2. The Small-World of Ownership and Control Table 4 presents empirical evidence on the small-world of ownership and control among UK ﬁrms using the methodology of Newman et al. (2001). The hypothesis under investigation: can the real-world data be adequately represented by a random graph with a prescribed degree distribution. The lower-half of the table contains information on the ‘ownership’ network: the unipartite ﬁrm and owner projections. The upper-half of the table contains information on the ‘control’ network: the unipartite board and director

52

MARTIN J. CONYON AND MARK R. MULDOON

Table 4. The Small-World of UK Ownership and Control.

The board Board projection Director projection Ownership Firm projection Owner projection

N

LCC

D

2,055 10,920

1,592 8,323

5.686 9.013

1,932 3,739

1,650 3,029

120.83 12.703

oDW

L

oLW

CD

oCDW

6.106 9.090

5.455 6.275

4.078 4.772

0.376 0.597

0.327 0.544

136.181 15.717

2.302 3.226

1.499 2.053

0.519 0.164

0.392 0.041

Source: Hemscott (2000).

projections. In each case we report the number of vertices (N), the number of vertices in the largest connected component (LCC), the degree (D), the average path length (L), and the clustering coefﬁcient (CD). The brackets (oW) are an expectations operator of the large-graph limits described by Newman et al. (2001). The data show that the largest connected component is a sizable fraction of the available vertices in the network (between about 76% and 85%). Even though most directors only have one board position, the few that have more than one prove to be more than sufﬁcient to link the majority of corporations together. Similarly, the typical shareholder owns stock in a comparatively small number of ﬁrms. However, the few that own shares in more than one ﬁrm are enough to link together the majority of ﬁrms in the data set. In itself, the presence of such a ‘giant component’18 in the ownership and control network, is a remarkable ﬁnding. The ‘degree’ of a vertex provides one indication of the importance of that actor in the network. Consider the board network. Table 4 shows that the typical board is connected to about six others – where connections arise from a shared director. The expected value of the statistic is about six too. At face value it seems that the real data is in accord with the random-graph model. The board’s degree is no greater or less than would be expected by chance (but see below). Similarly, the typical director is connected to about nine others, which, of course, includes members of his or her own board. Again, there is strong agreement between the actual data and theoretical prediction. Consider next the ownership network. Table 4 shows that the typical ﬁrm is connected to about 100 and 20 others, where here connectivity arises from a shared owner. Even so, the actual degree is slightly less than what would be expected by chance given the observed degree sequences.

Ownership and Control

53

Can the network of ownership and control in the United Kingdom be characterized as a small-world? Section 3 illustrated that small-worlds are characterized by relatively short path lengths and high clustering. Consider the path lengths for each of the four projections in Table 4. We observe much agreement between the actual data and the expected values arising from the NSW method. If anything, the actual values are slightly greater than the expected. It would, therefore, be tempting to conclude that path lengths are really no different from what we would expect from a random graph with a known degree distribution. The NSW method prompts us to the conclude that the world of ownership and control is not ‘small’. There are two important observations to make. First, we would stress that the short path length outcome is still quite a remarkable ﬁnding. The mean geodesic in any of the projections is a very small fraction of the number of vertices in the largest connected component. For instance, the mean geodesic (i.e., shortest distance) for the board projection is about 5.5 whereas the number of vertices is 1,592 – or about 0.4% of the vertices in the largest connected component. Indeed, for the owner projection, the observed geodesic is 3.2 and the number of vertices in the connected component is 3,029 (or about 0.1%). Such short path lengths can act as important and powerful routes for the spread of business practices, knowledge and ideas or even rumors (Cowan & Jonard, 2004).19 Second, the procedure advocated by Newman et al. (2001) only permits us to compare the real-world data to the expected value of the large-graph limit. We do not have any conﬁdence bounds, for example, to compare to our calculated statistic. The Chung and Lu (2002a, 2002b) method, on the other hand, permits us to simulate many random corporate worlds with a prescribed degree sequence. The outcome of the simulations gives us a range of values on the small-world statistics which we can compare the actual values to. The NSW clustering coefﬁcient results contained in Table 4 show broad agreement between actual and expected values for the board of director graphs, but not the ownership networks. For the board projection CD takes a value of 0.376 which is similar to the expected value ( ¼ 0.327). The director projection reveals a qualitatively similar result. The NSW analysis leads us to conclude that boards are not particularly cliquish. In contrast, the clustering coefﬁcient in the ownership network is greater than expected suggesting that ownership ties are more ‘clubby’ than expected by chance. For instance, in the ﬁrm projection of the ownership network CD takes a value of 0.519 which is greater than expected ( ¼ 0.392). A more pronounced conclusion is observed for the ownership projection.

54

MARTIN J. CONYON AND MARK R. MULDOON

4.3. Chung–Lu Simulation Results The results so far pertain to actual small-world statistics compared to their expected values using the machinery introduced by Newman et al. (2001). As noted their predictions are large-graph limits for a certain ensemble of randomly assembled social networks. Unfortunately, the procedure does not permit us to say anything about the conﬁdence intervals within which actual small-world values lie. To determine whether the world of UK ownership and control really is ‘small’ we simulated a set of corporate universes with a prescribed expected degree sequence. We followed the methods introduced by Chung and Lu (2002a, 2002b); Chung et al. (2003). Again, the main hypothesis under investigation is whether the actual data can be adequately represented by a random graph with a prescribed degree distribution. The results of the simulations are plotted in Figs. 3 and 4. Consider Fig. 3 which relates to the board of directors (corporate control). The upper-row contains information from the board projection, whereas the lower-row pertains to the director projection. Three network statistics are presented: the average degree (D), the path length (L) and the clustering coefﬁcient (CD). In addition, the frequency distribution arising Small-world of UK boards

11.5 12 12.5 Degree

13

Density 20 40 0

Density 5 10 0

Density .5 1 0 11

3.35

3.4

3.45 3.5 Length

3.55

Fig. 3.

14.5 Degree

15

.22

Clustering: Director Actual=0.60 and NSW=0.55

Density 20 40

Density 5 10

0

0 14

.19 .2 .21 Clustering

60

2 Density 1 1.5 .5 0 13.5

.18

Path Length: Director Actual=6.3 and NSW=4.8 15

Degree: Director Actual=9.0 and NSW=9.1

Clustering: Board Actual=0.38 and NSW=0.33 60

Path Length: Board Actual=5.5 and NSW=4.1 15

1.5

Degree: Board Actual=5.7 and NSW=6.1

3.85

3.9

3.95 4 Length

4.05

.38

.39

.4 .41 Clustering

.42

UK Corporate Boards. The Small-World of the UK Corporate Elite. Chung–Lu Random Graphs: 500 Simulations.

55

Ownership and Control Small-world of UK ownership

130 140 Degree

150

Density 20 30 0 .46

Path Length: Owner Actual=3.2 and NSW=2.0

Clustering: Owner Actual=0.16 and NSW=0.04

19

20

21 22 Degree

23

.52

Density 50 100

.48 .5 Clustering

0

0

0

Density 10 20

Density .2 .4 .6 .8

1

30

2.12 2.14 2.16 2.18 2.2 2.22 Length

150

120

Degree: Owner Actual=12.7 and NSW=15.7

Fig. 4.

10

Density 10 20 0 110

Clustering: Firm Actual=0.52 and NSW=0.39 40

Path Length: Firm Actual=2.3 and NSW=1.5 30

Density 0 .02 .04 .06 .08 .1

Degree: Firm Actual=120.8 and NSW=136.2

2.62 2.64 2.66 2.68 2.7 2.72 Length

.13

.135 .14 .145 Clustering

.15

UK Ownership. The Small-World of the Corporate Ownership. Chung–Lu Random Graphs: 500 Simulations.

from the CL random-graph simulations are given. We also report the actual and NSW expected values of each statistic. Consider the board projection. The calculated values of the degree, path length and clustering measures each lie outside the range predicted by the Chung–Lu simulations. A MannWhitney test conﬁrmed this. For example, H0: pr (CL clustering ¼ actual clustering) yielded z ¼ 1.729 with ProbW|z| ¼ 0.0839. The results are qualitatively similar for the director projection. Once again the calculated actual degree, path length and clustering measures lie outside the range predicted by the Chung-Lu random-graph model. In both projections we ﬁnd that the actual path lengths are longer than expected by chance. However, we ﬁnd that clustering is signiﬁcantly greater than predicted by the Chung-Lu random-graph model. We deduce that the world of the corporate elite is ‘small’ in the sense that it is more ‘clubby’ or ‘cliquish’ than would be expected by chance. These empirical results suggest the presence of additional social structure in the network of board control that is not accounted for by the random-graph model. Fig. 4 relates to the small-world of UK company ownership. The upperrow contains information from the ﬁrm projection, whereas the lower-row

56

MARTIN J. CONYON AND MARK R. MULDOON

pertains to the owner projection. The format of Fig. 4 is the same as in Fig. 3. In the ﬁrm projection the average degree is not signiﬁcantly different from that expected in a Chung–Lu random graph with prescribed degree sequence. (Mann–Whitney H0: pr (CL degree ¼ actual degree) z ¼ 1.169 with ProbW|z| ¼ 0.2426.) This is not the case in the owner projection, where the actual degree is less that predicted by the Chung–Lu simulations. The small-world statistics are both very different from that predicted by the random-graph model. In both the ﬁrm and owner projections the path lengths (L) are signiﬁcantly longer than that predicted by the Chung–Lu random-graph model. In this sense the world is not ‘small’: path lengths are not shorter than expected by chance. On the other hand the clustering coefﬁcient (CD) is signiﬁcantly greater than expected by the random-graph model. Once again, we deduce that the world of corporate ownership is ‘small’ but only in the sense that it is more ‘clubby’ or ‘cliquish’ than would be expected by chance. Firms are connected to other ﬁrms by means of a common shareholder more frequently than one would expect by chance. Again, the data suggest the presence of additional social structure in the ownership network that is not accounted for by the random-graph model.

4.4. Financial Institutions and the Small-World Previous research has argued persuasively that ﬁnancial institutions such as banks are crucial to understanding the topology of the corporate board network. Sociological research, inspired by Brandeis (1914) and Mills (1956), has argued that bankers may be appointed to boards as a way for ﬁnancial institutions to exercise power and inﬂuence over non-ﬁnancial institutions. An alternative perspective is that both ﬁnancial and nonﬁnancial institutions receive cross-fertilization beneﬁts by adding directors of each type to their respective boards. For example, see the analysis of the formation of the inner circle by Useem (1984). Davis et al. (2003) concluded, ‘we can derive two main conjectures regarding the underpinnings of the elite network. On the one hand, several authors point to the central importance of banks in ordering, or even creating, the network. In this view, the fact that the corporate elite is well-connected results from the presence of particular institutions at the core, acting as a switchboard connecting disparate directors. In contrast, others emphasize the unplanned nature of the network: members of the corporate elite all seem to know one another simply as an unintended consequence of increasing economic concentration’. Davis et al. (2003) present empirical evidence from the US showing

57

Ownership and Control

that during the 1960s and 1980s the most highly connected organizations were indeed banks. However, they also note that bank centrality seems to have eroded at the turn of the century. To test the importance of ﬁnancial institutions in the corporate network we conducted a very simple experiment. First, we took the list of ﬁrms and directors in the board network described in sub-Section 4.1 (call these ‘All ﬁrms’). We then excluded banks, ﬁnancial institutions and investment trusts from our initial list of ﬁrms yielding a new network (call this new graph ‘Non-ﬁnancial institutions’). We then re-calculated the graph statistics presented in Section 4. If ﬁnancial institutions are like non-ﬁnancial institutions then their removal would have little effect on the network measures. Conversely, if ﬁnancial institutions are ‘at the core, acting as a switchboard connecting disparate directors’ then we might expect to observe salient differences between the original ‘All ﬁrms’ board graph and the one without ﬁnancial institutions.20 In Table 5 we present empirical evidence based on a comparison of calculated network statistics to Newman et al. (2001) expected values. The upper-half replicates the board network results in Table 4. The lower-half contains evidence based on the graph which excludes ﬁnancial institutions. Disrupting the original network, by removing ﬁnancial institutions, has a number of consequences. In the board projection the mean degree is smaller, the average path length is longer and the mean clustering coefﬁcient is smaller. In the director projection we ﬁnd the mean degree is slightly greater, the average path length is longer and the mean clustering coefﬁcient is greater. We next simulated a set of corporate universes with a prescribed expected degree sequence using Chung and Lu (2002a, 2002b); Chung et al. (2003)

Table 5.

All ﬁrms Board projection Director projection Non-ﬁnancials Board projection Director projection

The Small-world of Corporate Boards: Including and Excluding Financial Institutions. N

LCC

D

oDW

L

oLW

CD

oCDW

2,055 10,920

1,592 8,323

5.686 9.013

6.106 9.090

5.455 6.275

4.078 4.772

0.376 0.597

0.327 0.544

1,398 8,503

761 4,846

3.806 9.201

3.961 9.237

6.565 7.166

4.863 5.601

0.341 0.679

0.278 0.665

Source: Hemscott (2000).

58

MARTIN J. CONYON AND MARK R. MULDOON

methods. We did so for the network of all ﬁrms and then separately for the network of non-ﬁnancial institutions (namely, those ﬁrms in the original graph excluding banks, investment trusts, etc.). The principal hypothesis under investigation is that the network statistics arising from the simulations are the same for both graphs. If banks and ﬁnancial institutions are key to the network, we would expect to reject this hypothesis. The results of the simulations are plotted in Fig. 5. The maintained hypothesis is rejected: network topology is inﬂuenced by the presence (or absence) of ﬁnancial institutions. Consider the board projection. The network of non-ﬁnancial ﬁrms has a smaller mean degree, longer average path length and smaller mean clustering coefﬁcient. In the director projection, the non-ﬁnancial network of ﬁrms has a lower mean degree, longer average path length and greater mean clustering coefﬁcient. The differences in the simulated network statistics between the ‘All Firm’ graph and the ‘Non-Financial Firms’ network are all signiﬁcant based on a Mann–Whitney U test. For example in the board projection, H0: pr (Clustering in all ﬁrms ¼ clustering in nonﬁnancial ﬁrms) yielded z ¼ 27.274 with ProbW|z| ¼ 0.000. We conclude that Small-world of the UK corporate elite Path-Length: Board

Clustering: Board

0

5

0 .5 1 1.5 2

10 15

0 20 40 60 80

Degree: Board

9

10

11 12 Degree

13

3.4

3.5 3.6 Path-Length

3.7

.16

.18 Cluster

.2

.22

kdensity Degree

kdensity Length

kdensity Cluster

kdensity Degree_NF

kdensity Length_NF

kdensity Cluster_NF

Path-Length: Director 20 40 60

0 13.5

14 14.5 Degree

15

Clustering: Director

0

5

0 .5 1 1.5 2

10 15

Degree: Director

13

.14

3.85 3.9 3.95 4 4.05 4.1 Path-Length

.38

.4

.42 .44 Cluster

.46

.48

kdensity Degree

kdensity Length

kdensity Cluster

kdensity Degree_NF

kdensity Length_NF

kdensity Cluster_NF

Fig. 5. The UK Board Network. Chung–Lu Random Graphs: 500 Simulations. Analysis on all Firms and Non-ﬁnancial Institutions. In the Figures ‘NF’ refers to ‘Non-Financial Institutions’.

59

Ownership and Control

ﬁnancial institutions are important and give rise to different network topologies. Speciﬁcally, in the board graph, the exclusion of ﬁnance institutions leads to lower network connectivity (smaller degree), longer path lengths and networks that are less ‘clubby’ (lower clustering coefﬁcient). Our research complements recent research by Neuman, Davis, & Mizruchi (2008) who analyzed the relationships among bank mergers, changes in boards and their networks. They ﬁnd that ‘while the largest banks have become even larger through mergers, their boards have stayed roughly the same size with the same pattern of connections, leaving banks relatively less central in the intercorporate network. And while global banks previously had more globally oriented boards, this is no longer the case, as the link between board networks and strategy has become more tenuous’. Our approach may provide a complementary explanation for their ﬁndings. If directors on banks boards are coming from a greater variety of other boards, it may become more difﬁcult to see the link between board members and corporate strategy.

5. CONCLUSIONS In this chapter we have presented a social network analysis of the ownership control of ﬁrms. Our chapter contributes to an emergent literature on smallworld models (Newman, 2003; Jackson, 2005). We make the following contributions. First, we argue that a social network approach to understanding the ownership and control of ﬁrms is relevant and germane. Such an approach usefully complements the canonical principal-agent model. In such agency models the central concern is with designing optimal contracts between a single owner and a single ﬁrm. Agency models are not generally used to study the complex labyrinth of connections that exist between ﬁrms. This is largely the domain of social network analysis. We conjecture that such network links are useful to study because their presence may promote the rapid diffusion of knowledge and ideas through a connected corporate network (Cowan & Jonard, 2004; Schilling & Phelps, 2005). Second, we asked whether the network of ownership and control can be characterized as a ‘small-world’? Small-worlds exist when ‘path lengths’ are relatively short and ‘clustering’ is relatively high. However, one needs to address the associated question: how ‘small’ do we expect that world to be? Simply calculating actual small-world measures alone does not tell us much as we still need to compare real data to an appropriate benchmark. We gave an account of how to compare actual social networks of ownership and

60

MARTIN J. CONYON AND MARK R. MULDOON

control with an ensemble of random graphs introduced by Chung and Lu (2002a, 2002b); Chung et al. (2003). Third, we provided empirical evidence for a network of UK ﬁrms. We found that in general the random-graph model is not particularly well-suited to describing the ownership and control of UK ﬁrms. Our simulations suggested measurable and signiﬁcant differences between the actual data and theoretical predictions. This implied the presence of additional social structure in the network of ownership and control that is not accounted for by the random-graph model. Only in some cases did we ﬁnd that the random-graph model seemed to predict with great accuracy the behavior of the real world. The evidence suggests that the world of UK ownership and control is, indeed, ‘small’. It is small in the sense that the measured path lengths in both the ownership and board projections are small compared with the number of available vertices in the largest connected component. However, the statistical evidence suggested these path lengths are actually not smaller than would be expected in a randomly assembled corporate universe with the prescribed degree sequence. On the other hand we found that the ownership and control network is decidedly more ‘clubby’ or ‘cliquish’ than would be expected by chance. Generally, the clustering measures in both the ownership and board graphs were greater than predicted by the Chung and Lu (2002a, 2002b); Chung et al. (2003) random-graph model. Boards tended to be connected to other boards by a shared director more frequently than one would observe by chance. In addition, ﬁrms are connected to other ﬁrms by means of a common shareholder more frequently than one would expect. There is important social structure in the network that is yet to be accounted for. It is in this context that future analytical research is important. If the corporate world is not easily represented as a random graph, then future research needs to focus on alternative explanations of the observed network structure. A potentially fruitful avenue is the class of theoretical models recently analyzed by (4) and others (such as Jackson & Rogers, 2005; Jackson & Wolinsky, 1996; Dutta & Mutuswami, 1997; Kim & Wong, 2007; Jackson & Watts, 2002). Finally, we extended our empirical analysis to investigate the role of ﬁnancial institutions. Previous research has argued that banks are important conduits creating connectivity in corporate networks. We found that ﬁnancial institutions are important and give rise to different network topologies. Speciﬁcally, we found in the board graph that the exclusion of ﬁnance institutions leads to lower network connectivity (a smaller degree was observed), longer path lengths and networks that are less ‘clubby’ (there

61

Ownership and Control

was a lower clustering coefﬁcient). Our preliminary results suggest the need for further research to examine the sensitivity of network statistics to the inclusion or exclusion of different types of actors in the network. Overall, our early-stage research has contributed to the corporate governance literature by presenting a small-world analysis of ownership and control. We hope this provides a stimulus for future graph-theoretic analyses of corporate governance phenomena.

NOTES 1. Each of these ideas is explained fully in Section 3. For example, one can imagine a bipartite graph as a social network consisting of two distinct types of actors. Here, we discuss two kinds of bipartite graph: an ‘ownership’ network, which consists of ﬁrms and their shareholders, and a ‘control’ network, which consists of boards and their directors. 2. Anglo-Saxon corporate governance differs from, say, continental European governance arrangements where two-tier (dual) boards and concentrated block holders are more common. Franks, Mayer, and Rossi (2005) provide an account of the evolution of ownership in the United Kingdom. Useem (1984) provides a sociological account of the inner circle of boards and directors. 3. And the idea was further popularized in the famous play by John Guare ‘Six Degrees of Separation’, Guare (1990). 4. The principal–agent paradigm covers any hierarchical task-delegated relationship. For example, the relation between manager and worker, landlord and tenant, client ﬁrm and its advisor, and insurance company and a car-driver or a home owner and a building contractor, etc. 5. The literature has distinguished between moral hazard and adverse selection problems. For instance, a moral hazard occurs if the actions taken by the agent are hidden from the principal (e.g. the owner cannot observe perfectly the CEOs effort or his diligence in selecting projects). Adverse-selection occurs when the agent’s true type is hidden from the principal (e.g. the owner does not know the true skills embodied in the CEO). 6. Extensions to the model are possible, for example to a principal multi-agent framework. But even here the focus remains on the design of the optimal contract rather than the importance of connections in the network structure. 7. A pair of vertices may be connected by several paths that share the shortest length. 8. Speciﬁcally, path lengths (L) and clustering coefﬁcients (CD) for each projection separately. 9. The data refer to year 2000. 10. Of the 29 board members listed in table, 10 hold director appointments at other publicly listed ﬁrms. 11. The case of Standard Chartered is slightly anomalous in this respect. Tan Sri Khoo Teck Puat was a wealthy Singapore national, until his death in 2004. His wealth was derived from share ownership in British bank Standard Chartered, which he bought up in the 1980s.

62

MARTIN J. CONYON AND MARK R. MULDOON

12. See Wilf (1990) for a discussion of generating functions. 13. Formulae for higher moments of the distribution, as well for the generating functions for sums of independent samples from distribution are also simply related to GðxÞ. Generating functions arising from empirical degree distributions are ﬁnite polynomials. 14. The set of papers discussed in this section may be seen as part of a broader sociological research agenda that has focused on managerial elites and interlocking directorships. A review of this literature has been provided by Pettigrew (1992). Important contributions to this stream of research include Mizruchi and Bunting (1981), Useem and Karabel (1986), Useem (1984), Haunschild (1993), Hallock (1997) and Pennings (1980). 15. Uzzi and Spiro (2005) examined the small-world among collaboration and creativity of artists who produced Broadway musicals from 1945 to 1989 using Newman et al. (2001) methods. 16. Hemscott – Hemmington Scott – is a leading international provider of quality business data, including information about corporate governance, and investor relations services. 17. Prior UK research suggests boards are a bit larger than this, but our data include even the smallest of boards – a population typically excluded from previous studies. In our data board size is an increasing function of ﬁrm size measured by market capitalization. 18. The term originated in the work of Erdo¨s and Re´nyi, who proved that if the expected degree z exceeds 1.0 in a Poisson random graph then, with high probability, the largest connected component contains the majority of vertices. 19. For a Poisson random graph the mean geodesic can be expressed as l ¼ logðNÞ=logðzÞ. In the case of the board projection this is approximately l ¼ logð1; 592Þ=logð6:65Þ ¼ 3:324. The Poisson random graph, like the random graph with a constrained degree distribution, predicts a short mean geodesic relative to the number of vertices. 20. The analysis compares ‘Non-Financial’ ﬁrms to the larger set of ‘All Firms’. One might instead compare the ‘Non-Financial’ network to one (or a group) derived from the full ‘All Firms’ network but deleting randomly selected ﬁrms to make the new set of ‘All Firms’ network contain the same number of ﬁrms as the ‘NonFinancial’ network. We leave this possibility for future research.

REFERENCES Baum, J. A., Van Liere, D. W., & Rowley, T. (2006). Between closure and holes: Hybrid network positions and the performance of U.K. investment banks. SSRN eLibrary. Available at http://ssrn.com/paper ¼ 952717 Baum, J. A. C., Shiplov, A. V., & Rowley, T. J. (2003). Where do small worlds come from? Industrial and Corporate Change, 12, 697–725. Berle, A., & Means, G. (1932). The modern corporation and private property. Chicago, USA: Commerce Clearing House. Bolloba´s, Be´lla. (2001). Random graphs (2nd ed.). Cambridge: Cambridge University Press.

Ownership and Control

63

Brandeis, L. D. (1914). Other peoples money: And how the bankers use it. New York, USA: Frederick A. Stokes. Burt, R. (1992). Structural holes: The social structure of competition. Cambridge: Harvard University Press. Chung, F., & Lu, L. (2002a). The average distance in random graphs with given expected degrees. Proceedings of National Academy of Sciences U.S.A., 99(25), 15879–15882. Chung, F., & Lu, L. (2002b). Connected compnents in a random graph with given degree sequences. Annals of Combinatorics, 6, 125–145. Chung, F., Lu, L., & Vu, V. (2003). The spectra of random graphs with given expected degrees. Proceedings of National Academy of Sciences U.S.A., 100(11), 6313–6318. Cohen, L., Frazzini, A., & Malloy, C. J. (2007). The small world of investing: Board connections and mutual fund eeturns, SSRN eLibrary. Conyon, M. J., & Muldoon, M. R. (2006). The small world of corporate boards. Journal of Business Finance and Accounting, 33, 1321–1343. Cowan, R., & Jonard, N. (2004). Network structure and the diffusion of knowledge. Journal of Economic Dynamics and Control, 28, 1557–1575. Cowan, R., & Jonard, N. (2008). Knowledge portfolios and the organization of innovation networks. In: J. Baum & T. Rowley (Eds), Network strategy: Advances in strategic management. Oxford UK: JAI/Elsevier. Dagnino, G., Levanti, G., & Li Destri, A. (2008). Evolutionary dynamics of interﬁrm networks: A complex systems perspective. In: J. Baum & T. Rowley (Eds), Network strategy: Advances in strategic management. Oxford UK: JAI/Elsevier. Davis, G. F., Yoo, M., & Baker, W. E. (2003). The small world of the corporate elite. 1982–2001. Strategic Organization, 1(3), 301–326. Doreian, P. (2006). Actor network utilities and network evolution. Social Networks, 28(2), 137–164. Doreian, P. (2008). Network location, actor utilities and network evolution. In: J. Baum & T. Rowley (Eds), Network strategy: Advances in strategic management. Oxford UK: JAI/ Elsevier. Dutta, B., & Mutuswami, S. (1997). Stable networks. Journal of Economic Theory, 76(2), 322–344. Erdo+ s, P., & Re´nyi, A. (1959). On random graphs. Publ. Math. Debrecen,, 6, 290–297. Erdo+ s, P., & Re´nyi, A. (1960). On the evolution of random graphs. Publ. Math. Inst. Hung. Acad. Sci., Ser. A,, 5, 17–61. Fama, E. F., & Jensen, M. C. (1983). Separation of ownership and control. Journal of Law and Economics, 26, 301–325. Franks, J. R., C. Mayer, & S. Rossi (2005), Ownership: Evolution and regulation, ECGI Working Paper 2005.15, European Corporate Governance Institute. Available at SSRN:http://ssrn.com/abstract ¼ 354381 Guare, J. (1990). Six degrees of separation: A play. New York: Vintage Books. Hallock, K. F. (1997). Reciprocally interlocking board of directors and executive compensation. Journal of Financial and Quantitative Analysis, 32, 331–344. Haunschild, P. R. (1993). Interorganizational imitation – the impact of interlocks on corporate acquisition activity. Administrative Science Quarterly, 38, 564–592. Jackson, M. O. (2007), The economics of social networks, In: R. Blundell, W. Newey, & T. Persson (Eds), Proceedings of the 9th world Congress of the econometric society (pp. 1–56). Cambridge: Cambridge University Press.

64

MARTIN J. CONYON AND MARK R. MULDOON

Jackson, M., & Rogers, B. (2005). The economics of small worlds. Journal of the European Economic Association, 3, 617–627. Jackson, M. O., & Watts, A. (2002). The evolution of social and economic networks. Journal of Economic Theory, 106(2), 265–295. Jackson, M. O., & Wolinsky, A. (1996). A strategic model of social and economic networks. Journal of Economic Theory, 71(1), 44–74. Jensen, M. C. (1993). The modern industrial revolution, exit and the failure of internal control systems. Journal of Finance, 48(3), 831–880. Kim, C., & Wong, K.-C. (2007). Network formation and stable equilibrium. Journal of Economic Theory, 133(1), 536–549. Kogut, B., & Walker, G. (2001). The small world of Germany and the durability of national networks. American Sociological Review, 66, 317–335. Kramarz, F., & D. Thesmar (2006). Social networks in the boardroom. SSRN eLibrary. Available at http://ssrn.com/paper ¼ 878678 McPherson, M., Smith-Lovin, L., & Cook, J. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415–444. Milgram, S. (1967). The small-world problem. Psychology Today, 2, 60–67. Mills, C. W. (1956). The power elite. New York: Oxford. Mizruchi, M. S., & Bunting, D. (1981). Inﬂuence in corporate networks. Administrative Science Quarterly, 26, 475–489. Neuman, E., Davis, J., & Mizruchi, M. (2008). Networks and industry consolidation in U.S. global banking, 1986–2004. In: J. Baum & T. Rowley (Eds), Network strategy: Advances in strategic management. Oxford UK: JAI/Elsevier. Newman, M. E. J. (2003). The structure and function of complex networks. SIAM Review, 45, 167–256. Newman, M. E. J., Strogatz, S. H., & Watts, D. J. (2001). Random graphs with arbitrary degree distributions and their applications. Physical Review E, 64, 026118. Newman, M. E. J., Watts, D. J., & Strogatz, S. H. (2002). Random graph models for social networks. Proceedings of National Academy of Science U.S.A., 99, 2566–2572. Nguyen-Dang, B. (2008), Does the Rolodex matter? Corporate elite’s small world and the effectiveness of boards of directors. SSRN eLibrary. Available at http://ssrn.com/ abstract ¼ 864184 Pennings, J. M. (1980). Interlocking directorates. San Francisco, California: Jossey-Bass. Pettigrew, A. M. (1992). On studying managerial elites. Strategic Management Journal, 13, 163–182. Robins, G., & Alexander, M. (2004). Small worlds among interlocking directors: Network structure and distance in bipartite graphs. Computational and Mathematical Organization Theory, 10, 69–94. Robins, G., Woolcock, J., & Pattison, P. (2005). Small and other worlds: Global network structures from local processes. American Journal of Sociology, 110, 894–936. Rowley, T., & Baum, J. (2008). The dynamics of network strategies and positions. In: J. Baum & T. Rowley (Eds), Network strategy: Advances in strategic management. Oxford, UK: JAI/Elsevier. Schilling, M. A., & C. C. Phelps (2005). Interﬁrm collaboration networks: The impact of small world connectivity on ﬁrm innovation. SSRN eLibrary. Available at http://ssrn.com/ abstract ¼ 564422

Ownership and Control

65

Tirole, J. (2005). The theory of corporate ﬁnance. Princeton and Oxford: Princeton University Press. Useem, M. (1984). The inner circle. New York, Oxford: Oxford University Press. Useem, M., & Karabel, J. (1986). Pathways to top corporate management. American Sociological Review, 51, 184–200. Uzzi, B., & Spiro, J. (2005). Collaboration and creativity: The small world problem. American Journal of Sociology, 111(2), 447–504. Walker, G. (2008). The rise of ecommerce as an epidemic in the small world of venture capital. In: J. Baum & T. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 3–29). Bingley, UK: JAI/Emerald Group. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge: Cambridge University Press. Watts, D. J. (1999a). Networks, dynamics, and the small-world phenomenon. American Journal of Sociology, 105, 493–527. Watts, D. J. (1999b). Small worlds: The dynamics of networks between order and randomness. Princeton, NJ: Cambridge University Press, Princeton University Press. Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of ‘‘small-world’’ networks. Nature, 393, 440–442. Wilf, H. S. (1990). Generating functionology. San Diego, CA, USA: Academic Press Professional, Inc.

EVOLUTIONARY DYNAMICS OF INTER-FIRM NETWORKS: A COMPLEX SYSTEMS PERSPECTIVE Giovanni Battista Dagnino, Gabriella Levanti and Arabella Mocciaro Li Destri ABSTRACT This chapter aims to identify the main determinants that deﬁne the architectural properties of network emergence and signiﬁcantly inﬂuence the dynamics underlying network evolution in time. The identiﬁcation and analysis of these determinants, as well as the dynamic processes tied to them, allows to appreciate the competitive bases and consequences of network morphology. To this purpose, using a complex systems perspective as an integrative conceptual approach, we represent networks as complex dynamic systems of knowledge and capabilities. We perform a comparative in-depth analysis of the processes underlying the emergence and evolution of STMicroelectronic’s global network and of Toyota’s supplier network in the US so as to allow an elucidatory empirical assessment of the theoretical representation elaborated in the article.

Network Strategy Advances in Strategic Management, Volume 25, 67–129 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25003-5

67

68

GIOVANNI BATTISTA DAGNINO ET AL.

INTRODUCTION During roughly the same time period (from the end of the 1980s to the beginning of the new millennium), Toyota Motor Company and STMicroelectronics were both engaged in setting up a complex network of inter-ﬁrm relationships. In both cases, the creation and evolution of the network was aimed to enable the focal ﬁrm to acquire and sustain a dominant competitive position in its industry. Though the industries in which these ﬁrms compete are different, they are similar in that strong capabilities in technology and operations are the fundamental aspects on which ﬁrm success rests. The scrutiny of the processes underlying the genesis and dynamics of these networks shows that in both cases there are three signiﬁcant phases through which they emerge and evolve and, furthermore, that within each phase the contents of the evolutionary process are the same. Despite the fact that the two cases mentioned above show numerous strategic similarities and an extremely similar process underlying their emergence and evolution, if we compare the resulting networks it becomes apparent that the differences in the structural outcome of the two endeavors is striking. In particular, Toyota Motor Company’s network is characterized by strong relations between densely connected ﬁrms, a strong network identity and unity of vision. In this case, the evolutionary pathway of the network is largely deﬁned by all the ﬁrms belonging to the network itself. STMicroelectronics’s network, on the other hand, includes a wide variety of actors, most of whom maintain relationships with the focal ﬁrm but are scarcely connected to one another. This case shows the emergence of a loosely coupled network in which STMicroelectronics plays the principal role in determining the directions towards which the network will further develop. The short example above draws attention to the following question: how come networks that have emerged and developed during the same time period, that are aimed towards the same objectives and whose evolutionary processes are composed of the same phases, have matured such stark opposite morphologies? In order to be able to answer questions like the one above, the strategic management ﬁeld needs to enhance its appreciation of the evolutionary dynamics of inter-ﬁrm networks. In fact, albeit the great advancements made in the last decade, existing research on networks appears mainly static (Ahuja, Soda, & Zaheer, 2007) as it looks at network structure more than at network processes. As a result, it has an incomplete comprehension regards

Evolutionary Dynamics of Inter-Firm Networks

69

how networks evolve over time and the fact that the way such evolution occurs can be the source of relevant advantages and distinctive competitive positions both for the single ﬁrms belonging to the network and for the inter-ﬁrm system per se. This chapter aims to pinpoint the main determinants that deﬁne the architectural properties of network emergence and that signiﬁcantly inﬂuence the idiosyncratic dynamics underlying network evolution in time. The identiﬁcation and analysis of these determinants, as well as the dynamic processes tied to them, also allows to appreciate the competitive bases and consequences of network morphology. To the extent that the processes underlying network emergence and evolution may be systematically inﬂuenced by the intentional actions taken by pivotal ﬁrms and, furthermore, considering the competitive consequences tied to different network morphologies, it becomes of interest for ﬁrm executives to identify a limited number of variables which may be leveraged and managed in order to direct the evolution of the network they participate in towards a speciﬁc strategic aim and coherently with the requirements of the competitive domain in which they compete. In order to make such a contribution, we have structured the chapter in two main parts: the ﬁrst is dedicated to the elaboration of a theoretical framework which consents to represent the main determinants and dynamic processes underlying inter-ﬁrm network evolution. The second part is aimed to confront the theoretical framework elaborated with the longitudinal qualitative analysis of two case studies in order to assess its capacity to deliver a satisfactory explanation of the emergence and evolution of interﬁrm networks. More in detail, in the ﬁrst part of the chapter (i.e., the next two sections), we use the holistic and multilevel logic provided by the complex system perspective (Morin, 1977; Prigogine & Stengers, 1984; Anderson, 1999) to integrate and extend the hints of two relevant management approaches: the knowledge-based theory of the ﬁrm, which focuses on the individual ﬁrm and explains the growing importance of its internal idiosyncratic knowledge and capability base in the generation and renewal of the ﬁrm’s competitive advantage (Nonaka, Toyama, & Nagata, 2000; Nonaka & Toyama, 2002); and the strategic networks approach, which underscores the necessity to extend the boundaries of strategic investigation from the single ﬁrm to the network of relationships in which ﬁrms are embedded, given that the value-generating capability increasingly rests at the network level rather than at the individual ﬁrm level.

70

GIOVANNI BATTISTA DAGNINO ET AL.

The integration of the two conceptual approaches above brings us to sketch an interpretative analytical framework of inter-ﬁrm networks which (a) considers the strategic network as a distinct conceptual macro-category that, by embracing and interconnecting a variety of idiosyncratic ﬁrms, originates a complex dynamic system of knowledge and capabilities; (b) underscores that, within an inter-ﬁrm network, we can identify three relatively distinct but complementary and coexisting levels of analysis: (1) the micro-systemic level, related to the single ﬁrm in the network; (2) the meso-systemic level, related to the various groups of densely connected ﬁrms within the network. These ﬁrms maintain particularly intense – dyadic or multiple – relationships vis-a`-vis those held with the other ﬁrms that belong to the network; (3) the macro-systemic level, which concerns the network system as a whole, and the relationships between the latter and the environment in which it operates. Each of the levels is in turn characterized by an idiosyncratic language, a speciﬁc pool of knowledge and capabilities and a semi-autonomous pace of evolution; (c) emphasizes that the evolutionary pathway of the inter-ﬁrm network, taken equally as a whole and in its single parts, stem from the dynamic processes which occur at each one of these levels and in the interactions that inescapably arise (partly in an intentional fashion and partly spontaneously) among its three different levels. These complex dynamic interactions are able to generate a superadditive expansion of the (synchronic and diachronic) cognitive potential within the inter-ﬁrm network. Once we have turned out the skeleton of our interpretive analytical framework, in the second part of the chapter (i.e., the few successive sections), we apply it to scrutinize the emergence and the evolutionary dynamics of the two business cases brieﬂy mentioned above – i.e., Toyota Motor Company’s endeavor to replicate its Japanese network of ﬁrst and second tier suppliers in the USA and the emergence of STMicroelectronics’s complex network spread worldwide. By adopting a multiple-case comparative approach, we use the framework elaborated in the ﬁrst part as an operational template to determine how closely empirical observations concerning the two cases chosen match it. Finally, we apply replication logic to draw a few analytical generalizations from the particular set of empirical results obtained. The chapter ends with a section of ﬁnal conclusions and the illustration of a number of interesting aspects and implications which emerge from this study and need further development.

Evolutionary Dynamics of Inter-Firm Networks

71

REPRESENTING INTER-FIRM NETWORKS THROUGH THE INTEGRATION OF THE KNOWLEDGE-BASED THEORY OF THE FIRM AND THE STRATEGIC NETWORK PERSPECTIVE: TOWARDS A COMPLEX SYSTEMS VIEW Shifting the Focus of Strategic Analysis from the Single Firm to the Network The emergence of the resource-based view of the ﬁrm in strategy studies during the 1990s may be seen as the ﬁrst systematic and comprehensive attempt to scrutinize the ﬁrm level sources of competitive advantage. As a by-product, analyses conducted in this vein also laid the groundwork for the elaboration of what may be considered a kind of ‘theory of the ﬁrm’ that is speciﬁc to the strategy ﬁeld (Mocciaro Li Destri & Dagnino, 2005). Stemming from these studies, as of the mid-1990s, the strategy ﬁeld saw the development of what is known as the knowledge-based theory of the ﬁrm (KBT). The latter approach is characterized by an emphasis on the role of knowledge as a determinant of ﬁrm competitive performance and the elaboration of more dynamic views of the ﬁrm. Independently from the elaboration of what may be considered a ‘theory of the ﬁrm’, at the end of the 1990s and through the dawn of the new millennium, a stream of research in strategy literature began to draw attention to the necessity to extend the boundaries of strategic investigation to the network of relationships in which ﬁrms are embedded. This stream is known as the strategic network perspective (SNP) and focuses attention towards the possibility for networks to provide participating ﬁrms with access to valuable resources, capabilities and knowledge. In the attempt to elaborate an analytical framework able to represent and capture the main determinants and processes underlying network evolution and emergence, whilst embracing both ﬁrm- and network-based sources of competitive advantage, in the sections that follow these two approaches will be succinctly illustrated and the potential for their integration will be explored. The Knowledge-Based Theory of the Firm Focusing on the individual ﬁrm and its internal idiosyncratic knowledge and capability base, the KBT explains the burgeoning relevance of knowledge in the generation and renewal of the ﬁrm’s competitive advantage. The initial formulation of the KBT adopted an essentially static perspective as it

72

GIOVANNI BATTISTA DAGNINO ET AL.

conceptualized the ﬁrm as an organization that develops superior capabilities to protect (Porter-Liebeskind, 1996), integrate and apply the knowledge residing within single individuals (Grant, 1996a). More recently, the KBT has evolved toward a dynamic perspective since it views the ﬁrm as an entity which continuously creates knowledge (Nonaka, 1994; Nonaka et al., 2000; Nonaka et al., 2002). In this second KBT perspective, the knowledge which a ﬁrm possesses or controls and the capability to create and deploy such knowledge are the primary sources of a ﬁrm’s sustainable competitive advantage. The second KBT perspective claims that knowledge is a dynamic and relational phenomenon generated by the activation of different levels of social interaction, involving the interplay between two epistemological dimensions of knowledge (tacit and explicit). Through the dynamic interactions among the individuals belonging to the ﬁrm (and between these individuals and the environment) that occur in various shared contexts1 (Nonaka & Toyama, 2002), ﬁrms manage to mobilize and organizationally amplify the individual tacit knowledge bases and to crystallize such knowledge at higher ontological levels (such as the group level, the organizational level and the inter-organizational level). In order to be able to shift the focus from the single ﬁrm to the inter-ﬁrm network, it is necessary to integrate this body of literature with studies regarding the external environment in which the ﬁrm operates and, in particular, with studies that consider the social and economic ties which reciprocally connect ﬁrms on a non-spot basis.

The Strategic Networks Perspective Moving from the recognition that the value-generating capability increasingly rests at the network level rather than at the individual ﬁrm level, the SNP underscores the necessity to extend the boundaries of strategic investigation to the network of relationships in which ﬁrms are embedded (Gulati, 1999; McEvily & Zaheer, 1999, Gulati et al., 2000). Strategic networks are networks in which enduring inter-organizational ties are strategically important for the ﬁrms embedded in them. They potentially provide participating ﬁrms with access to valuable resources, capabilities and knowledge, that are not fully owned or controlled by their internal organizations (Lavie, 2006), and with advantages from learning, scale and scope economies. In order to consider the role of the external sources of competitive advantage, the SNP introduces the notion of network resources (Gulati, 1999). The latter are resources that emerge from the ﬁrms’

Evolutionary Dynamics of Inter-Firm Networks

73

participation to the network, as they are inherent to the network rather than to the single ﬁrm. In addition, the SNP underscores that the ability of a single ﬁrm to beneﬁt from network resources originates from the interaction of three components: its own endowment of unique resources, capabilities and knowledge; its network position; and the structure of the network itself (Gulati et al., 2000; Zaheer & Bell, 2005). In sum, studies in this vein contribute signiﬁcantly to our understanding of the social structure in which ﬁrms become entrenched once they begin to participate to a network, whilst they do not unveil the elements and processes which characterize single ﬁrm existence and performance.

The Integration of the Knowledge-Based Theory of the Firm and the Strategic Network Perspective Though succinct, the description of the two branches of research in strategy studies offered above is sufﬁcient to outline their main traits and underscore that they are potentially complementary in the explanation of the essence and dynamics underlying inter-ﬁrm networks. The ﬁrst stream of research allows a dynamic representation of the single ﬁrm and of the processes which drive its evolution, as well as an analysis of the sources of ﬁrm-based competitive advantages. The second body of studies focuses attention on the different ties ﬁrms may create in order to interact on a stable basis with other counterparts and on the economic and competitive advantages these ties may deliver. This perspective, therefore, allows to contextualize the ﬁrm within a web of social connections that form a structure through which information and knowledge ﬂow. From a methodological vantage point, the possibility to integrate these two perspectives in order to gain a more complete view of inter-ﬁrm networks rests in the consideration that neither one of them is deterministic; i.e. outlines necessary and sufﬁcient conditions for explaining inter-ﬁrm networks in all their relevant aspects (Baum & Dutton, 1998). This potential for integration is further sustained by the common emphasis accorded to socially embedded interaction between agents (see Table 1). The integration of these two bodies of research, leads to the elaboration of a theoretical framework of inter-ﬁrm networks in which (a) attention is focused towards the role of knowledge exploitation and exploration, as – coherently with the view underlying the KBT – these

74

Table 1.

GIOVANNI BATTISTA DAGNINO ET AL.

Knowledge-Based Theory of the Firm and Strategic Networks Perspective Compared. Knowledge-Based Theory of the Firm

Level of analysis Unit of analysis Focus Limitations

Main contributions taken into account Commonalities

Strategic Networks Perspective

The individual ﬁrm

The ﬁrm in the network of relationships Knowledge and capabilities Inter-ﬁrm relationships Network resources Sources of ﬁrm-based Sources of network-based competitive advantages competitive advantages Does not consider the external Does not buttress a real shift sources of competitive from the single ﬁrm to the advantages network level of analysis Nonaka (1994); Nonaka Gulati (1999); McEvily and et al.(2000) and Nonaka and Zaheer (1999) and Gulati, Toyama (2002) Nohria, and Zaheer (2000) None of them is fully deterministic Common emphasis accorded to socially embedded interactions among agents

processes are considered the primary forces underlying ﬁrm and interﬁrm evolution, performance and survival; (b) the attention paid to knowledge and its dynamics in the KBT also allows to underscore the social nature of knowledge transfer and creation, as well as the crucial role played by shared contexts of interaction to support knowledge-generating processes; (c) the SNP provides the concepts relative to the social ties and contexts which span ﬁrm boundaries, allowing to represent the ﬁrm within a social structure of inter-ﬁrm relationships which support and facilitate the transfer of knowledge and information. This view enables ﬁrst to pinpoint the essential forces underlying the emergence of inter-ﬁrm networks. In particular, as the environment in which the ﬁrm operates becomes more turbulent, the driving forces triggering the ﬁrm’s search for rents rest on a twofold strategy. On the one hand, the ﬁrm tries to appropriate as much value as possible from its existing set of knowledge and capabilities. On the other hand, it aims to create and renovate the sources of its competitive advantages by means of learning new ways of managing existing sets of knowledge and capabilities, developing new sets of knowledge and capabilities and matching changeable

Evolutionary Dynamics of Inter-Firm Networks

75

environmental conditions with valuable knowledge and capabilities. The unremitting tension to generate new knowledge and to amplify the value of the existing one forces the single ﬁrm to cooperate with other ﬁrms and, therefore, to join the network as the participation to an inter-ﬁrm network allows the ﬁrm to access superior economic and cognitive opportunities. The framework obtained from the integration of the KBT and the SNP offers an interpretation of the dynamics underlying, and the strategic relevance of, the cognitive aspects related to inter-ﬁrm ties and shared contexts. Though this seems a signiﬁcant advancement in the comprehension of inter-ﬁrm dynamics, it is necessary to underscore that the framework outlined up to now draws its view of networks directly from the SNP. We argue that the latter perspective presents limitations which hinder the full appreciation of the processes underling network dynamics and performance. Limitations of the Strategic Networks Perspective In order to clarify these limitations, it is necessary to render explicit the view of inter-ﬁrm networks which is entrenched in the SNP. In particular, though the SNP has signiﬁcantly contributed to shift attention towards the empirical and theoretical relevance of ﬁrm networks within management studies, it may be argued that this research stream does not buttress a real shift from the single ﬁrm to the network level of analysis. In fact, studies conducted within the SNP focus attention on the ﬁrm and analyze how ‘a ﬁrm’s networks allow it to access key resources (knowledge and capabilities) from its environment’ (Gulati et al., 2000, p. 207). By considering the ﬁrm and its social context, the SNP underscores that the social ties in which a ﬁrm is embedded, thanks to its participation to the network, allow it to access and leverage informational advantages. These advantages consent ﬁrms to expand the perception of opportunities and the access to complementary resources necessary to grasp such opportunities. This perspective, however, limits the analysis of networks to coupling a single ﬁrm to its external social factors, and therefore fails to capture the distinctive dynamic aspects which underlie the evolution of the network as itself a complex system of knowledge and capabilities. In this respect, the SNP actually underestimates the social advantage provided by the complex inter-ﬁrm endowment of knowledge and capabilities. The latter advantage may be appreciated only if networks are considered as entities which are fully accomplished wholes able to perform autonomous strategic choices. Since the network is seen as key to access external valuable resources, but not as an entire whole fully accomplished to perform strategic choices by

76

GIOVANNI BATTISTA DAGNINO ET AL.

itself, this means that the greater part of the network literature actually supports a ﬁrm’s perspective more than a network perspective. In order for the theoretical framework elaborated in this chapter to overcome the limitations tied to the ﬁrm centered perspective which permeates the SNP, we turn to the complex systems theory and use it as an interpretive lens. The complex system theory allows us to transcend the mere integration of the KBT and the SNP, and to expand these perspectives and concepts towards the analysis of both the network as a whole and of the different levels which deﬁne the fundamental structure of the network. We suggest that this further reinterpretation of the bodies of literature chosen as conceptual bases of the framework proposed shifts the focus from the single ﬁrm to the network and its multiple relevant levels of analysis. This step is indispensable in order to capture the various different cognitive processes and social structures which systematically inﬂuence network evolution and performance, extending the ability to embrace both the ﬁrmand the network-based determinants of competitive advantages. Accordingly, in the following sections we proceed to describe the main traits of the complex system theory and to illustrate the representation of the inter-ﬁrm network resulting from the adoption of this theory as an interpretative lens through which to re-elaborate the KBT and the SNP.

THE COMPLEX SYSTEMS THEORY AS A HOLISTIC AND MULTILEVEL LOGIC TO SCRUTINIZING PHENOMENA Building on the tenets of the general systems theory (von Bertalanffy, 1969), the complex systems theory (CST) focuses on the properties, the structures and the evolutionary patterns of complex systems that operate in dynamic and potentially discontinuous environments. Developed primarily in biology and physics (Prigogine & Stengers, 1984; Maturana & Varela, 1987; Waldrop, 1992; Kauffman, 1993), thanks to its pervasive and interdisciplinary framework, the CST is progressively gaining an intriguing role also in management and organization studies (McKelvey, 1997, 1999; Cohen, 1999; Anderson, 1999; Axelrod and Cohen, 2000). The transfer of concepts stemming from the CST to strategic management requires attention in order to avoid uncritical applications and possibly misleading interpretations. Nonetheless, the application of the CST to strategic management has a signiﬁcant potential in informing (and

Evolutionary Dynamics of Inter-Firm Networks

77

transforming) research in the ﬁeld as it provides researchers with a valuable set of insights and tools that embody a distinctive point of view and suggest new kinds of questions. In this chapter, we stress that the CST is able to increase our understanding of the evolutionary dynamics of inter-ﬁrm networks, or how inter-ﬁrm networks emerge and evolve over time. Conceiving the ﬁrm network as a complex and dynamic system made of a variety of ﬁrms that, interacting, give shape to the evolutionary pathway of the system as a whole, we consider the network as a distinct conceptual macro-category that extends the economic potential of the single ﬁrms of which it is composed (Dagnino, 1999, 2004). In addition, as the evolutionary pathway of the network is due to the dynamic interactions among the different ﬁrms which belong to it, we underscore that an in-depth comprehension of the process of network evolution relies on the simultaneous consideration of the different, but nonetheless coexisting and coevolving, levels of interaction that occur inside the network; i.e., the ﬁrm level, the inter-ﬁrm group level and the network level.

Distinctiveness of Complex Systems Theory as Relates to Inter-ﬁrm Networks Conceptualizing and analyzing inter-ﬁrm networks in the light of the CST, we take advantage of the following properties of complex systems (see Table 2). (a) Emergence. Some patterns and properties of the ﬁrm network result from the spontaneous interactions of the ﬁrms participating in it, rather than being inﬂuenced by intentional managerially coordinated or controlled behaviors. (b) Self organization. The inter-ﬁrm network exhibits self-organizing behaviors accomplishing an endogenous dynamic process thanks to which it spontaneously becomes increasingly organized. Accordingly, the network continuously shapes and reshapes itself, modiﬁes its boundaries, creates and recreates its set of knowledge and capabilities in connection with environmental dynamics. (c) Path dependence. The way a network behaves depends on the interaction between the stimuli it receives and the structural elements that deﬁne its nature and state in a given moment in time and space. It is noteworthy that the structural state of the network is the product of the

78

GIOVANNI BATTISTA DAGNINO ET AL.

Table 2. Distinctiveness of Complex Systems as Relates to Inter-Firm Networks. Concept Drawn from Complex System Theory Emergent properties Self organization Path dependence Organizational closure Thermodynamic openness Complexity Coevolution

Short Description

Some patterns and properties of a complex systems result from the spontaneous interactions of their components Starting in a random state, complex systems usually evolve towards order instead of disorder Historical- and path-dependent contingencies inﬂuence the state and the behaviors of complex systems The organization of the complex systems is unchanged and identiﬁes the system per se Complex systems import energy from the external environment Complex systems are made of different and intertwined complex subsystems at different interacting levels The adaptation of complex systems emerge from the adaptive effort of their components attempting to improve their ﬁtness functions

accumulation of knowledge and capabilities that has occurred in the past. Therefore, it synthesizes past behaviors of the network itself. This means that historical contingencies play a role in inﬂuencing the state and behaviors of the network. (d) Organizational closure and thermodynamic openness. The ﬁrm network is at the same time organizationally closed and thermodynamically open; namely, it is an autonomous system. Closure refers to the order that deﬁnes the organization of the network (i.e. the set of relationships that connect the various components) and allows to identify the network per se, regardless of the network’s speciﬁc structure in any given moment in time and space. Openness concerns the energy exchanges of the network with the external environment (in terms of resources, knowledge and capabilities). As a consequence, the stimuli stemming from environmental dynamics can selectively activate structural changes inside the network (i.e., its adaptation) in order to preserve its organizational closure and to secure its survival over time.2 (e) Complexity. The inter-ﬁrm network is a complex system made of a set of independent ﬁrms (which are complex subsystems per se) that are connected to one another by feedback loops in order to create an ‘organized complex unity’ (Morin, 1977, 2001), a sort of unitas

Evolutionary Dynamics of Inter-Firm Networks

79

multiplex. Complexity entails considering simultaneously two antagonistic but complementary notions: the multiplicity of the single ﬁrms in the network and the ‘oneness’ of the whole network accomplished through its organization. As a result, the network is simultaneously something more and something less than the sum of the participating ﬁrms. It is something more, because properties that otherwise would not exist emerge through the organization of the parts. It is something less, as the organization imposes conditions that partially inhibit the potentialities of the parts. (f) Coevolution. By striving to improve a ﬁtness function over time, each ﬁrm in the network continuously adapts to the other participating ﬁrms in the network and to the external environment (Anderson, 1999; Lewin & Volberda, 1999; Volberda & Lewin, 2003).

Summing up, we maintain that the continuous interactions taking place within the inter-ﬁrm network rest on both internal and external stimuli. The internal stimuli derive from the coevolutionary processes activated by the intertwined and coadapting ﬁrms of the network. The external stimuli originate from the coevolution between the network as a whole and the external environment. Thanks to the self-organizing capabilities and emergent properties that the network possesses, these continuous interactions generate the evolutionary dynamics of both the network as a whole and the single ﬁrms embedded in it.

Whole-Parts Interaction The analysis of the inter-ﬁrm network regarded as a complex and dynamic system is grounded in the coexistence and the dynamic interactions between the whole and the parts (Baum, 1999; McKelvey, 1999). As a consequence, the adoption of the CST to conceptualize and scrutinize inter-ﬁrm networks allows us not only to curtail the relevant antagonism between holism3 and reductionism,4 but also to assign a role to both the single parts and the whole in the emergence and evolution of the inter-ﬁrm network (Fontana & Ballati, 1999). Thus, it becomes possible to appreciate the new and different properties which emerge from the interactions of autonomous ﬁrms and crystallize at different levels within the network. At the same time, the variety and the autonomy of the ﬁrms participating in the network that cannot be inferred from the analysis of the network as a whole are not lost; on the contrary, they are underscored.

80

GIOVANNI BATTISTA DAGNINO ET AL.

Holistic Logic More in detail, the holistic logic that the CST provides us with makes it possible to unveil the synergies that a set of different and interrelated ﬁrms that cooperate together can achieve. Taking advantage of its holistic logic, the CST framework is able to shed light on the relevance of the interactions occurring among the ﬁrms belonging to the network. These interactions are of crucial importance since it is through them that the evolutionary pathway of the network emerges. Additionally, the interactions among the ﬁrms take place at multiple levels within the network. At all levels the evolutionary dynamics are signiﬁcantly inﬂuenced by the interactions which come about at the lower and the higher levels of the ﬁrm collections within the same network (McKelvey, 1997). This multilevel logic is capable of scrutinizing complex organizations and paves the way to examine simultaneously the coexisting and coevolving levels of interaction which happen inside the network and the consequent cross-level effects. The integration and extension of the KBT and the SNP through the lens of the CST allows us to sketch and advance an interpretative analytical framework of the inter-ﬁrm network that looks at it as a complex dynamic system of knowledge and capabilities. In order to explain the evolutionary dynamics of the network over time, the framework calls attention to the cross-level effects between the different network levels and emphasizes the unique and relatively autonomous nature of the ﬁrm network. Accordingly, we maintain that it is possible to identify the network by means of its idiosyncratic complex bundle of knowledge and capabilities and its functional specialization vis-a`-vis the external environment.

THE INTER-FIRM NETWORK AS A COMPLEX DYNAMIC SYSTEM OF KNOWLEDGE AND CAPABILITIES Integrating concepts drawn from the KBT and the SNP by means of the adoption of the CST, we sketch an interpretative analytical framework that considers the inter-ﬁrm network a distinct conceptual macro-category. Embracing and interconnecting a variety of idiosyncratic ﬁrms, this macrocategory originates a complex and dynamic system of knowledge and capabilities which is able to stretch the cognitive scope both of the network as a whole and of the participating ﬁrms (Dagnino, 1999, 2004).

Evolutionary Dynamics of Inter-Firm Networks

81

More in detail, within the network a set of specialized ﬁrms cooperate in order to jointly achieve more efﬁcient, effective and timely processes of knowledge exploitation and exploration. Cooperative activities occur in various contexts of interaction scattered within the network. These contexts of interaction are a reinterpretation in cognitive terms of the nested bundle of inter-ﬁrm relationships. They identify shared contexts that possess speciﬁc time and space properties. Within these shared contexts, groups of ﬁrms exchange valuable existing knowledge and capabilities and/or co-produce new knowledge and capabilities.5 These different contexts of interaction intermingle dynamically and are reciprocally interconnected so as to form a single overarching shared space that embraces all the shared contexts and therefore the network as a whole. Within these speciﬁc contexts of interaction, the various participating ﬁrms are able to develop a set of capabilities. More in detail, they share information, knowledge and capabilities, interpret information and transform it into knowledge and produce new knowledge through the creation of new meanings and new contexts. The latter activity paves the way to the emergence of a common language that is integrated and shared among all the interacting ﬁrms.

Multiple Analytical Levels of the Inter-Firm Network In order to pinpoint the different levels of analysis that are crucial to achieve an adequate representation and interpretation of network dynamics, we look at the connective structure of the inter-ﬁrm network. This structure is made of a nested bundle of inter-ﬁrm relationships. Such relationships may be dyadic or may extend beyond multiple units so as to encompass all the ﬁrms belonging to the network. Their intensity depends on the goals they pursue and the coordination mechanisms on which they rest. From a knowledge-based perspective, inter-ﬁrm relationships are relevant to the extent that they contribute to the cognitive processes which occur in the network. We are therefore looking for the different shared context of interaction and their distinctive characteristics relative to the way knowledge is transferred and created. Given this criterion, in the complex and dynamic system of knowledge and capabilities which constitute inter-ﬁrm networks it is possible to single out three levels of analysis (epitomized in Fig. 1). (a) The micro-systemic level, which identiﬁes the knowledge and capabilities related to the single ﬁrm belonging to the network.

82

GIOVANNI BATTISTA DAGNINO ET AL.

Macro-systemic level

Meso-systemic level

Micro-systemic level

Fig. 1.

The Three Analytical Levels of the Inter-Firm Network and Their Mutual Interactions.

(b) The meso-systemic level, which includes the knowledge and capabilities associated with dense inter-ﬁrm groups within the network. (c) The macro-systemic level, which embraces the knowledge and capabilities available to all the ﬁrms participating in the network. The adoption of the holistic and multilevel logic that the CST provides us with consents to shed light on the dynamics underlying the networks’ evolution. In particular, it is able to grasp, on the one hand, the speciﬁc role of each of the three analytical levels identiﬁed above and, on the other, the effect of the interaction among these three levels on the evolutionary pathways undertaken by the network. More in detail, at the micro-systemic level, each ﬁrm autonomously tries to extract rents from both amplifying the value of the existing set of knowledge and capabilities that it possesses or controls and generating new valuable knowledge and capabilities. In order to achieve this goal, the ﬁrm carries out idiosyncratic and speciﬁc processes of knowledge exploitation

Evolutionary Dynamics of Inter-Firm Networks

83

and exploration. These processes are driven by its own set of specialized cognitive assets but also leverage the potential of the set of knowledge and capabilities available at the meso-systemic and the macro-systemic levels. The capability of the ﬁrm to take advantage of the cognitive assets shared at higher analytical levels is deeply associated with the capabilities to pick up, absorb and integrate these assets with the internal ones (Cohen & Levinthal, 1990; Grant, 1996b). At the meso-systemic level, idiosyncratic ﬁrms confront and interact with each other in shared contexts that are characterized by speciﬁc and deﬁnite cognitive goals. By means of mutual confrontation and interaction, the ﬁrms connected by strong ties in dense groups jointly achieve processes of learning and co-generation of knowledge, that lead to the emergence of a shared and integrated knowledge base and a common language. Moreover, as the ﬁrms work together and reciprocally adapt over time, they turn out to be more mutually coherent and isomorphic. As a result, they tend to reach marked heuristic homogeneity and a harmonious speciﬁcity that allow to carry out smoother and quicker sharing and transfer of both tacit and explicit knowledge (Dagnino, 1999). Additionally, repeated interactions between these ﬁrms are able to develop mutual commitment and an atmosphere of trust that foster open-ended cooperation, sharing of valuable (tacit and explicit) knowledge and curb the risk of opportunistic behaviors (Coleman, 1988; Uzzi, 1997; Kale, Singh, & Perlmutter, 2000; Dyer & Hatch, 2006). Accordingly, the meso-level contexts display more rapid and direct confrontation and comprehension among participating ﬁrms, availability to openly share valuable knowledge as well as higher commitment and motivation to cooperate, in order to jointly carry out processes of learning and co-generation of knowledge. Over time, these conditions reinforce each other in a virtuous circle that overcomes the barriers to tacit and idiosyncratic knowledge transfer and, more in general, increases the efﬁciency, the efﬁcacy and the speed of knowledge transfer, sharing and co-production within the meso contexts of interaction (Capaldo, 2007). At the macro-systemic level, heterogeneous and specialized ﬁrms and dense groups of ﬁrms that jointly form the network are intertwined and interact dynamically so as to yield a single shared space (in this vein Dorejan, 2008, conveys ‘the global network view’). Within this shared space, the connections among the participating organizations are weaker vis-a`-vis those that take shape at the meso-systemic level (Burt, 1992; McEvily & Zaheer, 1999; Baum, Calabrese, & Silverman, 2000; Zaheer & Bell, 2005). Accordingly, over time the interactions occurring within the shared space drive to the emergence of a common knowledge base and a shared

84

GIOVANNI BATTISTA DAGNINO ET AL.

language that are more general in relation to the ones developed within the ﬁrm groups at the meso-systemic level. By means of the peculiarities that the shared space exhibits (i.e., nested web of weak ties between idiosyncratic organizations and general joint knowledge bases and languages), within this space it is possible to achieve processes of transfer and sharing of information and of explicit and general knowledge that entail lower costs and higher speed. Additionally, thanks to the weak connections and the confrontation of a plurality of heterogeneous and specialized ﬁrms and inter-ﬁrm groups, the macro shared space displays signiﬁcant levels of variety and variability of the knowledge and the capabilities that reside within it. Thanks to the conditions above, the sets of knowledge and capabilities within the network tend to match (diachronically and synchronically) with the changeable environmental conditions and, therefore, to preserve and increase over time the value of the complex system of cognitive assets in relation to the environmental dynamics. Network Multilevel Architecture The interpretative analytical framework sketched above allows us to underscore that the inter-ﬁrm network has an indubitable multilevel structure. This structure takes shape vis-a`-vis the speciﬁc cognitive necessities and the characteristics of the competitive and socio-institutional domain in which the ﬁrm operates. By underscoring this multilevel architecture, it is possible to show how inter-ﬁrm networks are able to create a very efﬁcient, effective and ﬂexible structure supporting cognitive processes of both knowledge deployment and creation. In particular, it is able to leverage the beneﬁts of the strong connections associated with the creation of speciﬁc shared contexts at the meso-systemic level. Within these contexts, the availability to openly share valuable knowledge, the mutual commitment and the motivation to cooperate promote a virtuous cycle that results in superior processes of tacit and explicit knowledge transfer as well as the co-production of new knowledge and capabilities. Strong Ties and the Risk Inward Looking Myopia Whilst the meso-level of dense interactions supports the co-generation of complex knowledge between ﬁrms, the macro-level of weak interactions overcomes the limitations and risks related to using only strong ties within inter-ﬁrm networks (Uzzi, 1997; Capaldo, 2007). More in detail, a number of authors in the network literature (Burt, 1992, 1994; Uzzi, 1996, 1997; Rowley, Behrens, & Krackhardt, 2000; Hagedoorn & Duysters, 2002; Nooteboom, 2004) have shed light on the tendency for strong ties to give

Evolutionary Dynamics of Inter-Firm Networks

85

way to processes of homophily, whereby ﬁrms who interact intensely tend to become reciprocally isomorphic, leading to a decrease in the variety of their knowledge endowments. This circumstance emphasizes the risk6 that ﬁrms belonging to the same dense group may mature an inward looking myopia which limits the ﬁrms’ ability to sense the emergence of opportunities or threats connected to changes in the external environment as well as its capacity to adapt to the changes perceived. Inter-ﬁrm networks curb the aforementioned weaknesses associated to the strong ties of the meso-systemic level, by balancing them with an intertwined web of weak connections among a variety of heterogeneous and specialized ﬁrms and groups of ﬁrms at the macro-systemic level. The contribution weak ties make to the knowledge generation process within ﬁrm networks rests both in their capacity to maintain and nurture ﬁrm heterogeneity and the differentiation of knowledge and competences within the network, as well as being a valid support for highly efﬁcient, effective and rapid processes of information sharing and brokerage and for the transfer of explicit and general knowledge between member ﬁrms. We argue that the adoption of the CST as an epistemological base to reinterpret and integrate the KBT and the SNP consents, not only to reveal the layered and changeable structure of inter-ﬁrm networks, but also to ﬁnd the rationale underlying the emergence and evolution of this structure and of the network itself. In particular, the network’s initial structure emerges from the sets of knowledge and capabilities that reside within it and from the cognitive goals which are ﬁxed at the outset. At each level, a semiindependent process of knowledge evolution occurs; the knowledge and capabilities created at each level are rendered available also to the other levels, sparking off further developments and contributing to their evolutionary process. These processes shape and endogenously stimulate further evolutions of the system and contribute to determine the effects of exogenous stimuli on the system too. Firm- and network-based determinants of competitive advantage coexist and interact within the different and interconnected analytical levels of the network. By means of the self-organizing capabilities and emergent properties that the network possesses, this interaction engenders the economic and cognitive performance of the network as a whole and of its parts, as well as the network potential to develop and evolve over time. Network Morphology and Multilevel Evolutionary Interactions The morphology of a network in any given moment time and point in space may be intended as the reﬂection of the cognitive characteristics of the

86

GIOVANNI BATTISTA DAGNINO ET AL.

domain in which the network operates. In particular, according to the cognitive characteristics of the domain in which the network is operating, we would suppose to ﬁnd that the role of the different levels changes in order to better ﬁt and support the speciﬁc necessities tied to the transfer of information and knowledge, the signaling of different knowledge bases and the knowledge creation processes which are crucial in that domain. Accordingly, it is worthwhile taking into account that the strategic interactions among the three analytical levels aim to implement top-down and bottom-up ﬂows of information and knowledge (see Fig. 2). The two kinds of ﬂows and interactions are characterized by different evolutionary intensity and frequency. The top-down ﬂows refer to the knowledge and information that are available to all the ﬁrms participating in the network (at the macro-systemic level) and to the ﬁrms establishing inter-ﬁrm groups (at the meso-systemic level). Their activation depends on the ﬁrm and group capabilities to absorb these pieces of knowledge and information and integrate them with the internal ones and on their motivations to engage in behaviors that are efﬁcient at the micro- and meso-systemic levels. These types of behaviors also spark mutual fertilizations and reciprocal stimuli that consent to increase the efﬁciency, the efﬁcacy and the rapidity of

Top-down flows of knowledge and information Meso-systemic level

Micro-systemic level

Macro-systemic level Micro-systemic level ***** Bottom-up flows of knowledge and information

Micro-systemic level

Meso-systemic level

Macro-systemic level

Macro-systemic level

Fig. 2.

The Flows of Knowledge and Information among the Three Analytical Network Levels.

Evolutionary Dynamics of Inter-Firm Networks

87

cognitive processes that are accomplished at higher levels of the inter-ﬁrm network. The establishment of bottom-up inter-level ﬂows of knowledge and information entails two inescapable crucial problems: (a) the availability of the single ﬁrms and groups to openly share their valuable knowledge with other units and groups; (b) the nature of the transfer process, that especially with reference to tacit knowledge requires the creation of speciﬁcally geared shared contexts (that are designed to be functional to predeﬁned learning purposes). As a result, the activation of these ﬂows is strongly connected to: (i) the accomplishment of regular knowledge-sharing activities; (ii) the capabilities of the single ﬁrms and groups to sense the opportunities associated with intra-network interactions; iii) their motivation to activate the potential intra-network connections. Accordingly, it is worth mentioning that the ﬁrms embedded in the network play a critical role when they act as strategic brokers that are actually able to sew together the different information, knowledge and capabilities that reside within the three network levels.

RESEARCH METHODS AND DATA In this chapter our aim is to increase the understanding of the main determinants that deﬁne the architectural properties of network emergence and signiﬁcantly inﬂuence the dynamics underlying network evolution in time. In order to achieve this goal, we have sketched an interpretative analytical framework of the inter-ﬁrm network as a complex dynamic system of knowledge and capabilities. We now apply this framework to scrutinize the emergence and evolutionary dynamics of two business cases: Toyota Motor Company and STMicroelectronics.

Theoretical Sampling The selection of the two cases under inspection relies on the basic principles of theoretical sampling (Pettigrew, 1990). Theoretical sampling suggests that the relevant cases are selected more than for statistical reasons on the basis of their relevance to our research questions and of their ability to replicate the analytical framework that has been developed (Glaser & Strauss, 1967, Mason, 1996).7 We maintain that the study of cases is an appropriate research strategy to help us understand the phenomenon under investigation

88

GIOVANNI BATTISTA DAGNINO ET AL.

for three main reasons. First, the nature of our research question requires a process theory explanation of the temporal order and sequences in which a discrete set of events leads to an observable outcome. Understanding how networks evolve over time and why they evolve in a given way entails the scrutiny of the temporal ordering and the patterns of interaction among the participating ﬁrms. Second, using a qualitative process approach we take into account the context where the phenomenon unfolds. This leads to the consideration of multiple levels of analysis that, at ﬁrst sight, are somewhat difﬁcult to separate from one another. Third, collecting process data from several sources of empirical evidence, we attempt to document as thoroughly as possible the sequence of events pertinent to the evolutionary dynamics of the networks analyzed (Yin, 1994; Langley, 1999). Summing up, the study of cases allow us to transcend the surface description of the network phenomenon so as to penetrate the inner logic behind its observed temporal unfolding and, therefore, to recognize the underlying generative mechanisms that drive the emergence and evolution of networks. In the ﬁrst part of this chapter, taking a deductive approach, we have adopted the CST in order to sketch an interpretative analytical framework of the inter-ﬁrm network in which it is possible to integrate and expand the theoretical contributions of the KBT and the SNP. We use this framework as an operational template to determine how closely empirical observations concerning the two selected cases match it, or to assess the extent to which the framework developed contributes to a satisfactory explanation of the emergence and the evolution of the ﬁrm networks studied. Thanks to the multiple-case comparative approach, we apply replication logic to draw a few analytical generalizations from the particular set of empirical results obtained from the study of the two selected cases.

Multiple Cases with and Embedded Design More in detail, we select sample cases from two different industry settings (i.e., the automotive industry and the semiconductor industry). This research strategy provides a better opportunity to detect possible differences and commonalities pertinent to the mechanisms underlying the emergence and the evolutionary dynamics of networks that operate in environments characterized by different innovation paces. Accordingly, if the ﬁndings of both cases match with the theoretical template, by using replication logic we

Evolutionary Dynamics of Inter-Firm Networks

89

will achieve a more general domain within which our results can be generalized and, in turn, strengthen the external validity of the developed analytical framework (Yin, 1994; Van de Ven & Poole, 2002). In addition, rather than on the whole global Toyota’s network of suppliers, our research strategy led us to focus attention on the network that connects Toyota to its ﬁrst-tier suppliers as speciﬁcally concerns the Georgetown plant, Kentucky (USA). Incidentally, in this respect the study of Toyota case differs from the ST case that entails the analysis of its entire global network. This option rest on the consideration that, as it reproduces the same functions, structures, mechanisms and goals, the US-based Toyota’s network is nothing else that a miniature model of the whole global Toyota’s network of suppliers. Thanks to this option we are able, on the one hand, to surmount the empirical difﬁculties related to: (a) the extensive lifetime of the overall Toyota’s network, which spans from 1937 to 2007, vis-a`-vis the ST one, which has originated in 1987; (b) the massive volume of data that a comprehensive scrutiny of the Toyota’s overall network would require in relation to its intricate conﬁguration architecture. The intricacies above could jeopardize the internal validity of this study. On the other hand, since the two networks at hand are typiﬁed by similar overall time span, the research choice we have taken secures the full comparability of the miniature model of Toyota’s network with ST’s network. Due to the stratiﬁed nature of inter-ﬁrm network architectures, we use multiple-case analysis with an embedded design. Each case study involves a multilevel structure that produces a scrutiny ranging over and across the three previously identiﬁed levels of analysis (i.e., the micro-systemic level, the meso-systemic level and the macro-systemic level). Coherently, we investigate the cross-level effects that occur within the network.

Temporal Bracketing In addition, we use a temporal bracketing strategy by decomposing in each case the time scale into successive periods. This type of temporal decomposition offers captivating opportunities for structuring process analysis. Speciﬁcally, it consents to carry out both within-case comparisons across subsequent periods and cross-case comparisons that sustain the internal and the external validity of the study (Eisenhardt, 1989; Langley, 1999).

90

GIOVANNI BATTISTA DAGNINO ET AL.

Sources of Data In the scrutiny of the two chosen cases, we employ multiple data collection methods in order to combine them via triangulation of evidence. This strategy enhances the construct validity of our applied ﬁeld research investigation. The development of converging lines of inquiry starting from a variety of sources of information allows more convincing and accurate ﬁndings (Eisenhardt, 1989; Yin, 1994). The collection of data includes: (a) documentary information, gathered from previous studies, research reports, books, scientiﬁc journals, business press articles, the Internet and others; (b) archival records, collected from the archives of the ﬁrms participating in the networks under investigation and (c) personal interviews, consisting in telephone and face-to-face interviews with key informers and executives. The information we have gathered refers to all the three levels of analysis of the inter-ﬁrm network. It is noteworthy that, in order to reduce the problems associated with managing a wide and unstructured data bulk, we have codiﬁed all the collected data and developed both an electronic and a paper database, allowing us to insure the reliability of our study.

THE STMICROELECTRONICS CASE STMicroelectronics N.V. (from now on ST) was created in 1987 from the merger of two European State-owned ﬁrms: the Italian ‘SGS Microelettronica’ and the French ‘Thomson Semiconducteurs’.8 Since its formation, ST pursued an aggressive growth strategy, investing heavily in R&D, establishing strategic alliances and industry partnerships, building up an integrated presence in the world’s major economic regions, and honing one of the world’s most efﬁcient manufacturing operations. This strategy has allowed ST to rapidly become a global leader in developing and delivering semiconductor solutions across the wide spectrum of microelectronics applications. Interestingly enough, crucial to the success of ST is the knowledge network that the ﬁrm has developed worldwide over time. This knowledge network consists of a web of strategic alliances and partnerships with key customers (that are generally world leaders in markets driven by the power of semiconductors), major suppliers and other semiconductor industry manufacturers, as well as joint R&D programs with leading universities and research centers throughout the world. ST’s network is aimed at creating a

Evolutionary Dynamics of Inter-Firm Networks

91

complex system of knowledge and capabilities that enables ST to offer cutting-edge solutions to customers in all segments of the electronics industry. In order to exemplify the emergence and the evolutionary dynamics of ST’s network and the rationale underlying its materialization and evolution over time, the temporal bracketing strategy we have chosen to pursue has driven us to split the investigation period into three temporal phases (i.e., 1987–1991; 1992–1999 and 2000–2006). Equally for motives of investigational parsimony and information enhancement, the strategy of decomposing the time scale into successive periods appears to be a particularly fertile one in this applied ﬁeld research investigation.

Phase I: 1987–1991 Starting its business in 1987, ST had to tackle two focal problems. First, the new management found economic and ﬁnancial difﬁculties that were inherited from previous managerial logics. Second, they found themselves located in the ‘wrong place’. Since they were headquartered in Italy and France, they felt geographically detached from the places in which key technologies and knowledge in the semiconductor industry were being developed, as well as from where the main ﬁrms using microelectronics applications were situated. The lack of technological and market knowledge in their home countries forced ST to become a global player. More in detail, they adopted a twofold strategy looking for the potential of knowledge trapped in pockets of local expertise scattered worldwide, and searching for customers in distant locations and in very different semiconductor applications. Additionally, as ST searched throughout the world for new clients, they dealt with the limitations associated with the use of standard semiconductors in satisfying a wide range of idiosyncratic customer needs. Accordingly, they developed awareness of the opportunity of neighboring the production of standard semiconductors with the creation of integrated and dedicated chips which could perform sets of functions needed for speciﬁc applications (the so-called System-on-Chip or SoC).9 In order to produce integrated system chips, ST needed to combine their technological and manufacturing knowledge and capabilities related to silicon with the knowledge and capabilities of the speciﬁc systems embedded within their customers. The need to integrate cognitive assets stemming from different sources, conducted ST to build strategic alliances with key customers in several industries. ST’s customer ﬁrms were generally global

92

GIOVANNI BATTISTA DAGNINO ET AL.

leaders in key industries or of particular applications, such as: Seagate (United States) and Western Digital (United States) for disk drivers; Nortel (Canada) for telecommunications; Bosch (Germany) for automotive electronics; Thomson Multimedia (France) for video applications and Pioneer (Japan) for consumer electronics. Aimed at co-developing application-speciﬁc products, these alliances provided ST with unique access to: (a) technological knowledge that was tacit and dependent on the speciﬁc context of the application it had to serve; and (b) market knowledge associated with the main microelectronics products. Later on, the knowledge and capabilities developed jointly with (or by means of) key customers located in different countries and specialized in a variety of applications were used to produce system chips for other customers with slightly different applications worldwide, to manufacture standard products and to develop next generation application-speciﬁc products. Additionally, these sources of knowledge were complemented by cooperating with leading universities (located in both the US and EU) and other major research institutions (such as CEA, IMEC and France Telecom R&D), as well as participating in European research programs (JESSY, 1989–1996). Summing up, in the course of Phase I (1987–1991), ST started to develop a set of emergent, quasi-spontaneous connections that were generally in isolation or unrelated to each other. These connections laid the foundation for the emergence of a virtuous circle between the creation of new knowledge and capabilities and the exploitation of existing knowledge and capabilities (Doz, Santos, & Williamson, 2001).

Phase II: 1992–1999 Leveraging the base of relational capabilities (Powell, Koput, & SmithDoerr, 1996; Lorenzoni & Lipparini, 1999; Kale et al., 2000) they had developed during the ﬁrst phase, ST was able to reinforce the existing strategic alliances with key customers and lengthen the set of learning collaborations with other leader ﬁrms (such as Alcatel, Gemplus, IBM, HP, ATI Technologies, Ford, and others). More in detail, each ST alliance was characterized by speciﬁc and clear cognitive goals and a precise schedule. In order to manage each connection, ST created a dedicated unit that was designed to be functional for predeﬁned speciﬁc learning purposes. Also, the ﬁrms cooperating with ST decided to shape a joint team that gathered at ﬁxed dates (generally every 6–9 weeks). During these regular meetings, the

Evolutionary Dynamics of Inter-Firm Networks

93

members of the joint team became accustomed to juxtapose and combine the results stemming from the work carried out by each ﬁrm and to assess in progress the achievement of the predetermined cognitive goals. Additionally, ST started establishing joint research and technological codevelopment projects with other manufacturers of the semiconductor industry; such as Philips NV (of the Netherlands) and Siemens (of Germany). The main driver of cooperation among competitors was the perception of a common challenge; i.e., the difﬁculty of the so-called ‘game against nature’ associated with the miniaturization of the chips and the integration on a single die of entire systems with much functionality. This challenge exceeded individual capabilities and required competing ﬁrms to collaborate. By means of collaboration, these ﬁrms were able to reduce the costs and share the risks associated with huge R&D investments, as well as to gain access to new technological and manufacturing knowledge and capabilities. During the second phase, ST continued and extended joint R&D programs with a variety of funding sources: (a) European research programs, such as JESSY (1989–1996), that later lead to MEDEA (1997–2001); (b) European technology platforms, such as ENIAC and ARTEMIS; (c) collaboration with leading universities (located in the US, Europe and Asia) and other major research institutions (such as CEA, IMEC and France Telecom R&D). Additionally, ST took part in international consortia and associations (such as the International 330 mm Initiative, International SEMATECH, International Technology Roadmaps for Semiconductors) aimed at collectively building technology frameworks capable of supporting the ﬁrms’ efforts to proceed along their technological trajectory. ST also formed joint development programs with key suppliers (such as Applied Materials, ASM Lithography, Canon, Hitachi and others) and leading Electronic Design Automation (EDA) tool producers (such as Cadence, CoWare and Synopsys) in order to jointly deﬁne the technical standards of the facilities and equipments that ST needed to manufacture their products. Finally, in 1998, ST signed a joint venture agreement with Shenzhen High Tech Industrial Company Ltd (SHIC) to build a back-end assembly and test plant in Shenzhen, China. Summing up, in the course of Phase II, a portfolio of strategic alliances and collaborations was built allowing ST and its partners to mutually achieve more efﬁcient, effective and timely processes of knowledge exploitation and exploration. Notwithstanding that, it is noteworthy that ST’s alliance portfolio was made of ﬁrms that usually did not interact with

94

GIOVANNI BATTISTA DAGNINO ET AL.

each other as the connections that were established among ST and their partners were not driven by a general coordination purpose of knowledge sharing. Consequently, at that time, ST’s embryonic network was not yet a complex system of knowledge and capabilities.

Phase III: 2000–2006 In the course of the third phase (2000–2006), the analysis of the ST portfolio of strategic alliances underscores the creation of an intertwined web of ﬁrms mutually interacting by means of a ﬁrst-class blend of both weak and strong ties. The interactions among ST and the ﬁrms participating in the ST’s network led to the emergence and clear deﬁnition of a complex system of knowledge and capabilities and to its evolution over time. More in detail, during Phase III ST persevered in their strategy of strengthening and widening the portfolio of strategic alliances and learning collaborations. These alliances mainly consisted of (a) projects of product co-development with key customers (Table 3); (b) programs of R&D with other semiconductor industry manufacturers and also main customers (Table 4); (c) projects of co-development of equipments and EDA tools with major suppliers10; (d) European research programs,11 European technology platforms12 and international consortia (Table 4) and (e) research programs with leading universities13 and research centers.14 Comparing Tables 3 and 4, which draw the evolutionary pathways of the different alliances throughout the phase under investigation, it is possible to detect that each alliance over time tends to be revised and extended in its learning scope in order to respond to the new challenges in the semiconductor industry. Additionally, over time ST and its partners established new learning collaborations that included ﬁrms participating in different pre-existing alliances. While ST and allied ﬁrms were working together in the context of the relationships they had established (i.e., at what we have called the ‘mesosystemic level’), they were developing a set of common knowledge and capabilities, reciprocal commitment and an atmosphere of mutual trust, as well as a web of weak connections at the macro-systemic level. This situation, on the one hand, allowed for the emergence of superior capabilities of (tacit and explicit) knowledge transfer and the co-generation

95

Evolutionary Dynamics of Inter-Firm Networks

Table 3. Main Clients of ST. Markets Communications

Computer peripherals

Automotive

Digital consumer

Industrial and multisegment

Source: ST documents.

Clients Agilent Alcatel – Lucent Ericsson Humax Kyocera Motorola Acer Agilent Technologies Creative Technology Delta HP/Compaq Bosh Conti Daimler Chrysler Delphi Denso Harman Agilet Technologies Bose Corporation Echostar LG Electronics Grundig Hugues Kenwood Matsushita American Power Conversion Astec Autostrade Delta Gemplus IBM Liton

Nokia Nortel Networks Philips Safran Siemens Thomson IBM Lexmark

Focus Applications Wireless Connectivity Mobile phone Portable multimedia Imaging Networking Data storage Printers

Samsung Seagate Maxtor Western Digital Hella Marelli Pioneer Sirius Valeo Visteon Olympus Philips Pioneer Samsung Scientiﬁc Atlanta Sony Thomson Vestel Magnetek Nagra Oberthur Philips Schlumberger Siemens Toppan

Powertrain Safety Car body Car multimedia

Set-top boxes High deﬁnition DVD Digital and HD TV Audio

Power supply Motor control Lighting Metering Smartcards

OMAPI standard

MIPI

NANOCMOS project

MINATEC IDEAs laboratory Crolles2 Alliances/CEA

Environmental 4

SMIA

2002

2003

2004

2004

2004

2004 ST, HDICf ST, Hynix

2004

ST, Philips, Inﬁneon Technologies, Freescale ST, Nokia

ST, Philips, Freescale, CEA

ST, Texas Instruments, ARM, Nokia (founding membersc) ST, Philips, Inﬁneon Technologies, CEA Leti,d IMECe ST, France Telecom, CEA Leti

2004

2004

Crolles2 Alliances

2002

ST, Philips, Motorola (Freescaleb), TSMC ST, Texas Instruments

ST, Philips, Inﬁneon Technologies ST, Hitachi ST, Philips, TSMCa

Environmental 4 Joint venture SuperH Crolles 2

2001 2001 2002

ST, Philips

Members

ST, Ovonyx

Crolles 2

Names

Research contract for development of 45 and 32 nm CMOS technologies Development standard for lead-free electronics packaging and promote a greener industry Speciﬁcation of Standard Mobile Imaging Architecture Joint venture for digital TV software (Shanghai, China) Joint venture to build a front-end memory manufacturing facility in Wuxi City (China)

Jointly building an advanced 12-inch (300 mm) wafer pilot fab Licensing and joint development program (Flash memory, MOS logic and other applications) Development standard for lead-free products Joint venture to develop RISC microprocessors Cooperate on process alignment for 90 nm CMOS generation and beyond Development CMOS technologies from 90 to 32 nm chip on 300 mm wafers Establishment of an open standard for wireless applications Deﬁnition of open standard for hardware and software interfaces in mobile terminals Project to propel Europe to limit of CMOS technology Creation of a joint multidisciplinary laboratory

Objectives

Alliances, R&D Programs and Consortia Initiated in the Course of ST’s Phase III.

2000

2000

Years

Table 4.

96 GIOVANNI BATTISTA DAGNINO ET AL.

ST, France Telecom ST, Freescale

2006

2006

Extending the scope of their joint semiconductor R&D activities to include R&D related to wafer testing and packaging Enhancing collaboration across libraries and SoC (LIPP) Consortium to determine a complete solution for overlay measurements for 45 nm generation USLI devices and below Development of a common memory subsystem to lower cost for phone makers Consortium for controlling leakage power in NanoCMOS SoCs Team to diagnose avian ﬂu using rapid detection point-of-need lab-on-chip R&D partnership on secure mobile platform and SIM card IC architectures Broad technology agreement for automotive application

Source: ST documents. a TSMC (Taiwan Semiconductor Manufacturing Company) b Freescale is a subsidiary of Motorola c MIPI is a non-proﬁt corporation that includes handset manufacturers, semiconductor companies, hardware peripheral manufacturers and operating system vendors d CEA (French Atomic Energy Commission), Leti (Laboratory of electronics and information technologies) is a laboratory of the CEA e IMEC is a Belgian researcher institute f HDIC (Shanghai High Deﬁnition Digital Innovation Ltd.) g CLEAN is a consortium of 14 European partners composed of semiconductor vendors, EDA vendors, and renowned academic and research centers

ST, Veredus Laboratories

2006

ST, othersg

CLEAN

ST, KLA – TENCOR, IMEC, QWED

2006

SOCOT

2005

ST, Philips, Freescale

ST, Intel

Crolles2 Alliances

2005

ST, Philips, Freescale

2005

Crolles2 Alliances

2005

Evolutionary Dynamics of Inter-Firm Networks 97

98

GIOVANNI BATTISTA DAGNINO ET AL.

of new knowledge within each alliance. On the other hand, it paved the way to the formation of a network-speciﬁc shared space (at what we have called the macro-systemic level) that consented to accomplish efﬁcient, effective and rapid information sharing and explicit and general knowledge transfer among the ﬁrms embedded in it. In addition, this smoother and faster transfer and sharing of information and knowledge supported the sparking of mutual fertilizations and stimuli within ST’s network and the identiﬁcation of the potentialities associated with establishing both new strong ties at the meso-systemic level and weak ties at the macro-systemic level. The activation of the potentialities above pursued to respond to the opportunities and/or the threats connected to the recurrent changes in the semiconductor industry, and drove the dynamic evolution of ST’s network. Summing up, it is possible to maintain that, during Phase III, among ST and its partners, an intertwined web of dense (at the meso-systemic level) and weak (at the macro-systemic level) ties and, therefore, a complex system of knowledge and capabilities emerged. The interactions occurring among the three different analytical levels within ST’s network generated the evolutionary pathways of the ST’s network as a whole, of ST’s strategic alliances and of ST and its allied ﬁrms taken in isolation.

Overview of ST’s Network Strategy The preceding analysis underscores that, in order to achieve and maintain the cutting-edge technological frontier in the semiconductor industry, ST needed to access, develop and integrate a wide set of different (tacit and explicit) knowledge and capabilities, as well as to continuously adapt these sets to the changeable environmental conditions. Accordingly, the aim of fulﬁlling its cognitive necessities by means of the accomplishment of effective and efﬁcient processes of knowledge transfer and co-production, has driven ST to shape different types and levels of ties with other organizations characterized by idiosyncratic and speciﬁc knowledge endowments (i.e., key customers, major suppliers, other semiconductor industry manufacturers, leading universities and research centers worldwide). Consequently, we are in the position to maintain that the nature of the cognitive domains underlying the semiconductor industry crucially affects the structure of ST’s network. It is noteworthy that the pattern of ST’s network structure depends, not only on its cognitive domains, but also on deliberate strategic choices taken by ST. On the one hand, ST have chosen not to limit the search for

99

Evolutionary Dynamics of Inter-Firm Networks

knowledge to locations in which the bulk of technological and market knowledge related to the semiconductor industry is notably concentrated (i.e., Silicon Valley and Japan), but looked for knowledge trapped in pockets of local expertise scattered throughout the world. On the other hand, it has decided to produce integrated and customized system chips, instead of standardized components. This option implied that ST had to mobilize and integrate idiosyncratic and specialized knowledge from a variety of different sources, spread across various sites worldwide. This twofold strategy differs from the strategies mainly implemented by the other major semiconductor industry manufactures (e.g., Intel) and, by providing state-of-art dedicated solutions to customers in all segments of the global electronic industry, it allowed ST to rapidly turn into one of the key leaders in the global semiconductor industry. Accordingly, ST have regularly managed to outperform the market since its inception. Commencing in 1987 through 2006, the sales of ST have grown at a compounded annual growth rate (from now on CAGR) of 14% compared to 11% of the semiconductor industry as a whole.15 In the annual rankings of the worldwide semiconductor companies,16 starting from the 14th rank in 1987, ST reached the 13th position at the end of the Phase I; the 9th position at the end of the Phase II and the 5th position at the end of the Phase III (see Table 5). More in detail, on the basis of provisional 2006 results published by iSupply, ST at the end of the period of time under scrutiny (the year 2006) were the number one global semiconductor supplier in industrial products, the number two in analog products and the number three in wireless, automotive electronics and NOR Flash. Furthermore, in the course of the third phase the revenues of ST arising from strategic partners have increased from 2,087 million US dollars in 1999 to 4,050 million US dollars in 2006, with a CAGR1999/2006 of 10%.

Table 5.

1987 1991 1999 2006

Economic Performance of STMicroelectronics.

Net Revenues (Million US $)

Rank in Total Semiconductor Marketa

851.0 1,347.0 5,056.3 9,854.0

14 13 9 5

Sources: Gartner Dataquest Corp., iSuppli Corp. a Ranking by revenues

100

GIOVANNI BATTISTA DAGNINO ET AL.

THE TOYOTA CASE Toyota Motor Company Ltd. (from now on Toyota) was established in 1937, when Kiichiro Toyoda adapted the knowledge stemming from a detailed ﬁeld study of Ford’s conveyor system to the small production volumes of the nascent Japanese automobile market. Since its founding, Toyota has been driven by a set of guiding principles aimed at ensuring the best quality and reliability of their products and the respect for people and the environment, on the one hand, and the reduction of the in-process inventory and the production of precise quantities of preordered items with a minimum waste, on the other hand. The lean production philosophy (the so-called ‘Toyota way’ – Monden, 1998; Fujimoto, 1999; Dyer, 2000) allowed Toyota to rise from the ashes of industrial upheaval in post-war Japan and turn into the world’s largest automobile manufacturer. Commencing its expansion into the American automotive market in the late 1950s, through the decades Toyota have steadily built a solid reputation for high customer service and satisfaction so that their sales ﬁgures rival today those of US domestic automakers and have in fact reached in 2006 the most prominent role in the world markets dethroning GM for the ﬁrst time in over 80 years. In 1984, thanks to a joint venture with GM (called NUMMI), Toyota embarked in the production of vehicles on the US soil. Afterward, in 1988 they decided to establish their ﬁrst fully owned manufacturing company in Georgetown, Kentucky, which was named ‘Toyota Motor Manufacturing Kentucky Inc.’ (from now on TMMK). Interestingly, the transfer of Toyota’s assembly plants from Japan to the US entailed the transplant, not only of the production system internal to Toyota’s organization, but also of the associated network of supplier relationship (Florida & Kenney, 1993; Adler, Fruin, & Liker, 1999). With the aim of scrutinizing the formation and the evolutionary pathway of the highly idiosyncratic network that tightly connects Toyota to their ﬁrst-tier suppliers in the Kentucky plant, we have decided to analyze a period of time which covers some 12 years. The period starts with the establishment of the ﬁrst Toyota plant in the US in 1988 and ends in 2000, when Toyota’s network eventually reached its maturity stage. In addition, consistent with the temporal bracketing research strategy, we have decomposed the period under investigation in three relevant temporal phases (i.e., 1988–1991; 1992–1993 and 1994–2000).

Evolutionary Dynamics of Inter-Firm Networks

101

Phase I: 1988–1991 When in 1988 Toyota built its ﬁrst US plant in Georgetown (Kentucky), in order to effectively implement its production system, it needed to ﬁnd ﬁrsttier suppliers able to match the peculiarities that epitomized its highly idiosyncratic production system. More in detail, the Toyota Production System (from now on TPS) was based on two focal concepts. The ﬁrst was called ‘jidoka’ (that could be loosely translated as ‘automation with human touch’), which meant that, when a problem occurred, the equipment stopped immediately, preventing defective products from being produced. The second was the idea of ‘just-intime’, which meant that each process produced only what was needed by the next process in an unremitting ﬂow. The simultaneous use of the above concepts allowed Toyota to manufacture the products required by its customers in the precise quantities desired at a given point in time. Moreover, the search for continuous improvement (the so-called kaizen) ensured high-quality products and services that were able to meet a wide variety of customer demands around the world (Cusumano, 1985). Achieving the pull logic behind the TPS entailed choosing suppliers that shared Toyota’s basic idea of combining ﬂexibility and high quality. Precisely, the lack of synchronization between the ﬂows of components from the suppliers and the requests originating from the manufacturing process, as well as the presence of defective components, caused the stop of the production line. Accordingly, Toyota evaluated the aspiring ﬁrst-tier suppliers on the basis of their availability to take part in long-term supply relationships and to carry out a process of joint growth and continuous learning and improvement based on reciprocal commitment and mutual trust (MacDufﬁe & Helper, 1997). Seemingly, Toyota’s way of conceptualizing and managing the relationships with its suppliers differed markedly from that of the established US automobile makers, which were traditionally based on price and arm’s length relationships. Notwithstanding this situation, Toyota did not take on the characteristics of the American supplier relationships. Rather it created a radically different system typiﬁed by a nested network of connections with its suppliers, consistent with its philosophy of continuous learning and relentless improvement (Florida & Kenney, 1993). To start with, Toyota established with its suppliers a set of bilateral longterm ties with annual price reviews and supported them to adopt the lean production technique bestowing initial high prices. Later on, Toyota

102

GIOVANNI BATTISTA DAGNINO ET AL.

attempted to replicate the routines and the learning processes that had been successful in creating an efﬁcient and effective knowledge-sharing network in Japan. In 1989, Toyota set up a supplier association between its ﬁrst-tier suppliers, the so-called Bluegrass Automotive Manufacturers Association (BAMA). BAMA was directed to share information and explicit knowledge on production techniques and to promote socialization among the ﬁrms participating in it (Dyer & Nobeoka, 2000). The supplier association tried to achieve its goals by means of (a) general meetings, in which, every other month, the suppliers and Toyota gathered and shared information and explicit and general knowledge on the TPS and on market trends; (b) committees, speciﬁcally designed to facilitate the transfer of explicit knowledge on critical topics for all the ﬁrms in the network, such as inventory cost reduction, quality improvement and safety. Regular committee meetings convened six times each year. Additionally, within the committees training programs and visits to the best practice plants, as well as an annual conference were organized. Summing up, in the course of Phase I, starting from a set of bilateral relationships, Toyota created a number of spaces of socialization. These spaces were shared by all the suppliers and allowed Toyota to efﬁciently and effectively transfer information and explicit knowledge associated with the implementation of its production system to all the ﬁrms evolved. This situation drove to the creation of a base of shared explicit knowledge among Toyota’s nest of suppliers. Moreover, the suppliers participating in BAMA and its activities began to reciprocally interact and, therefore, a connective structure of weak ties began to emerge.

Phase II: 1992–1993 In order to foster the sharing of both tacit and explicit knowledge, in 1992 Toyota established the Toyota Supplier Support Center (TSSC). TSSC was an organizational unit meant to help solve operational problems both at Toyota’s and at suppliers’ plants. More in detail, TSSC facilitated the transfer of tacit knowledge that resided in Toyota and improved supplier productivity and quality by providing them direct on-site assistance through consulting teams. These consulting teams were built up of experienced Toyota personnel with in-depth knowledge of the principles and practices of

Evolutionary Dynamics of Inter-Firm Networks

103

the TPS. The durability of the teams at the supplier plant depended on the nature of the problem they had to face and on the learning capabilities of the relevant supplier (Dyer & Hatch, 2006). It is worth mentioning that TSSC assistance was available to suppliers free of charge, but it entailed the acceptance of a strict norm of reciprocal knowledge sharing. In order to take advantage of the specialized assistance, each supplier had to agree to open up its plants to the other Toyota suppliers. Additionally, as Toyota provided free assistance, training and instruction for suppliers, through the exploitation of which these supplier ﬁrms were able to increase their productivity and quality, over time a feeling of indebtedness and trust towards Toyota emerged within the suppliers. In addition, a collection of tacit and explicit knowledge and capabilities on the TPS that were common to Toyota and its suppliers was developed. This situation led to the establishment of strong and intense bilateral connections between Toyota and its suppliers. Moreover, the latter developed awareness of the opportunities associated with the sharing of explicit and tacit knowledge. Summing up, in the course of Phase II, a set of strong dyadic ties among Toyota and its suppliers forcefully emerged. These ties allowed smoother one-way tacit knowledge transfer from Toyota to its suppliers and paved the way to the reciprocal sharing of knowledge. This set of strong connections was complemented by weak ties that joined the suppliers by means of BAMA and its activities. Nonetheless, the suppliers did not start interacting among themselves in order to reciprocally exchange knowledge under the guidance of a general coordination purpose. Consequently, at the end of this phase, Toyota had not yet generated a complex network of knowledge and capabilities.

Phase III: 1994–2000 During the third phase (1994–2000), the analysis of Toyota’s collection of ﬁrst-tier suppliers permits to observe the emergence of an intertwined web of ﬁrms that cooperate by sharing existing knowledge and capabilities as well as generating new knowledge and capabilities. These cooperative activities drove to the formation of a complex system of knowledge and capabilities and fueled its evolution over time. More in detail, in order to foster the reciprocal sharing of tacit and of explicit knowledge among its suppliers and the co-production of new knowledge, in 1994 Toyota created a web of learning teams, labeled Plant

104

GIOVANNI BATTISTA DAGNINO ET AL.

Development Activity (PDA) core groups. These teams were composed of different and specialized suppliers that shared similar sets of knowledge and capabilities regard the TPS. The creation of a web of learning groups was aimed to support the suppliers which had decided to take part in the program with productivity and quality improvements by means of the activation of speciﬁc learning contexts. In accordance with the TSSC, every year they chose a critical theme (associated to the implementation of the TPS) to scrutinize. The in-depth analysis of the predeﬁned topic entailed carrying out a succession of visits to the supplier plants in order to jointly develop proposals for upgrading the problem under investigation. At the end of each year, all the learning groups used to gather at their annual conference to share what they had learned by performing the joint activities. Over time, the interactions occurring within the learning groups led to the creation among the participating ﬁrms of a common set of tacit and explicit knowledge and capabilities, the evolution of a shared language, as well as to the establishment of sentiments of reciprocal commitment and mutual trust. This situation laid the foundations that allowed achieving superior capabilities of (tacit and explicit) knowledge transfer and co-production of new speciﬁc knowledge. In addition, the learning teams were usually being reorganized every three years. This allowed Toyota to generate variety and variability of the knowledge and capabilities that resided within these units. Accordingly, the rotation of the ﬁrms belonging to the learning teams was able to overcome the risks associated with the emergence within these groups of myopias that limited the ﬁrms’ capability to sense the emergence of opportunities or threats connected to changes in the external environment and to react to the changes perceived. Given the above visualization of Toyota’s network of ﬁrst-tier suppliers, it is possible to maintain that the interactions of the different participating ﬁrms over time drove to the emergence of a complex network of knowledge and capabilities. More in detail, the interactions occurred via learning processes activated at both the meso-systemic and the macro-systemic levels. At the meso-systemic level, they developed within the learning teams. At the macro-systemic level, the interconnections occurred within the general BAMA meetings, the supplier association committees, the consulting teams and the annual conference of the learning teams. The intertwined web of weak and strong ties that joined Toyota and its suppliers, allowed the creation of a strong network identity (Dyer & Nobeoka, 2000) among the different ﬁrms and the incentive to openly share valuable knowledge within the different analytical levels of the network.

Evolutionary Dynamics of Inter-Firm Networks

105

Overview of Toyota’s Network Strategy Accordingly, Toyota’s complex network of suppliers achieved superior processes of exploitation of the knowledge and capabilities relative to the TPS and of the production knowledge and capabilities of the participating suppliers. Similarly, it permitted to carry out efﬁcient, effective and timely processes of creation of new technical knowledge at the mesosystemic level, in order to continuously improve the productivity and the quality of the network system in accordance with the challenges of the automotive industry. Subsequently, the knowledge generated at the meso-systemic level was shared at the macro-systemic levels sparking mutual fertilization among all the ﬁrms embedded in the network. The unremitting tension towards learning and improvement drove the evolutionary dynamics of Toyota’s network as a whole, of Toyota themselves and of its ﬁrst-tier suppliers. The in-depth analysis performed heretofore underscores that the aim of effectively and efﬁciently exploiting and exploring the knowledge associated with the implementation and the management of the TPS, drove Toyota to shape different types of ties between the ﬁrm and its suppliers, as well as the set of interaction mechanisms and spaces. Accordingly, we are in the position to afﬁrm that the nature of the cognitive domains characterizing the auto industry, in general, and the Toyota production system, in particular, strongly affects the structure of Toyota’s network of suppliers. Interestingly enough, the conﬁguration of the network structure depends not only on the cognitive domains, but also on deliberate choices of Toyota. More in detail, the study of Toyota’s transplant to the Kentucky site clearly indicates that Toyota have chosen to not conform to the prevailing US organizational model and the practices associated with supplier relationships. Quite the opposite, they have acted on the environment to create the resources and conditions required to replicate the routines and learning processes that had been successful in creating an efﬁcient and effective knowledge-sharing network in Japan.17 In order to creatively respond to the deﬁciencies of the US environment as regards the delivery and quality requirements of the Japan-like just-in-time system, in the early days of the Georgetown plant Toyota encouraged its ﬁrst-tier Japanese suppliers to locate in the US, by ﬁnancing and helping them to set up US branches. Afterwards, however Toyota turned its attention to US-owned suppliers and they worked intensely with them to implant the TPS fully and to accelerate the diffusion of the

106

GIOVANNI BATTISTA DAGNINO ET AL.

relevant knowledge. Accordingly, we maintain that the success of the Toyota’s transplant was neither natural nor automatic; it hinged on the strategic actions that Toyota have taken to transform existing patterns of inter-ﬁrm relationships in the US in light of its cognitive and functional necessities. It is valuable observing how the creation and evolution of the network that connects Toyota and its ﬁrst-tier suppliers in Kentucky’s Georgetown plant affected the economic performance of TMMK. During the period under scrutiny, the production of the Georgetown factory (see Table 6) has consisted of two models of sedan, i.e. Camry (since 1988) and Avalon (starting in 1994), a model of minivan (i.e., Sierra minivan, starting in 1997) and a variety of power trains (i.e., Axle, 4-cylinder engines, V6 engine starting in 1989). Should we consider that the two models of sedan, i.e. Camry and Avalon, have produced merely in the Georgetown plant, we can maintain that their sales in the US market portray the economic performance of TMMK. From 1988 to 2000, the sales of Camry in the US have grown at a CAGR of 6.95%. More in detail, the CAGR of the Camry’s sales in relation to each of the three phases has been: 9.0% in the ﬁrst phase; 4.57% in the second phase and 6.48% in the third phase (see Table 7). Interestingly, in the course of the ﬁnal period examined, Camry has managed to be the ﬁrst best sold sedan in the US for four years in a row (1997–2000). The production of the Avalon model has taken place only during Phase III of our analysis and its sales have increased from 1994 to 2000 at a CAGR of 48.28% (in Table 7).

Table 6. Production of Toyota Motor Manufacturing, Kentucky, Inc. (TMMK).

1988 1991 1993 2000

TMMK Vehicle Production (Units Per Year)

TMMK Power Train Production (Units Per Year)

18,527 240,242 284,599 495,429

–a 153,811 206,382 479,405

Sources: Toyota. a Power train production starts in 1989 (181 units)

107

Evolutionary Dynamics of Inter-Firm Networks

Table 7.

Economic Performance of Toyota Motor Manufacturing, Kentucky, Inc. (TMMK). Camry Sales in the US (Units Per Year)

1988 1991 1993 2000

187,000 264,000 299,700 448,162 Avalon Sales in the US (Units Per Year)

1994 2000

6,603 104,078

Sources: Toyota.

DISCUSSION AND CONCLUSION In the ﬁnal section, we use the framework entrenched in the complex systems analysis developed in the ﬁrst part of the chapter as a supporting operational template to determine how closely empirical observations concerning the two cases selected match it; i.e., to assess the extent to which the framework developed contributes to a suitable explanation of the emergence and the evolution of the two inter-ﬁrm networks under scrutiny. Using a multiple-case comparative approach, we apply replication logic to draw a few analytical generalizations from the particular set of empirical results obtained from the scrutiny of the cases under investigation. We eventually gather a few implications of the analysis performed for strategy theory and managerial practice. Our path of investigation is organized as follows. First, we will examine concisely the network dynamics that characterize the two cases at hand. Second, by juxtaposing the cases, we will identify the major commonalities (and differences) that exist between the evolutionary paths and structural conﬁgurations of the two networks analyzed. Third, on the basis of the tips obtained, through the use of replication logic we will draw four propositions from our empirical results. These propositions allow to harvest some analytical generalizations regards the fundamental dimensions which deﬁne network structures and the dynamic relationships which underlie network structure emergence, evolution and performance. The four propositions represent a starting point for future research regarding the dynamic

108

GIOVANNI BATTISTA DAGNINO ET AL.

processes underlying network emergence and evolution. Fourth, we will single out a few implications of the study performed both for forthcoming studies in the network strategy vein and for the intentional management of networks in practice. Lastly, we gather the limitations and conclusions of the chapter. We start by considering ST’s complex network.

The Complex Dynamic Network that Links ST with Its Strategic Partners As previously illustrated, with the purpose of achieving technological dominance in the semiconductor industry and in order to co-generate various types of knowledge, ST started to develop a set of dyadic and multiple relationships with an array of co-aligned actors: (a) key customers, (b) competitors, (c) major suppliers, (d) local leading universities (such as Stanford in Silicon Valley) and (e) other research centers worldwide. Ultimately, ST has established a complex dynamic network, within which the ﬂexible division of work underlying its innovation processes ﬁnds fulﬁlment in a way that allows all partner ﬁrms to reach the edge of the knowledge frontier in particular specialized activities. More speciﬁcally, the evolutionary path of ST’s network depends on its capacity to continuously create, combine, transfer and renew the vast variety of valuable knowledge that characterizes the turbulent industry environments in which its partner ﬁrms are embedded. The capacity to nurture a high variety of knowledge bases within the network is assured by the prevalent role of the macrosystemic level, whilst a fundamental role of guidance regards the processes of knowledge combination and transfer is played by ST at the microsystemic level.

The Complex Dynamic Network that Interconnects Toyota and the Group of Its First-Tier Suppliers Unlike ST’s network, Toyota’s one is mainly a vertical network of ﬁrms that has been built up over time from a set of simple dyadic relationships between each supplier and the core ﬁrm (i.e., Toyota) to a complex system of multiple nested ties. Some of these connections (such as the Supplier Association and the Committee Meetings) refer to the macro-systemic level, while others (like the Voluntary Learning Teams and the Supplier Association Committees) relate to the meso-systemic level. The strong role accorded within Toyota’s network to highly interconnected ties and their

Evolutionary Dynamics of Inter-Firm Networks

109

dynamic interactions appear extremely well suited to exploit the existing (tacit and explicit) knowledge inside the network and to foster the relentless co-generation of new knowledge and applications between ﬁrms belonging to the same wide embracing meso-systemic level.

Commonalities and Differences between ST’s and Toyota’s Multilevel Networks Comparing the results obtained from the empirical study of the ST and Toyota cases, we are in the position to identify the complementary and differential contributions to the evolutionary dynamics of the network that the three analytical levels display in each of the two cases. At a ﬁrst glance, we could say that, while in Toyota’s case we confront ‘a complete vertical network’, in ST’s case we tackle ‘a partial vertical network’. This situation may be considered a consequence of the two strikingly different industries in which they operate (e.g., high-tech vs. more established, rapid vs. slow growth rates, industrial market vs. consumer market, and so on). However, if we dig deeper, this contention does not seem sufﬁciently convincing. In fact, whereas ST’s and Toyota’s networks seemingly represent the tale of two highly distinctive network worlds, their scrutiny allows to uncover the Pandora’s Box that illuminates the path towards a more dynamic exploration of network conﬁguration and architectures. This is actually a tale of one world. By considering the interactions among the three network levels, we can eventually purport that the nature of the cognitive domains underlying, respectively, the semiconductor and automobile industry decisively and similarly affects the architectures of ST’s and Toyota’s networks (Tables 8 and 9). Accordingly, the investigation of the emergence and evolution of the network that links ST to its strategic partners allows to underscore that the crucial contribution of the macro-systemic level is to support the identiﬁcation of the potentialities associated to establishing, modifying, reinforcing or breaking up strong connections at the meso-systemic level. This provision is achieved by means of efﬁcient, effective and rapid information sharing and explicit and general knowledge transfer among the ﬁrms participating in the network. The activation of the potentialities which emerge at the macro-level mainly depends on the ST’s capacity (at the micro-systemic level) to perceive and exploit them in order to match the challenges recurrently arising in the semiconductor industry. The contribution that the meso-systemic level provides is to overcome the barriers to tacit

110

GIOVANNI BATTISTA DAGNINO ET AL.

Table 8.

The Properties of the Complex Systems of Knowledge and Capabilities of ST and Toyota. ST’s Network

Toyota’s Network

Emergent properties

The interactions that take place (in part intentionally and in part spontaneously) among the different ﬁrms participating in the ST’s network at the mesoand the macro-systemic levels, pave the way to the creation of new knowledge and capabilities as well as the emergence of novel patterns of knowledge transfer and sharing

Self organization

In order to render the network’s sets of knowledge and capabilities coherent with the rapid innovation pace of the semiconductor industry, the existing alliances are frequently revised and extended in their learning scope (and if it is necessary they are discontinued), as well as other alliances are established with new valuable ﬁrms. As a result, a high dynamics characterizes both the ﬁrms participating in the ST’s network and the pool of knowledge and capabilities residing within the micro-, meso- and macrosystemic levels

Over time, the mechanisms and contexts/spaces of interaction deliberately implemented by Toyota at the meso- and the macro-systemic levels of the network are complemented with spontaneous and informal contacts among suppliers aimed at generating new opportunity of knowledge sharing and creation. As a result, the Toyota’s network relies less on its focal ﬁrm to direct and facilitate the knowledge exploitation and exploration activities Once the Toyota’s network has emerged, the composition of the participating ﬁrms as well as the mechanisms and contexts/spaces of interaction implemented within the network tend to be stable over time

Path dependence

The way ST’s network has emerged strongly depends on the initial ST’s knowledge endowment and the lack of speciﬁc technological

This situation is connected, on the one hand, with the Toyota’s choice to develop long-term supply relationships. On the other hand, it is related to the moderate time pressure of innovation of the auto industry and a longer life cycle of automotive products in comparison with the semiconductor ones The way Toyota’s network replication has emerged is deeply affected by the experience, the knowledge and capabilities that

111

Evolutionary Dynamics of Inter-Firm Networks

Table 8. (Continued )

Organizational closure

ST’s Network

Toyota’s Network

assets. This situation pushes ST to search worldwide the knowledge it needs Over time, the accumulated experience and the shared bases of knowledge and capabilities developed at the meso- and the macro-systemic levels, affect network behaviors generating a tendency to look for and coproduce the critical knowledge preferably inside the network

Toyota has accumulated establishing its network of ﬁrsttier suppliers in Japan Additionally, the shared experience, the strong network identity and the common bases of knowledge and capabilities developed at the meso- and the macro-systemic levels, inﬂuence network behaviors generating a tendency to look for and coproduce the critical knowledge assets mainly inside the network The idiosyncratic collection of relationships that connect Toyota and its ﬁrst-tier American suppliers permits to pinpoint the Toyota’s network per se The sets of suppliers belonging to this network and of its mechanisms and contexts/spaces of interaction were essentially stable over time. Nevertheless, changes in these sets are possible and they do not affect the Toyota’s network existence and identity The stimuli stemming from environmental dynamics activate changes inside the Toyota’s network, that are geared to secure its survival over time The Toyota’s network is composed of specialized autonomous suppliers (that are complex subsystems per se). These ﬁrms are connected to one another at three interacting levels As a result, a thorough scrutiny of the Toyota’s network requires the simultaneous consideration of the three interacting levels

The idiosyncratic set of relationships that connect ST and its strategic partners allows us to identify the ST’s network per se This network exists and can be distinguished although, over time, the participating ﬁrms can vary and different strategic alliances and learning collaborations follow one another

Thermodynamic openness

Complexity

The stimuli stemming from environmental dynamics activate changes inside the ST’s network, that are geared to secure its survival over time ST’s network is made of idiosyncratic autonomous ﬁrms (that are complex subsystems per se). These ﬁrms are connected to one another at three interacting levels As a result, a thorough scrutiny of the ST’s network requires the simultaneous consideration of the three interacting levels

112

GIOVANNI BATTISTA DAGNINO ET AL.

Table 8. (Continued )

Coevolution

ST’s Network

Toyota’s Network

By striving to extract rents from processes of knowledge exploitation and exploration, each ﬁrm in the ST’s network continuously adapts to the other participating ﬁrms and to the external environmental. As each ﬁrm adapts, it drives changes in the ﬁtness landscape of the other ﬁrms. The interaction of the coupled landscape’s changes and the joint adaptations of the single ﬁrms constantly generates the dance of coevolution of both the ST’s network as a whole and the ﬁrms embedded in it

By striving to extract rents from processes of knowledge exploitation and exploration, each ﬁrm in the Toyota’s network continuously adapts to the other participating ﬁrms and to the external environmental. As each ﬁrm adapts, it drives changes in the ﬁtness landscape of the other ﬁrms. The interaction of the coupled landscape’s changes and the joint adaptations of the single ﬁrms constantly generates the dance of coevolution of both the Toyota’s network as a whole and the ﬁrms embedded in it

and idiosyncratic knowledge transfer and to achieve smother and faster processes of knowledge co-generation within shared contexts (i.e., its tight portfolio of strategic alliances) that are characterized by speciﬁc and unambiguous learning goals. Also at the meso-systemic level the deﬁnition of the learning goals is a critical step that is usually guided by ST’s idiosyncratic capabilities (at the micro-systemic level) to seize the opportunities associated to the development and the maintenance of a portfolio of strategic alliances. In this case, therefore, it becomes apparent that the micro-systemic level (and in particular the focal ﬁrm – i.e., ST) plays a crucial role vis-a`-vis the macro- and the meso-systemic levels, which continuously take vital lymph from the focal ﬁrm’s guidance. Consequently, we are in the position to conﬁrm that ST’s micro-level ability to unremittingly create and recreate its network coherently with the shifting environmental conditions is the key contribution to grasp the network’s evolutionary pathway (Fig. 3). The scrutiny of the complex dynamic network that interconnects Toyota and the group of its ﬁrst-tier suppliers consents us to pinpoint, within the macro-systemic level, the existence of a collection of different idiosyncratic mechanisms and shared spaces that are aimed at coordinating and integrating Toyota and its suppliers. By way of repeatedly using the

113

Evolutionary Dynamics of Inter-Firm Networks

Table 9.

A Comparison of the Network Dynamics in the Two Cases Examined.

STMicroelectronics

Toyota Motor Company

Phase 1: 1987–1991 Emergent meso-level connections with key customers, research institutions and universities Objective: integrate a variety of cognitive assets, co-develop new technical tacit and explicit knowledge Characteristics: meso-level connections unrelated to each other

Phase 1: 1988–1991 Creation of spaces of socialization at the macro-systemic level shared by all suppliers Objective: one-sided transfer of information and explicit knowledge from Toyota to its suppliers Characteristics: weak ties at the meso and the macro-systemic levels

Phase 2: 1992–1999 Building/strengthening a portfolio of strategic alliances with key customers, competitors, major suppliers, universities and research centers Objective: joint research and technological co-development projects Characteristics: lack of a general coordination purpose of knowledge sharing

Phase 2: 1992–1993 Emergence of strong dyadic ties among Toyota and its suppliers Objective: smoother one-way tacit and explicit knowledge transfer from Toyota to its suppliers Objective: smoother one-way tacit and explicit knowledge transfer from Toyota to its suppliers

Phase 3: 2000–2006 Creation of an intertwined web of heterogeneous ﬁrms mutually interacting by means of a ﬁrst-class blend of both weak and strong ties Objective: efﬁcient effective and timely processes of knowledge exploitation and exploration Characteristics: ﬂexible division of innovative work that allows all partner ﬁrms to reach cutting-edge in a particular specialized activity

Phase 3: 1994–2000 Establishment of an intertwined web of specialized suppliers repeatedly interacting within shared contexts and spaces Objective: cooperate to reciprocally sharing existing knowledge and generating new knowledge Characteristics: strong network identity and high inter-ﬁrm motivation to cooperate; incessant tension towards learning and relentless improvement

mechanisms and spaces at hand, strong network identity and high inter-ﬁrm motivation to cooperate emerge at the macro-systemic level. This condition allows for the achievement of efﬁcient and effective reciprocal sharing, transfer and co-generation of capabilities and knowledge related to the TPS at the meso-level. More in detail, the latter processes take shape under the guidance of a general coordination purpose; i.e., the incessant tension

114

GIOVANNI BATTISTA DAGNINO ET AL. Phase I Meso-systemic level

Micro-systemic level

Phase II Macrosystemic level

Meso-systemic level

Micro-systemic level

Phase III Macrosystemic level

Meso-systemic level

Micro-systemic level

Fig. 3.

A Representation of the ST’s Multilevel Network along the Three Phases.

Evolutionary Dynamics of Inter-Firm Networks

115

towards learning and improvement to meet the challenges in the automotive industry. This tension ensures the coexistence and co-prosperity of the network as a whole and of the single ﬁrms participating to it. Additionally, the contribution provided by the meso-systemic level to the evolution of Toyota’s network consents effective, efﬁcient and timely co-production of new speciﬁc knowledge as well as the transfer of the tacit capabilities and knowledge that reside locally in its suppliers. It is worth mentioning, that once Toyota’s complex network had emerged and the different mechanisms and contexts of interaction at both the meso-systemic and the macro-systemic levels had been implemented (by Toyota), the main contribution of the micro-systemic level to the network’s evolutionary path has been to oversee the correct functioning of these mechanisms and contexts. In association to the role of the focal ﬁrm, the role of the macro-systemic level is that of creating a strong inter-ﬁrm network identity and transferring information and explicit knowledge to and from the external environment and through the network as a whole. It is, however, at the meso-systemic level that the main cognitive processes take place in this network. In Toyota’s case, the meso-systemic level is composed of wide embracing contexts that are strongly autonomous vis-a`-vis the other levels. Furthermore, the knowledge co-produced within the meso-level is crucial to Toyota’s competitive position and steers the evolutionary pathway undertaken by the network as a whole (see Fig. 4). Consequently, the ability of the mechanisms and spaces implemented at the meso-systemic level to continuously foster learning and improvement within the network vis-a`-vis the changes in the automotive industry is the central clue that characterizes the evolutionary pathway of Toyota’s network. On the ground of the former considerations, and applying replication logic to our empirical results, we propose the following four propositions which, rather than targeting to forge immediately testable hypotheses, contain analytical generalizations based on the inductive–deductive reasoning of this study. These propositions seem interesting starting points for further theoretical and empirical research regarding the dynamics underlying the emergence and evolution of inter-ﬁrm networks. Proposition 1 – The Fundamental Dimensions of Network Structures in a Dynamic View. Fully ﬂedged complex networks typically posses both strong and weak ties or, in other terms, they develop a multilevel structure. The weight of strong ties via-a`-vis that of weak ties and the role of the focal ﬁrms characterize and deﬁne the idiosyncratic structure of each network.

116

GIOVANNI BATTISTA DAGNINO ET AL. Phase I Macro-systemic level

Meso-systemic level

Micro-systemic level

Phase II Macro-systemic level

Meso-systemic level

Micro-systemic level

Phase III Macrosystemic level

Meso-systemic level

Micro-systemic level

Fig. 4.

A Representation of the Toyota’s Multilevel Network along the Three Phases.

Evolutionary Dynamics of Inter-Firm Networks

117

According to this proposition, fully developed networks will always possess a multilevel structure which includes and is composed of both strong and weak ties, as well as single ﬁrms at the micro-level. This view breaks away from the dichotomic view of inter-ﬁrm relations as being either strong or weak. Though if considered in isolation, single ties between different ﬁrms are either strong or weak, by widening the picture in order to consider the complexity of network structures it becomes apparent that ﬁrms interconnect and evolve network structures in which both strong and weak ties are present. Also, curtailing the holistic/reductionist dilemma and focusing on the cognitive aspects tied to inter-ﬁrm performance and evolution, the integrated complex system/knowledge-based view proposed allows to appreciate and underscore that strong ties and weak ties contribute in fundamentally different ways to the cognitive dynamics that occur within networks – supporting information and explicit knowledge transfer, in the ﬁrst case, and the sharing of complex tacit knowledge and the co-production of new knowledge, in the second case – and single focal ﬁrms may play a more or less relevant role in guiding these cognitive processes. In order for complete cognitive cycles to be efﬁciently and effectively carried out, the processes which rest on strong ties, on the one hand, and on weak ties, on the other, must all be present. However, depending on the cognitive goals the network aims towards or is stimulated to point to, the relative importance of one level vis-a`-vis the other will change, therefore leading to distinctive morphologies of network structures that foster different knowledge sharing and learning processes and bring each network to develop idiosyncratic cognitive capabilities over time. Proposition 2 – The Main Variable which Inﬂuences Network Structure Conﬁguration. There is a direct correlation between the architectural conﬁguration of inter-ﬁrm networks and the cognitive characteristics of the competitive domain in which the network operates. In particular, the relevant trait of the cognitive domain in which the network operates may be intended as the ‘cognitive scope’, which must be effectively and efﬁciently spanned by inter-ﬁrm networks (or by single ﬁrms) in order to be able to compete successfully in the business environment under consideration. According to this proposition, network structures, though idiosyncratic to each single inter-ﬁrm system, may be rendered intelligible if correlated to the characteristics that deﬁne the competitive dynamics and the key strategic factors of the environment in which the network operates. In particular,

118

GIOVANNI BATTISTA DAGNINO ET AL.

network structures should be correlated to the variety and complexity of different knowledge bases and pieces of information that must be spanned, combined and synthesized in order to create the knowledge which is key to compete with success in the industry considered. In essence, the cognitive scope which ﬁrms or inter-ﬁrm networks must master in industries in which competition rests on radical innovations is larger than the cognitive scope of the processes underlying knowledge production in industries in which competition is based on incremental innovations. In the latter case, in fact, the knowledge bases which must be confronted and brought to new synthesis tend to cover a more focused ﬁeld but tend to go in greater depth. In this vein, van Liere, Koppius, and Vervest (2008) focus on the network actors’ cognitive ability to see the network. This is an antecedent to our framework: actors’ ability to see their networks will clearly affect how they perceive the competitive domain. Thus, networks that operate in industries characterized by high innovation rates and in which competitive dynamics depend on the capacity to develop radical innovations, tend to foster the continuous renovation of the variety of knowledge bases and stimulate creativity through the adoption of structures in which macro-systemic ties are largely prevalent vis-a`-vis the meso-systemic level. In order to direct the evolutionary pathway of the network, though, a dominant role must be played by focal ﬁrms at the micro-systemic level. It is at this level that variety and opportunity sensing occurs and a fundamental coordination activity is carried out. On the other hand, networks which operate in industries characterized by less turbulent innovation rates and in which competition is based on the capacity to carry out incremental innovations and operational efﬁciency, foster rapid knowledge exploitation and co-generation of new knowledge towards clearly deﬁned and widely shared goals through dense structures characterized be the strong presence of mesosystemic ties. In this case, once the system is in place, the focal ﬁrms’ role, at the micro-systemic level, is reduced to that of a monitor, who checks the correct functioning of the institutions created in order to support the processes underlying the systems’ cognitive processes and evolutionary path. Proposition 3 – The Causal Antecedents of Network Structure Emergence. The causal antecedents of network structures are given by the different knowledge bases of the ﬁrms belonging to the network and their role in relation to: (a) the cognitive characteristics of the domain in which the network operates; and (b) the (cognitive) aim of the inter-ﬁrm system as a whole.

Evolutionary Dynamics of Inter-Firm Networks

119

This proposition is logically consequential to the ﬁrst two stated above and, thus, does not seem to need detailed explanation. Nonetheless, it is important to underscore that this proposition does not lead to consider network structures as exogenously determined. In fact, single participating ﬁrms’ knowledge bases are exogenously given at the beginning of the networks’ formation, but are subsequently endogenously generated through the various phases of the networks’ evolution in time also in correspondence to the latter’s outcomes and performance. The degree to which each participating ﬁrms’ knowledge base may be developed is, however, neither exogenous nor completely self determined. Rather, the position each actor occupies within a given network (though initially due to its precedent history and it’s knowledge base) will inﬂuence the amount and type of knowledge it will be able to absorb and develop and, therefore, the position it will be able to gain in time. Hagedoorn and Frankort (2008) offer support for Proposition 3. Their argument that network embeddedness, which increases over time, leads to a desire to form non-local ties to break this embeddedness and thus alter the structure is germane to the idea that individual network ﬁrm’s knowledge bases are endogenously generated via the various evolutionary phases. In addition, their contention supports the core claim about the endogeneity of network change we advance in this paper. Proposition 4 – The Dynamics Underlying Network Stability and Instability. Network structures will tend to be stable as long as they are adequately coherent with the cognitive characteristics of the competitive domain in which the ﬁrm operates. Therefore, instability of network structures is a consequence of changes in the cognitive characteristics of the competitive domains in which networks operate. Once again it seems important to underscore the role of endogenous mechanisms in the evolution of network structures. To this regard, it is to be noted that, while changes in the competitive arena may be a consequence of exogenous factors, networks are themselves players in the competitive ﬁeld and, as such, contribute (at times even crucially) to determine the changes that occur in the industry in which they operate. The capacity of leading networks to set (or reset) the rules of the competitive game does not, however, exempt them from structural change and instability, which appears an important underresearched area (see also Doreian, 2008 and Amburgey, Al-Laham, Tzabbar, & Aharonson, 2008). This latter consideration may be appreciated by considering the passage from one phase to another in industry life cycles. The passage from a phase of

120

GIOVANNI BATTISTA DAGNINO ET AL.

development to one of maturity, for example, may push an inter-ﬁrm network towards the adoption of a structure that enhances efﬁciency (though this may drive out variety), even if the network in consideration is the leading player in the ﬁeld and has elaborated the dominant design in the industry considered.

Implications and Limitations On the basis of the conceptual and qualitative analysis performed on ST and Toyota’s networks, the chapter bears the seeds for identifying various implications for both strategy theory and practice. The chapter ends with the consideration of the limitations of the inquiry proposed. Implications for Network Strategy Theory Early studies in the network approach have observed, either that there are few studies that employ longitudinal data to analyze networks (McPherson, Smith-Lovin, & Cook, 2001) or that most studies of network structure are cross-sectional (Burt, 2000). Other studies echoed that most network research has taken an individual-level perspective missing the opportunity to illuminate the structure of collective action (Salancik, 1995) and that little attention has been given to the evolution of entire networks (Powell et al., 2005). In the conscious attempt to overcome the limitations mentioned above, the study reported in this chapter focuses on the cognitive aspects tied to network performance and evolution in time. The empirical part of the chapter uses longitudinal data which is aimed to shed light on the processes underlying the evolution of the network as a whole and of its most signiﬁcant layers and actors. The inherent dynamic nature of this study is due not only to the adoption of the complex system theory as an epistemological base, but more essentially to the notion of knowledge, knowledge sharing and knowledge creation processes underlying it. In essence, the factor that renders change in networks endogenous and that allows to pinpoint the fundamental dimension underlying complex systems dynamics in the case of inter-ﬁrm networks is tied to processes of ‘real learning’ (Hahn, 1973) or, more simply, of knowledge creation. In our understanding, the theoretical framework elaborated and the results obtained in this study proffer to further developments that may allow us to comprehend the common traits in the current competitive and technological ecosystems and help interpret and guide strategic actor’s

Evolutionary Dynamics of Inter-Firm Networks

121

inter-ﬁrm behaviors. In particular, we have singled out three basic implications, which come to disclose avenues for further research that the evolutionary network perspective heretofore outlined has the potential to encourage: (a) the assessment of the relationships between the cognitive characteristics of the particular domain in which the network operates and the comparative role of strong and weak ties; (b) a more satisfactory understanding of the network-based determinants of competitive advantage; (c) the combination of the individual-level perspective with the consideration of the evolution and performance of entire networks. Firstly, by looking at the comparative vigor of strong and weak ties at the different network levels over time, we have commenced to insinuate the idea that both strong and weak ties are present in all fully ﬂedged complex networks though, depending on the cognitive characteristics of the domain in which the network operates, either one of the other tends to play a more crucial role in the orientation of the networks’ evolutionary path. The differential value of such approach rests in the conﬁrmation that the presence of a mix of strong and weak ties is the rule rather than the exception and that there is ﬂexibility in calibrating the weight of the different ties in relation to the network’s cognitive requirement. Second, by means of a qualitative ﬁeld methodology designed to grasp and explain the origin and structuring over time of inter-ﬁrm networks, we have eventually paved the way for a more satisfactory understanding of the network-based determinants of competitive advantage. By visualizing that the loci of competitive advantage rest in the network rather than in the single ﬁrm, we have been able to unveil the nature of the cognitive (rentgenerating) synergies that emerge from the multiple interactions among the various ﬁrms participating to the network. Consistently with this argument, Rowley and Baum (2008) suggest that partnering with direct competitors will alter both the ﬁrms’ future partnering behavior and their network positions. Third, the adoption of a multilevel view, which compares and combines coherently individual- and collective-level perspectives, in the analysis of networks allows to draw different actions, strategies and behaviors single ﬁrms or groups of ﬁrms belonging to networks may follow within a unitary framework of evaluation. This allows to capture reasons underlying ﬁrm choices and resulting network structures which may be difﬁcult to comprehend otherwise. The criterion underlying network structures and

122

GIOVANNI BATTISTA DAGNINO ET AL.

their evolution is tied to the cognitive objectives and resources of the network. In this sense, the efﬁciency, efﬁcacy and performance of the network as a whole (or the ‘global network view’) prevails on the interests of any single participating ﬁrm, as from the well-being of the system depends that of each of its participating ﬁrm. This multilevel perspective may, for example, help answer questions tied to problems of individual agency or opportunistic behaviors within networks. For example, if actors that span structural holes can use this position to beneﬁt themselves as they trade information, favors etc., why do structural holes remain unﬁlled? In our view, the distinctive role assigned to the loose macro-level context in the generation of knowledge variety and opportunity creation has proved to be particularly important in the ST case – and, by generalization, we suggest, in all those cases in which competition rests on the capacity to create radical innovations. Individual ﬁrms’ rent-seeking behaviors which ﬁll the gaps at the macro-level would be detrimental to the innovative capacity of the system as a whole. Dilemmas like the one above can be solved only through the adoption of a dynamic and multilevel approach. Implications for Network Strategy Practice Though the idea of network emergence and evolution which may be drawn from this study is certainly not deterministic and allows ample space to emergence and spontaneity, it does consider factors which exert systematic inﬂuence on network processes and performance. As such, these factors gain relevance for practice. In fact, by acting on these factors, managers may to some extent intentionally direct and design network structures and processes. In particular, the suggestions which may be gained by this study for the management of networks are: (a) to link strategy and performance of the network as a whole and of the participating ﬁrms to network structure design; (b) to consider network structure as multilevel and correlate its design to the characteristics of the competitive domain in which it operates, and, in particular, to the cognitive scope which must be spanned to innovate successfully in the domain considered; (c) to gain a clear comprehension of the role and position of the focal ﬁrm itself within the network in each of its evolutionary phases; (d) to possess a criterion appropriate to assess the stability or instability of network structures and evaluate whether changes (or the absence of changes) are the positive and natural consequence of the evolution of

Evolutionary Dynamics of Inter-Firm Networks

123

the network in relation to its external environment, or if rather they are initial signs of inadequacy and crisis that it is necessary to act upon. Limitations and Conclusion As regards the limitations of this chapter we ﬁrst raise the question, inherently inescapable in qualitative research, of the generalizability of issues originating from the comparative study of only two cases of networks, though in depth and extensive. We argue that there is always the necessity to extend the number and range of cases and industries to make the conceptual generalizations more solid and grounded. The study therefore remains exploratory in essence. In this regard, the use of a well-balanced mix of instruments and methodologies, characteristic of comparative static and dynamic analyses, could sustain the relative strength of the generalizability issue. Second, we contend that a considerable extension of this study could entail the longitudinal scrutiny of ﬁrms operating in the same industry (i.e., ST and Intel or Toyota and Chrysler). This would allow for the study of network competition which is an intriguing and virtually missing issue in network research. Lastly, this study is focused entirely on networks whose formation and evolution is based on the need and desire to cooperate in the knowledge sharing and creation processes. However, it must be underscored that not all networks emerge and exist for knowledge-related purposes. Networks formed between venture capitalists, for example, seem to be based on risksharing issues rather than knowledge sharing. The framework elaborated in this study is hence of little use in all those cases in which knowledge sharing is not the main underlying rational behind network formation and evolution.

NOTES 1. Nonaka et al. (2000) refer to the contexts of interaction that take place within the ﬁrm as ba. Ba are shared contexts, deﬁned in time and space, within which individuals may relate to one another: (a) sharing information and knowledge; (b) interpreting information and transforming it into knowledge and (c) producing new knowledge through the conversion of meanings and contexts. The latter activities drive to the elaboration of a common language and build a platform of knowledge and competences, which are integrated and shared among the individuals interacting in the ba (Nonaka et al., 2000). 2. Organizational closure allows to deﬁne the so-called cognitive dominion of the network. The cognitive dominion of a complex system is the set of the interactions that the system can embrace without endangering its own organization and without

124

GIOVANNI BATTISTA DAGNINO ET AL.

loosing its own identity. Loosing organizational closure means the disintegration of the system per se (Ceruti, 1994). 3. The adoption of a holistic perspective entails analyzing a phenomenon as an interrelated and integrated unicum resulting from the interaction of its parts. In this perspective, the properties of the single parts that make the phenomenon become indistinct, on the one hand, and the interactions among the parts and the properties emerging via them become signiﬁcant on the other. 4. The adoption of a reductionist perspective entails analyzing a phenomenon through an approach that consists in a sequence of phases, such as: (a) breaking the phenomenon up into single elements; (b) scrutinizing the single elements in isolation from the other parts; and (c) moving upwards from the properties of the elements to the general properties of the phenomenon. This perspective overlooks identifying the properties that emerge through the interactions between the parts; i.e., the properties that are new as regards to those of the single elements. 5. In each context a variety of mechanisms are arranged in order to achieve the coordination and the integration of the participating ﬁrms. These mechanisms consist in the institutionalization of formal and informal rules and norms, incentive systems, joint decision processes for problem solving, negotiation mechanisms, linking-pin roles and units, formal and informal relations of authority, and so on. Each of these mechanisms displays peculiarities which make its adoption more efﬁcient and effective under speciﬁc conditions, mainly associated with the goal of the context itself (i.e., the sharing of tacit/explicit knowledge and/or the co-production of tacit/explicit knowledge) (Levanti & Mocciaro Li Destri, 2006). 6. We stress that homophily (Powell et al., 2005) in dense interﬁrm connections is a risk – not a necessity – and may never actually occur in speciﬁc empirical settings. In fact, whereas for the individual strong ties and intense interactions with one or a limited number of other individuals brings to a homogeneity of views which renders isomorphic behaviors highly probable, the same reasoning does not necessarily apply at higher ontological levels and, in particular, at the ﬁrm level. It seems possible to imagine that the risk of isomorphism at the ﬁrm level depends on the variety of activities it carries out, on the breadth of its ﬁelds of activity, as well as its organizational characteristics. 7. According to Mason (1996, pp. 93–94), ‘theoretical sampling means selecting groups or categories to study on the basis of their relevance to your research questions, your theoretical position and analytical framework, your analytical experience, and most importantly the explanation or account which you are developing’. 8. It is worth noting that, at its onset, the company was labeled SGS-Thomson. 9. When in 1991 ST implemented their Total Quality Management Program, they chose to adopt the following mission label: ‘to offer strategic independence to out partners worldwide as a proﬁtable and viable broad-range semiconductor supplier’. 10. Such as Air Liquide, Applied Materials, ASM Litography, Canon, HewlettPackard, KLA – Tencor, LAM Research, MEMC, Schlumberger, Teradyne, Wacker, Co Ware, Synopsis, UMC and Cypress Semiconductors. 11. Such as MEDEAþ (the pan-European program for advanced cooperative research and development in microelectronics technologies and its applications) and

Evolutionary Dynamics of Inter-Firm Networks

125

ITEA, which later led to ITEA2 (the pan-European program for advanced precompetitive R&D for software-intensive systems and services). 12. Such as ENIAC (European Nanoelectronics Initiative Advisory) and ARTEMIS (Advanced Research and Technology for Embedded Intelligence and Systems). 13. Located in the UK (Bristol and Newcastle), Italy (Bologna, Catania, Milan, Pavia and Turin), France (Grenoble, Marseille, Toulouse and Tours), in the US (Carnegie Mellon, Stanford, Princeton, Berkeley, UCSD, and UCLA) and Singapore. 14. Such as CEA, IMEC and France Telecom R&D. 15. Sources: STMicroelectronics and World Semiconductor Trade Statistics (WSTS). 16. The rankings are compiled by Gartner Dataquest Corp. and iSuppli Corp. taking into account the revenues of the semiconductor ﬁrms. 17. We are aware that this line of inquiry is at odd with much of the organization theory literature. Conventional organization theory literature suggests that the environment has a strong direct effect on organizations. Accordingly, it is acknowledged that it is generally difﬁcult to transfer organizations between dissimilar environments and as well that, once transferred, organizations tend to take on characteristics of the new environment and/or of organizations with which they interact (Di Maggio & Powell, 1983). In fact, a recent management practiceoriented stream (Adler et al., 1999), used case studies and large-scale surveys to explain in-depth the process of transferring and transforming the best Japanese Management Systems (JMS) by both Japanese and US owned ﬁrms. While the most successful of Japanese manufacturing plants in the US rely on home country management techniques, they have had to adapt them to ﬁt US conditions. Similarly, the growing number of US ﬁrms that are adopting these techniques to strengthen their own positions face a considerable challenge in transforming them to ﬁt local conditions. But despite the hurdles ﬁrms face, the evidence obtained robustly indicates that key aspects of JMS are successful and transferable in the US.

ACKNOWLEDGMENTS Earlier versions of this paper were presented at the Advances in Strategic Management Conference on ‘Strategic Networks’, Toronto May 24–25, 2007, at the 25th DRUID Celebration Conference on ‘Entrepreneurship and Innovation’, Copenhagen, June 17–20, 2008 and in research seminars at IESE Business School, LUISS University of Rome and the University of Palermo. For valuable discussions and kind support, we wish to thank the participants in these venues and, in particular, Africa Arin˜o, Ron Burt, Joel Baum, Bruno Cassiman, Francesco De Leo, Fabrizio Ferraro, Joan Enric Ricart and Tim Rowley. For assistance in the data collection process, we acknowledge Cristina Di Gesu` of STMicroelectronics.

126

GIOVANNI BATTISTA DAGNINO ET AL.

REFERENCES Adler, P. S., Fruin, M. W., & Liker, J. K. (1999). Remade in America. Transplanting and transforming Japanese management systems. Oxford: Oxford University Press. Ahuja, G., Soda, G., & Zaheer, A. (2007). The genesis and dynamics of networks. Organization Science – Call for Papers for 10/31/2008. BPS-Net. November. Amburgey, T. L., Al-Laham, A., Tzabbar, D., & Aharonson, B. (2008). The structural evolution of multiplex organizational networks: Research and commerce in biotechnology. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 171–209). Bingley, UK: JAI/Emerald Group. Anderson, P. (1999). Complexity theory and organization science. Organization Science, 10(3), 216–232. Axelrod, R., & Cohen, M. D. (2000). Harnessing complexity. Organizational implications of a scientiﬁc frontier. New York: Basic Books. Baum, J. A. C. (1999). Whole-part coevolutionary competition in organizations. In: J. A. C. Baum & B. McKelvey (Eds), Variation in organization science: In honor of Donald T. Campbell. London: Sage. Baum, J. A. C., Calabrese, T., & Silverman, B. S. (2000). Don’t go it alone: Alliance network composition and startups’ performance in Canadian biotechnology. Strategic Management Journal, 27, 267–294. Bertalanffy, von L. (1969). General systems theory. New York, NY: G. Braziller. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge MA: Harvard University Press. Burt, R. S. (1994). Structural holes and good ideas. American Journal of Sociology, 110, 349–399. Burt, R. S. (2000). Decay functions. Social Networks, 22, 1–28. Capaldo, A. (2007). Network structure and innovation: The leveraging of a dual network as a distinctive relational capabilities. Strategic management Journal, 28, 585–608. Ceruti, M. (1994). Constraints and possibilities. The evolution of knowledge and knowledge of evolution. Lausanne: Gordon & Breach. Cohen, M. (1999). Commentary on the ‘Organization Science’ special issue on complexity. Organization Science, 10(3), 373–376. Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35, 128–152. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, 95–120. Cusumano, M. A. (1985). The Japanese automobile industry: Technology and management at Nissan and Toyota. Cambridge, MA: Harvard University Press. Dagnino, G.B. (1999). The system of business enterprises as a complex dynamic network of resources and competences. Academy of Management Best Papers Proceedings, Chicago. Dagnino, G. B. (2004). Complex systems as key drivers for the emergence of a resource and capability-based interorganizational network. Emergence: Complexity and Organization, 6(1–2), 61–69. Di Maggio, P., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational ﬁelds. American Sociological Review, 48, 147–160.

Evolutionary Dynamics of Inter-Firm Networks

127

Doreian, P. (2008). Actor utilities, strategic action and network evolution. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 247–271). Bingley, UK: JAI/Emerald Group. Doz, Y., Santos, J., & Williamson, P. (2001). From global to metanational. How companies win in the knowledge economy. Boston, MA: Harvard Business School Press. Dyer, J. H. (2000). Collaborative advantage. Winning through extended enterprise suppliers networks. Oxford: Oxford University Press. Dyer, J. H., & Hatch, N. W. (2006). Relation-speciﬁc capabilities and barriers to knowledge transfers: Creating advantage through network relationships. Strategic Management Journal, 27, 701–719. Dyer, J. H., & Nobeoka, K. (2000). Creating and managing a high-performance knowledge-sharing network: The Toyota case. Strategic Management Journal, 21, 345–367. Eisenhardt, K. M. (1989). Building theories from case study research. Academy of Management Review, 14(4), 532–550. Florida, R., & Kenney, M. (1993). Beyond mass production: The Japanese system and its transfer to the US. New York: Oxford University Press. Fontana, W., & Ballati, S. (1999). Complexity. Complexity, 4(3), 14–16. Fujimoto, T. (1999). The evolution of a manufacturing system at Toyota. Oxford: Oxford University Press. Glaser, B. G., & Strauss, A. L. (1967). The discovery of grounded theory. Chicago: Aldine. Grant, R. M. (1996a). Toward a knowledge-based theory of the ﬁrm. Strategic Management Journal, 17(Winter Special Issue), 109–122. Grant, R. M. (1996b). Prospering in dynamically-competitive environments: Organisation capability as knowledge integration. Organization Science, 7(4), 375–387. Gulati, R. (1999). Network location and learning: The inﬂuence of network resources and ﬁrm capabilities on alliance formation. Strategic Management Journal, 20(5), 397–420. Gulati, R., Nohria, N., & Zaheer, A. (2000). Strategic networks. Strategic Management Journal, 21(Special Issue), 203–215. Hagedoorn, J., & Duysters, G. (2002). Learning in dynamic inter-ﬁrm networks: The efﬁcacy of multiple contacts. Organization Studies, 23, 525–548. Hagedoorn, J., & Frankort, H. (2008). The gloomy side of embeddedness: The effect of overembeddedness on partnership formation. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: advances in strategic management (Vol. 25, pp. 503–530). Bingley, UK: JAI/ Emerald Group. Hahn, F. (1973). On the notion of equilibrium in economics. Cambridge, UK: Cambridge University Press. Kale, P., Singh, H., & Perlmutter, H. (2000). Learning and protection of proprietary assets in strategic alliances: Building relational capital. Strategic Management Journal, 21, 217–237. Kauffman, S. A. (1993). The origins of order: Self-organization and selection in evolution. New York: Oxford University Press. Langley, A. (1999). Strategies for theorizing from process data. Academy of Management Review, 24(4), 691–710. Lavie, D. (2006). The competitive advantage of interconnected ﬁrms: An extension of the resource-based view. Academy of Management Review, 31(3), 638–658.

128

GIOVANNI BATTISTA DAGNINO ET AL.

Levanti, G., & Mocciaro Li Destri, A. (2006). Governance structures and knowledge sharing and creation in inter-ﬁrm networks. Paper presented at the Strategic Management Society Conference, Vienna. Lewin, A. Y., & Volberda, H. W. (1999). Prolegomena on coevolution: A framework for research on strategy and new organizational forms. Organization Science, 10(5), 519–534. Lorenzoni, G., & Lipparini, A. (1999). The leveraging of interﬁrm relationships as distinctive organizational capabilities: A longitudinal study. Strategic Management Journal, 20, 317–338. MacDufﬁe, J. P., & Helper, S. (1997). Creating lean suppliers: Diffusing lean production through the supply chain. California Management Review, 39(4), 118–151. Mason, J. (1996). Qualitative reasoning. London: Sage. Maturana, H., & Varela, F. (1987). The tree of knowledge. Boston: Shambala. McEvily, B., & Zaheer, A. (1999). Bridging ties: A source of interﬁrm heterogeneity in competitive capabilities. Strategic Management Journal, 20(12), 1133–1156. McKelvey, B. (1997). Quasi-natural organization science. Organization Science, 8(4), 352–380. McKelvey, B. (1999). Avoiding complexity catastrophe in coevolutionary pockets: Strategies for rugged landscapes. Organization Science, 10(3), 294–321. McPherson, M., Smith-Lovin, L., & Cook, J. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415–444. Mocciaro Li Destri, A., & Dagnino, G. B. (2005). The development of the resource-based ﬁrm between value appropriation and value creation. Advances in Strategic Management, 22, 153–188. Monden, Y. (1998). Toyota production system: An integrated approach to just-in-time. Norcross GA: Engineering & Management Press. Morin, E. (2001). La me´thode V. L’humanite´ de l’humanite´. Paris: Editions du Seuil. Morin, E. (1977). La me´thode I. La nature de la nature. Paris: Editions du Seuil. Nonaka, I. (1994). A dynamic theory of organizational knowledge creation. Organization Science, 5(1), 14–37. Nonaka, I., & Toyama, R. (2002). A ﬁrm as a dialectical being: Towards a dynamic theory of a ﬁrm. Industrial and Corporate Change, 11(5), 995–1009. Nonaka, I., Toyama, R., & Nagata, A. (2000). A ﬁrm as knowledge-creating entity: A new perspective on the theory of the ﬁrm. Industrial and Corporate Change, 9(1), 1–20. Nooteboom, B. (2004). Inter-ﬁrm collaboration, learning and networks. An integrated approach. London: Routledge. Pettigrew, A. M. (1990). Longitudinal ﬁeld research on change. Theory and practice. Organization Science, 1(3), 267–292. Porter-Liebeskind, J. (1996). Knowledge, strategy and the theory of the ﬁrm. Strategic Management Journal, 17(Winter Special Issue), 93–107. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41, 116–145. Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network dynamics and ﬁeld evolution: The growth of interorganizational collaboration in the life sciences. The American Journal of Sociology, 110(4), 1132–1205. Prigogine, I., & Stengers, I. (1984). Order out of Chaos: Man’s new dialogue with nature. London.

Evolutionary Dynamics of Inter-Firm Networks

129

Rowley, T. J., & Baum, J. A. C. (2008). The dynamics of network strategies and positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 641–671). Bingley, UK: JAI/Emerald Group. Rowley, T., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21, 369–386. Salancik, G. R. (1995). Wanted: A good network theory of organization. Administrative Science Quarterly, 40(2), 345–349. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Van De Ven, A. H., & Poole, M. S. (2002). Field research methods. In: J. A. C. Baum (Ed.), Companion to organizations. Malden, MA: Blackwell. van Liere, D., Koppius, O., & Vervest, P. (2008). Strengthening of bridging positions: Network horizon and network horizon heterogeneity. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 595–639). Bingley, UK: JAI/Emerald Group. Volberda, H. W., & Lewin, A. Y. (2003). Co-evolutionary dynamics within and between ﬁrms: From evolution to co-evolution. Journal of Management Studies, 40(8), 2111–2136. Waldrop, W. M. (1992). Complexity: The emerging science at the edge of order and chaos. New York: Touchstone. Yin, R. K. (1994). Case study research (2nd ed). Thousand Oaks, CA: Sage. Zaheer, A., & Bell, G. G. (2005). Beneﬁting from network position: Firm capabilities, structural holes, and performance. Strategic Management Journal, 26, 809–825.

THE ROLE OF DYADIC MULTI-DIMENSIONALITY IN THE EVOLUTION OF STRATEGIC NETWORK TIES Julie M. Hite ABSTRACT Dyadic multi-dimensionality informs the variation that exists within and between network ties and suggests that ties are not all the same and not all equally strategic. This chapter presents a model of dyadic evolution grounded in dyadic multi-dimensionality and framed within actor-level, dyadic-level, endogenous, and exogenous contexts. These contexts generate both strategic catalysts that motivate network action and bounded agency that may constrain such network action. Assuming the need to navigate within bounded agency, the model highlights three strategic processes that demonstrate how dyadic multi-dimensionality underlies the evolution of strategic network ties.

Network Strategy Advances in Strategic Management, Volume 25, 133–170 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25004-7

133

134

JULIE M. HITE

INTRODUCTION Underlying a strategic network is the dynamic nature of the network’s most basic elements – the individual network actors and their ties. Network literature has consistently called for better explanations at this micro-level of analysis which focuses on network actors and ties (Granovetter, 1979; Kilduff & Tsai, 2003), given that even small changes at the micro level of network ties can signiﬁcantly affect evolution and change within the larger network (Amburgey & Al-Laham, 2008; Doreian, 2008). Explanations of strategic networks at the dyadic level, however, require understanding the systemic and multi-dimensional nature of network ties. A network dyad consists of two actors and the tie that connects them. Network or dyadic ties, as ties between two network actors, can refer either to a single relation deﬁned for a given purpose (e.g. advice relation) or to a set of multiple relations between the actors. Network ties exist not only within a larger network system, but also within a dyadic system that includes the two actors and the tie. Ties themselves also constitute critical and dynamic systems that continuously change and evolve due to the accumulation of small changes within the dyad. The dyadic system is multidimensional, which creates the basis for a broad range of dyadic variation. This dyadic variation represents an important concept in explaining how strategic network choice may inﬂuence the evolution of network ties, as network actors seek within this variation to select, develop and retain critical strategic dyadic dimensions. Dyadic multi-dimensionality is found within four contexts of ties: actorlevel, dyadic-level, endogenous, and exogenous. Each context is richly multidimensional and provides sources of variation within and between network ties and thus potential explanations for the evolution of network ties. These multiple contexts generate strategic catalysts that motivate network actors to seek dyadic change while, at the same time, contributing to the bounded agency within which network actors must navigate. Thus the evolution of network ties throughout the dyadic life cycle is a dynamic process in which network actors navigate multi-dimensional contexts and bounded agency to facilitate strategic network action. Strategic network research at macro network levels often treats network ties as a ‘‘black box’’ rarely accounting for types or degrees of tie variation. Micro-level network research often acknowledges and examines different types of ties and their effects, thus recognizing variation between ties; however, much of this research treats only a single relation or content ﬂow, discounts the valued or directed nature of ties, or aggregates potential

The Role of Dyadic Multi-Dimensionality

135

multi-dimensionality into dichotomous factors (i.e. strong and weak ties). As a result, important dyadic variation that may inﬂuence the strategic evolution and outcomes of network ties has not been fully addressed. Strategic network research is beginning to look more deeply at dyadic multidimensionality in dyadic content, structure and governance (e.g. Elfring & Hulsink, 2007; Hite, 2003, 2005). Multi-dimensional perspectives are necessary to gain richer strategic explanations (Wasserman & Faust, 1994) and to understand how to strategically manage the complexity of network tie evolution. Consequently, the multi-dimensional nature of the network ties and their contexts must be accounted for to provide meaningful theoretical explanations for the creation, evolution and outcomes of strategic networks. This chapter presents a model of dyadic evolution that is grounded in dyadic multi-dimensionality, framed within actor-level, dyadic-level, endogenous and exogenous contexts. The model suggests the dual role of these contexts in generating both strategic catalysts that motivate network actors to seek change in their network ties and the bounded agency that may constrain such network action. Assuming the need to navigate within bounded agency, the model highlights three strategic processes that may underlie strategic evolution of network ties.

DYNAMIC EVOLUTION OF DYADIC NETWORK TIES Over the past decade awareness of the dynamic nature of networks and interest in the evolution of network ties has been increasing. Over a decade ago Wasserman and Faust’s (1994) classic Social Network Analysis did not address the dynamic nature of networks except in a brief comment that ‘‘occasionally a researcher is interested in how ties in a network change over time’’ (p. 55). The majority of network research has focused on static, crosssectional views of networks and their ties, which do not take into account their dynamic changing natures. Indeed, a common critique of social network research has been its neglect in recognizing the dynamic nature of network ties (Emirbayer, 1997; Moody, McFarland, & Bender-deMoll, 2005). Network researchers are now taking a more dynamic view of networks by increasingly addressing both theoretical and empirical issues of how networks change and develop over time (Borgatti & Foster, 2003; Contractor, Wasserman, & Faust, 2006; Hoang & Antoncic, 2003; Moody et al., 2005), as well as how various processes of network dynamics create,

136

JULIE M. HITE

sustain and dissolve network structure at the dyadic level (e.g. Elfring & Hulsink, 2007; Hite, 2005; Uzzi & Lancaster, 2003; Uzzi, Spiro, Guimera, & Amaral, 2006). This dynamic view requires a longitudinal perspective rather than a static ‘‘snapshot in time y to grasp the dynamic processes that shape this reality over time’’ (Emirbayer & Goodwin, 1994, p. 1413). In response, analysis methods, visualization techniques, and software such as SIENA have been designed to facilitate analysis of longitudinal network data (Moody et al., 2005; Snijders, 1996, 2001, 2005b). Understanding dynamic networks begins with the network’s most basic elements – the individual network actors, their dyadic ties and their ‘‘relational dances’’ (Moody et al., 2005, p. 1231). ‘‘Just as the architect needs to understand what makes buildings stay up rather than fall down, the network strategist needs to understand the elementary processes through which network structure comes about’’ (Koka, Madhavan, & Prescott, 2006, p. 721). Over time, as micro-level network changes occur, these ‘‘small, incremental changes accumulate to the point at which substantial macro-level transformations of structure occur’’ (Wasserman, Scott, & Carrington, 2005, p. 6). The accumulation of many small and continuous changes across multiple actors and their ties over time results in dynamic and systemic network change (Snijders, 2005b) that can signiﬁcantly and strategically affect the structure, content, governance, and strategic outcomes of the network. Dyadic change clearly occurs when network actors add or drop ties to accomplish strategic purposes (Elfring & Hulsink, 2007) or when actors themselves enter or leave the network. However, a more ﬁne-grained perspective of the evolution of network ties suggests that the tie itself is a continuously evolving, multi-dimensional system. Change occurs within the dyad – within each of the two actors and within their tie. For example, a tie may become more relationally or structurally embedded over time (Hite, 2005) or take on additional purposes and become more multiplex (Robins & Pattison, 2005). Network ties, as continuously changing multi-dimensional systems, can be strategic to the extent that network actors intentionally design and manage them to facilitate ﬁrm performance (Gulati, Nohria, & Zaheer, 2000a; Kilduff, Tsai, & Hanke, 2006; Pillai, 2006). Adding evolution to the discussion of strategic network ties implies that variation, a core concept in evolution, exists among network actors and network ties that may inform the creation and selection of ties as well as their development and retention (McKelvey & Aldrich, 1983). Fig. 1 presents a broad model of the evolution of network ties that assumes such variation is found in the

137

The Role of Dyadic Multi-Dimensionality

External Environment ous

en xog

) ects

Eff

(E

Network System

ts)

ffec

en

dog

(En

E ous

Dyadic Contexts •ACTOR-LEVEL •DYADIC-LEVEL -Content -Structure -Governance -Life Cycle Phase •ENDOGENOUS NETWORK -Network Structure -Network Content •EXOGENOUS INFLUENCES -Organizational Context -Strategic Environment

Dyadic Change

Bounded Agency Motivation for Dyadic Change

Strategic Network Processes

-Content -Structure -Governance -Life Cycle Phase

•Layering •Leveraging •Timing

Strategic Catalysts

Fig. 1.

Framework of Dyadic Evolution.

inherent multi-dimensionality of dyadic ties. Four dyadic contexts frame the sources of dyadic multi-dimensionality and generate strategic catalysts which motivate network action. At the same time, however, these contexts also generate bounded agency which moderates or constrains the strategic actions of network actors. Navigating within bounded agency, strategic actors may use the variation from dyadic multi-dimensionality to facilitate network action and evolution through three strategic processes: layering, leveraging, and timing. The evolution of network ties then, in turn, affects these dyadic contexts, creating antecedent contexts for future dyadic change. Over time the evolution of a network tie may be usefully framed from the perspective of the dyadic life cycle. Thus the evolution of network ties reﬂects systemic changes within the dyad and its contexts facilitated by strategic processes that both draw upon and then affect dyadic multi-dimensionality. Network research has begun to take into account and describe dyadic multi-dimensionality and its potential relationship to the evolution of network ties (e.g. Bliemel & Maine, 2006; Chung-hoon, Hite, & Hite, 2005, 2007; Elfring & Hulsink, 2007; Hite, 2003, 2005; Uzzi & Gillespie, 2002; Uzzi & Lancaster, 2003; Uzzi et al., 2006).

138

JULIE M. HITE

MULTI-DIMENSIONALITY WITHIN FOUR DYADIC CONTEXTS The evolution of network ties does not occur in a vacuum; rather, dyadic contexts matter. Ties are situated within speciﬁc contexts of conditions that pre-exist dyadic change and can affect whether, how, and when dyadic changes occur (Elfring & Hulsink, 2007; Hite, 2005; Hite & Hesterly, 2001). Understanding the evolution of network ties requires identifying, explaining, and accounting for the multiple dimensions of these antecedent contexts which affect how ties may differ from each other and, more important for tie evolution, how ties themselves may differ from one point in time to another. For both processes, dyadic multi-dimensionality generates the variation underlying potential evolutionary mechanisms. Dyadic contexts are found at four different levels of analysis: the actor, the dyad itself, the endogenous network, and the exogenous environment. The following discussion will identify potential sources of multi-dimensionality within these contexts, with particular emphasis on the dyadic context. Actor-Level Context The actor-level context focuses on the network actors as a source of dyadic multi-dimensionality. In strategic networks, thus in the scope of this chapter, a network actor may represent an individual acting as an agent for the ﬁrm or represent the ﬁrm itself (Emerson, 1981; Hite, 2003; Zaheer, McEvily, & Perrone, 1998). Individual actors demonstrate multi-dimensionality due to differing goals, orientations, experience, and capabilities. For example, entrepreneurial orientation affects tie formation efforts (Elfring & Hulsink, 2003, 2007), and network horizons differ based on actor experience (Anderson, Hakansson, & Johanson, 1994; Doreian, 2008; Holmena & Pedersena, 2003). Thus network actors ‘‘are likely to differ in their ability to notice and respond to’’ (Kilduff et al., 2006, p. 1040) changes in their networks. Firm-level actors may demonstrate multi-dimensionality in terms of the ﬁrm’s founding conditions, strategic choices, ﬁrm’s life cycle stage, institutional knowledge and capabilities. For example, ‘‘not all emerging ﬁrms are equally endowed in terms of initial network connections and these differences matter’’ (Hite & Hesterly, 2001, p. 283). In addition, Elfring and Hulsink (2007) explain that two key founding conditions that can affect the development of the ﬁrm’s network ties are whether the entrepreneur is an industry insider or outsider and what type of innovation the ﬁrm introduces.

The Role of Dyadic Multi-Dimensionality

139

Considering Borgatti and Foster’s (2003) caution that different types of actors may ‘‘have different capabilities and the relations have different meanings’’ (p. 1001) to network actors, explanations of the evolution of network ties need to account for the multi-dimensional attributes of network actors and relax assumptions of actor homogeneity (Contractor et al., 2006). Not only are network actors multi-dimensional, they are also by nature dynamic. Changes occur within network actors as they learn, imitate, and experiment as well as shift their network strategies to facilitate adaptation to internal and external environments. For example, Zaheer et al. (1998) demonstrate that a network actor may change from individualto ﬁrm-level actor. Thus the multi-dimensional and dynamic nature of network actors creates a critical context that must be taken into account in explanations of dyadic network change.

Dyadic-Level Context In the dyadic-level context, the ‘‘element or unit in social space is not the individual but the ties that connect them’’ (Wasserman et al., 2005, p. 6). Dyadic ties consist of multi-dimensional characteristics which ‘‘cannot be reduced to the properties of the individual agents’’ (Scott, 2000, p. 3), but rather reﬂect the interdependence of actor behavior (Robins & Pattison, 2005). Multi-dimensionality within the dyad itself creates idiosyncratic ties that can contribute to competitive advantage (Dyer & Singh, 1998) and the variation that can facilitate tie evolution. Dyadic multi-dimensionality is found in the tie’s content, structure, and governance (Amit & Zott, 2001; Hoang & Antoncic, 2003), as well as in the dyadic life cycle. Multi-Dimensionality of Dyadic Content Dyadic ties represent potential bridges, conduits, or pipes through which different types of content may ﬂow or be exchanged (Elfring & Hulsink, 2007; Hite & Hesterly, 2001; Podolny, 2001). Tie content deﬁnes the tie relation and the purpose of the dyadic connection. Network actors create and manage ties to acquire, provide, or exchange content, and the choice of dyadic partner affects potential differences in type, quantity, and quality of tie content. Multi-dimensionality is found in the different types of content that may ﬂow across the tie, as well as its quality and quantity (Hite, 2003; McEvily & Marcus, 2005). Three dimensions of tie content type are seen in the ﬂow of resources, opportunities, and legitimacy (Elfring & Hulsink, 2007). First,

140

JULIE M. HITE

ﬂow of tie content can provide many types of resources (tangible and intangible), such as information, knowledge, and advice (e.g. Borgatti & Cross, 2003; Shane & Cable, 2002), ﬁnancial capital (e.g. Steier, 2000; Uzzi, 1999), social capital (Coleman, 1990), physical and human resources (e.g. Hite et al., 2006), and relational content such as emotional/social support (e.g. Bruderl, Preisendorfer, & Ziegler, 1992; Hite, Williams, & Baugh, 2005; Shane & Stuart, 2002). Second, tie content can provide opportunities, as reﬂected in entrepreneurial network literature (Hoang & Antoncic, 2003). Third, tie content can convey legitimacy through reputation, trust, expectations of normative behavior, and signaling functions (e.g. Human & Provan, 2000; Podolny, 2001; Stuart, Hoang, & Hybels, 1999). While network research has generally examined a single content or relation between actors (e.g. alliance, friendship, advice, communication, innovation, etc.), dyadic ties are more richly understood as multiplex – comprising multiple relations co-existing within a single tie (Koehly & Pattison, 2005; Lazega & Pattison, 1999; Mohrman, Galbraith, & Monge, 2004). Multiplexity exists when more than one type of content ﬂows across the tie such that ‘‘several types of relationships come together’’ (Koehly & Pattison, 2005, p. 162): for example, a multiplex tie may function as ‘‘a link across which two or more kinds of knowledge are shared’’ (Mohrman et al., 2004, p. 43). Thus multiple purposes may be served within a single tie, which further increases its multi-dimensionality. If the multiplex nature of dyadic ties is compared to a rope with many different colored strands, adding tie content or purposes to the dyadic tie is like adding another strand to the rope: increasing both multiplexity and tie strength. Relational embeddedness and strength of ties (Hite, 2003; Uzzi, 1996) are both grounded in the idea that multiplexity of tie content provides the potential strategic leverage for dyadic ties to evolve towards greater relational embeddedness (Hite, 2005). Multiplexity of tie content, an example of dyadic multi-dimensionality, provides a foundational explanation for variation within and between ties.

Multi-Dimensionality of Dyadic Structure The nature of tie structures also reﬂects how dyadic multi-dimensionality creates variation within and between ties and thus informs the strategic nature of dyadic ties and their evolution. Dyadic structure, which explains how network content ﬂows between two network actors, can be framed in terms of structures of exchange, connection, and interaction.

The Role of Dyadic Multi-Dimensionality

141

Structures of Dyadic Exchange. Structures of dyadic exchange address the infrastructure through which tie content is exchanged between two network partners. If dyadic ties are viewed as analogous to pipes (Elfring & Hulsink, 2007; Hite & Hesterly, 2001; Podolny, 2001), tie exchange structures include dimensions of diameter, pressure, valves, and directionality. Dyadic diameter, a reference to width, suggests that ties may vary in the extent of content they carry. Larger diameters imply capacity for greater amounts or greater multiplexity of tie content ﬂow. Dyadic pressure refers to the force under which tie content ﬂows, suggesting ties may differ in amount of content ﬂowing at a given point in time. Variation may also be found in patterns of dyadic pressure over time. Dyadic valves refer to the mechanisms that regulate the ﬂow of tie content and can begin, facilitate, encourage, moderate, and ebb these ﬂows. Directionality of tie content ﬂow is the aspect of dyadic exchange structure most commonly addressed in network literature. One-way ties reﬂect a transfer of network content from an actor to a dyadic partner and, given their assymetry, may be less sustainable than two-way ties with mutual exchange (Emerson, 1981). Two-way ties reﬂect a mutual transfer of network content; however, a reciprocal or mutually beneﬁcial two-way exchange may not be evident within a single dyadic tie unless potential multiplex and asymmetric exchange are taken into account (Lazega & Pattison, 1999). Multiplex exchange structures facilitate the ﬂow of multiplex tie content. Although one type of content may ﬂow asymmetrically, dyadic ties with multiplex exchange structures are likely to demonstrate a two-way, mutually beneﬁcial exchange structure when sufﬁcient multiplexity of tie content is taken into account (Koehly & Pattison, 2005). Asymmetric exchange structures, which reﬂect differences between the tie content provided and the content received, may be represented in terms of differential diameter, pressure, timing, or mutual beneﬁt. Asymmetric exchange structures, however, may actually be mutually beneﬁcial when a longitudinal perspective accounts for multiple exchanges over time. A special case of multiplex and asymmetric exchange structures is found in relationally embedded ties (Hite, 2003), in which tie content may be provided in exchange for maintenance of the dyadic tie itself or with the expectation that a mutual beneﬁt will be achieved through a future reciprocal exchange – likely of a different content nature. Thus the tie exchange structure may appear to facilitate one-way and asymmetric transfer of tie content when, within the frame of multiplexity and relational embeddedness, the tie structure actually facilitates mutually beneﬁcial two-way exchange. Dyadic exchange structures constitute an important example of dyadic

142

JULIE M. HITE

multi-dimensionality and, therefore, the potential variation within and between network ties. Structures of Dyadic Connection. Dyadic multi-dimensionality is also found in the structures of dyadic connection. In addition to tie content and exchange, an important strategic purpose of dyadic ties is their functional role in connecting strategic actors to the larger network structure. Four structures of dyadic connection further demonstrate the multi-dimensionality of ties and their potential variation: embedding, bridging structural holes, bridging multiple networks, and facilitating structural formality. The ﬁrst structure of dyadic connection, a dyadic embedding structure, exists when a dyadic tie facilitates greater cohesion within the larger network. A dyadic embedding structure occurs when the dyadic tie connects network actors who also have other local network ties, such as in closing a triad. The structural role of this redundant tie increases network cohesion and density, as well as the structural embeddedness of the dyadic actors. Dyadic embedding structures may strengthen the dyadic tie itself, as well as the surrounding ties. A second dyadic connection structure, a dyadic bridging structure, is found when a dyadic tie functions as a bridge to span or ﬁll structural holes (Burt, 2000, 2001; Coleman, 1990). This structure occurs when a tie connects a network actor to a new dyadic partner, particularly when no other tie path or functional tie path currently exists. Thus a dyadic bridging structure connects actors and ﬁlls a structural hole within the larger network, increasing the actor’s network reach. Yet as not all structural holes are the same (Doreian, 2008), further variation also exists within this dyadic connection structure. A third dyadic connection structure, a multiplex dyadic bridging structure, is found when a dyadic tie creates a bridge to strategically link actors across multiple networks (Doreian, 2008; Hansen, Mors, & Lovas, 2005; Hite et al., 2006; Lomi, 2002; Soda, Zaheer, & Carlone, 2008). Dyadic connections that bridge multiple networks are facilitated by multiplexity, such as addition of new tie content to an existing tie (Ettlinger, 2003; Hite et al., 2005; Koehly & Pattison, 2005), as well as by brokering, such as a dyadic partner brokering the actor into a new network. For example, while network actors may already exchange technological information, one actor may bring the dyadic partner into the ﬁrm’s network of suppliers. Thus a new purpose and ﬂow of tie content is added to the dyadic tie, creating a multiplex connection between the two actors that also serves as a crossnetwork link. This bridging structure reﬂects the co-evolution of ties and

The Role of Dyadic Multi-Dimensionality

143

structure, demonstrating how different networks can affect one another (Soda et al., 2008). Bridging multiple networks may increase dyadic cohesion, strengthen the surrounding dyadic ties, and become a catalyst for creating additional ties across the two networks. Structural formality, a fourth type of structural connection, is ‘‘the degree to which [a dyadic partner] is connected formally to the organization within its recognized boundaries’’ (Chung-hoon et al., 2005, 2007). Low structural formality is demonstrated when the network actors are not formally acting as agents for their ﬁrms. As structural formality increases, the legitimacy of the tie increases in regard to speciﬁc functions. As a result, the tie becomes more formally recognized by the actors’ respective organizations, and may shift from individual to ﬁrm level tie (Zaheer et al., 1998). For example, evidence of ties with higher structural formality may be found with contracts, board memberships and alliances. In addition, Chung-hoon et al. (2005, 2007) found that enduring donor–recipient relations can reﬂect high levels of structural formality. Thus structural formality, which may be related to tie loyalty, governance, organizational control, and inﬂuence on decision making, represents an important dimension of the structure of dyadic connection. These four structures of dyadic connection represent strategic dimensions of ties which provide variation within and between ties that can facilitate dyadic evolution. While network literature addresses the underlying theoretical concepts represented in each of these dyadic connection structures, it is important to also recognize these dyadic connection structures as representing different dimensions of tie structure demonstrating dyadic multi-dimensionality and therefore another source of potential tie variation. Structures of Dyadic Interaction. The structure of dyadic ties is also evident in the multi-dimensional interaction processes of the two network actors, which explain how they exchange tie content. Network ties differ along several dimensions of dyadic interaction, which can explain how and why variation occurs both within and between ties as well as over time. Dimensions of dyadic interaction structures include frequency and duration of the interaction, ease of the interaction, effort put forth for the dyadic partner and perceived quality of the interaction (Hite, 2003). Critical dimensions are also found in communication and negotiation processes such as styles, media, timing, and location. The culture of the tie also represents multiple dimensions of dyadic interaction, in terms of values, norms, and expectations. With ﬁrm-level actors, dyadic interaction processes may

144

JULIE M. HITE

also include level of codiﬁed information regarding exchange, continuation of interaction across multiple agents, and diversity (homophily vs. heterogeneity) of institutional knowledge and capabilities. The multi-dimensionality within structures of dyadic interaction informs how dyadic ties can create strategic advantages. For both individual and ﬁrm actors, this aspect of dyadic structure would clearly contribute to variation in the strategic capabilities of network ties and therefore may create differential network advantages (Borgatti & Cross, 2003; Sparrowe & Liden, 1997). Combining dyadic multi-dimensionality of both content and structure allows for even richer examination of the variation that can exist within and between network ties and suggests that network actors may be able to strategically seek dyadic change along these multiple dimensions to enhance the nature and value of their ties. The accumulation of dyadic changes within these dimensions could result over time in the evolution of an actor’s network tie from one state to another. Multi-Dimensionality of Dyadic Governance Dyadic governance represents a critical aspect of the dyadic context of strategic ties and also clearly reﬂects their multi-dimensionality. Two key dimensions of dyadic governance are found in the type or form of different governance mechanisms. The literature suggests that network ties are generally governed by market or contractual mechanisms (Bradach & Eccles, 1991; Williamson, 1979) or by relational mechanisms (Dyer & Singh, 1998; Powell, 1990; Zaheer & Venkataraman, 1995). Both market governance and relational governance occur in many forms, adding further to the multi-dimensionality involved. For example, relational governance may be described in terms of the type, extent and mutuality of trust within the dyadic tie. Speciﬁcally, concepts of relational embeddedness suggest that relational governance can be based on the potential strengths of three different types of trust: personal, competency, and social (Hite, 2003). In Hite’s (2003) typology of relational embeddedness model, full embeddedness would provide the strategic advantage of having all three types of trust, thus beneﬁting from more effective governance than any one type of trust alone may provide. Network ties may also rely on multiple or redundant governance mechanisms that represent combinations of both market and relational governance (Poppo & Zenger, 2002; Rowley, Behrens, & Krackhardt, 2000). For example, while under relational governance trust may replace formal contracting in providing oversight for the dyadic partners (Bradach & Eccles, 1991; Dyer & Singh, 1998; Zaheer & Venkataraman, 1995). Zaheer

The Role of Dyadic Multi-Dimensionality

145

et al. (1998) demonstrate that trust may be initially based on a social relationship and later be enhanced by more formal contracting. Poppo and Zenger (2002) also suggest that even in the presence of relational governance, dyadic governance may beneﬁt from contracts as mechanisms regulating information and communication. The proposition that both types of governance mechanisms can operate within a given dyad – in different degrees, at different times, and for different purposes – clearly suggests that dyadic variation may result from the simultaneous existence of multiple dimensions of dyadic governance. Thus the multi-dimensionality of dyadic governance generates the necessary dyadic variation to allow network actors to be strategic in their development and adaptation of effective dyadic tie governance.

Dyadic Life Cycle The dyadic life cycle represents yet another dimension of the dyadic context. Mirroring the organizational life cycle, the evolution of network ties can also be reﬂected in ﬁve different phases of dyadic change: (1) emergence, (2) development, (3) maintenance, (4) decay, and (5) dissolution (Amburgey & Al-Laham, 2008; Doreian, 2006, 2008). Each phase of dyadic evolution represents a different stage of development, suggests speciﬁc strategic opportunities and challenges, and reﬂects potential changes in the content, structure, and governance of ties (Doreian, 2008; Hite, 2005; Hoang & Antoncic, 2003). During the tie emergence phase, the dyadic context reﬂects the search for as well as selection and entry of ties into the network. In the tie development phase, network ties experience increased levels of change as they strengthen, increase in relational embeddedness, and align for greater strategic ﬁt. The third phase, management, represents the continual management of the dyadic equilibrium (Doreian, 2006) once the tie has developed to provide effective strategic content, structure, and governance. The fourth phase of dyadic life cycle, decay, occurs as a tie moves away from a previously more developed or functional state. However, decay is not synonymous with tie dissolution or death, which is represented by the last life cycle stage in which a tie is dropped from the network. Dyadic ties that are dissolved may represent failed emergence, failed strategic functionality, or ﬁnal devolution. Therefore dyadic multi-dimensionality is reﬂected in the life cycle stages of network ties. As network actors seek dyadic changes, these accumulative changes may provide the basis for the tie’s evolution (or de-evolution) from one stage to another.

146

JULIE M. HITE

Summary of the Dyadic-Level Context The rich multi-dimensional nature of the dyadic context is demonstrated in the diversity among network actors, the multiplexity of dyadic tie content, the forms of dyadic tie structures, the types of dyadic tie governance, and the different phases of the dyadic life cycle. Dyadic multi-dimensionality helps to explain the variation across dyads – why ties are not all the same and not all equally strategic (Bliemel & Maine, 2006; Granovetter, 1985; Hite, 2003) – as well as the variation with ties over time. While at the macro network level, dyadic ties are often treated as a ‘‘black box’’; the multidimensional nature of the dyadic context must be accounted for to provide meaningful theoretical explanations for the creation, evolution and outcomes of strategic networks. The multi-dimensional nature of the dyadic context was addressed in Granovetter’s (1973) conceptualization of strength of ties and in Granovetter’s (1985) and Uzzi’s (1996) research on embeddedness, which combine elements of both content and structure. The study of embeddedness has provided an important foundation for acknowledging the multidimensionality of the dyadic context, particularly in terms of relational or social embeddedness (e.g. Hite, 2003; Nahapiet & Ghoshal, 1998; Uzzi, 1996, 1997, 1999; Uzzi & Gillespie, 2002; Uzzi & Lancaster, 2003), structural embeddedness (e.g. Feld, 1997; Moody & White, 2003), and the combination of both (e.g. Chung-hoon et al., 2007; Moran, 2005; Rowley et al., 2000). Yet in strategic network literature, both strength of ties and embeddedness are still generally treated as dichotomous variables (Evald, Klyver, & Svendsen, 2006; Hite, 2003), generally aggregating dyadic multidimensionality to create dichotomous labels of strong or weak, embedded or not embedded. In support of a better understanding of dyadic multidimensionality, Wasserman and Faust (1994) caution against aggregating multiple tie relations, as doing so risks the loss of potentially richer strategic explanations. Dyadic ties are formally proposed to be multi-dimensional in Hite’s (2003) typology of relational embeddedness. This typology integrates content (e.g. multiple relations), structure (e.g. asymmetric exchange, dyadic processes, structural embeddedness), and governance (e.g. types of trust) to capture a multi-dimensional view of the nature and range of relational embeddedness among dyadic ties. The typology allows for the examination of the range of multi-dimensionality as well as its aggregation into a more parsimonious analysis, generating the necessary variation to move beyond a dichotomous view of relational embeddedness and capture a richer understanding of the multi-dimensionality of the dyadic context.

The Role of Dyadic Multi-Dimensionality

147

Endogenous and Exogenous Contexts While the main focus of the discussion on multi-dimensionality is on the dyadic context, the dyad also resides within multi-dimensional endogenous and exogenous contexts. The dyad’s endogenous context is represented by the patterns and processes of the larger network’s current content, structures, and governance. Within this context, dyadic multi-dimensionality can be generated by other actors and dyads in the network or the structural position of the dyad within this larger network structure (Doreian, 2008), as well as by the network’s history, purpose, and culture (Emirbayer & Goodwin, 1994; Kilduff & Tsai, 2003). The exogenous context of a dyadic tie extends beyond a network perspective to multidimensional forces in the dyad’s macro environment. This exogenous context represents a multitude of potential dimensions including the tie’s ﬁrm, strategic environment, community, and society. While these exogenous forces are often beyond the direct control of network actors, their dyadic ties are clearly affected by this external context. As with actor and dyadic contexts, the multi-dimensionality of endogenous and exogenous contexts also creates variation among network ties. Variation resulting from these macro contexts provides an important basis for the strategic network choices of actors as they seek dyadic change. These four dyadic contexts – actor, dyadic, endogenous, and exogenous – create unique settings and environments underlying dyadic multidimensionality – hence variation – both within and across ties. Intriguing strategic questions arise from the interaction of these multiple contexts, such that for any given dyad these multiple potential contexts are simultaneously at play. The dyadic complexity is increased as two actors, along with the tie itself, engage in one dyadic dance across these multiple contexts. Examining and accounting for these four contexts, particularly the dyadic context, provides an informative framework for strategically understanding and managing within the multi-dimensional nature of dyadic ties. The resulting variation may facilitate understanding of which types of evolution of strategic network ties may occur – as well as when and under what conditions.

THE DUAL ROLES OF DYADIC CONTEXTS The four dyadic contexts occupy contrasting roles in the evolution of network ties. As indicated in Fig. 1, they provide sources of strategic catalysts that can motivate network actors to seek dyadic change. However,

148

JULIE M. HITE

these same contexts also generate sources of constraint that bound the agency or autonomy of the network actor in pursuing dyadic change. Thus understanding the evolution of network ties requires clear speciﬁcations of the dyadic contexts within which the tie is operating to better understand the sources of potential strategic catalysts and bounded agency.

Strategic Catalysts Network theory and strategic network research increasingly assume that network actors have agency and can act with strategic intent to accomplish their purposes (Borgatti & Foster, 2003; Kilduff & Tsai, 2003; Kilduff et al., 2006; Koka et al., 2006; Mizruchi, 2004). Indeed, Pillai (2006) asserts that network actors need to have a ‘‘conscious strategy of seeking advantageous position in a network and preserving the same’’ (p. 140), and Kilduff and Tsai (2003) argue that networks are ‘‘responsive to the ongoing aspirations and efforts of individual actors’’ (p. 114). Such strategic network action presumes that network actors know strategic goals (Kilduff & Tsai, 2003), the current state of the network, and the ideal network conﬁgurations needed to achieve the goals. Under this scenario of agency, choice, control, and strategic intent, network actors purposefully build new ties and develop and leverage existing ties to accomplish their strategic purposes and beneﬁt ﬁrm performance (Hite, 2005; Hite & Hesterly, 2001; Kilduff & Tsai, 2003). Network ties can serve strategic purposes as a result of their content, structure, and governance. The dyadic multi-dimensionality of ties generates the variation that can provide strategic network actors with a wide range of opportunities to craft, create, manage, and adapt – under bounded agency – different combinations of tie content, structures, and governance to pursue their strategic purposes. From the relational view, their multi-dimensional nature provides ties with potential strategic advantages and relation-speciﬁc capabilities (Dyer & Singh, 1998). When network actors take strategic action to change dyadic ties, this action is often in response to a catalyst. A variety of catalysts exist within each of the dyadic contexts that motivate network actors to take strategic action, providing foundations for explaining why evolution of network ties may occur. Strategic Catalysts from the Actor-Level Context Individual network actors are themselves complex systems that can generate catalysts for dyadic change as they continuously adapt in response to each

The Role of Dyadic Multi-Dimensionality

149

other to create a sustainable dyadic tie (Kilduff & Tsai, 2003; Kilduff et al., 2006; Krackhardt & Kilduff, 1990). Actor-level catalysts, which may include an individual’s attitudes, preferences, emotions, and self-interest (Kilduff et al., 2006; Snijders, 2005b), may motivate change that is selfserving, in whole or in part, that may not necessarily be strategic for the ﬁrm (Blyler & Coff, 2003; Doreian, 2006). For ﬁrm-level actors, catalysts for dyadic change can include the ﬁrm’s strategic goals and needs as well as evaluation of dyadic and organizational outcomes. Strategically seeking to meet the changing needs and challenges across the ﬁrm’s life cycle stage may result in dyadic change (Elfring & Hulsink, 2007; Evald et al., 2006; Hite & Hesterly, 2001; Larson & Starr, 1993). For example, organizational survival and growth goals may motivate network actors to create and adapt ties ‘‘to align the mix of weak and strong ties to the changing needs of a venture over time’’ (Elfring & Hulsink, 2007, p. 1853). Organizational mission, goals, strategies, and performance present critical catalysts that motivate network actors to engage in goal-directed dyadic change (Doreian, 2006; Kilduff & Tsai, 2003). For example, the ﬁrm life cycle stage represents an important dimension of the actor-level context; however, it is the speciﬁc performance-driven goals and needs that arise within this context that represent the catalysts. Goals guide the pro-active strategic design and adaptation of the content, structure, and governance of dyadic ties to increase their value and beneﬁt for the ﬁrm (Elfring & Hulsink, 2007; Rowley et al., 2000; Snijders, 2005b). For example, given that dyadic ties provide access to resources, opportunities, and legitimacy (Elfring & Hulsink, 2007; Gulati & Gargiulo, 1999; Hite & Hesterly, 2001) and that different types of ties provide different strategic advantages (Elfring & Hulsink, 2007; Pillai, 2006), strategic network actors can intentionally seek to develop the content, structure, and governance of their dyadic ties to best obtain these beneﬁts (Evald et al., 2006; Hite, 2005; Rowley et al., 2000). The strategic goals of the ﬁrm can act as the catalyst that motivates dyadic change processes that result in network patterns similar to the evolution, renewal, and revolution patterns found at the egocentric level by Elfring and Hulsink (2007). Strategic Catalysts from the Dyadic-Level Context The dyadic context also generates catalysts that motivate network actors to seek dyadic change. Given that a tie represents a negotiated relationship between two actors, these actors may seek dyadic change when the nature of the tie changes along dimensions of content, structure, governance, or life cycle. If tie content is no longer sufﬁcient, an actor may seek to increase

150

JULIE M. HITE

the ﬂow of content, change the type of content, weaken the tie, or acquire a supplementary tie. If the tie structure shifts, weakens, or becomes too asymmetric, actors may seek to adjust tie structure to optimize dyadic outcomes. If a tie requires additional governance, an actor may add mechanisms to protect the exchange. Another strategic catalyst from the dyadic context may occur if a tie results in detrimental or negative consequences for ﬁrm performance (LaBianca & Brass, 2006). These results can motivate an actor or a ﬁrm (as a network actor) to strategically drop the tie or allow it to weaken and loosen. Thus change within the multiple dimensions of the dyad itself represents a potential source of catalysts for dyadic change. Strategic Catalysts from Endogenous and Exogenous Contexts Endogenous network changes and inﬂuence create catalysts that can motivate actors to seek dyadic change. Structural changes are continuously occurring within the larger network due to dyadic changes among an actor’s other ties or indirect ties. Changes in existence or strength of other ties within the larger network can affect the extent of a given tie’s local cohesion or structural embeddedness, as well as the tie’s reach and centrality position. Changes in the content, structure, or governance of other ties in the network may also have ripple effects that require dyadic adaptation by a network actor in order to align with the new network environment. Given the interdependent nature of ties, potential catalysts for dyadic change also result from the inﬂuence of the network’s cultural norms and expectations and from the extent of relational embeddedness and social capital within the network. The exogenous environment also generates catalysts that motivate actors to seek dyadic change: for example the ﬁrm’s environment, economic and socio/political climates, technological innovations, and industry events (Koka et al., 2006; Madhavan, Koka, & Prescott, 1998). Madhavan et al. (1998) propose that such environmental effects can reinforce or loosen structure at the network level; given that network change initially occurs with dyadic-level change, the ﬁrst effects of exogenous shocks to the network system would likely be to reinforce or loosen dyadic ties (Kilduff et al., 2006). Emirbayer and Goodwin (1994) describe exogenous catalysts from the environment as ‘‘push’’ forces that motivate the adaptation of network ties to the ﬁrm’s environment. Lastly, serendipity in the external environment also acts as a catalyst that presents both motivations and opportunities for dyadic change (Kilduff & Tsai, 2003).

The Role of Dyadic Multi-Dimensionality

151

With the focus on strategic networks, catalysts for dyadic change from each of the four contexts are ﬁltered through the network actor who chooses whether and how to respond. Continual strategic response and adaptation to catalysts from dyadic contexts may provide useful explanations for many of the processes involved in evolution of network ties. In their response to catalysts, network actors intentionally, within bounded agency, craft their network action to change dyadic ties to better ﬁt and respond to the ﬁrm’s organizational needs, goals, strategies, and environment. These strategic actions may then also serve as catalysts for future network action. Bounded Agency Another underlying assumption of network theory is that actors navigate a careful balance between agency and constraint as they pursue intended network strategies. Thus strategic network research often includes issues of actor agency in creation, examination, and evolution of network structures (Emirbayer & Goodwin, 1994; Mizruchi, 2004; Podolny & Page, 1998). Discussions of network actor agency generally fall along the continuum of agency (proactive choice) and path dependence (constraint) (Stinchcombe, 1965). Actor agency, at one end, suggests that the network actor has the freedom to choose to make network changes. Path dependence, at the other end, suggests that the actor’s choice is fully constrained. Within this continuum is the middle ground of bounded agency (Evans, 2002) that recognizes that the realities of both choice and constraint raise ‘‘a number of boundaries or barriers that circumscribe and sometimes prevent the expression of agency’’ (Evans, 2002, p. 262). Bounded agency implies intentional, strategic action constrained by real limits. As such, bounded agency aligns with a model of modiﬁed rational choice of individual action (Doreian, 2008; Friedman & McAdam, 1992) and assumes that action is multiply determined (Emirbayer & Goodwin, 1994) as well as multiply-constrained (Kim, Oh, & Swaminathan, 2006). While network actors make choices with strategic intent, they do so while strategically navigating within bounded agency. The evolution of network ties results from actions taken by network actors who navigate within real limitations of bounded agency; therefore network actors may be constrained in their ability to create, maintain, change, or drop their dyadic ties (Gulati, 1995b; Podolny & Page, 1998). Paradoxically, bounded agency may also facilitate focused, relevant action due to both push and pull factors that can enhance motivation to network action as strongly as constraints may restrict

152

JULIE M. HITE

action (Emirbayer & Goodwin, 1994). Bounded agency’s push factors compel and require network action, such as tie creation, whereas its pull factors encourage and facilitate more focused network action. In both cases, actor agency is bounded by factors that lead toward an action rather than merely limit expression of choice. Factors that may bound the agency of network actors, by either limiting or focusing action, can be identiﬁed within each of the four dyadic contexts. Bounded Agency from the Actor-Level Context Whether a network actor functions as an individual, an individual acting as an agent for the ﬁrm, or the ﬁrm itself, actor-level constraints contribute to bounded agency. Rational choice aligned with strategic or purposive rationality suggests that network actors are maximizing, utility while strategically calculating the costs and beneﬁts of various network actions (Doreian, 2008; Emirbayer & Goodwin, 1994; Friedman & McAdam, 1992). This rational choice model suggests a source of actor-level constraint in that the values of network actors guide their needs, preferences, goals, and ultimately their criteria for rationality. Thus even when a speciﬁc network action may clearly provide speciﬁc beneﬁts, individuals actors may constrain their own actions due to their subjective super-ordinate values and previous commitments. In addition to this subjective rationality, bounded rationality explains actor-level constraints that bound the agency of network actors due to ambiguity, absorptive capacity, and perception (Kilduff et al., 2006; Simon, 1972). Actors’ perceptions, a key source of bounded agency, are inﬂuenced by attributes, cognitions, personalities, histories, and preferences. Network literature has examined actors’ perceptions of their networks, including the factors associated with network perceptions, the accuracy of these perceptions, and changes in network perceptions as actors learn (e.g. Casciaro, 1998; Hite et al., 2006; Krackhardt & Kilduff, 2002; Mehra, Smith, Dixon, & Robertson, 2006). In alignment with social selection models, understanding the evolution of network ties cannot be complete without allowing and accounting for such actor-level differences and their effects on network constraint. For example, how do network actors perceive the network structure within which they are situated or anticipate the effects of network changes they initiate? From a structural perspective, network actors can best see their own direct networks and likely have fair estimates of their egocentric networks, yet may have only a sketchy, at best, understanding of the whole network. Thus network actor agency is bounded by actors’ perceptions of the network such that they can only attempt to

The Role of Dyadic Multi-Dimensionality

153

inﬂuence the larger network from their own myopic view. Network actors have ‘‘subjective perceptions of the structures they have to negotiate which affect how they act’’ (Evans, 2002, p. 252), thus agency is bounded by the perceptual horizon of the network actor (van Liere, Koppius, & Vervest, 2008). Evans (2002) also points out that actors have ‘‘subjectively-perceived frames for action and decision,’’ which are ‘‘shaped by the experiences of the past, the chances present in the current moment and the perceptions of possible futures’’ (p. 262). As a result, the perceptions of network actors contribute to bounded rationality which, in turn, bounds their agency in strategic network action. Another critical source of bounded agency stems from the agency of the dyadic partner. Actors may be limited in their ability to pursue strategic network action due to the choices that result from the subjective and bounded rationality of their dyadic partner. Navigating bounded agency due to the actor-level context may require an actor to have an accurate understanding of both self and dyadic partner. However, actor-level constraints alone provide an underdetermined explanation for bounded agency. Bounded Agency from the Dyadic-Level Context From the dyadic-level context, the characteristics and components of the pre-existing tie can create constraints on strategic action and therefore affect future dyadic change and development (Greve & Salaff, 2003; Gulati, 1995a; Gulati & Gargiulo, 1999; Hite, 2005). For example, strategic interdependence between two ﬁrms and the social context resulting from prior alliances can affect decisions regarding change within the network tie (Gulati, 1995b). The current state of the tie may also bound an actor’s agency if the tie is unable to change to serve a needed purpose, to provide necessary content, or to be effectively governed. The current dyadic life cycle stage can also create constraints in the time necessary to further develop a tie: that is, a tie in an early emergent or developmental stage may not be ready or able to move to a more mature stage. Within the dyadic-level context, given the interdependence of actors and actions (Wasserman & Galaskiewicz, 1994), navigating bounded agency to make dyadic change may also require negotiated network action between the two network actors (Snijders, 2005a). While a partner’s choice may create an actor-level constraint, the inability to effectively communicate or negotiate with the partner creates dyadic-level constraints on strategic network action. In the case of relationally embedded ties, dyadic-level constraints to network action have been referred to as one of the dark

154

JULIE M. HITE

sides of embeddedness (Granovetter, 1985); such restraints may occur, for example, when tie maintenance becomes more highly valued than the potential dyadic change. Bounded Agency from the Endogenous and Exogenous Contexts Given that dyadic ties are situated within a larger network and possibly within multiple networks, based on the assumption that the network structure of ties shapes action (Evans, 2002; Wellman, 1983), it may be concluded that endogenous constraints that bound actor agency result from the current network structures and processes. Bounded agency results from being embedded and involved in a patterned and structured network of ties (Emirbayer & Goodwin, 1994; Wellman, 1983, 1988). Endogenous network constraints can be framed from micro, macro, and historical perspectives (Burt, 1987; Contractor et al., 2006; Kilduff & Tsai, 2003). Micro-level endogenous constraints result from the direct interactions of network actors with other actors and ties within their egocentric networks. At this level, the network actors in a dyad often need to take into account other local network ties and the agency of other actors, particularly under conditions of high structural or relational embeddedness or under both conditions (Chung-hoon et al., 2005, 2007; Granovetter, 1985; Moran, 2005; Rowley et al., 2000). The direct and egocentric networks of the dyadic actors may also create further constraints due to overembeddedness (Uzzi, 1996) or to a high ratio of relationally embedded ties. Such micro-level endogenous constraints on dyadic change occur due to other nearby ties that bound the agency of network actors. The macro-level network also generates endogenous constraints resulting from the structural conﬁgurations of the broader network, such as cohesion, structural holes, and centralities, as well as from the culture and evolution within the network itself (e.g. Burt, 1987; Gulati, Nohria, & Zaheer, 2000b). However, these macro-level constraints may be harder for a network actor to detect as they may be beyond the actor’s perception or network horizon (van Liere et al., 2008). Therefore, a network actor may not acknowledge these constraints or take them into strategic account. As a result, the inﬂuence of endogenous constraints at the macro level may be less likely to be identiﬁed as such and therefore be less likely to functionally bound actors’ agency than constraints at the micronetwork level. In networks where actors are assumed to behave strategically, however, one of the challenges of explaining bounded agency requires analyses to ‘‘move the focus from structures onto individuals without losing the perspective of structuration’’ (Evans, 2002, p. 252). Thus while

The Role of Dyadic Multi-Dimensionality

155

network theory posits endogenous constraints on actors due to the whole network structure, the question remains as to how and why macro level constraints may actually bound network actors’ agency in seeking dyadic change. The current micro-macro network structure results from the aggregated historical evolution of network ties and the previous bounded agency of network actors. Ironically, while the structural evolution may have been inﬂuenced in the past by actor agency, the current network structure presents a critical source of endogenous constraint that now bounds actors’ agency (Kilduff & Tsai, 2003; Snijders, 2005b). In this vein, Emirbayer and Goodwin (1994) suggest ‘‘that intentional, creative human action serves in part to constitute those very social networks that so powerfully constrain actors in turn’’ (p. 1413), suggesting a co-evolution of actor agency and network structure (Snijders, Steglich, & Schweinberger, 2007). For example, Gulati (1995a, 1995b) found that the pre-existing alliance network structure inﬂuences the new alliances that network actors will seek. As a result, the evolution of network ties is, by nature, a path-dependent process in which network actors are inﬂuenced by the past, which they cannot undo. Yet in this process they are only ‘‘inﬂuenced but not determined by structures’’ (Evans, 2002, p. 248), so they retain current bounded agency to determine their future actions. Bounded agency also results from exogenous constraints on network action that reside outside of or beyond the actor or the network itself, including ‘‘additional networks of relations among the actors’’ (Contractor et al., 2006, p. 690) (see also Emirbayer & Goodwin, 1994). Consideration of exogenous constraints assumes that additional multiple and complex sets of intertwining factors inﬂuence and constrain network actors at multiple levels (Contractor et al., 2006). At the exogenous level, network actors may be constrained by macro inﬂuences such as industry events (Madhavan et al., 1998), economics (Neuman, Davis, & Mizruchi, 2008), transportation, communication, and physical geography. For example, Barney (2004) argues for ‘‘a geographically-embedded view of relations’’ (p. 325). Many exogenous network constraints can also be framed under the broad heading of cultural inﬂuences (Emirbayer & Goodwin, 1994). For example, strategic networks have been examined within and across a variety of cultures, including national cultures (e.g. Barney, 2004; Faust, Entwisle, & Rindfuss, 2003; Hite et al., 2006), industry cultures (e.g. Baum, Calabrese, & Silverman, 2000; Chung-hoon et al., 2005; Dyer & Singh, 1998; Lockie, 2006), and political cultures (e.g. Bryson & Kelley, 1978; Stevenson & Greenberg, 2000).

156

JULIE M. HITE

In summary, each of the four dyadic contexts both motivates strategic action and at the same time generates potential constraints on network action. Network action is indeed a multiply determined, socially situated process (Emirbayer & Goodwin, 1994; Evans, 2002) that is ‘‘not driven exclusively by human agency, but rather is deeply structured as well by several other ‘environments’ of action’’ (Emirbayer & Goodwin, 1994, p. 1443). Snijders (2005a) suggests four potential types of action within such bounded agency: (1) ineffective action when the constraints are too large, (2) inﬂuential action when the actor changes the constraint, (3) navigated action when the actor ﬁnds a different path to a similar end, and (4) negotiated action when the actor negotiates with the constraint for mutual beneﬁt. Therefore, strategic network actors can and do intentionality navigate within the constraints of bounded agency to shape social structures and action to beneﬁt the ﬁrm (Emirbayer & Goodwin, 1994; Emirbayer & Mische, 1998).

STRATEGIC PROCESSES OF DYADIC CHANGE The evolution of strategic network ties is grounded in the concept of dyadic multi-dimensionality and the resulting variation within and between ties. Building upon this variation, network actors may use strategic processes to shape the dyadic change and the evolution of their network ties (Amburgey & Al-Laham, 2008; Doreian & Stokman, 1997). In analyzing strategic change processes, one must assume the concept of dynamic stability in networks in which ‘‘actors build from the stability of networks to incorporate network change’’ (Kilduff et al., 2006, p. 1037) (see also Robins & Pattison, 2005). Therefore strategic processes of dyadic change include both a point of current stability from which the network actor can initiate change and the variation inherent in dyadic multi-dimensionality. Strategic processes of laying, leveraging, and timing contribute to explanations of evolution of network ties, speciﬁcally in terms of dyadic content, structure, governance and life cycle.

Layering Processes Layering is a process in which evolution occurs as new layers are added to network ties, reﬂecting Kilduff et al.’s (2006) description of networks ‘‘as layer upon layer of relations, built up over time’’ (p. 1039). Layering

The Role of Dyadic Multi-Dimensionality

157

processes align with evolutionary processes of selection and retention (McKelvey & Aldrich, 1983) as network ties are built ‘‘brick by brick,’’ strategically selecting and retaining layers to build effective ties. Layering is analogous to adding new strands to rope: in effect changing and potentially strengthening the tie with each new strand. Uzzi’s (1996) description of embedded ties as encompassing problem solving, ﬁne-grained information transfer, and trust suggests such multi-layered ties. Hite’s (2003) typology of relational embeddedness proposes layering components of personal relationship, dyadic interaction, and social capital, positing that layering different types of trust may provide more effective relational governance. Examples of layering are found in homophily, knowledge sharing, and asymmetric exchange. First, homophily is, in essence, layering perceived similarities onto the tie (Amburgey & Al-Laham, 2008; McPherson, SmithLovin, & Cook, 2001), contributing to the development of the tie. Second, knowledge sharing layers new shared knowledge on the tie as dyadic partners learn from and problem solve with each other. Third, asymmetric exchange generates layering as dyadic partners alternately provide content to the tie and these one-way transfers build layers over time. Layering implies strategic implications for the evolution of strategic network ties. Layering increases multiplexity as one content relation is layered upon another, suggesting that multiplexity itself evolves. Larson and Starr (1993) use this multiplexity argument early in the organizational network literature to explain that emerging organizations add business exchange relations to previously existing personal relations and vice versa. Hite (2003) suggests that the multiplexity of relationally embedded dyads differs in both content and extent, as not all relationally embedded ties have the same types or number of relational components layered within the tie. The order in which layers are added may also have strategic implications, given Moody et al.’s (2005) temporal process of relational sequence. For example, different layering paths may occur as actors add relational components to network ties, and these paths have strategic governance implications as each component is associated with a speciﬁc type of trust (Hite, 2003, 2005). Thus the order in which dyadic layers are added suggests an associated order of trust development, in that different types of trust may be available within the tie at various stages of development. At the opposite end, de-layering can also occur if dimensions are removed or dropped from the tie, analogous to removing a strand from a rope or a brick from a wall. Hite (2005) notes examples in which relational attributes were removed or no longer available within the dyad. Layering processes help explain the evolution of network ties,

158

JULIE M. HITE

given dyadic multi-dimensionality, in that the content, structure, and governance of ties can change over time as network actors add or remove layers from the tie.

Leveraging Processes Network actors may strategically facilitate the evolution of dyadic ties through leveraging processes. Leverage, as a principle of physics, indicates that a lever (strategic action) can hinge upon a fulcrum (pre-existing element) to create movement or change. Thus leveraging is an exponential process in which actors make use of pre-existing elements to facilitate dyadic change that may not be possible through additive layering processes alone. For example, strategic network literature addresses the leveraging of resources, tacit knowledge, and technical competencies (e.g. Collins & Hitt, 2006; Hansen & Lovas, 2004; Wilson & Appaiah-Kubi, 2002). From a network perspective, actors can leverage both the dyadic tie and other network ties to facilitate dyadic change. Leveraging the Dyadic Tie Dyadic multi-dimensionality provides a rich variety of fulcrum sources for leveraging tie development. Thus one dimension of the tie is used to build another. In Hite (2005), the development of relational embeddedness demonstrated four such leveraging processes facilitating dyadic change (Hite, 2005). Network actors leveraged personal affect to create ease of dyadic interaction and to motivate brokering. They leveraged interaction effort to develop increased social capital in terms of obligations (e.g., norms, expectations, and reciprocity) and leveraged personal sociality to create social capital in terms of brokering opportunities. These leveraging processes built new dyadic dimensions as a result of pre-existing dimensions. Thus network ties may evolve to become increasingly relationally embedded and to provide greater potential for future leveraging processes. Leveraging processes can be seen in homophily and reciprocity. Once perceptions of homophily are layered within the tie, these perceptions can be leveraged to facilitate multiple dimensions within the dyad, such as personal relationship, more effective interaction, and social capital. Levering through reciprocity is seen in asymmetric exchange. While layering provides the ﬁrst network ﬂow in one direction, the dyadic partner responds to and leverages off the ﬁrst one-way transfer to reciprocate the exchange. This is a leverage process because reciprocity assumes the presence of a ﬁrst transfer. Therefore, the dyadic partner’s interaction leverages from the initial ﬂow.

The Role of Dyadic Multi-Dimensionality

159

With continued asymmetric reciprocity, the search for the ever-elusive symmetry then functions as the fuel in the dyadic engine to maintain the tie and keep exchanges occurring. Over time, the leverage process of reciprocity increases the duration of interactions and the potential development of additional dyadic dimensions. Leveraging processes can also utilize multiplexity in which one tie can serve multiple purposes and, therefore, a single tie may co-exist within multiple networks (Hite et al., 2005). For example, a tie may exist in a technological-innovation network, a resource-sharing network, and a friendship network. Network actors can leverage this dyadic multiplexity across different networks by building from an existing tie to develop a new network relation. For example, a network actor can bring a dyadic tie from a supplier network into an innovation network to help with design issues. This inclusion within another network leverages the pre-existing tie to create new relations and content ﬂows within the dyad, thus increasing the multiplexity of the tie. Multiplexity also explains leveraging of latent dyadic ties (Hite, 2003). For strategic networks, latent ties exist between actors who are not currently interacting for business purposes, although the tie may still have dimensions of social capital and personal relationship. For example, if a dyadic partner has once been active in the actor’s network, the network actor can tap into this latent tie to bring the dyadic partner into the ﬁrm’s current network. The actor may also bring this latent tie into a new network for other business purposes. This leveraging process is seen in Gulati’s (1995a, 1995b) ﬁnding that ﬁrms tend to create alliances with pre-existing dyadic ties. Entrepreneurs leverage their pre-existing ties from other networks, such as family, friends, and former work colleagues, by bringing a new purpose into the tie such that the dyad now serves a business function for the emerging ﬁrm (Hite & Hesterly, 2001; Larson & Starr, 1993). Leveraging latent ties is also suggested by the Organizational-Donor Integration Model (Chunghoon et al., 2005, 2007), which proposes leveraging strategies for the development of enduring external donor relations in higher education institutions. This model suggests that pre-existing structural or relational embeddedness can be leveraged through integration mechanisms to cause donor ties to become more relationally embedded, have more structural formality, or both. Additional types and extents of trust may also result from strategically leveraging pre-existing dyadic dimensions to further develop the tie (Hite, 2003; Jones & George, 1998). Therefore, dyadic multidimensionality provides critical sources of leverage that can facilitate further development and evolution of network ties.

160

JULIE M. HITE

Leveraging Other Network Ties and Resources Leveraging other network ties can also facilitate dyadic tie evolution. For example, brokering provides leverage as actors build from other stable ties to create or strengthen ties (Podolny & Page, 1998; Robins & Pattison, 2005). During ﬁrm emergence, network actors often leverage dyadic ties in pre-existing networks to broker new dyadic ties (Elfring & Hulsink, 2007; Hite, 2003, 2005; Hite & Hesterly, 2001; Hite, Hite, Rew, Jacob, & Mugimu, 2005; Larson & Starr, 1993). An actor’s strong ties may ‘‘provide the link to the wider social context and act as a mechanism to invoke weak ties’’ (Jack, 2005, p. 1234). Leveraging these other ties can also facilitate the transfer of legitimacy and social trust from one dyad to another (Elfring & Hulsink, 2007; Hesterly, Jones, & Madhok, 1998; Hite, 2003). However, not all network actors may have pre-existing network ties that can be effectively leveraged. For example, in ‘‘rapidly growing industries like biotechnology, where so many new organizations enter each year without ties’’ (Podolny & Page, 1998, p. 70), an actor’s current ties may not be relevant to the business of the new ﬁrm. Under these conditions network actors must leverage their own pre-existing resources, knowledge, capabilities, and social skills to develop dyadic ties. For example, entrepreneurs can leverage access to public resources such as the Internet and industry lists, as well as take advantage of conferences, workshops, and networking events to meet new potential dyadic partners (Elfring & Hulsink, 2007). They can facilitate both the creation and evolution of dyadic ties by providing information and resources, making interaction easier, putting out extra effort, and creating socializing and brokering opportunities for the dyadic partner (Hite, 2005). The ability to leverage other network ties and resources is exponentially related to the actor’s number of ties and extent of resources: that is, a greater number of ties and resources increases the leveraging options and opportunities available to the network actor. With only a few ties, the network actor would have to rely repeatedly on the same few ties. With only a few resources, the network actor would have less to offer dyadic partners. Thus initially an actor’s capability, availability, and willingness to participate in leveraging strategies may be limited. As network actors increase their ties and resources, it becomes easier to participate in leveraging processes to create and develop dyadic ties. Dyadic leveraging processes necessarily assume a temporal perspective, given that a dimension existing in one time period is used to develop a dimension in a later time period. A further assumption is that ties become layered with different

The Role of Dyadic Multi-Dimensionality

161

dimensions and that these dimensions can be used to facilitate additional dimensions. In turn, these processes serve to change strategic dimensions of network ties and facilitate their evolution.

Temporal Processes An evolutionary approach to strategic network change necessarily acknowledges cascading change over time due to temporal processes. Network ties can and do change along dimensions of their content, structure, and governance through the accumulation of multiple aggregated changes (Snijders, 2005b). Moody et al. (2005) describe four temporal dimensions necessary for understanding dynamic change in network relations: pace, sequence, concurrency, and types of ties. Network actors may be able to strategically affect dyadic change by adapting these temporal processes. First, relational pace, which involves the rate of change within a tie, is described in terms of ‘‘levels (fast, slow), change (accelerating, decelerating), or stability (cascades, jumps and starts, etc.)’’ (Moody et al., 2005, p. 1209). Actors may be able to manage the relational pace of dyadic interaction and change. For example, increasing interaction frequency strengthens ties (Granovetter, 1973, 1985; Hite, 2003), and changing the timing of layering and leveraging processes may affect the rate of dyadic change. The concept of managing relational pace may also shed light on punctuated equilibrium (Miller & Friesen, 1980) at a dyadic level. Second, relational sequence addresses the order of the development of relations within a dyad (e.g. Hite, 2005; Larson & Starr, 1993). Actors may strategically plan the sequence of layering and leveraging processes with the intention of managing a speciﬁc path of dyadic evolution (Hite, 2005). Third, relational concurrency assumes that different relations may overlap within a given time period yet not in others, thus temporally enriching the concept of multiple relations. As a result, actors may be able to leverage dyadic multiplexity by utilizing the temporal overlaps between multiple network relations (Lazega & Pattison, 1999). Lastly, different types of ties demonstrate these temporal interactions differently in terms of their pace, sequence, and concurrency. As a result, ‘‘when viewed in continuous time, networks may develop by spurts or build slowly and steadily, or they may reﬂect repeated ritual behaviors that mix moments of order and chaos’’ (Moody et al., 2005, p. 1209). These temporal processes inform the dynamic stability of networks in which certain dyadic dimensions can be stable in time while others are in

162

JULIE M. HITE

a state of change (Kilduff et al., 2006). Theoretical explanations of the evolution of network ties must address why, how and under what conditions these temporal dimensions and their combinations matter in their effects on micro-level network change. Given that not all ties are the same (Hite, 2003), actors who are adept at strategically managing these temporal processes may be able to strategically customize tie differences in pace, sequence, and concurrency (Moody et al., 2005). Dyadic multi-dimensionality and the resulting tie variation are at the heart of strategic layering, leveraging, and timing processes. These processes may provide rich explanations of how strategic network actors can navigate bounded agency to seek dyadic change. Layering, leveraging, and temporal processes – and their combinations – may provide useful strategies enabling network actors to intentionally seek change within the content, structure, and governance of their network ties and thus facilitate tie evolution. Of the three processes, layering provides important but simple explanations, while leveraging and timing have greater potential for richer theoretical explanations. Theoretical explanations of the evolution of network ties need to carefully balance the role of dyadic multi-dimensionality with the complexities of temporal change while providing realistic strategic implications within both the control and bounded agency of strategic network actors. Both the endogenous macro-network system and the external environment can be affected by dyadic level change. In turn, the resulting endogenous and exogenous inﬂuences affect the future context of the tie, and the stage is set for continued, dynamic dyadic evolution.

CONCLUSION Understanding the dynamic nature of network ties requires a richer set of theories and constructs than used in historically static network analysis. Additional theoretical richness for examining the mechanisms and processes of dyadic network change can be found in the use of multiple theoretical perspectives (Contractor et al., 2006; Doreian, 2006; Monge & Contractor, 2001). Explanations of the evolution of network ties need to address dyadic multi-dimensionality and examine ways that actors can strategically navigate bounded rationality to facilitate dyadic change (e.g. content, structure, governance) to better serve their strategic purposes. Given that the evolution of network ties involves a relational change between two network actors, explanations of dyadic change must recognize that such change consists of and occurs through ‘‘streaming relational events y [that]

The Role of Dyadic Multi-Dimensionality

163

unfold as a continuous social process’’ (Moody et al., 2005, p. 1210). Thus the evolution of strategic network ties across the dyadic life cycle is a dynamic process in which network actors navigate simultaneous contexts and bounded agency to facilitate strategic action and affect change in network ties. The study of strategic networks requires understanding the dynamic nature of dyadic multi-dimensionality and the potential systemic effects of change within the micro-level network. The micro-level network is not simply a black box; rather, its multi-dimensionality provides the critical variation that enables strategic network actors to facilitate the evolution of dyadic content, structure, and governance to create systemic change within strategic networks. For example, the evolution of the strength of ties and relational embeddedness builds upon and results from the inherent multidimensionality of dyadic ties (Hite, 2003; Kenis & Knoke, 2002; Larson & Starr, 1993; Uzzi, 1999). Dyadic multi-dimensionality challenges tacit assumptions that all actors or all ties are the same. Rather, accounting for variation within and between dyadic ties provides critical foundations for explaining the resulting variation in network structures, actor positions, and strategic outcomes. Dyadic multi-dimensionality represents an important aspect of how strategic network choice can inﬂuence dyadic evolution as network actors seek within dyadic variation to select, develop, and retain critical dimensions of network ties. This chapter argues that the multi-dimensional nature of dyadic ties matters in understanding the evolution of strategic network ties. The multi-dimensionality within the dyadic contexts – actor-level, dyadic-level, endogenous, and exogenous – both generates strategic catalysts to motivate actors to seek dyadic change and creates forces that bound agency and thereby constrain network action. Given the need to navigate the inevitable bounded agency, three strategic processes were highlighted through which actors can facilitate strategic dyadic change. Any discussion of evolution or change from one state to another must be founded on the elements that can change and the factors that can motivate change. Dyadic multidimensionality clearly encompasses multiple contexts and dimensions of network ties, particularly in terms of dyadic content, structure, governance, and life cycle stage. Accounting for dyadic multi-dimensionality enhances explanations for how and why strategic advantages may be distributed differentially across different network ties, how dyadic network structures may evolve, and how both strategic catalysts and bounded agency may shape the strategic action of network actors to inﬂuence the evolution of their dyadic ties and network structures.

164

JULIE M. HITE

Future strategic network research must account for dyadic multidimensionality to better inform the evolution of network ties. In particular, the multi-dimensional nature of network ties stands to inform how and why network ties can provide speciﬁc strategic advantages in terms of access to partners, information, innovation, and resources. In addition, from a more dynamic perspective, dyadic multi-dimensionality can inform how network actors can strategically facilitate the evolution of their ties to further improve their network position as well as to improve critical dyadic strategic advantages.

REFERENCES Amburgey, T. L., & Al-Laham, A. (2008). The structural evolution of multiplex organizational networks: Research and commerce in biotechnology. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 171–209). Bingley, UK: JAI/Emerald Group. Amit, R., & Zott, C. (2001). Value creation in E-business. Strategic Management Journal, 22, 493–520. Anderson, J. C., Hakansson, H., & Johanson, J. (1994). Dyadic business relationships within a business network context. Journal of Marketing, 58(4), 1–15. Barney, K. (2004). Re-encountering resistance: Plantation activism and smallholder production in Thailand and Sarawak, Malaysia. Asia Paciﬁc Viewpoint, 45, 325–339. Baum, J. A. C., Calabrese, T., & Silverman, B. S. (2000). Don’t go it alone: Alliance networks and startups’ performance in Canadian biotechnology. Strategic Management Journal, 21(3), 267–294. Bliemel, M. J., & Maine, E. M. A. (2006). The impact of embeddedness on the performance of new technology-based ﬁrms (NTBFS): A literature review. Segal Graduate School of Business working paper. Unpublished manuscript. Blyler, M., & Coff, R. W. (2003). Dynamic capabilities, social capital, and rent appropriation: Ties that split pies. Strategic Management Journal, 24(7), 677–686. Borgatti, S. P., & Cross, R. (2003). A relational view of Information seeking and learning in social networks. Management Science, 49, 432–445. Borgatti, S. P., & Foster, P. C. (2003). The network paradigm in organizational research: A review and typology. Journal of Management, 29(6), 991–1013. Bradach, J. L., & Eccles, R. G. (1991). Price, authority and trust: From ideal types to plural forms. In: G. Thompson, J. Frances, R. Levacic & J. Mitchell (Eds), Markets, hierarchies and networks: The coordination of social life (pp. 277–292). London: Sage Publications. Bruderl, J., Preisendorfer, P., & Ziegler, R. (1992). Survival chances of newly founded business organizations. American Sociological Review, 57, 227–242. Bryson, J., & Kelley, G. (1978). A political perspective on leadership emergence, stability, and change in organizational networks. Academy of Management Review, 3, 713–723. Burt, R. S. (1987). Social contagion and innovation: Cohesion versus structural equivalence. American Journal of Sociology, 92, 1287–1335.

The Role of Dyadic Multi-Dimensionality

165

Burt, R. S. (2000). The network structure of social capital. In: R. I. Sutton & B. M. Staw (Eds), Research in organizational behavior (Vol. 22). Greenwich, CT: JAI Press. Burt, R. S. (2001). Structural holes versus network closure as social capital. In: N. Lin, K. S. Cook & R. S. Burt (Eds), Social capital: Theory and research. Hawthorne, NY: Aldine de Gruyter. Casciaro, T. (1998). Seeing things clearly: Social structure, personality, and accuracy in social network perception. Social Networks, 20(4), 331–351. Chung-hoon, T. L., Hite, J. M., & Hite, S. J. (2005). Searching for enduring donor relationships: Evidence for strategies and factors in an integrated fund raising model. CASE International Journal of Educational Advancement, 6, 34–53. Chung-hoon, T. L., Hite, J. M., & Hite, S. J. (2007). Organizational integration strategies for promoting enduring donor relations in higher education: The value of building inner circle network relationships. CASE International Journal of Educational Advancement, 7, 2–19. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Harvard University Press. Collins, J. D., & Hitt, M. A. (2006). Leveraging tacit knowledge in alliances: The importance of using relational capabilities to build and leverage relational capital. Journal of Engineering and Technology Management, 23(3), 147–167. Contractor, N. S., Wasserman, S., & Faust, K. (2006). Testing multitheoretical, multilevel hypotheses about organizational networks: An analytic framework and empirical example. Academy of Management Review, 31, 681–703. Doreian, P. (2006). Actor network utilities and network evolution. Social Networks, 28(2), 137–164. Doreian, P. (2008). Actor utilities, strategic action and network evolution. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 247–271). Bingley, UK: JAI/Emerald Group. Doreian, P., & Stokman, F. N. (1997). Evolution of social networks. Amsterdam: Gordon and Breech Publishers. Dyer, J. H., & Singh, H. (1998). The relational view: Cooperative strategy and sources of interorganizational competitive advantage. Academy of Management Review, 23(4), 660–679. Elfring, T., & Hulsink, W. (2003). Networks in entrepreneurship: The case of high-technology ﬁrms. Small Business Economics, 21, 409–422. Elfring, T., & Hulsink, W. (2007). Networking by entrepreneurs: Patterns of tie formation in emerging organizations. Organization Science, 28, 1849–1872. Emerson, R. M. (1981). Social exchange theory. In: M. Rosenberg & R. H. Turner (Eds), Social psychology (pp. 30–65). New York: Basic Books. Emirbayer, M. (1997). Manifesto for a relational sociology. American Journal of Sociology, 103, 281–317. Emirbayer, M., & Goodwin, J. (1994). Network analysis, culture and the problem of agency. American Journal of Sociology, 99(6)1411–1154. Emirbayer, M., & Mische, A. (1998). What is agency? American Journal of Sociology, 103, 962–1023. Ettlinger, N. (2003). Cultural economic geography and a relational and microspace approach to trusts, rationalities, networks, and change in collaborative workplaces. Journal of Economic Geography, 3(2), 145–171. Evald, M. R., Klyver, K., & Svendsen, S. G. (2006). The changing importance of the strength of ties throughout the entrepreneurial process. Journal of Enterprising Culture, 14, 1–26.

166

JULIE M. HITE

Evans, K. (2002). Taking control of their lives? Agency in young adult transitions in England and the new Germany. Journal of Youth Studies, 5, 245–269. Faust, K., Entwisle, B., & Rindfuss, R. (2003). Accounting for social network variability among villages: Kinship and economic ties in Nang Rong, Thailand. Paper presented at the International Social Network Conference, Cancun, Mexico. Feld, S. L. (1997). Structural embeddedness and stability of interpersonal relations. Social Networks, 19(1), 91–95. Friedman, D., & McAdam, D. (1992). Collective identity and activism: Networks, choices and the life of a social movement. In: A. D. Morris & C. M. Mueller (Eds), Frontiers in social movement theory (pp. 156–173). New Haven, CT: Yale University Press. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360–1380. Granovetter, M. S. (1979). The theory gap in social network analysis. In: P. W. Holland & S. Leinhardt (Eds), Perspectives on social network research (pp. 501–518). New York: Academic Press. Granovetter, M. S. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91(3), 481–510. Greve, A., & Salaff, J. W. (2003). Social networks and entrepreneurship. Entrepreneurship, Theory & Practice, 28, 1–22. Gulati, R. (1995a). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38(1), 85–113. Gulati, R. (1995b). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1439–1498. Gulati, R., Nohria, N., & Zaheer, A. (2000a). Guest Editors’ introduction to the special issue: Strategic networks. Strategic Management Journal, 21, 199–201. Gulati, R., Nohria, N., & Zaheer, A. (2000b). Strategic networks. Strategic Management Journal, 21, 203–215. Hansen, M. T., & Lovas, B. (2004). How do multinational companies leverage technological competencies? Moving from single to interdependent explanations. Strategic Management Journal, 25(8–9), 801–822. Hansen, M. T., Mors, M. L., & Lovas, B. (2005). Knowledge sharing in organizations: Multiple networks, multiple phases. Academy of Management Journal, 48(5), 776–793. Hesterly, W. S., Jones, C., & Madhok, A. (1998). Trust: A typology and its applications in economic exchange.Unpublished manuscript. Hite, J. M. (2003). Patterns of multi-dimensionality among embedded network ties: A typology of relational embeddedness in emerging entrepreneurial ﬁrms. Strategic Organization, 1(1), 9–49. Hite, J. M. (2005). Evolutionary processes and paths of relationally-embedded network ties in emerging entrepreneurial ﬁrms. Entrepreneurship Theory and Practice, 29, 113–144. Hite, J. M., & Hesterly, W. S. (2001). The evolution of ﬁrm networks: From emergence to early growth of the ﬁrm. Strategic Management Journal, 22(3), 275–286. Hite, J. M., Hite, S. J., Jacob, W. J., Rew, W. J., Mugimu, C. B., & Nsubuga, Y. K. (2006). Building bridges for resource acquisition: Network relationships among headteachers.

The Role of Dyadic Multi-Dimensionality

167

In Ugandan private secondary schools. International Journal of Education Development, 26, 495–512. Hite, J. M., Hite, S. J., Rew, W. J., Jacob, W. J., & Mugimu, C. B. (February 17, 2005). Resource network centrality and school performance: An examination of secondary schools in Mukono, Uganda. Paper presented at the International Network of Social Network Analysis Sunbelt Conference, Redondo Beach, CA. Hite, J. M., Williams, E. J., & Baugh, S. B. (2005). Multiple networks of public school administrators: An analysis of network content and structure. International Journal of Leadership in Education, 8, 91–122. Hite, J. M., Williams, E. J., Hilton, S. C., & Baugh, S. C. (2006). The role of administrator characteristics on perceptions of innovativeness among a network of public school administrators. Education and Urban Society, 38, 160–187. Hoang, H., & Antoncic, B. (2003). Network-based research in entrepreneurship: A critical review. Journal of Business Venturing, 18, 165–187. Holmena, E., & Pedersena, A. C. (2003). Strategizing through analyzing and inﬂuencing the network horizon. Industrial Marketing Management, 32, 409–418. Human, S. E., & Provan, K. G. (2000). Legitimacy building in the evolution of small ﬁrm mutilateral networks: A Comparative study of success and demise. Administrative Science Quarterly, 45(2), 327–369. Jack, S. L. (2005). The role, use and activation of strong and weak network ties: A qualitative analysis. Journal of Management Studies, 42(6), 1233–1259. Jones, G. R., & George, J. M. (1998). The experience and evolution of trust: Implications for cooperation and teamwork. Academy of Management Review, 23(3), 531–546. Kenis, P., & Knoke, D. (2002). How organizational ﬁeld networks shape interorganizational tie-formation rates. Academy of Management Review, 27(2), 275–293. Kilduff, M., & Tsai, W. (2003). Social networks and organizations (1st ed.). London: SAGE Publications. Kilduff, M., Tsai, W., & Hanke, R. (2006). A paradigm too far? A dynamic stability reconsideration of the social network research program. Academy of Management Review, 31(4), 1031–1048. Kim, T. Y., Oh, H., & Swaminathan, A. (2006). Framing interorganizational network change: A network inertia perspective. Academy of Management Review, 31(3), 704–720. Koehly, L. M., & Pattison, P. (2005). Random graph models for social networks: Multiple relations or multiple raters. In: P. J. Carrington, J. Scott & S. Wasserman (Eds), Models and methods in social network analysis (pp. 162–191). Cambridge, UK: Cambridge University Press. Koka, B. R., Madhavan, R., & Prescott, J. E. (2006). The evolution of interﬁrm networks: Environmental effects on patterns of network change. Academy of Management Review, 31(3), 721–737. Krackhardt, D., & Kilduff, M. (1990). Friendship culture: The control of organizational diversity. American Anthropologist, 92, 142–145. Krackhardt, D., & Kilduff, M. (2002). Structure, culture and Simmelian ties in entrepreneurial ﬁrms. Social Networks, 24(3), 279–290. LaBianca, G., & Brass, D. J. (2006). Exploring the social ledger: Negative relationships and negative asymmetry in social networks in organizations. Academy of Management Review, 31, 596–614.

168

JULIE M. HITE

Larson, A. L., & Starr, J. A. (1993). A network model of organization formation. Entrepreneurship: Theory and Practice, 17(2), 5–15. Lazega, E., & Pattison, P. (1999). Multiplexity, generalized exchange and cooperation in organizations. Social Networks, 21, 67–90. Lockie, S. (2006). Networks of agri-environmental action: Temporality, spatiality and identity in agricultural environments. Sociologia Ruralis, 46(1), 22–39. Lomi, A. (2002). Networks across organizations and markets. Paper presented at the Academy of Management, Denver, Colorado. Madhavan, R., Koka, B. R., & Prescott, J. E. (1998). Networks in transition: How industry events (re)shape interﬁrm relationships. Strategic Management Journal, 19(5), 439–459. McEvily, B., & Marcus, A. (2005). Embedded ties and the acquisition of competitive capabilities. Strategic Management Journal, 26, 1033–1055. McKelvey, B., & Aldrich, H. (1983). Populations, natural selection and applied organizational science. Administrative Science Quarterly, 28(2), 101–128. McPherson, J. M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415–444. Mehra, A., Smith, B. R., Dixon, A. L., & Robertson, B. (2006). Distributed leadership in teams: The network of leadership perceptions and team performance. Leadership Quarterly, 17(3), 232–245. Miller, D., & Friesen, P. H. (1980). Momentum and revolution in organizational adaptation. Academy of Management Journal, 23, 591–614. Mizruchi, M. S. (2004). Network theory. In: K. C. George Ritzer, et al. (Eds), Handbook of sociological theory (pp. 534–540). London: SAGE. Mohrman, S., Galbraith, J. R., & Monge, P. R. (2004). Final report of research contract 84882: Network attributes impacting the generation and ﬂow of knowledge within and from the basic science community. Albuquerque, NM: Sandia National Laboratories and the Ofﬁce of Science, Research Contract 84882: The organizational and managerial factors that contribute to the stream of value from the basic research of the ofﬁce of science. Monge, P. R., & Contractor, N. S. (2001). Emergence of communication networks. In: F. M. J. L. L. Putnam (Ed.), New handbook of organizational communication (pp. 440–502). Thousand Oaks, CA: Sage. Moody, J., & White, D. R. (2003). Structural cohesion and embeddedness: A hierarchical conception of social groups. American Sociological Review, 68, 103–127. Moody, J., McFarland, D., & Bender-deMoll, S. (2005). Dynamic network visualization. American Journal of Sociology, 110, 1206–1241. Moran, P. (2005). Structural vs. relational embeddedness: Social capital and managerial performance. Strategic Management Journal, 26(12), 1129–1151. Nahapiet, J., & Ghoshal, S. (1998). Social capital, intellectual capital, and the organizational advantage. Academy of Management Review, 23(2), 242–266. Neuman, E. J., Davis, G. F., & Mizruchi, M. S. (2008). Networks and industry consolidation in U.S. global banking, 1986–2004. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 213–248). Bingley, UK: JAI/Emerald Group. Pillai, K. G. (2006). Networks and competitive advantage: A synthesis and extension. Journal of Strategic Marketing (London), 14, 129(117). Podolny, J. M. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107, 33–60.

The Role of Dyadic Multi-Dimensionality

169

Podolny, J. M., & Page, K. (1998). Network forms of organization. Annual Review of Sociology, 24, 57–76. Poppo, L., & Zenger, T. (2002). Do formal contracts and relational governance function as substitutes or complements. Strategic Management Journal, 23(8), 707–725. Powell, W. W. (1990). Neither market nor hierarchy: Network forms of organization. Research in Organizational Behavior, 12, 295–336. Robins, G., & Pattison, P. (2005). Interdependencies and social processes: Dependence graphs and generalized dependence structures. In: P. J. Carrington, J. Scott & S. Wasserman (Eds), Models and methods in social network analysis (pp. 192–214). Cambridge, UK: Cambridge University Press. Rowley, T. J., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21(3), 369–386. Scott, J. (2000). Social network analysis (2nd ed.). London: Sage Publications. Shane, S., & Cable, D. (2002). Network ties, reputation, and the ﬁnancing of new ventures. Management Science, 48(3), 364–382. Shane, S., & Stuart, T. E. (2002). Organizational endowments and the performance of university start-ups. Management Science, 48, 154–171. Simon, H. A. (1972). Theories of bounded rationality. In: C. B. McGuire & R. Radner (Eds), Decision and organization (pp. 161–176). New York: American Elsevier. Snijders, T. A. B. (1996). Stochastic actor-oriented dynamic network analysis. Journal of Mathematical Sociology, 21, 149–172. Snijders, T. A. B. (2001). The statistical evaluation of social network dynamics. Sociological Methodology, 361–395. Snijders, T. A. B. (February 16–20, 2005a). Analyzing dynamics of non-directed social networks. Paper presented at the SUNBELT XXV International Social Network Conference, Redondo Beach, CA. Snijders, T. A. B. (2005b). Models for longitudinal network data. In: P. J. Carrington, J. Scott & S. Wasserman (Eds), Models and methods in social network analysis (pp. 215–247). Cambridge: Cambridge University Press. Snijders, T. A. B., Steglich, C. E. G., & Schweinberger, M. (2007). Modeling the co-evolution of networks and behavior. In: K. V. Montfort, J. Oud & A. Satorra (Eds), Longitudinal models in the behavioral and related sciences. Philadelphia, PA: Lawrence Erlbaum Associates. Soda, G., Zaheer, A., & Carlone, A. (2008). Imitative behavior: Network antecedents and performance consequences. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 531–560). Bingley, UK: JAI/Emerald Group. Sparrowe, R. T., & Liden, R. C. (1997). Process and structure in leader-member exchange. The Academy of Management Review, 22(2), 522–552. Steier, L. (2000). Entrepreneurship and the evolution of angel networks. Organization Studies, 21(1), 163–193. Stevenson, W. B., & Greenberg, D. (2000). Agency and social networks: Strategies of action in a social structure of position, opposition, and opportunity. Administrative Science Quarterly, 45(4), 651–678. Stinchcombe, A. L. (1965). Organizations and social structure. In: J. G. March (Ed.), Handbook of organizations (pp. 153–193). Chicago: Rand-McNally. Stuart, T. E., Hoang, H., & Hybels, R. (1999). Interorganizational endorsements and the performance of entrepreneurial ventures. Administrative Science Quarterly, 44, 315–349.

170

JULIE M. HITE

Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Uzzi, B. (1999). Embeddedness in the making of ﬁnancial capital: How social relations and networks beneﬁt ﬁrms seeking ﬁnancing. American Sociological Review, 64(4), 481–505. Uzzi, B., & Gillespie, J. J. (2002). Knowledge spillover in corporate ﬁnancing networks: Embeddedness and the ﬁrm’s debt performance. Strategic Management Journal, 23(7), 595–618. Uzzi, B., & Lancaster, R. (2003). Relational embeddedness and learning: The case of bank loan managers and their clients. Management Science, 49(4), 383–399. Uzzi, B., Spiro, J., Guimera, R., & Amaral, L. (August 11, 2006). Emergence: The Dynamics of Network Formation. Paper presented at the American Sociological Association, Montreal, Quebec, Canada. van Liere, D. W., Koppius, O. R., & Vervest, P. H. M. (2008). Network horizon: An information-based view on the dynamics of bridging positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 595–639). Bingley, UK: JAI/Emerald Group. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge, UK: Cambridge University Press. Wasserman, S., & Galaskiewicz, J. (1994). Advances in social network analysis: Research in the social and behavioral sciences. Thousand Oaks: SAGE Publications. Wasserman, S., Scott, J., & Carrington, P. (2005). Introduction. In: P. J. Carrington, J. Scott & S. Wasserman (Eds), Models and methods in social network analysis (pp. 1–7). Cambridge: Cambridge University Press. Wellman, B. (1983). Network analysis: Some basic principles. Sociological Theory, 1, 155–200. Wellman, B. (1988). Structural analysis: From method and metaphor to theory and substance. In: B. Wellman & S. D. Berkowitz (Eds), Social structures: A network approach (pp. 19–61). Cambridge, UK: Cambridge University Press. Williamson, O. E. (1979). The governance of contractual relations. Journal of Law and Economics, 22(2), 233–261. Wilson, H. I. M., & Appaiah-Kubi, K. (2002). Resource leveraging via networks by hightechnology entrepreneurial ﬁrms. Journal of High Technology Management Research, 13(1), 45. Zaheer, A., McEvily, B., & Perrone, V. (1998). Does trust matter? Exploring the effects of interorganizational and interpersonal trust on performance. Organization Science, 9(2), 141–159. Zaheer, A., & Venkataraman, S. (1995). Relational governance as an interorganizational strategy: An empirical test of the rate of trust in economic exchange. Strategic Management Journal, 16, 373–392.

THE STRUCTURAL EVOLUTION OF MULTIPLEX ORGANIZATIONAL NETWORKS: RESEARCH AND COMMERCE IN BIOTECHNOLOGY Terry L. Amburgey, Andreas Al-Laham, Danny Tzabbar and Barak Aharonson ABSTRACT Inter-organizational alliances and the networks they generate have been a central topic in organization theory over the last decade. However, network analyses per se have been static. Even when information over time has been available, the temporal component has been set aside or aggregated to the end point of the study. Substantially more research has been conducted on organizations initiating inter-organizational relationships. The organization-level research has been decidedly dynamic in nature. However, organization-level research has largely examined the structural characteristics of the networks generated by organizational actions. Work combining network-level and organization-level phenomena has been rare and, to our knowledge, no research including the effects of organizationlevel actions on the evolution of network-level phenomena has occurred. In this chapter we use more than 6000 R&D alliances and more than 6500 M&D alliances initiated by more than 1000 biotech ﬁrms in the U.S. Network Strategy Advances in Strategic Management, Volume 25, 171–209 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25005-9

171

172

TERRY L. AMBURGEY ET AL.

over a 30 year period to construct quarterly networks. We test 13 hypotheses linking the actions of the ﬁrms to changes in network structure. Utilizing hazard-rate models we test the effects of institutional status, positional status (centrality), and structural status (coreness) of ﬁrms on their propensity to form ties with different structural consequences. Our research indicates that both R&D and M&D networks in U.S. biotechnology are developing a distinct core/periphery structure over time. Furthermore, we ﬁnd support for a process of preferential attachment wherein organizations are more likely to form ties with organizations of similar institutional and structural status. Furthermore, we ﬁnd evidence for cross effects, for example attachment processes that enfold across the two networks.

INTRODUCTION We are interested in the emergence and structural evolution of interorganizational networks. Although this research is focused upon a speciﬁc feature of structure (connectivity) and a speciﬁc empirical setting [R&D and M&D (marketing & distribution) agreements in the ﬁeld of biotechnology] our ultimate goal is more general. We believe that evolutionary models should have several features: dynamic change over time, history dependence, multiple levels of analysis, and both modiﬁcation and replacement processes (Amburgey & Singh, 2002, p. 327). We develop and test an ecological approach to emergence and structural evolution that incorporates these features, a network ecology model so to speak. We are not the ﬁrst scholars to examine this topic, there has been some exemplary work done, particularly in the recent past (e.g. Baum et al., 2004; Gulati & Gargiulo, 1999; Powell, Koput, White, and Owen-Smith, 2005; Rowley, Greve, Rao, & Baum, 2005; Stuart, 1998; Uzzi, Guimera, Spiro, & Amaral, 2005; Walker, Kogut, & Shan, 1997). Virtually all work in this area incorporates dynamic change over time and much of it also incorporates history dependence. However we believe that much remains to be done. We believe that attention to multiple levels of analysis and both modiﬁcation and replacement processes add important insights to our knowledge of network emergence and evolution. We also believe that the analysis of the co-evolutionary patterns in multiplex networks and the interdependencies between multiple networks is of crucial importance.

Research and Commerce in Biotechnology

173

Our research is closely related to, and extends, other recent work on the evolution of organizational networks, particularly the work by Uzzi et al. (2005) and Powell et al. (2005). The similarity is based on our common interest in the effects of different types of ties and in the overall organizational ﬁeld. Uzzi and his colleagues believe that it is important to focus on how types of ties affect the emergent structure of a network (2005, p. 3). We also believe that it is import to focus on the effects of different types of ties. However, our work deﬁnes types of ties in terms of structural consequences rather than in terms of the nature of the partners. We also extend and augment the work of Uzzi and his colleagues with regard to how lower level network processes produce network level outcomes. For example they point out that the size of the largest component systematically grows from fragmented to braided to densely overlapping to fully connected networks (Uzzi et al., 2005, p. 6). Our work focuses explicitly upon the role of connectivity and provides a ﬁne grained analysis of the creation, consolidation, and growth of network components in addition to other evolutionary processes. Like Powell and his colleagues, we are interested in the evolution of the organizational ﬁeld of biotechnology. However, we take a very different approach. Powell et al. (2005) combine different types of ties (e.g., R&D and M&D agreements) but distinguish between ties on the basis of partner type in a fashion similar to Uzzi et al. (2005). In contrast, our research covers the entire organizational ﬁeld and includes the same types of partners but treats the R&D and M&D networks as separate entities and explores the interaction between the two. We believe that the evolution of the organizational ﬁeld is driven by ties deﬁned in terms of structural consequences rather than ties deﬁned in terms of the nature of the partners. Our focus is consistent with what Hite (2008, pp. 9–10) refers to as the structures of dyadic connection. Evaluating ties on the basis of their consequences is not new; consider the large literature on structural holes and bridging ties. A focus on network consequences of ties rather than the node consequences of ties is crucial in understanding how the actions of nodes affect network structure rather than network demography. We believe that our work extends and augments the insights provided by Powell and his colleagues by providing additional insights into the evolutionary consequences of preferential attachment based on homophily and the logic of accumulation. Our general approach to network evolution is guided by Doreian and Stokman (1997). Network evolution involves structural change over time driven by network processes. Network processes, in turn, are series of events

174

TERRY L. AMBURGEY ET AL.

that create, sustain and dissolve network structures. Ultimately, interorganizational networks result from the formation and termination of ties with other organizations (Gulati & Gargiulo, 1999; Rowley et al., 2005; Uzzi et al., 2005). Thus we believe that the organizational level of analysis is the most fundamental. This hardly unique, most work on inter-organizational networks is done at the organizational level. However we examine the dynamics of ties with regard to their effects on network structure. Our work includes the dynamics of both the organizational and the network level. We believe that this is both unique and crucial. A conceptual diagram of our approach is shown in Fig. 1. In this research we distinguish between two levels of analysis: the organizational and the network level (c.f. Amburgey & Al-Laham, 2007 for a more elaborated conceptualization). The overall network consists of a component substructure, for example local regions with varying degrees of connectedness and varying degrees of cohesion. The network components and the organizations are linked by two network processes. The ﬁrst relationship (1) describes a network process that entails the structural consequences of tie formation (and deletion) by organizations on the local region of the network or the network as a whole (Hite, 2008, pp. 9–10). For example the cohesion of a local region (component) will be inﬂuenced by this process. Similarly, the consolidation of two components into a single larger component is a consequence of tie formation at the organizational level. The second relationship (2) describes a network process linking attributes of the local network neighborhood and the global network structure to the propensities of ties with different structural consequences. For example, the cohesion of the component that a ﬁrm is embedded within should affect the propensity to form different types of ties. Similarly, the number of components in the network is likely to affect the propensity to form different

Network – Level Component Structure (Size and Cohesion) Structural consequences of tie formation

Propensity consequences of context

Organizational – Level Tie-Formation (and Dissolution)

Fig. 1.

Conceptual Frame of Reference.

Research and Commerce in Biotechnology

175

types of ties. This second process is what Hite (2008, p. 15) refers to as the endogenous context.

THE ORGANIZATIONAL FIELD OF BIOTECHNOLOGY Biotechnology is characterized by the application of a wide variety of scientiﬁc disciplines to a number of different commercial applications. Biotechnology ﬁrms face a central challenge in creating new knowledge and building up internal learning capabilities for integrating, transforming and applying that knowledge into products and services (Hagedoorn, 2002; Powell, Koput, and Smith-Doerr, 1996). The unique characteristics of ﬁrms in the biotech industry in general include a long and complex product development and approval cycle, heavy reliance upon basic scientiﬁc research and a set of very heterogonous technologies with the potential to transform various application ﬁelds. The knowledge stock of biotechnology ﬁrms represents a conﬂuence of heterogeneous scientiﬁc disciplines, very unlike traditional ﬁrms. The attempts of biotechnology ﬁrms are not only directed towards generating new products, but also new methods and processes to discover new products. To achieve commercial success, ﬁrms rely on highly complex and speciﬁc knowledge which is still emerging, unlike the mature knowledge structure of other industries. Biotechnology ﬁrms follow business models that rely on highly complex and speciﬁc knowledge that is emerging, unlike the business models of other industries that rely less on new scientiﬁc research. Therefore, the competence of biotechnology ﬁrms is applied research devoted to the exploitation of speciﬁc scientiﬁc discoveries; not basic research as is conducted in public organizations, nor the engineering capabilities and marketing system needed for large scale production and distribution (Gambardella, 1995). Biotechnology ﬁrms’ specialized knowledge and business models often means that they must partner with non-proﬁt-seeking organizations, such as government agencies and university departments, as well as larger, older proﬁt-seeking ﬁrms in the pharmaceutical or chemical sectors. These different types of organizations from different sectors generally have different goals with respect to the dissemination of knowledge exchanged or produced within an alliance, as well as widely different incentive structures. This disparity in knowledge, incentives and assets has led to a division of labor between public organizations, biotechnology ﬁrms, and pharmaceutical ﬁrms.

176

TERRY L. AMBURGEY ET AL.

As a result, the complementary knowledge and assets held by each type of organization can then be consolidated through inter-organizational relationships such as strategic alliances (Gambardella, 1995, pp. 147–148). This consolidation can produce beneﬁts not available to organizations operating alone. Furthermore, the mutual knowledge dependency has led to a network of R&D collaboration between varieties of organizations, where small biotech start ups play an important role in this new locus of innovation (Hagedoorn & Roijakkers, 2002, p. 242). Although biotechnology ﬁrms are skilled at R&D, they lack the capital and infrastructure to market and distribute products globally. As a consequence, the mutual dependency between biotech ﬁrms and pharmaceutical companies has also led to a network of M&D collaborations.

Strategic Alliances in Biotechnology Strategic alliances within the ﬁeld of biotechnology have taken two fundamental forms: collaborative research and development, and marketing or distribution agreements. Collaborative research agreements are focused upon the initial stages of innovation, the creation of knowledge and the development of goods and services capable of being commercialized. From an early stage, biotech ﬁrms have to build up the technological capabilities required not only to participate in the industry, but even more to lead the innovation race in the industry (Amburgey, Dacin, & Singh, 1996). Due to the radical nature of the various emerging technologies in the biotechnology ﬁeld, discrete technological trajectories prevail, being reﬂected in sequential research projects that carry a high competence destroying risk (Casper, 2000). The technological trajectories in biotechnology are extreme volatile, leading to a high failure rate of research projects in the various stages of the product development processes. For example, technologies seen as quite exotic a few years ago, such as the cloning of target strains of DNA for lab work (PCR), are now widely available (Casper, 2000, p. 898). The ongoing scientiﬁc progress therefore leads to continuous technological adaptation pressures for biotech ﬁrms. To be viable, ﬁrms must build up the technological competencies required for a fast conversion of knowledge from an innovative idea or a scientiﬁc discovery into codiﬁable and marketable innovations. These pressures lead to the adoption of cooperative relationships with other organizations (Grant & Baden-Fuller 1995; Hamel, 1991; Khanna, Gulati, & Nohria, 1998; Kogut 2000; Lyles 1994; Mowery, Oxley, and Silverman, 1996; Powell et al., 1996).

177

Research and Commerce in Biotechnology

To convert scientiﬁc knowledge into commercial products or services, a biotechnology ﬁrm conducts applied research devoted to the exploitation of speciﬁc scientiﬁc discoveries. However, once products are approved, they must be marketed and distributed. The second type of strategic alliance is M&D agreements, mostly with larger pharmaceutical companies. These agreements are required for all but the very largest of biotechnology companies since marketing and distribution of life science products is extremely expensive. M&D alliances provide access to complementary assets critical to successful commercialization, such as market access, marketing and distribution infrastructure, production facilities and/or expertise (Baum, Calabrese, & Silverman, 2000; Pisano, 1990; Silverman & Baum, 2002). This is particularly true of collaboration with pharmaceutical and chemical ﬁrms, who excel in product commercialization (Arora & Gambardella, 1990). Established pharmaceutical companies have long-standing routines and competencies to manage a new drug through the regulatory process and then to market it via their sales force. In addition, large pharmaceutical companies tend to have the resources to ﬁnance this costly and timeconsuming part of the development process, and they are often short of innovative products in their own research pipeline (Rothaermel & Deeds, 2004). Within recent years the degree of embeddedness of biotech ﬁrms within the M&D network is increasing, reﬂecting the gradual shift from upstream exploration of new ideas and scientiﬁc discoveries to downstream exploitation of products and solutions (Baum et al., 2000; Rothaermel & Deeds, 2004; Silverman & Baum, 2002). Fig. 2 depicts an idealized model of the Research Institutes & Biotech Firm’s Universities Area of Area of Expertise Expertise

Basic Research

Gene Identificat ion

Target Identification & Validation

Pharmaceutical Firm’s Area of Expertise

Lead Identifi cation

PreClinical Trials

Clinical Trials

Manufac -turing

Upstream Collaboration

Downstream Collaboration

• R&D network:

• M&D network • Production, marketing & distribution and licensing

• Licensing and collaborative research

Fig. 2.

Value Chain of the Biotech-Industry.

Sales & Marketing

178

TERRY L. AMBURGEY ET AL.

biotech industries value chain, and the different types of networks under observation in our study. We will now turn to the discussion of structural characteristics of these networks.

THE EMERGENCE AND EVOLUTION OF NETWORK CONNECTIVITY There are a variety of structural characteristics that characterize networks. Some of the more popular are connectivity, diameter, average path length, clustering coefﬁcient, and degree distribution (Albert & Barabasi, 2002). We focus on connectivity. One of the most fundamental characteristics of a network (or graph) is connectedness (Wasserman & Faust, 1994, p. 109). Connectedness is arguably the most salient feature of a network; networks are deﬁned by the connections of entities. Moreover, most of the structural features of networks are predicated on connectedness. For example the diameter, average path length, etc., are only meaningful for a network where all nodes are reachable. A network or graph is connected if there is a path (however long) between every pair of nodes (Wasserman & Faust, 1994). If the network is disconnected then rather than one large network, it consists of two or more components. Fig. 3 provides a graphic rendition of the biotechnology R&D network in the United States in the last quarter of 1983 (the ﬁrst quarter with sufﬁcient R&D agreements to say that a network exists). Fig. 4 provides a similar graphic rendition of the biotechnology M&D network in the same quarter. The circles represent organizations in the network; biotech ﬁrms, government agencies, pharmaceutical companies, etc. The lines connecting the organizations represent R&D or M&D agreements between the organizations in effect at the beginning of that quarter. As the ﬁgures indicate, rather than one large network, there are a number of disconnected components. As the ﬁgures show, the R&D network in the last quarter of 1983 consists of 17 separate components that are unconnected with any other component; the M&D network consists of 7 unconnected components. Clearly the most prominent structural feature of the networks at this time is fragmentation. Line graphs representing the number of components in the two networks are given in Fig. 5. The line graphs in Fig. 5 indicate that in the R&D network there is an early sharp increase in the number of components. This is followed by a

Research and Commerce in Biotechnology

Fig. 3.

179

Example of Components in the U.S. Biotech R&D Network (Last Quarter of 1983).

protracted period where the number ﬂuctuates from period to period but with no discernable time trend. Then towards the end of the observation window there is a period of decline. In the M&D network there is an early sharp increase in the number of components. This is followed by a protracted period where the number ﬂuctuates from period to period with a smaller upwards time trend. Given that both networks exhibit fragmentation, what other structural trends are occurring? In Figs. 3 and 4 saw that some components were dyads, some were triads, and a few contained more members; there is diversity in the sizes of the components. Fig. 6 provides time trends in normalized entropy in the R&D and M&D networks. Normalized entropy provides a measure of the heterogeneity in the sizes of the network components. The computational formula for normalized heterogeneity is provided in the data and methods section of the chapter since we use it as a variable in our empirical analyses. In general, a high value of entropy represents an evenness of component sizes while a low value represents heterogeneity in component sizes. As Fig. 6 indicates both the networks exhibit a sharp reduction in the entropy measure: in both cases the networks are developing a structure that consists of a few large components and a

180

TERRY L. AMBURGEY ET AL. 128 4071

4778

3485 2314

263

333 3005

4630 2402

206

3513 3039

654

474 4010

199

4224

4014

4587

3013 3049

4095 56 3056 4005

Fig. 4.

Example of Components in the U.S. Biotech R&D Network (Last Quarter of 1983).

greater number of small components. In other words both networks exhibit structural differentiation characterized by a core and a periphery. The development of structural differentiation in the two networks is also visible when a k-core decomposition is conducted (Fig. 7). The k-core measure indicates the cohesiveness of the components in the net, based upon the degree centrality of the nodes in the component. A k-core is a component in which each node is adjacent to at least a minimum number k, of the other nodes in the component (see Wassermann & Faust, 1994, p. 266). Thus, the larger the k-value, the higher the degree of connectedness of the node under observation, and the larger the cohesiveness of its component. In Figs. 8 and 9 have calculated the k-core values of the R&D and M&D network. We have summed the number of nodes in the respective k-core strata over the time periods under observation (time quarters). Turning to the R&D network ﬁrst (see Fig. 8), there are several aspects of interest.

181

0

20

40

60

Research and Commerce in Biotechnology

0

20

40 quarter

rand_num_comp

Fig. 5.

60

80

mand_num_comp

Number of Components.

First, the number of layers increases over time, indicating an increasing structural differentiation of the net. At present, there are six cohesive layers in the R&D network. Second, the number of organizations is unequally distributed across the layers. The largest absolute number of nodes as well as the steepest increase over time takes place in the 1-core layer. This layer represents the periphery of the network, where organizations are only loosely connected to each other. The lowest number of nodes is located in the 6-core area of the network, which represents the innermost core. This is the layer with the highest connectivity and robustness due to multiple linkages between nodes. A different picture emerges when we turn to the M&D network strata (see Fig. 9). First, there are fewer layers than in the R&D network, indicating a signiﬁcantly lower degree of structural differentiation of the M&D network. At present, there are only four cohesive layers in the M&D network. Turning to the developmental pattern over time, we see a strong increase in the number of nodes in the outer layer – the periphery – of the network. The number of nodes that are only loosely connected to each others is still growing with a high rate. The remaining layers are signiﬁcantly smaller in

TERRY L. AMBURGEY ET AL.

0

.2

.4

.6

.8

182

0

20

40 quarter rand_norm_ent

Fig. 6.

60

80

mand_norm_ent

Size Homogeneity of Components.

size, and are more or less stable over time. Thus the periphery exhibits a signiﬁcantly different dynamic then the inner core of the M&D network. What do these trends show us? In both settings the networks are fragmented throughout the observation periods but there is the emergence of a core/periphery structure. The core and the periphery consist of different layers of cohesion, as indicated in the k-core decomposition. Second, there is an ecological process where new ﬁrms are founded and others pass from the scene. This process forms the fundament for a network ecology where new components are created and others depart, where some grow in size while others shrink. The ecology thus involves structural change over time driven by network processes, and these processes, in turn, are series of events that create, sustain and dissolve network structures (c.f. Fig. 1). Graph theory suggests that these structural features and trends can, under the right circumstances, arise in networks constructed randomly (Albert & Barabasi, 2002). However these circumstances do not appear to hold in these settings. Moreover a wealth of academic research and the observations of participants indicate that both R&D and M&D agreements are purposive, not random.

Research and Commerce in Biotechnology

Fig. 7.

183

Coreness Strata: k-Core Decomposition.

What network processes are associated with the number of components and their size distribution? Just as population size is a function of differential entries and exits (Baum & Mezias, 1992; Lomi, 1995), network structure is a function of the pattern of creation and deletion of ties among organizations (Doriaen & Stokman, 1997; McEvily & Zaheer, 1999). We will describe these types of ties and their structural consequences in the following section.

STRUCTURAL CONSEQUENCES OF TIE FORMATION Tie Formation and the Development of a Core/Periphery Structure How does the creation of ties among organizations in the organizational ﬁeld of biotechnology affect the number, size distribution, and cohesion of components and the development of an overall core/periphery structure? Taking biotech ﬁrms as our focal point, the trajectory of the development

TERRY L. AMBURGEY ET AL.

0

200

400

600

800

184

0

20

40 quarter rand_k1sum rand_k3sum rand_k5sum

Fig. 8.

60

80

rand_k2sum rand_k4sum rand_k6sum

Coreness Strata – R&D Network.

depends on the nature of the ties. All agreements can be categorized in one of four ways: (a) the creation of a bridge to another component, (b) the creation of a new component, (c) the creation of a pendant to an existing component, or (d) the creation of an additional intra-component tie (see Fig. 10). The ﬁrst three types have a direct consequence on the number and size distribution of components in the network, the last has an effect on the cohesion of components. The creation of a bridging tie has a substantial effect because it has consequences for both the number of components and the size distribution of the components. For example assume that two components in the network have six members and four members respectively. One of the organizations in the ﬁrst component forms a bridging tie with an organization in the second component. As a result there is a reduction in the number of components and an increase in size in the resulting single component. As a process in network ecology, the formation of bridging ties produces the exit of two components and the entry of a single larger component (as mentioned previously, network processes are series of events that create, sustain and dissolve network structures).

185

0

200

400

600

Research and Commerce in Biotechnology

0

20

40 quarter mand_k1sum mand_k3sum

Fig. 9.

60

mand_k2sum mand_k4sum

Coreness Strata – M&D Network.

Fig. 10.

Type of Network Ties.

80

186

TERRY L. AMBURGEY ET AL.

The creation of an entirely new component also has consequences for both the number of components and the size distribution of the components. For example if two organizations not previously in the network establish a tie with one another then a new component has been added and the small size of the new component also affects the size distribution. This is another entry process for components in the network ecology. If an existing component adds a node that was previously not part of the network and has no ties to any other node (frequently referred to as a pendant) then the size distribution is altered but not the number of components. Thus if we have a component with six members and one of those members forms a tie with an organization that was previously not in the network then the component grows in size by one but the number of components does not change. This is a transformative process in the network ecology as it represents the growth of a component. Finally, it is frequently the case that an organization forms a tie with another organization that is a member of the same component. The creation of this additional tie between component members does not directly affect the number of components or the size distribution of components. However it does affect the internal structure of the component by increasing its cohesion and robustness. What factors produce different rates of tie formation with various structural consequences? The development of our hypotheses in the following subsection is focused upon a process of preferential attachment. Our arguments revolve around three factors driving preferential attachment: positional status, institutional status, and structural position.

HYPOTHESES Preferential attachment is a long standing construct in network theory, extending at least as far back as the random-biased model developed in the 1950s (Skvoretz, Fararo, & Agneesens et al., 2004). In contrast to random networks, these types of models, while allowing for some degree of randomness in the formation of ties, posit the existence of some preference or bias in the formation of ties. Our approach to preferential attachment is the well established concept of homophily, a tendency of organizations to choose ties with organizations of similar characteristics (Doreian & Stokman, 1997; McPherson & Smith-Lowen, 1987). Given that an overwhelming body of empirical literature demonstrates that the foundation

Research and Commerce in Biotechnology

187

of strategic alliances is predominantly complementarity how would the concept of homophily work? We take it as a given that R&D and M&D alliances are founded on technological or ﬁnancial complementarities between the partners. If only one organization has the appropriate complementary assets then that’s where the tie is likely to be formed. However in many instances there are multiple choices for partnering. Our basic argument states that, given more than one appropriate partner, biotech ﬁrms will tend to form ties with other organizations that are similar in status. Homophily can operate on the basis of many different kinds of similarity, such as size, collaborative distance, or ﬁnancial attributes (Gulati, 1995a, 1995b; Powell et al., 2005; Rowley et al., 2005). We focus on status similarity because we believe that status will be perceived as being linked to important factors such as resource endowments as well as endorsement effects (Stuart, Hoang, & Hybels, 1999). This leads to several speciﬁc hypotheses about the status of the organization and the rates at which different types of ties are formed.

Positional Status and Preferential Attachment Our ﬁrst set of hypotheses focus on the positional or structural status (eigenvector centrality) of a biotech ﬁrm in the R&D and M&D networks (Podolny, 2001, 1994, 1993). Positional status is a conceptualization of status which is founded upon the network position of an actor. Previous work on the formation of ties has demonstrated that status similarity increases the probability of a new tie between organizations, and thus drives the evolution of the network (c.f. Gulati & Gargiulo, 1999). Previous work however has not distinguished between different types of ties, and their consequences for the structure of the net. We argue that biotech ﬁrms with high positional status should be more likely to form bridging ties and intracomponent ties in both networks. The centrality of a biotech ﬁrm is a function of the size of the ﬁrm’s component – ﬁrms in large components will tend to have greater centrality scores than ﬁrms in smaller components. If the most similar (with regard to centrality) organizations are in the same component, this will lead to the formation of intra-component ties. If the next most similar (with regard to centrality) organizations are located in other components of roughly the same size, this will lead to the formation of bridging ties. Thus we argue that ﬁrms will form ties with other organizations of similar centrality. In some cases, the most similar

188

TERRY L. AMBURGEY ET AL.

organizations will be in the same component. In other cases the most similar organizations will be in other components of equivalent size. The creation of a pendant tie involves a tie between a higher centrality ﬁrm and a ﬁrm currently not part of the network (thus by deﬁnition, a ﬁrm with a centrality value of zero). Although a lack of other technological choices may lead a high status ﬁrm to form a pendant tie this should be the least likely to occur. Finally, the creation of a new, isolated, component can only occur among organizations that are not currently in the network. Thus, by deﬁnition, this type of tie cannot occur with a high positional status ﬁrm; the creation of new components can only occur among organizations with a zero centrality score. These four processes work independently from the type of network, for example R&D and M&D networks. We state the following hypotheses. Hypothesis 1. The higher the positional status of a biotech ﬁrm, the greater the likelihood of a bridging tie. Hypothesis 2. The higher the positional status of a biotech ﬁrm, the greater the likelihood of an intra-component tie. Hypothesis 3. The higher the positional status of a biotech ﬁrm, the smaller the likelihood of a pendant tie.

Institutional Status and Preferential Attachment Although the positional status is our preferred method for assessing organizational status, ownership status provides us with another alternative measure which is institutional or reputational in nature rather than structural. The decision to go public is a watershed event in the evolution of the ﬁrm, of a level comparable to founding or acquisition of the ﬁrm. Substantial changes in governance take place after the ﬁrm goes public. The management of the ﬁrm is now answerable to a large number of public investors in the ﬁrm, in contrast with a small set of ﬁnanciers and venture capitalists who previously owned the ﬁrm. In addition, the board of directors typically increases in size and includes a signiﬁcant proportion of outside directors, who are neither initial ﬁnanciers, nor venture capitalists, nor managers. Substantially higher levels of disclosure are needed, including the ﬁling of ﬁnancial statements, and other information with the Securities

Research and Commerce in Biotechnology

189

and Exchange Commission (SEC). In addition to these disclosures, there is ongoing monitoring of the ﬁrm by investment ﬁrms and investors who have much more information on the ﬁrm than what was available when the ﬁrm was privately held (Stuart et al., 1999). To formulate our hypotheses we argue that publicly owned ﬁrms are generally seen as higher in status than privately owned ﬁrms. However, a tendency towards homophily does not mean that public ﬁrms are not only more likely to form ties to public ﬁrms. As a matter of fact, many of the desirable partners in the organizational ﬁeld of biotechnology are not ﬁrms at all, but instead public sector organizations such as universities or research institutes. We propose that homophily can affect preferential attachment in two different ways. First, publicly owned ﬁrms can prefer to collaborate with other publicly owned ﬁrms. Second, public ﬁrms can also prefer to collaborate with public sector organizations with greater structural status. This leads to an additional three hypotheses that parallel our ﬁrst three hypotheses. Hypothesis 4. Public ownership increases the likelihood of a bridging tie. Hypothesis 5. Public ownership increases the likelihood of an intracomponent tie. Hypothesis 6. Public ownership decreases the likelihood of a pendant tie. Our ﬁnal hypothesis focuses on the creation of a new component. Although publicly owned ﬁrms can create new components we argue that they are unlikely to do so. As discussed above, the creation of a new component requires that both partners are not in the network prior to the formation of the tie. We believe that it is unlikely that a publicly owned biotechnology ﬁrm has not yet formed any research or marketing and distribution ties, and thus is a network isolate (Gulati & Gargiulo, 1999; Stuart, 1998). Moreover, we believe that their higher institutional status makes them less likely to ally with isolates to form a new component, unless the partner is also a publicly owned ﬁrm with higher institutional status rather than structural status. We state the following hypothesis: Hypothesis 7. Public ownership decreases the likelihood of a component creating tie.

190

TERRY L. AMBURGEY ET AL.

Structural Position and Preferential Attachment So far the prominence of a node in the network was determined by nodal characteristics, especially the differential status of the nodes direct partner (eigenvector centrality). Our ﬁnal set of hypotheses is based upon tie characteristics reﬂecting the structural position of a node within a cohesive network subgroup (i.e., within a component). Cohesive subgroups are subsets of actors among whom there are relatively strong, direct or intense ties (Wasserman & Faust, 1994, p. 249). In social network analysis, the notion of subgroup is formalized by the general property of cohesion among subgroup members based on speciﬁed properties of the ties among the members. However, since the property of cohesion of a subgroup can be quantiﬁed using several different speciﬁc network properties, cohesive subgroups can be formalized by looking at different properties of the ties among subsets of actors. We will concentrate on adjacency as a property of cohesive subgroups. This approach is based on restrictions on the minimum number of nodes adjacent to each node in a subgroup of the network (c.f. Wassermann & Faust, 1994, p. 263). Since this measure is quantiﬁed by the degree of the node in a graph, this subgroup method focuses on nodal degree. Subgroups based on nodal degree require nodes to be adjacent to relatively numerous other subgroup members. However, unlike a clique deﬁnition that requires all members of a cohesive subgroup to be adjacent to all other subgroup members, this alternative requires that all subgroup members be adjacent to some minimum number of other subgroup members (c.f. Wassermann & Faust, 1994, p. 263). Our measure of cohesive subgroups based upon nodal degree will be the k-core of the node under observation. As discussed previously, the k-core measure indicates the cohesiveness of the subgroups (i.e., components) in the net, based upon the degree centrality of the nodes in the subgroup. A k-core is a subgroup in which each node is adjacent to at least a minimum number k, of the other nodes in the subgroup (see Wassermann & Faust 1994, p. 266). Thus, the larger the k-value of a node, the larger the cohesiveness of the subgroup (i.e., the component) in which he is embedded. Given our observation that the most cohesive area of the two networks represent the respective network-core, a high k-value of a node reﬂects its coreness in the network. Our hypotheses will address two consequences of a node’s coreness. Our ﬁrst set of hypotheses will address within-network processes, especially the consequences of coreness for structural homophily. Our second set of

Research and Commerce in Biotechnology

191

hypotheses will address across-network processes, especially the consequences of coreness for positional substitution.

Within-Network Processes: Coreness and Preferential Attachment The analysis of k-core values of network nodes leads – in its simplest way – to the distinction between members of a cohesive subgroup versus nonmembers, for example nodes that do not belong to any cohesive subgroup. In a more reﬁned interpretation, the coreness value indicates the distance of ﬁrms towards the core of the network, its most dense and most cohesive region. Coreness thus indicates membership in a dense, cohesive and robust subgroup of the network. Furthermore, coreness reﬂects connectivity. A high coreness value is based upon a high nodal-degree, for example the higher the number of ties of the node, the higher its coreness (k-core value). Therefore, the larger the coreness of a node, the less likely the node will become disconnected from the subgroup by the removal of a single tie. A biotech ﬁrm with a high coreness value will thus reside in a cohesive and robust area of the network. Similar to the advantages positional status of a biotech ﬁrm has for its access to critical resources – such as knowledge – its coreness will reﬂect the ease of accessing the information and resource rich core of the network (Doreian, 2008, p. 19). Thus, we expect a high coreness of a ﬁrm to have similar beneﬁts as a high centrality. The above described processes of homophily will therefore work in a similar way. A high coreness value signals a structural position in favour of accessing valuable resources, such as knowledge in the R&D network, or marketing capabilities in the M&D network, making the biotech ﬁrm a preferred partner for linkages. This process of preferential attachment will work in a similar way in both the R&D and M&D networks. Summing up, we argue that ﬁrms will form ties with other organizations of similar coreness. In some cases, the most similar organizations will be in the same component, in other cases the most similar organizations will be in other components. If the most similar (with regard to coreness) organizations are in the same component, this will lead to the formation of intra-component ties. If the next most similar (with regard to coreness) organizations are located in other components of roughly the same size, this will lead to the formation of bridging ties. We therefore hypothesize that biotech ﬁrms with high coreness should be more likely to form bridging ties and intra-component ties in both networks.

192

TERRY L. AMBURGEY ET AL.

Hypothesis 8. The closer a biotech ﬁrm is to the core of the network, the greater the likelihood of a bridging tie. Hypothesis 9. The closer a biotech ﬁrm is to the core of the network, the greater the likelihood of an intra-component tie. Similarly, we also hypothesize that biotech ﬁrms with high coreness should be more likely to form pendant ties. We believe that there are two reasons for this assumption. First, coreness based upon nodal degree indicates – as we have discussed earlier on – a history of active collaboration. The higher the overall coreness of a ﬁrm, the more ties it has formed. There is a vast of prior research indicating that ﬁrms with prior ties will form new ties of any kinds with an increasing probability (Anand & Khanna, 2000; Gulati, 1995b; Rothaermel & Hoang, 2005). The underlying reason is the development of a general capability of the ﬁrm to manage alliances in at least two ways: ﬁrst, by improving the efﬁciency of the partner selection process, and second by improving the efﬁciency of the task fulﬁllment within the alliance itself (c.f. Inkpen & Dinur, 1998). Prior ties will therefore make it more likely that the ﬁrm will cooperate with any kind of partner in the network, regardless of the structural position of that organization. Second, from the point of view of the pendant ﬁrm, a biotech ﬁrm with a high coreness is a more attractive partner than a peripheral ﬁrm (Gulati & Gargiulo, 1999; Stuart, 2000). It will thus link preferentially with core ﬁrms. We state: Hypothesis 10. The closer a biotech ﬁrm is to the core of the network, the greater the likelihood of a pendant tie.

Across-Network Processes: Coreness and Substitution Effects Our ﬁnal set of hypotheses addresses the assumption that a ﬁrm’s coreness will inﬂuence the probability to form ties across the industries value-chain. As discussed previously, the two prevailing networks in the biotech industry reﬂect the division of labor across the value chain (c.f. Fig. 2): R&D networks foster the joint exploration of scientiﬁc discoveries and innovative solutions, whereas M&D networks aim at joint exploitation of these discoveries. This reﬂects a sequence whereas ﬁrms engage in R&D

Research and Commerce in Biotechnology

193

collaborations ﬁrst; followed by M&D collaborations as soon as there is a solution ripe to market. M&D thus follows R&D in the biotech industry (c.f. Rothaermel & Deeds, 2004). This sequence will therefore entail ﬁrms to be involved in the two networks sequentially. However, studies examining the industry evolution in biotech have pointed to the fact that as the evolution enfolds, ﬁrms reach a stage where they market products, and simultaneously engage in R&D (c.f. Baum et al., 2000; Rothaermel & Deeds, 2004; Silverman & Baum, 2002). This is due to the long and complex product development cycle, where ﬁrms constantly have to ﬁll their ‘‘research pipeline’’ to ensure a stream of marketable targets. It follows that as the industry evolves ﬁrms can be involved in both networks simultaneously. What consequences does this evolutionary pattern have for our hypotheses? As ﬁrms participate in both networks, we cannot assume that their structural position in one network will be independent from the structural position in the other network (Neuman, Davis & Mizruchi, 2008). There are several lines of argumentation that we want to draw. Given the sequential pattern that we have outlined above, we assume that a ﬁrm’s structural position in the R&D network will inﬂuence its position in the M&D network. Second, given the simultaneous pattern that we have outlined above, we assume that a ﬁrm’s position in the M&D network will spill over to its subsequent position in the R&D network. Thus, status and position in one network are not independently from the other network, and ﬁrms can substitute for status and position among networks. These substitution effects will lead to the same consequences of structural position for the formation of different types of ties that we have outlined above. We thus state: Hypothesis 11. The closer a biotech ﬁrm is to the core of one network, the greater the likelihood of a bridging tie in the other network. Hypothesis 12. The closer a biotech ﬁrm is to the core of one network, the greater the likelihood of an intra-component tie in the other network. Hypothesis 13. The closer a biotech ﬁrm is to the core of one network, the greater the likelihood of a pendant tie in the other network.

194

TERRY L. AMBURGEY ET AL.

DATA AND METHODS Data Sources and Structure There are two longitudinal datasets used in this study, the ﬁrst one covering the network of R&D agreements and the second, the network of M&D agreements. The data used in the study is derived from a dataset of 847 new U.S. biotechnology ﬁrms founded during the period 1973–1997. The signiﬁcance of the beginning date was the major Cohen– Boyer breakthrough of 1973 involving recombinant DNA using plasmids. We used two primary sources to compile the sample. The ﬁrst was the Bioscan database published by Oryx Press (1984–2000). The second source was the U.S. Companies Database compiled by the North Carolina Biotechnology Center (now Bioworld). We purged the consolidated list of companies of all ﬁrms which were founded before 1973, those which were not U.S. ﬁrms, or which were non-independent entities (subsidiaries, divisions and joint ventures) to arrive at 847 companies. These ﬁrms were observed from 1973 to 2000. During the period studied, 235 ﬁrms exited from the industry through liquidation or acquisition. These data, and other sources, were used to construct an event history for each company. Event histories are data structures that include information on the number, timing, and sequence of the events that are being examined. Our data on biotechnology ﬁrms includes, in addition to strategic alliances, date of incorporation or authorization to do business, liquidations, name changes, mergers or acquisitions, patents, initial public offerings, secondary public offerings, and private placements of equity. Each ﬁrm’s history began at the time of its incorporation or qualiﬁcation to do business, and ended at the time of an event or at the end of the month, whichever came ﬁrst. The organization’s second spell began on the following day and ended at the time of an event or the end of the month. This pattern continued until the ﬁrm exited through failure or acquisition or until the end of the observation period, and such spells were coded as right censored, allowing time-varying covariates to be updated throughout the ﬁrm’s history at monthly intervals. A wide variety of sources were used to augment the information in our two primary sources. We examined the legal archives on the LexisNexis service to obtain exact dates of incorporation or qualiﬁcation to do business as well as dates of mergers, acquisitions, and changes of name. A search of the news archives on Lexis-Nexis (including specialized outlets such as Biotechnology News Watch) was used to identify dates of

Research and Commerce in Biotechnology

195

events. The online archives of Recombinant Capital and Knowledge Express also provided supplemental information on strategic alliances. Information from the U.S. Patent and Trademark Ofﬁce was the primary source for the assignment of patents; including CD-ROMs and online searches of the PTO web site. Where only the month and year of an event could be determined, the day was set at the midpoint of the month to minimize timing errors. The event histories used in the analyses cover the period 1973–2000. These data, and other sources, were used to construct an event history for each company. Event histories are data structures that include information on the number, timing, and sequence of the events that are being examined. The focus of our research attention is on the biotechnology ﬁrms. Therefore, other organizations in the organizational ﬁeld, such as universities, government laboratories, pharmaceutical companies, etc. enter our data only as strategic alliance partners for biotech ﬁrms. No other data has been collected on the partner organizations.

Construction of Quarterly Networks The information on R&D and M&D agreements was used to construct a series of quarterly networks. The quarterly networks for both types of agreements cover the period 1983–2003, since there is no effective R&D or M&D network in existence prior to 1983. UCINET VI was used to compute, for each quarterly network, the number of network components, normalized entropy, component membership, and coreness strata for each organization in the network. The information on component membership was used to categorize the formation of each alliance in the event histories as a bridging tie, pendant tie, intra-component tie, or component creating tie in the appropriate network. For single partner agreements the procedure was straightforward. The component membership of the partner in the previous quarter was compared to the component membership of the focal ﬁrm in the previous quarter. If both organizations were previously in the network but in different components the tie was a bridging tie. If both organizations were previously in the network and in the same component, the tie was an intracomponent tie. If one of the organizations was previously in the network but the other was not then the tie was a pendant tie. Finally, if neither of the organizations was previously in the network, the tie was a component creating tie. Not all agreements consist of dyads. In the U.S. data there

196

TERRY L. AMBURGEY ET AL.

are a number of agreements with a focal ﬁrm that involves two partners. The evaluation of structural consequences was performed across all of the partners.

Measures The dependent variables are the tie formation and tie termination rates l(t). Each rate is deﬁned as lðtÞ ¼ lim½qðt; t þ DtÞ=Dt; Dt ! 0 where q is the discrete probability of the ﬁrm initiating or terminating that type of tie between t and (tþDt), conditional on the history of the process up to time t. This rate summarizes the information on the intervals of time between successive events, with higher values of the rate corresponding to shorter times between events and vice versa.

Independent Variables Three types of independent variables were used in the analysis: attributes of the population, attributes of the local network, and attributes of individual ﬁrms. Population variables included population density, corporate patents granted, and counts of research alliances by other ﬁrms. Population density was deﬁned as the total number of biotechnology ﬁrms at the beginning of a calendar year. The density variable was adjusted to reﬂect the disappearance of ﬁrms through failure, acquisition, or merger. The annual number of corporate patents granted in genetic engineering was also included, and was used to measure cumulative patent activity. The U.S. Patent and Trademark Ofﬁce publish information on the date of every patent issued, and this source was used to construct the number of patents granted to biotech ﬁrms each year. Annual counts of research alliances among all biotechnology ﬁrms in the sample were used to capture competitive rivalry produced by cooperative strategies. These counts were adjusted to remove each ﬁrm’s alliances and patents. The quarter variable was included to allow for time dependence. The second set of variables consists of global attributes of the network. The ﬁrst network variable is the number of isolated components in the network. The second is normalized entropy. Normalized entropy represents the degree of distributional equality, the extent to which the components are

197

Research and Commerce in Biotechnology

of relatively equal or unequal size. It is computed as follows: H¼

n X ½ðPk =PÞ n lnðPk =PÞ k¼1

H n ¼ H= ln n ðnormalizedÞ Where n is the total number of subgroups present in the population, Pk is the size of the kth subgroup, and P is the total population of all subgroups included in the index. The third set of variables measure attributes of the local network neighborhood, the components and the k-cores. We determined the number of nodes (organizations) and the number of ties in each component for each quarter. These variables, typically used to construct a ratio were used separately as measures of component size and tie volume. We prefer to evaluate the number of members (size) and the number of ties separately for two reasons. The ﬁrst is the belief that size is an interesting feature in its own right, particularly the distribution of sizes. The second is the recognition that maximum density is most likely to occur with dyads. This feature of the ratio measure has special consequences for component destruction. Since components are destroyed only when all of the ties connecting the members are terminated, dyads are most likely to be destroyed since only one tie need be removed for the component to be destroyed. If maximum density always occurs with dyads we can ﬁnd a misleading relationship between density and the rate of component destruction. The structural position of the ﬁrms was identiﬁed through a k-core decomposition of the quarterly networks. The fourth set of variables measured attributes of the ﬁrms themselves. The ﬁrst such variable was age, measured as the number of days since the founding or qualiﬁcation of the ﬁrm. The second was the cumulative number of prior patents. This variable was updated whenever a patent was granted. The third was the public/private ownership status of the ﬁrm. The fourth was cumulative number of prior research alliances, calculated for each ﬁrm. The ﬁfth was quarterly the Bonacich (eigenvector) centrality score for each organization in the network. This indicator can formally be deﬁned as: st ða; BÞ ¼

1 X

aBk Rkþ1 1: t

k¼0

In this expression, a is a scaling coefﬁcient, B is a weighting parameter that can range between zero and the absolute value of the inverse of the value of

198

TERRY L. AMBURGEY ET AL.

the maximum eigenvalue of the sociomatrix Rt, 1 is a column vector where each element has the value ‘‘1,’’ and st is also a column vector where element Si, t denotes the status of biotech organization i. Given this speciﬁcation, a biotech ﬁrm’s position (status) is a function of the number and the position of the ﬁrms with which it forms cooperative research agreements. In turn, the position of these partners is the function of the number and the position of their partners, and so on. The B parameter is set equal to the reciprocal of the maximum eigenvalue.

Model The event series was modeled as a stochastic point process (Amburgey, 1986). The alliance rates l(t) were each speciﬁed as an exponential function of the independent variables and a set of parameters capturing the effects of the variables on the founding rate such that: lðtÞ ¼ expðbX t Þ Parameters were estimated using maximum likelihood with the STATA program. The estimation procedure clustered observations by ﬁrm to reduce the impact of unobserved ﬁrm-speciﬁc effects (White, 1982). The signiﬁcance levels of the parameters were evaluated by examination of t-ratios, whereas the goodness-of-ﬁt of the different models was evaluated by examination of likelihood ratio statistics.

RESULTS Table 1 provides descriptive statistics for the independent variables used in the analyses of formations. Table 2 provides parameter estimates and standard errors of estimates for each of the four models of the formation rates of R&D agreements. Each of the models is statistically signiﬁcant at a very high level compared to a baseline null model. Table 3 provides parameter estimates and standard errors of estimates for each of the four models of the formation rates of M&D agreements. Each of the models is statistically signiﬁcant at a very high level compared to a baseline null model.

199

Research and Commerce in Biotechnology

Table 1(a). Variable 1. Density 2. Total patents 3. Total alliances 4. Quarter 5. R&d x components 6. M&d x components 7. R&d norm. entropy 8. M&d norm. entropy 9. Age 10. Prior patents 11. Prior R&D 12. Prior M&D 13. Publicly traded 14. R&D centrality 15. M&D centrality 16. R&D kcore 17. M&D kcore 18. R&D comp. nodes 19. M&D comp. nodes 20. R&D comp. ties 21. M&d comp. ties

Descriptive Statistics.

Mean

Std. Dev.

Min

Max

542.31 535.26 533.18 46.29 35.79 48.54 0.12 0.28 2867.58 4.74 3.90 1.91 0.33 1.77 1.28 1.45 0.60 467.29 107.64 41.12 61.20

83.19 16.17 238.40 18.16 6.98 11.94 0.124 0.128 1846.52 16.70 7.82 4.66 0.47 4.86 5.26 1.73 1.00 550.92 221.41 226.67 126.83

268 10 43 9 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0

612 1688 934 75 45 61 0.72 0.64 9823 357 80 44 1 97.33 100 6 4 1381 704 2787 517

DISCUSSION This chapter reports results from an ongoing research project examining the evolution of structure in inter-organizational networks in the ﬁeld of biotechnology. Our conceptual frame of reference integrates organizationlevel, subgroup, and overall network-level processes, considering the dynamics of several levels of analysis. To our understanding, network evolution involves structural change over time driven by network processes; and network processes, in turn, are series of events that create, sustain and dissolve network structures. Analyzing data on the complete biotech population in the U.S. for the period 1973–2000 for R&D and M&D networks our research shows that these networks are characterized by the development of a core and periphery structure. Descriptively this process is shown by the development of several distinct indicators over time. First, the number of components in the quarterly R&D networks and the normalized entropy, a measure of the heterogeneity in the sizes of the network components indicate a trend

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.

1.0 0.48 0.55 0.83 0.15 0.93 0.82 0.85 0.40 0.14 0.21 0.16 0.17 0.08 0.06 0.29 0.20 0.31 0.25 0.51 0.24

1.0 0.09 0.84 0.61 0.57 0.42 0.57 0.46 0.31 0.25 0.15 0.21 0.08 0.04 0.30 0.18 0.53 0.31 0.18 0.25

1.0 0.34 0.45 0.55 0.40 0.41 0.12 0.09 0.06 0.07 0.07 0.02 0.03 0.10 0.09 0.20 0.10 0.28 0.18

1.0 0.30 0.85 0.66 0.83 0.51 0.26 0.27 0.18 0.22 0.08 0.05 0.34 0.22 0.59 0.34 0.34 0.30

1.0 0.07 0.02 0.16 0.20 0.21 0.11 0.05 0.09 0.03 0.01 0.11 0.05 0.39 0.15 0.12 0.10

1.0 0.83 0.76 0.41 0.16 0.22 0.16 0.18 0.09 0.06 0.31 0.21 0.32 0.27 0.56 0.26

1.0 0.82 0.32 0.13 0.17 0.13 0.13 0.07 0.05 0.24 0.17 0.11 0.20 0.68 0.19

Descriptive Statistics.

1.0 0.41 1.0 0.17 0.25 1.0 0.21 0.27 0.53 1.0 0.16 0.32 0.34 0.71 1.0 0.17 0.22 0.28 0.51 0.44 1.0 0.07 0.08 0.25 0.59 0.41 0.32 1.0 0.05 0.19 0.21 0.50 0.73 0.28 0.44 1.0 0.29 0.21 0.38 0.72 0.55 0.62 0.53 0.37 1.0 0.19 0.34 0.27 0.53 0.77 0.49 0.31 0.56 0.60 1.0 0.34 0.31 0.16 0.15 0.09 0.14 0.03 0.01 0.17 0.11 1.0 0.27 0.35 0.27 0.46 0.61 0.42 0.20 0.39 0.51 0.79 0.22 1.0 0.42 0.16 0.05 0.08 0.07 0.06 0.05 0.02 0.13 0.09 0.04 0.09 1.0 0.25 0.33 0.24 0.45 0.60 0.41 0.21 0.39 0.50 0.78 0.20 0.97 0.09 1.0

Table 1(b).

200 TERRY L. AMBURGEY ET AL.

Firm Level

Network L.

Population L.

Table 2.

R&D centrality

Public ownership

Prior M&D alliances

Prior R&D alliances

Prior patents

Age (Days)

Normalized entropy

Isolated components

Time (No. of Quarter)

Total patents granted (Cum.) Total alliances (Cum.)

Population density

Variables

0.00191 (.00326) 0.00057 (.00101) 0.00143 (.00097) –0.08676 (.02785) 0.06302 (.01650) 0.20810 (.93368) –0.00041 (.00010) –0.00309 (.01037) 0.02348 (.01531) –0.13573 (.04547) 0.70085 (.20214) 0.00779 (.01018)

Model 1 (Bridging Ties)

0.00277 (.00178) 0.00081 (.00029) 0.00186 (.00032) 0.03903 (.01289) 0.03538 (.00905) 2.13978 (.63740) –0.00034 (.00005) –0.01355 (.00798) 0.02520 (.01032) –0.02181 (.02620) 0.71211 (.13035) 0.00864 (.01137)

Model 2 (Pendant Ties)

0.00267 (.00149) 0.00008 (.00018) 0.00226 (.00019) 0.01978 (.00875) 0.00239 (.00769) 1.36632 (.41596) –0.00031 (.00003) –0.00635 (.00280) 0.03434 (.00683) –0.47667 (.01612) 0.33789 (.08882) 0.01412 (.00479)

Model 3 (IntraComponent Ties)

–0.00344 (.00528) –0.00209 (.00123) 0.00025 (.00105) 0.05871 (.04308) –0.01683 (.01632) 2.92081 (.71778) –0.00037 (.00009) –0.06261 (.07067) –0.28323 (.11161) 0.33133 (.18290) 1.16000 (.23064)

Model 4 (Component Creation Ties)

The Effects of Population, Network, Component and Firm Level Variables on the Consequences of Network Tie Formation: R&D.

Research and Commerce in Biotechnology 201

¼ signiﬁcant at o0.05

Number of nodes in the component Number of ties in the component Number of events Chi2 Degrees of freedom Probability value

M&D coreness

R&D coreness

M&D centrality

Variables

Model 2 (Pendant Ties)

0.00901 (.01080) 0.21388 (.05007) 0.04696 (.08170) –0.00000 (.00009) –0.00009 (.00026) 924 560.44 17 Po.001

Model 1 (Bridging Ties) 0.02396 (.00936) 0.44499 (.07337) 0.41445 (.13688) –0.00016 (.00014) –0.00021 (.00026) 360 1109.41 17 Po.001

Table 2. (Continued )

0.01593 (.00454) 0.60023 (.02829) –0.05284 (.04974) 0.00022 (.00006) –0.00194 (.00098) 2828 2531.37 17 Po.001

Model 3 (IntraComponent Ties)

129 131.84 13 Po.001

–2.34247 (1.21165)

–0.02747 (.06374)

Model 4 (Component Creation Ties)

202 TERRY L. AMBURGEY ET AL.

Firm Level

Network Level

R&D centrality

Public status (Dummy)

Prior M&D alliances

Prior R&D alliances

Prior patents (No.)

Age (Days)

Normalized entropy

Number of components

Time (No. of Quarter)

Total patents granted (Cum.) Total alliances (Cum.)

Population density

Variables

0.00176 (.00652) –0.00096 (.00091) 0.00013 (.00086) 0.00609 (.03142) 0.05240 (.03384) 6.16457 (3.56400) 0.00009 (.00009) –0.01488 (.02659) 0.00023 (.02080) 0.08721 (.03539) 1.0473 (.50246) 0.00348 (.01626)

Model 5 (Bridging Ties)

Model 7 (Intracomponent Ties) 0.00664 (.00242) 0.00049 (.00043) 0.00151 (.00043) –0.08334 (.02403) 0.00054 (.01542) 2.80298 (.82055) 0.00013 (.00007) –0.04995 (.00608) 0.00333 (.00827) 0.02532 (.01993) 0.57840 (.19825) 0.01224 (.00979)

Model 6 (Pendant Ties) 0.00471 (.001978) –0.00017 (.00029) 0.00067 (.00204) –0.02473 (.07406) 0.01413 (.01128) 3.59433 (6.6144) –0.00014 (.00004) –0.00596 (.00631) –0.00532 (.00948) 0.06528 (.01631) 0.53382 (.15450) 0.00329 (.00773)

1.17287 (.18766) –0.01743 (.04252)

0.00440 (.00364) –0.00052 (.00073) 0.00176 (.00080) –0.01987 (.02468) –0.02006 (.01746) 3.31020 (.77794) –0.00018 (.00005) –0.01497 (.01675) –0.01112 (.02868)

Model 8 (Component Creation Ties)

The Effects of Population, Network, Component, and Firm Level Variables on the Consequences of Network Tie Formation: M&D.

Population Level

Table 3.

Research and Commerce in Biotechnology 203

Number of nodes in the component Number of ties in the component Number of events Chi2 Degrees of freedom Probability value

¼ signiﬁcant at o0.05

Component Level

M&D coreness

R&D coreness

M&D centrality

Variables

Model 6 (Pendant Ties) –0.04904 (.01699) 0.28897 (.04754) 0.77669 (.08422) –0.00134 (.00066) 0.00192 (.00103) 1091 4396.71 17 Po.001

Model 5 (Bridging Ties) –0.03777 (.01434) 0.38722 (.11310) 1.50036 (.11310) –0.00484 (.00204) –0.00493 (.00541) 109 650.31 17 Po.001

Table 3. (Continued )

–0.00958 (.01360) 0.30674 (.05176) 0.78308 (.11302) 0.001627 (.001325) 0.00244 (.00109) 633 1344.97 17 Po.001

Model 7 (Intracomponent Ties)

209 164.24 12 Po.001

0.06003 (.08000)

Model 8 (Component Creation Ties)

204 TERRY L. AMBURGEY ET AL.

Research and Commerce in Biotechnology

205

towards a few large components surrounded by a larger number of small components. Second, both networks are characterized by increasing differentiation within the large central components as shown by the increasing number of ‘‘coreness’’ strata. We believe that evolutionary models should have several features: dynamic change over time, history dependence, multiple levels of analysis, and both modiﬁcation and replacement processes. What does a network ecology perspective show us? There is deﬁnitely dynamic change over time. Regarding the number of components, we see an initial sharp rise in the number of components followed by a long reduction in the number of components. In both cases, the early period is characterized by new ﬁrms entering as isolates and entering the network through the formation of isolated components. However, even as the number of new components proliferates, the networks begin to develop a core/periphery structure as a small group of components begin to grow in size. Moreover, these larger components develop a differentiated internal structure. There are multiple levels of analysis and cross level effects as well. One example can be seen in the growth of components. This growth occurs in two ways. First there is the consolidation of components through the establishment of bridging ties which simultaneously reduce the number of components and increase the size of the resulting component. The second method of growth is the addition of pendants where an isolate is linked to an organization already in the network. The development of a core-periphery is largely produced by differential rates of the different types of events which occur at the organizational level. Contrary to our ﬁrst set of hypotheses (H1–H7), both methods of growth appear to be substantially driven by organizations with higher institutional status, consistent with the logic of accumulation (c.f. Powell et al., 2005). In both networks, public organizations with high status are more likely to form the bridging ties that consolidate components and to participate in the addition of pendants. However, public organizations are also more likely to bring isolates into the network through the creation of new components. We thus conﬁrm Hypothesis 4 and 5, and disconﬁrm Hypothesis 6 and 7. With regard to structural status we see different patterns in the R&D versus M&D networks. Turning to the R&D network ﬁrst, we only get support for our second hypotheses (H2), but no support for H1 and H3. The structural status of ﬁrms does not inﬂuence their propensity to form bridging or form pendant ties. However, ﬁrms with a high eigenvector centrality form more intra-component ties, thereby increasing the density and robustness of their component.

206

TERRY L. AMBURGEY ET AL.

In sharp contrast stand the ﬁndings with regards to the M&D network. Here the estimators for all three types of ties are signiﬁcant and negative. We thus conﬁrm H3, but disconﬁrm H1 and H2. Centrality in the M&D network therefore reduces ﬁrm’s propensities to form ties with speciﬁc structural consequences. We ﬁnd a process of history dependence as well. Turning to our second set of hypotheses (H8, H9, and H10) the ﬁndings clearly conﬁrm our assumptions for both networks. Biotech ﬁrms that move towards the core of the network are more likely to form ties that shape the evolution of the network. The higher the coreness of ﬁrms within their network, the higher the probability that they form bridging and intra-component ties, and link isolates to the net. Core ﬁrms thus consolidate the network, thereby further increasing the density and robustness of the core. These ﬁndings indicate that there is a process of preferential attachment based upon structural homophily. A high coreness value signals a structural position in favour of accessing valuable resources, such as knowledge in the R&D network, or marketing capabilities in the M&D network, making the biotech ﬁrm a preferred partner for linkages. This process of preferential attachment works in a parallel fashion in both the R&D and the M&D network. Turning to our third set of hypotheses (H11, H12, and H13), the ﬁndings partially conﬁrm cross effects, for example effects across the two networks. The substitution effects that we have assumed work particularly well in the M&D network. The closer ﬁrms are in the core of the R&D network, the more they are able to leverage this advantage, and form ties that change the structure of the M&D network. Therefore, ﬁrms that are deeply embedded in research and development are able to inﬂuence the evolution of the marketing and distribution network to a signiﬁcant extent, besides those ﬁrms that are core in the M&D network. The opposite, however, only works partially. Firms that are core in the M&D network are only partially able to drive the evolution of the R&D network. The substitution effect only holds for bridging ties: ﬁrms in the core in M&D consolidate the R&D net by linking components to a larger subgroup, but they do not increase the density of the components or link isolates. To sum up, it seems clear that preferential attachment is an important element in the emergence and evolution of network structure and that attachment based upon institutional status as well as structural position plays a crucial role. Furthermore, there is evidence that the network position (i.e., coreness) of ﬁrms inﬂuences their propensity to form ties which in turn

Research and Commerce in Biotechnology

207

alter the network structure and over time the network position of ﬁrms. Our research indicates that biotechnology R&D and M&D networks in the United States are developing a distinct core/periphery structure over time. Our results – although mixed – provide support for a process of status based preferential attachment (homophily) where organizations tend to form ties with similar organizations, for example organizations of similar status. This tends to occur both when the measure of status is positional, institutional or structural, although with deviations in the M&D network.

REFERENCES Albert, R., & Barabasi, A.-L. (2002). Statistical mechanics of complex networks. Reviews of Modern Physics, 74, 47–97. Amburgey, T. L. (1986). Multivariate point processes in social research. Social Science Research, 15, 190–206. Amburgey, T. L. & Al-laham, A. (2007). Islands in the Net. Structural evolution of R&D networks in Biotechnology. Working Paper. Amburgey, T. L., Dacin, T., & Singh, J. V. (1996). Learning races, patent races, and capital races: Strategic interaction and embeddedness within organizational ﬁelds. In: J. Dutton & J. A. C. Baum (Eds), Advances in strategic management (pp. 303–322). Greenwich: JAI Press. Amburgey, T. L., & Singh, J. V. (2002). Organizational evolution. In: J. A. C. Baum (Ed.), Companion to organizations (pp. 327–343). Malden: Blackwell Publishers, Inc. Anand, B. T., & Khanna, T. (2000). Do ﬁrms learn how to create value? The case of alliances. Strategic Management Journal, 295–315. Arora, A., & Gambardella, A. (1990). Complementarity and external linkages: The strategies of the large ﬁrms in biotechnology. The Journal of Industrial Economics, 38, 361–379. Baum, J. A. C., Calabrese, T., & Silverman, B. (2000). Don’t go it alone: Alliance network composition and startup’s performance in Canadian biotechnology. Strategic Management Journal, 21, 267–294. Baum, J. A. C., & Mezias, S. (1992). Localized competition and organizational failure in the manhattan hotel industry. 1898–1990. Administrative Science Quarterly, 37, 580–604. Casper, S. (2000). Institutional adaptiveness, technology policy, and the diffusion of new business models: The case of German biotechnology. Organization Studies, 21, 887–914. Doreian, P. (2008). Actor utilities, strategic action and network evolution. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 247–271). Oxford, UK: JAI/Elsevier. Doreian, P., & Stokman, F. N. (1997). Evolution of social networks. Amsterdam: Gordon and Breech Publishers. Gambardella, A. (1995). Science and innovation: The U. S. pharmaceutical industry during the 1980s. Cambridge: Cambridge University Press. Grant, R. M., & Baden-Fuller, C. (1995). A knowledge-based theory of inter-ﬁrm collaboration. Proceedings of the 1995 Academy of Management, 17–21.

208

TERRY L. AMBURGEY ET AL.

Gulati, R. (1995a). Competition for competence and interpartner learning within international strategic alliances. Strategic Management Journal, 12, 83–103. Gulati, R. (1995b). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from. American Journal of Sociology, 104, 1439–1493. Hagedoorn, J. (2002). Inter-ﬁrm R&D partnerships: An overview of major trends and patterns since 1960. Research Policy, 31, 477–492. Hamel, G. (1991). Competition for competence and interpartner learning within international strategic alliances. Strategic Management Journal, 12, 83–103. Hite, J. (2008). The evolution of strategic dyadic network ties: Strategically navigating bounded agency within multidimensional and dynamic dyadic relationships. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 133–170). Oxford, UK: JAI/Elsevier. Inkpen, A. C., & Dinur, A. (1998). Knowledge management processes and international joint ventures. Organization Science, 9, 454–468. Khanna, T., Gulati, R., & Nohria, N. (1998). The dynamics of learning alliances: Competition, cooperation and relative scope. Strategic Management Journal, 19, 193–210. Kogut, B. (2000). The network as knowledge: Generative rules and the emergence of structure. Strategic Management Journal, 1, 405–425. Lomi, A. (1995). The population ecology of organizational founding: Location dependence and unobserved heterogeneity. Administrative Science Quarterly, 40, 111–144. Lyles, M. (1994). The impact of organizational learning on joint venture formation. International Business Review, 459–468. McEvily, B., & Zaheer, A. (1999). Bridging ties: A source of ﬁrm heterogeneity in competitive capabilities. Strategic Management Journal, 20, 1133–1156. McPherson, G. M., & Smith-Lowen, L. (1987). Homophily in voluntary organizations. Status distance and the composition of face to face groups. American Sociological Review, 52, 370–379. Mowery, D., Oxley, J., & Silverman, B. (1996). Strategic alliances and interﬁrm knowledge transfer. Strategic Management Journal, 17, 77–91. Neuman, E. J., Davis, G. F., & Mizruchi, M. (2008). Networks and industry consolidation in U.S. Global Banking, 1986–2004. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 211–245). Oxford, UK: JAI/Elsevier. Pisano, G. P. (1990). The R and D boundaries of the ﬁrm: An empirical analysis. Administrative Science Quarterly, 35, 153–176. Podolny, J. (1993). A status-based model of market competition. American Journal of Sociology, 98, 829–872. Podolny, J. (1994). Market uncertainty and the social character of economic exchange. Administrative Science Quarterly, 39, 458–470. Podolny, J. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107, 33–60. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41, 116–145.

Research and Commerce in Biotechnology

209

Powell, W. W., Koput, K. W., White, D. R., & Owen-Smith, J. (2005). Network dynamics and ﬁeld evolution: The growth of interorganizational collaboration in the life sciences. American Journal of Sociology, 110, 116–145. Rothaermel, F. T., & Deeds, D. L. (2004). Alliance type, alliance experience and alliance management capability in high-technology ventures. Journal of Business Venturing, 21, 429–460. Rothaermel, F. T., & Hoang, H. (2005). The effect of general and partner-speciﬁc alliance experience on joint R&D project performance. Academy of Management Journal, 48, 332–345. Rowley, T. J., Greve, H., Rao, H., & Baum, J. A. C. (2005). Time to break up: Social, and instrumental antecedents of ﬁrm exits from exchange cliques. Academy of Management Journal, 48, 499–520. Silverman, B. S., & Baum, J. A. C. (2002). Alliance based competitive dynamics. Academy of Management Journal, 45, 791–806. Skvoretz, J., Fararo, T. J., & Agneesens, F. (2004). Advances in biased net theory: Deﬁnitions, derivations, and estimations. Social Networks, 26, 113–139. Stuart, T. E. (1998). Network positions and propensities to collaborate: An investigation of strategic alliance formation in a high technology industry. Administrative Science Quarterly, 43, 668–698. Stuart, T. E. (2000). Interorganizational alliances and the performance of ﬁrms: A study of growth and innovation rates in a high-technology industry. Strategic Management Journal, 21, 791–811. Stuart, T. E., Hoang, H., & Hybels, R. C. (1999). Interorganizational endorsements and the performance of entrepreneurial ventures. Administrative Science Quarterly, 44, 315–349. Uzzi, B., Guimera, R., Spiro, J., & Amaral, L. (2005). Emergence: The dynamics of network formation. Unpublished manuscript. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8, 109–125. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge: Cambridge University Press. White, H. (1982). Maximum likelihood estimation of misspeciﬁed models. Econometrica, 50, 1–25.

INDUSTRY CONSOLIDATION AND NETWORK EVOLUTION IN U.S. GLOBAL BANKING, 1986–2004 Eric J. Neuman, Gerald F. Davis and Mark S. Mizruchi ABSTRACT This chapter analyzes the relations among bank mergers, changes in boards and their networks, and changes in the global footprint of merging banks. We examine all mergers involving U.S. banks with foreign branches between 1986 and 2004. We ﬁnd that while the largest banks have become even larger through mergers, their boards have stayed roughly the same size with the same pattern of connections, leaving banks relatively less central in the intercorporate network. And while global banks previously had more globally oriented boards, this is no longer the case, as the link between board networks and strategy has become more tenuous. Because global banks were particularly prone to merging, the average commercial bank in the U.S. is now far more domestically oriented than ﬁrms in most other industries. American banks have thus become more domestic at the same time that the rest of American industry has grown much more global.

Network Strategy Advances in Strategic Management, Volume 25, 211–245 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25006-0

211

212

ERIC J. NEUMAN ET AL.

Banks had networking strategies long before anyone in business used the term network. In 1914, Louis Brandeis described the ‘‘endless chain’’ created when banks such as JP Morgan & Co. sent their ofﬁcers to serve on the boards of client ﬁrms (Brandeis, 1914). The largest banks’ executives collectively served on dozens of corporate boards. George F. Baker of First National personally served on 22 boards – far outpacing even Vernon Jordan in his heyday. Bank-centered networks were not limited to the United States. In describing the contemporary German system, Lenin (1916, pp. 41–42) stated that ‘‘a very close personal union is established between the banks and the biggest industrial commercial enterprises y through the acquisition of shares, through the appointment of bank directors to the Supervisory Boards (or Boards of Directors) of industrial and commercial enterprises, and vice versa.’’ Banks provided one-stop-shopping for control of industry, a ﬁnding with real practical value for a prospective chief executive with Lenin’s orientation. In the U.S. and elsewhere, banks continued to hold their central position for decades, their boards comprised of executives from the best-connected corporations and non-proﬁts (Mizruchi, 1982). Whether as a means to gather high-level intelligence to guide investment choices (Mintz & Schwartz, 1985) or a device to signal legitimacy to prospective corporate clients (Davis & Mizruchi, 1999), board networks were central to bank strategies for most of the twentieth century. Since the early 1980s, there have been fundamental changes in the banking industry. Advances in information and communication technologies, deregulation, and the consequent expansion of ﬁnancial markets have undermined the need for commercial banks, as creditworthy borrowers increasingly turned to ﬁnancial markets for debt. Consolidation of the banking industry, long prevented by the peculiarities of America’s federal system of regulation, began in earnest in the 1990s. Banks responded to new opportunities and challenges in diverse ways, including geographic expansion and industrial diversiﬁcation. Some, such as NationsBank (now known as Bank of America) and Bank One, went on acquisition binges, growing from regional to super-regional to near-national, with a focus on operating retail branches in ever-broader territories. Others, such as Mellon Bank and Bankers Trust (the latter now owned by Deutsche Bank), diversiﬁed into businesses such as investment banking and mutual funds. And still others, such as Citibank and the Bank of Boston, expanded their reach into global markets, by buying or expanding branch networks outside the U.S. Sometimes these strategies collided, as the largest banks were folded into an elite group of mega-banks. NationsBank, for example – previously a primarily domestic institution based in North Carolina – ended

Industry Consolidation and Network Evolution in U.S. Global Banking

213

up with a vast network of overseas branches due to its acquisitions of BankAmerica and FleetBoston. Prior work suggests that networks are both a cause and a consequence of bank strategies. As banks retreated from the corporate lending business during the 1980s and 1990s, they also shrank their boards and curbed their traditional practice of recruiting ‘‘celebrity directors’’ (Davis & Mizruchi, 1999). Having a well-connected board was a consequence of pursuing a strategy of lending to business. As commercial banks adopted a more ﬁnancial service-oriented approach, their boards came to look more like those of investment banks. Once in place, however, a board can have a considerable inﬂuence on strategy. Well-connected banks, for example, were quicker to globalize in the 1960s than their more peripheral peers, suggesting that so-called celebrity directors brought actionable intelligence to bear on bank decision making (Mizruchi & Davis, 2004). In this chapter, we link networks and strategy in the context of the ongoing bank merger wave.1 We are particularly interested in understanding how the global footprint of American banks has changed as the number of participants has declined. Several of the most venerable names in global banking have disappeared or shifted identities, leaving only three signiﬁcant players in the global market: Citibank, JP Morgan Chase, and Bank of America. We address four questions: First, do the boards of global banks look different from those of domestic banks? Second, how do boards of global banks change after an acquisition? Third, how has the position of global banks in intercorporate networks changed as a result of industry consolidation? And fourth, how does the composition of the board inﬂuence a bank’s approach to globalization following mergers? In addressing these questions, this chapter contributes to understanding the organizational dynamics underlying ﬁnancial globalization and the link between networks and strategy, especially with regard to ﬁrm-level and industry-level endogenous drivers of network change. Consistent with Amburgey, Al-Laham, Tzabbar, and Aharonson (2008) and Hite (2008), we show the changing nature of network ties and structures over time. Our ﬁndings also provide an informative contrast with those of Conyon and Muldoon (2008), who show that ﬁnancial institutions continued to play a critical role in the intercorporate network in the U.K. as late as 2000.

U.S. BANKS AND GLOBALIZATION Until very recently, ﬁnancial institutions in the United States were fragmented both geographically and industrially. Geographically,

214

ERIC J. NEUMAN ET AL.

commercial banks (that is, banks that gather deposits from savers and lend funds to businesses and other borrowers) were organized on a state-by-state level and regulated both by the Federal Reserve Bank (a national entity) and state-level ofﬁcials. Despite having names like ‘‘Bank of America,’’ from 1832 until the early 1980s, banks (with rare exceptions) did not operate branches in more than one state within the U.S., and many states limited local banks to only a single branch (Marquis & Lounsbury, 2007). Geographic regulation was eased in the 1980s, and in 1994 the RiegleNeal Act lifted restrictions on interstate banking, prompting a substantial industry consolidation that created ‘‘super-regional’’ banks. Until recently, commercial banking (accepting deposits and making loans) had been strictly separated from investment banking (underwriting securities, buying, and selling stocks and bonds) and insurance by the Glass-Steagall Act of 1933. Banking laws passed during the U.S. Civil War (in 1863) created a national currency and prevented nationally chartered banks from owning stock. Commercial banks became increasingly involved with investment banking during the early part of the twentieth century, and by 1930 nearly half of new securities offerings went through afﬁliates of commercial banks. This intertwining was blamed by some for worsening the Great Depression; Glass-Steagall, which was passed during this period, forbade commercial banks from owning and dealing in securities. Thus, Morgan Stanley (an investment bank) split off from JP Morgan (which became exclusively a commercial bank). Despite years of debate and efforts by banks for repeal, Glass-Steagall stayed in effect until 1999. Commercial banks such as JP Morgan and Bankers Trust were granted exceptions by the Federal Reserve Bank for brokerage and bond underwritings beginning in the 1980s and early 1990s, while conversely some investment banks were allowed to do commercial lending beginning in 1994. Glass-Steagall was ﬁnally repealed in 1999 in the wake of the merger between Citicorp and Travelers (an insurance company that also owned Salomon Smith Barney, a large investment bank). Citigroup, as the new ﬁrm was named, subsequently spun off its insurance business. American banks followed a distinctive path to globalization compared to banks in Europe and elsewhere. London grew to become a major international ﬁnancial center during the Victorian Era, and by the turn of the twentieth century it was home to outposts of dozens of foreign banking houses, including Schroder, Rothschild, Kleinwort, and JP Morgan. Indeed, ﬁnance continues to serve an outsize role in the British economy as other sectors of industry have faded (c.f. Conyon & Muldoon, 2008). Amsterdam also served as a global ﬁnancial center, along with a number of other trading

Industry Consolidation and Network Evolution in U.S. Global Banking

215

capitals. In contrast to their European counterparts, nationally chartered American banks were not allowed to open overseas branches until 1913, although other forms of foreign business were allowed. As late as 1960, only eight U.S. banks had any foreign branches (led by Citibank’s predecessor, First National City Bank) – a stark contrast with American manufacturers and oil companies, which had administered expansive global operations for decades. The domestic orientation of American banks changed in the late 1960s and 1970s, as the banks rapidly expanded their overseas operations. A number of causes were behind this timing. First, a tight money supply made it attractive to gather ‘‘Eurodollar’’ deposits via London branches.2 Second, some banks followed their large clients overseas to capture the clients’ non-domestic business. Third, by the mid-1970s, leading banks were making roughly half their proﬁts overseas, and Chase Manhattan was earning more than three-quarters of its proﬁts outside the U.S. (Hallow, 1993). This encouraged other banks to follow the example set by Citibank and Chase, and by 1980, 150 of them had overseas branches, spanning dozens of countries. The foreign assets of U.S. banks increased 100-fold between 1960 and 1980, and even the smallest banks often had foreign branches. To take one example, Colonial Bank of Waterbury, Connecticut (1980 population: 103,266) operated branches in the Cayman Islands (1974–1985) and in London (1981–1983), despite being only the 152nd largest bank in the United States in 1980. The industry suffered a dramatic reversal of fortune in 1982 following the Mexican debt crisis. American banks had come to be among the predominant lenders to low-income countries, replacing states as the largest source of funds. By 1982, these banks had extended billions of dollars of loans to states whose capacities to repay them were questionable. In late 1982, Mexico suspended its debt payment, leading to a contagion of default among Latin American countries and a crisis for their creditors. The subsequent period is known as ‘‘the lost decade’’ in economic development circles, as capital ﬂows to low-income countries abruptly halted, ultimately to be replaced with market-based ﬁnancing (Larosiere, 2005). By the end of the decade, the banks that had been the most aggressive globalizers were those hardest hit. Congressman John Dingell claimed in 1991 that Citibank was ‘‘technically insolvent’’ and ‘‘struggling to survive.’’ American banks responded to these crises by retrenching, beginning an exodus from overseas branching that lasted from the early 1980s into the mid-1990s. As Fig. 1 shows, the number of foreign branches of U.S. banks declined steadily from 1986 through 1994. The decline in the number of U.S.

216

ERIC J. NEUMAN ET AL.

Total Number of Branches

900 800 700 600 500 400 300 200 100

19 8

6 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04

0

Year

Fig. 1.

Cumulative Ownership of Foreign Branches by U.S. Banks, 1986–2004.

Total Number of Banks

50 45 40 35 30 25 20 15 10 5

19

86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04

0

Year

Fig. 2.

Number of U.S. Banks Operating Foreign Branches, 1986–2004.

banks with at least one foreign branch was even sharper. As Fig. 2 demonstrates, this decline was especially pronounced in the late 1980s and early 1990s. Prospects for international banking appeared to improve in the mid1990s, however. As Fig. 1 illustrates, the mid- to late-1990s saw the total number of branches rise sharply, peaking at 814 in 2000 (although the number of banks with foreign branches continued to decline, albeit more

217

Industry Consolidation and Network Evolution in U.S. Global Banking

Total Number of Countries

80 78 76 74 72 70 68 66

04

03

20

02

20

20

00 01 20

99

20

19

97 98 19

96

19

95

19

94

19

93

19

92

19

91

19

19

89 90 19

88

19

87

19

19

19

86

64

Year

Fig. 3.

Number of Countries with At Least One Branch of U.S. Bank, 1986–2004.

slowly than in earlier years). A number of geopolitical changes took place in this era that in theory offered U.S. banks new venues for business. The dissolution of the former Soviet Union, the opening of markets such as China and Vietnam, and the vast expansion in cross-border trade all created opportunities for American banks to expand their global operations. Yet, as Fig. 3 illustrates, the number of countries in which U.S. banks operated rose and fell rather modestly during this period, varying between a low of 69 in 1992 and a high of 78 in 2000. The list of countries entered during this period, shown in Table 1, provides little clear evidence that these new opportunities were the driving force behind this brief expansion in overseas branching. Entry into China (1987), Russia (1995), South Africa (1995), Vietnam (1995), and Bulgaria (2000) appear alongside New Zealand (1987), Australia (1994), Canada (2000), and Israel (2000) – countries for which there was no obvious exogenous explanation for entry. A closer examination reveals that much of the growth in the industry resulted from increased contestation between large, internationally focused banks in areas of existing operation. Most of the expansion came in the form of additions to banks’ existing branch networks, particularly in South America. For example, Argentina witnessed the equivalent of a branching arms race between Citibank and Bank of Boston. Between 1996 and 1999, Citibank grew from 45 to 96 branches (a 113% increase) and Bank of Boston from 44 to 139 branches (a 216% increase). These new Argentinean

218

ERIC J. NEUMAN ET AL.

Table 1. Year 1987 1990 1993 1994 1995

1996 1998 2000

Countries Entered by U.S. Banks (By Year). Countries Entered China New Zealand Costa Rica Fiji Australia Nicaragua Bangladesh Mexico Russia South Africa Vietnam Lebanon Netherlands Antilles Algeria Cameroon Bulgaria Canada Israel

Year in which a U.S. bank opened a branch in a country when there

had previously been no branches of U.S. banks in that country.

branches alone account for 87% of the overall growth in foreign branching during this period. Focusing on the number of banks operating foreign branches helps clarify some of these changes. Fig. 2 shows that the number of banks with foreign branches monotonically decreased from 46 in 1986 to 22 in 2004. As Fig. 4 shows, this decline in the number of banks led to an increased concentration in the proportion of foreign branches operated by the three most involved banks. By 2004, just 3 of the remaining 22 banks with foreign branches accounted for over 85% of the total branches. While these changes in foreign branch banking were happening, the banking industry’s consolidation began in earnest. The number of banks declined from over 10,000 in 1986 to under 7,000 in 2000, while the concentration of assets increased substantially. What was particularly striking in this movement was that much of the consolidation happened among the very largest banks, including those that were most globally oriented. Fig. 5 shows the family trees of the three largest banks, which emerged from 13 already large banks in just 14 years – between 1990 and

219

Industry Consolidation and Network Evolution in U.S. Global Banking 90% 85% 80% 75% 70% 65% 60% 55%

93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04

92

19

91

19

90

19

89

19

88

19

87

19

86

19

19

19

85

50%

Year

Fig. 4.

Percentage of All Foreign Branches Owned by the Three Most Global U.S. Banks, 1986–2004.

First Chicago

13

First Chicago NBD Bank One 1998

NBD Banc One

1

JP Morgan 2004 Chase JP Morgan JPM Chase 2001

27

26

4

Chase

2

Chase 1996

Chemical

6

Chemical 1991

How 13 big banks became three behemoths 1990-2004

Manufacturer’s 8 Hanover Fleet

16

Fleet Boston 1999 Bank of Boston

2 Bank of America

15

2004

Bank of America 3 Bank of America 1992 Security Pacific

Bank of America 1998

3

Citigroup

Fig. 5.

1

Nation’s Bank

5

7

Bank Consolidation, 1990–2005 (Numbers in Circles are Ranks in 1990).

220

ERIC J. NEUMAN ET AL.

2004. Notably, all of the pre-merger constituents except Bank One were among the most signiﬁcant international banks in the United States. This merger movement provides a fruitful context within which to understand the changing shape of bank globalization. We shall do this by examining the pre- and post-merger strategies of merging global banks.

THEORETICAL CONSIDERATIONS We have described two colliding forces in the 1990s: consolidation of the banking industry, and conﬂicting pressures around globalization. We argue here that the bank consolidation movement is an informative context in which to unpack the reciprocal role of networks and strategy in driving the process of corporate globalization. That is, mergers among banks, which may be based on divergent strategies, provide a context in which to explore how networks shape – and are shaped by – strategic moves. In this section, we describe some of the theoretical accounts given for banks’ global expansion and retrenchment; the place of board networks in strategy; why branch networks are an apt place to examine the effects of consolidation; and how we approach the merger movement. We ﬁrst provide a word on why we study bank branching as our primary outcome measure. Although there are several ways in which a bank can operate in a foreign country – including contracting with other banks to act as agents (known as correspondent banking), opening a representative ofﬁce, or creating a foreign subsidiary – branching is perhaps the most signiﬁcant commitment to a market. Branches can do ‘‘banking proper,’’ that is, taking in deposits and making loans, and they are legally part of the parent bank. Thus, their contributions show up on the parent company’s balance sheet: deposits gathered by branches can be used elsewhere in the world, and branches can draw on the capital of the parent bank to make loans. As such, we see branches as the most consequential presence of a bank, and a suitable measure of overseas expansion by banks. Why did American banks globalize their branch networks as they did? Prior research gives a number of possible accounts for U.S. banks’ initial forays into foreign markets, and how they ended up where they did (see Mizruchi & Davis, 2004). The ﬁrst account is that banks followed their customers. Just as vendors in the Japanese auto industry mirrored the locational choices of their major customers (Martin, Swaminathan, & Mitchell, 1998), banks might follow their corporate clients overseas, the better to meet the clients’ needs. Given the peculiar federal regulation of

Industry Consolidation and Network Evolution in U.S. Global Banking

221

American banking, we might expect to see Detroit banks locating branches near automakers’ foreign plants, or St. Louis banks opening branches near beer and pet-food manufacturing facilities. National City Bank of Minneapolis, for instance, was founded in 1964 by Lowell and Dwayne Andreas, who in 1972 went on to run agricultural giant Archer Daniels Midland (while maintaining seats on the National City board). Archer Daniels Midland developed signiﬁcant interests in Eastern Europe, so it is more than a coincidence that in 1979 National City became the ﬁrst – and so far only – American bank to operate a branch in Hungary. This branch was closed immediately after National City was acquired by M&I Bank in 2001. A second possible reason for banks’ opening of foreign branches is that the banks’ board members might have interests in or knowledge of particular countries. Although a bank’s directors are often local business executives, their companies may or may not be customers of the bank. Nonetheless, such individuals can provide generic business intelligence about the attractiveness of different foreign markets. Third, some observers have suggested that banks followed their competitors into particular markets. Banking folklore holds that Walter Wriston of Citibank was almost single-handedly responsible for the globalization of American banking, and as we noted earlier, the vanguard banks found overseas banking highly proﬁtable during the 1970s. Mizruchi and Davis (2004) found that as highly central banks moved overseas, other central banks were those most likely to follow. Given that the most well-connected banks were likely to see one another as competitors as well as peers, the ﬁnding suggested that banks might have been following their most direct competitors in establishing overseas branches (Henisz & Delios, 2001). Although several accounts exist for banks’ decisions to open foreign branches, the reasons for banks’ decisions to ‘‘de-globalize’’ have received little attention. In fact, we are unaware of any theory of ‘‘strategic retrenchment’’ associated with the decline in overseas branching. One way to consider the issue of de-globalization is to assume that it is simply the converse of globalization: as a bank’s customers reduce their foreign operations, for example, one might expect the bank to follow suit. The problem with this account is that for most American businesses, globalization followed a monotonically non-decreasing pattern: the average foreign sales of U.S. companies – outside of those in a few service industries (such as retail) or intrinsically local industries (such as utilities) – has risen virtually every year since the 1980s. There is little sign that companies outside of banking have been abandoning their foreign operations en masse.

222

ERIC J. NEUMAN ET AL.

An alternative possibility is that banks may have simply been outcompeted outside their domestic markets. One of the unintended consequences of American bank globalization is that the American banks trained a generation of their own competitors. While the World Bank notes the beneﬁts to foreign markets of hosting U.S. banks, in part due to the fact that American bankers can train those in the host nations, the other side of this story is that the local bankers trained by American banks can create effective local competitors to their former employers. This suggests that the longer a branch operates, the more likely it will be to generate its own competition. There is one phenomenon, however, that has been shown to inﬂuence banks’ decisions to open foreign branches, and that we believe might also have played a role in their decisions to retrench: the banks’ network ties, as indicated by their connections to other ﬁrms through their boards of directors. Corporate boards and the networks they create play a role, we argue, in both reﬂecting and shaping bank strategy. Board interlocks have been shown to affect strategies among a wide range of non-ﬁnancial corporations, from acquisitions (Haunschild, 1993) to takeover defense policies (Davis, 1991) to philanthropy and political activity (see Mizruchi, 1996 for a review of the literature on interlocks). These effects may be even more pronounced among banks. Mintz and Schwartz (1985) noted, for example, that banks recruit executives and other directors from wellconnected corporate boards speciﬁcally to guide their broad investment choices. Bank boards are much larger and better-connected than other boards, Mintz and Schwartz argued, because they provide high-level intelligence on broad trends in industry, what Useem (1984) called ‘‘business scan.’’ Thus, although directors might not be in the trenches giving advice on particular loans or on the adoption of speciﬁc strategies, they can provide insight into long-term trends affecting business through their service on other boards. In the realm of globalization, this might take the form of advice about speciﬁc countries or regions (as in, for example, ‘‘Newly wealthy businesspeople in Malaysia like American brands, and might be a good market for asset-gathering’’). Banks with more globally oriented directors may thus be more prone to expanding or maintaining global operations themselves. Similarly, banks seeking to globalize may be more likely to recruit directors from global businesses. In the same way that interlocks might have affected banks’ decisions to open foreign branches, they might also have affected the banks’ decisions to close those branches. Just as a bank’s CEO might hear of overseas opportunities from a director at a board meeting, he or she might also be

Industry Consolidation and Network Evolution in U.S. Global Banking

223

exposed to information suggesting that a country is an unsuitable place to conduct business. Although we believe that such events occurred and may explain some of the retrenchment that American banks experienced, there is another process that we believe may have been more important: the extensive merger activity among banks over the past two decades. This merger activity, we suggest, provides a context in which the networks within which banks were situated affected their behavior. Mergers are a particularly appropriate context for examining the link between networks and strategy. As we have seen, strategies may lead to the creation of networks, through the recruiting of directors. At the same time, networks may also lead to the development of strategies, given that a ﬁrm’s directors can shape the ﬁrm’s strategic direction. Mergers are a punctuating event in this process, a stock-taking for strategy. First, when large banks merge, the resulting entity – including its board – is typically a hybrid. The board of JP Morgan Chase, which resulted from the mergers of Manufacturers Hanover, Chemical, Chase Manhattan, and JP Morgan, included directors from each of its predecessors. Who stays and who goes after the merger is a reﬂection of the new entity’s priorities, and is thus indicative of the strategy of the new ﬁrm. Second, mergers bring to the forefront the question of which facilities will stay and which will be jettisoned. For banks, this includes decisions about the establishment, maintenance, or closure of particular branches, including those overseas. We anticipate that changes in ﬁrm strategies resulting from a merger will be observable within approximately two years. Our research strategy, then, is to examine the boards and global branch networks of merging banks before and after the merger to explore how consolidation affects globalization (or the reduction thereof). Speciﬁcally, we are interested in the ways in which board networks and branch networks mutually inﬂuence one another.

DATA Sample. We operationalize global banks as all publicly traded, U.S.-based commercial banks that operated at least one foreign branch (outside the Bahamas and Cayman Islands) between 1986 and 2004. There were 68 such banks. In our analysis speciﬁc to merged banks, we included only banks that had at least one foreign branch and that engaged in a merger, as either the acquiring or target ﬁrm, between 1986 and 2004. This criterion yielded a total of 37 banks, spanning 23 mergers. These mergers are listed in the table

224

ERIC J. NEUMAN ET AL.

in Appendix A. The total number of banks is less than two times the number of mergers because several banks engaged in more than one acquisition. We selected 1986 as our starting point because previous work has established this as the dawn of the bank merger wave (Rhoades, 2000). We end our analysis in 2004 to allow time after the acquisition for the acquiring bank to decide what to do with its acquired branches. Board data. One component of our analysis involves a comparison of the network centrality and board size of global banks versus non-global banks and non-bank ﬁrms. For this analysis we collected data on the boards of all corporations traded on the Nasdaq and the New York Stock Exchange in 1987, 1992, 1997, and 2002, as compiled by Compact Disclosure. This includes 3,736 ﬁrms (with a total of 34,468 directors) in 1987; 3,658 ﬁrms (with 32,434 directors) in 1992; 5,715 ﬁrms (with 45,582 directors) in 1997; and 4,760 ﬁrms (with 39,992 directors) in 2002. For before-and-after comparisons of merging banks, we used proxy statements ﬁled with the Securities and Exchange Commission (SEC) as well as annual reports and Compact Disclosure when proxy statements were unavailable. Branch data. Branch data came from the Federal Reserve Board following a Freedom of Information Act request. We submitted to the Fed identifying information for each branch operated by a U.S. bank between 1986 and 2004.3 In response, we received information on (a) the branch start date and, if applicable, the branch end date; (b) the branch location (including the address, city, and country); and (c) parent bank identiﬁcation information, including any changes in ownership, for 1,447 branches owned by the aforementioned 68 global banks. Corporate data. Corporate data, including information on bank ﬁnancial characteristics and geographic segment data for non-ﬁnancial ﬁrms, came from the Compustat database, accessed via WRDS. Speciﬁcally, the geographic segment data were used to calculate the amount of international experience that each outside director of a particular bank had at his or her own company. We operationalize a global board member as one whose company has international sales comprising at least 10% of its total sales. Merger data. Merger data were gleaned from the Federal Reserve Board, the Federal Financial Institutions Examination Council’s National Information Center, and archival searches of press releases and articles in the popular press. We recorded the names of the acquired and acquiring bank of each merger, information on the surviving bank, and the date of the merger. These data also appear in Appendix A.

225

Industry Consolidation and Network Evolution in U.S. Global Banking

FINDINGS How are the Boards of Global Banks Different? Table 2 presents a comparison of the size and two centrality measures – degree (the number of interlocks) and eigenvector centrality – of global banks versus non-global banks and non-ﬁnancial corporations. The data in Table 2 indicate that global bank boards were strikingly different from other boards in 1987. First, the boards of global banks were much larger: the median-sized board of a bank with at least one foreign branch had 18 members in 1987, compared to 12 for non-global (or ‘‘domestic’’) banks and 8 for non-ﬁnancial ﬁrms. Second, the boards of global banks were far more central than those of non-global banks or non-ﬁnancial ﬁrms. The median eigenvector centrality of global banks was 4.59, compared to 0.14 for both non-ﬁnancial ﬁrms and domestic banks, indicating that the directors that global banks recruited served on more central boards than those recruited by domestic banks. In data not shown in Table 2, global banks had an average of 5 ‘‘received ties’’ (i.e., executives of outside ﬁrms) compared to 2.3 received ties for non-isolate domestic banks. The median degree of centrality (the average number of total ties) was also far higher among global banks (21.5) than it was for non-ﬁnancial ﬁrms (2) and domestic banks (1). Perhaps most importantly from our perspective, we ﬁnd that the outside directors serving on bank boards came from more globally oriented Table 2. Board Size and Centrality of Non-Financials, Banks, and Global Banks. Comparisons of Boards: Median Values of Board Size, Eigenvector Centrality, and Degree Centrality, 1987–2002 1987 Non-ﬁnancials

Banks

Global banks

Size Eigenvector centrality Degree centrality Size Eigenvector centrality Degree centrality Size Eigenvector centrality Degree centrality

8 0.14 2 12 0.14 1 18 4.59 21.5

1992 8 0.19 2 11 0.08 1 17 5.13 20

1997 7 0.09 3 10 0.05 0 16 3.57 18

2002 8 0.18 3 10 0.04 0 17 4.1 18.5

226

ERIC J. NEUMAN ET AL.

companies: the average foreign sales of global bank directors’ home companies (also not shown in the table) was 11.9% in 1987, compared to 8.4% for domestic banks. Cumulatively – given that global banks had substantially more received ties – the outside directors of global banks had signiﬁcantly greater ‘‘global experience.’’

How do Bank Boards Change in the Context of Mergers? When one bank acquires another, there are a number of theoretically plausible strategies that the surviving bank could follow when selecting its board members. One possibility is to keep all board members from both banks. This ‘‘keep them all’’ strategy, especially if applied over repeated acquisitions, would yield an extremely large, ultra-central board. The opposite approach would be for the acquiring bank to demonstrate its control over the post-acquisition bank by keeping only directors from its own board. In between these two extremes, the acquiring bank could add to its own board a few select directors from the acquired bank’s board, possibly to smooth over tensions related to the acquisition but also because those directors may be highly respected or provide the acquiring bank with insight into the affairs of the acquired bank. Yet another option would be for a more equitable split between the banks, with approximately half of the surviving board coming from the board of each bank in the merger. This approach gives the best appearance of a ‘‘merger of equals’’ and also allows the acquiring bank to reinvigorate its board with new relationships and new ideas. We ﬁnd that among mergers of global banks, the ‘‘equitable split’’ and ‘‘acquiring control’’ strategies are dominant. Of the mergers listed in Appendix A, the vast majority adopted one of these two approaches. (Two mergers – RepublicBank acquiring Interﬁrst and Bank of America acquiring Security Paciﬁc – can be characterized as ‘‘keep them all,’’ though the massive boards that resulted did not last long. Each surviving bank’s board was reduced by at least 50% over the next two years.) This is consistent with the ﬁndings in Table 2 that the board size across global banks slightly decreased over our period of study. As an illustration of how these strategies are applied, in Figs. 6 and 7 we depict the board composition histories of Bank of America and JP Morgan Chase, the two banks responsible for most of the substantial mergers in our sample. Each bank acquired four other banks during our sample period and at one time or another each employed the two dominant strategies for

Industry Consolidation and Network Evolution in U.S. Global Banking

227

30

Number of Directors

25 20

FleetBoston (Old) Bank of America Boatmen's Bank C&S/Sovran NCNB/NationsBank/BofA

15 10 5

19

86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06

0

Year

Fig. 6.

Board Composition of Bank of America (Previous Names Include North Carolina National Bank, NCNB, and NationsBank), 1986–2007.

30

Number of Directors

25 20

Bank One JP Morgan (Old) Chase Manufacturers Hanover Chemical/Chase/JPM

15 10 5

19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06

0

Year

Fig. 7.

Board Composition of JP Morgan Chase (Previous Names Include Chemical Bank and Chase Manhattan), 1986–2007.

228

ERIC J. NEUMAN ET AL.

adding directors from acquired banks. Furthermore, in the years between their acquisitions both banks allowed their board sizes to decrease through attrition. The banks therefore exhibit a surprisingly similar pattern of waxing and waning of their boards over this 21-year period. The banks differ, though, in how long their acquired directors remained on their boards. Bank of America added 29 directors from banks it acquired, and the 22 who had left by 2007 did so after an average of 2.8 years. Of the 32 acquired directors that JP Morgan Chase added, the 23 that have since departed averaged 4.9 years of service. This culminates in substantial differences in the ﬁnal composition of the two boards. Upon its fourth acquisition, Bank of America’s acquired directors remained a minority (9 of 19 directors). For JP Morgan Chase, the situation was the opposite; after its fourth acquisition, only 3 of the 16 directors on JP Morgan Chase’s 2004 board had never served on the board of one of the banks that it acquired.

How Does the Position of Global Banks in the Intercorporate Network Change? Table 2 also shows that as the merger movement surged through the industry, the boards of global banks remained distinctive from boards of domestic banks and of non-ﬁnancials. Although the size and centrality of global banks’ boards decreased slightly over time, by 2002 the boards of global banks were still far larger and more central than both their domestic counterparts and non-ﬁnancial ﬁrms. The boards of the global banks remained nearly twice as large as those of domestic banks (a median size of 18 vs. 10) and more than twice as large as non-ﬁnancial boards (18 vs. 8). As in the earlier years, the centrality of global banks remained far above those of domestic banks and non-ﬁnancials. Changes in the network position of global banks are also illustrated in Figs. 8 and 9, which show the received-tie networks of global banks in 1987 and 2002. Node size is indexed to the extent of a ﬁrm’s foreign operations. For banks, this represents the size of the foreign branch network. For nonbanks, the size of the node reﬂects the percentage of the ﬁrm’s sales from outside the U.S. Two things are notable about these ﬁgures. First, in 1987, there is a strong correlation between the magnitude of a bank’s foreign branch network and the average level of foreign operations of its received ties. This was no longer true in 2002. In the earlier period, the more global banks recruited directors

Industry Consolidation and Network Evolution in U.S. Global Banking

229

Fig. 8. Received Tie Network for Global Banks, 1987 (Banks in Circle, Non-Banks in Square. Node Size Represents Magnitude of International Operations).

from more globalized ﬁrms (or at least executives of U.S. ﬁrms with more international revenues). Second, in the 1987 network, most banks with substantial international operations were tied into a single component, and every major international bank had a second-degree tie with another bank. AT&T executives served on the boards of both Manufacturers Hanover and Chase Manhattan. Amoco ofﬁcers served on the boards of Chase, Continental Illinois, and First Chicago. Exxon executives were on the boards of JP Morgan and Chemical, and Xerox executives were on the boards of Chase, Citicorp, and State Street Boston. By 2002, in contrast, there was not a single instance of such second-degree ties, and the overall network had become far more sparse. Global banks still had more ‘‘international’’ received ties than their domestic counterparts (18.4% foreign sales vs. 11%), but there were fewer of them (3.3 ties on average compared to 1.7 for domestic banks), and they did not provide conduits to other global banks.

230

ERIC J. NEUMAN ET AL.

Fig. 9. Received Tie Network for Global Banks, 2002 (Banks in Circle, Non-Banks in Square. Node Size Represents Magnitude of International Operations).

The Global Reach of Banks after Acquisition To what extent do global banks retain their foreign branches after they are acquired by another bank? The data in Table 3 indicate that the vast majority of foreign branches owned by banks that were acquired remained open even two years after the banks’ acquisition. Of the 561 total foreign branches held within one month of a parent branch being acquired, 454 (80.9%) remained with the bank two full years afterward. Still, there was considerable variation in the experiences of acquired banks. Some small branch networks (such as those of Crocker National Bank and National City Bank Minneapolis) were completely shuttered, while others (including First RepublicBank Dallas and First Fidelity) were untouched. Large networks, too, saw a range of effects on the life chances of their branches. Over half of JP Morgan’s 21 foreign branches were closed within one year of its acquisition by Chase, but all but two of Bank of Boston’s 191 branches remained in operation two years after being acquired

Acquired Bank Name

CROCKER NATIONAL BANK INTERFIRST BK DALLAS NA FIRST REPUBLICBANK DALLAS RAINIER NB IRVING TC FIRST PENNSYLVANIA BK NA NEW BANK OF NEW ENGLAND NA MANUFACTURERS HAN TC SECURITY PACIFIC NB BOSTON SAFE DEPOSIT & TC CONTINENTAL BK NBD BANK FIRST FIDELITY BK NA CHASE MANHATTAN BK NA FIRST INTERSTATE BK OF CA CORESTATES BANK NA BANK OF AMER NT&SA FIRST NB OF CHICAGO BANKBOSTON NA REPUBLIC NB OF NY PACIFIC BK NA MORGAN GUARANTY TC NATIONAL CITY BK MINNEAPOLIS WACHOVIA BK NA FLEET NA BK RIGGS BK NA

5/31/1986 6/6/1987 7/30/1988 1/1/1989 10/7/1989 3/5/1990 7/14/1991 12/31/1991 4/22/1992 5/21/1993 9/1/1994 12/1/1995 1/2/1996 3/31/1996 4/1/1996 4/28/1998 7/23/1998 10/2/1998 10/1/1999 12/31/1999 3/1/2000 12/31/2000 8/1/2001 9/1/2001 4/1/2004 5/14/2005

1 3 2 1 9 1 1 27 9 1 6 7 1 53 1 5 42 10 191 10 1 21 1 1 155 1

30 Days Prior 1 3 2 1 9 1 1 27 9 1 6 7 1 53 1 5 41 10 191 10 1 21 1 1 155

Day of Acq.

2 1 9 1 1 27 2 1 5 7 1 53 4 41 10 191 10 21 1 1 155

4 41 10 191 10 21 1 1 155

60 Days After

2 1 9 1 1 27 2 1 5 7 1 53

1

30 Days After

21 1 1 155

4 41 10 191 10

2 1 9 1 1 27 2 1 5 5 1 53

90 Days After

10

155

1 155

4 41 10 191 8

2 1 8 1 1 21 2 1 3 3 1 42

1 Year After

21

4 41 10 191 10

2 1 9 1 1 21 2 1 3 3 1 42

180 Days After

Surviving Branches of an Acquired Bank Before and After Acquisition.

Acq. Date

Table 3.

122

9

4 33 10 189 5

2 1 8 1 1 21 2 1 2 3 1 39

2 Years After

Industry Consolidation and Network Evolution in U.S. Global Banking 231

232

ERIC J. NEUMAN ET AL.

by Fleet. In all, 19 of the 26 acquired networks saw at least one branch closed within two years. Table 4 presents data similar to those in Table 3, except that the focus is on the number of countries in which the acquired banks operated, as opposed to the number of branches. Similar to the ﬁndings in Table 3, the data here demonstrate that in the vast majority of cases, the acquired bank did not exit countries in which it had previously had branches, even two years after its acquisition by another bank. On the other hand, the foreign branch networks for 18 of the 26 acquired banks saw at least one country exited within two years. The ﬁndings in these two tables indicate that most acquired banks experienced some reduction in their foreign branching, but in most cases foreign branches remained, even after the bank was acquired. Table 5 presents data on the effect of acquisition on the global presence on bank boards. Each row of the table represents a bank acquisition. The ﬁrst column (after the acquisition date) represents the pre-acquisition name of the acquired bank, followed by the number of the acquired bank’s board members who were executives of non-ﬁnancial corporations with more than 10% of sales outside the U.S. one year prior to its acquisition. The next column gives the name of the bank that made the acquisition, followed by its number of pre-acquisition ‘‘global board members’’ (also one year prior to making the acquisition). The ﬁnal column gives the name of the surviving bank subsequent to the acquisition, followed by the surviving bank’s number of global board members one year subsequent to the acquisition. The data in the table indicate that the acquired banks had slightly fewer global directors on their board than did their acquirers (2.35 vs. 2.91). This difference was not statistically signiﬁcant, however. Directors with international experience on the surviving banks’ boards increased to an average of 3.26 as boards were reconﬁgured to include members of the acquired bank. This is marginally statistically signiﬁcantly different from the number of global directors of the acquired bank, t(22) ¼ 1.74, p ¼ .10, but it is not statistically signiﬁcantly different from the number of global directors that had been serving on the acquiring bank’s board. Overall, the ﬁndings indicate that the acquisitions of banks had a detectable, but minor, effect on the banks’ maintenance of their foreign branches.

DISCUSSION In this chapter we have examined the relationship between interlocking directorate networks and strategy as it pertains to the foreign branching

5/31/1986 6/6/1987 7/30/1988 1/1/1989 10/7/1989 3/5/1990 7/14/1991 12/31/1991 4/22/1992 5/21/1993 9/1/1994 12/1/1995 1/2/1996 3/31/1996 4/1/1996 4/28/1998 7/23/1998 10/2/1998 10/1/1999 12/31/1999 3/1/2000 12/31/2000 8/1/2001 9/1/2001 4/1/2004 5/14/2005

Acq. Date

Table 4.

CROCKER NATIONAL BANK INTERFIRST BK DALLAS NA FIRST REPUBLICBANK DALLAS RAINIER NB IRVING TC FIRST PENNSYLVANIA BK NA NEW BANK OF NEW ENGLAND NA MANUFACTURERS HAN TC SECURITY PACIFIC NB BOSTON SAFE DEPOSIT & TC CONTINENTAL BK NBD BANK FIRST FIDELITY BK NA CHASE MANHATTAN BK NA FIRST INTERSTATE BK OF CA CORESTATES BANK NA BANK OF AMER NT&SA FIRST NB OF CHICAGO BANKBOSTON NA REPUBLIC NB OF NY PACIFIC BK NA MORGAN GUARANTY TC NATIONAL CITY BK MINNEAPOLIS WACHOVIA BK NA FLEET NA BK RIGGS BK NA

Acquired Bank Name

1 3 2 1 8 1 1 17 8 1 6 5 1 29 1 5 27 8 16 9 1 19 1 1 13 1

30 days Prior 1 3 2 1 8 1 1 17 8 1 6 5 1 29 1 5 27 8 16 9 1 19 1 1 13

Day of Acq.

2 1 8 1 1 17 2 1 5 5 1 29 4 27 8 16 9 19 1 1 13

4 27 8 16 9 19 1 1 13

60 Days After

2 1 8 1 1 17 2 1 5 5 1 29

1

30 Days After

19 1 1 13

4 27 8 16 9

2 1 8 1 1 17 2 1 5 3 1 29

90 Days After

9

13

1 13

4 27 8 16 7

2 1 7 1 1 16 2 1 3 2 1 26

1 Year After

19

4 27 8 16 9

2 1 8 1 1 16 2 1 3 2 1 26

180 Days After

5

8

4 26 8 16 4

2 1 7 1 1 16 2 1 2 2 1 25

2 Years After

Global Footprint of Acquired Bank Before and After Acquisition (i.e., The Number of Countries Spanned by the Acquired Banks’ Branches).

Industry Consolidation and Network Evolution in U.S. Global Banking 233

IRVING BANK CORP FIRST PENNSYLVANIA CORP BANK OF NEW ENGLAND (FAILED) MANUFACTURERS HANOVER CORP SECURITY PACIFIC BANCORPORATION CONTINENTAL BANK CORP NBD BANCORP

10/7/1989 3/5/1990 7/14/1991

1/2/1996

9/1/1994 12/1/1995

4/22/1992

12/31/1991

FIRST FIDELITY BANCORPORATION

RAINIER BANCORPORATION

1/1/1989

6/6/1987

CROCKER NATIONAL CORPORATION INTERFIRST CORPORATION

Acquired Bank

5/31/1986

Acq. Date

2

2 6

3

4

1 2 5

0

2

1

#

FIRST UNION CORP

BANKAMERICA CORP FIRST CHICAGO CORP

BANKAMERICA CORP

FIRST REPUBLICBANK CORPORATION SECURITY PACIFIC BANCORPORATION BANK OF NEW YORK COMPANY CORESTATES FINANCIAL CORP FLEET/NORSTAR FINANCIAL GROUP CHEMICAL BANKING CORP

WELLS FARGO & CO

Acquiring Bank

1

2 7

3

4

8 6 1

3

0

1

#

BANKAMERICA CORP FIRST CHICAGO NBD CORPORATION FIRST UNION CORP

BANKAMERICA CORP

FIRST REPUBLICBANK CORPORATION SECURITY PACIFIC BANCORPORATION BANK OF NEW YORK COMPANY CORESTATES FINANCIAL CORP FLEET/NORSTAR FINANCIAL GROUP CHEMICAL BANKING CORP

WELLS FARGO & CO

Surviving Bank

2

2 6

6

4

8 2 1

5

0

1

Table 5. Number of Received Ties from Firms with At Least 10% International Sales for Acquired and Acquiring Banks (1 Year Before the Acquistion) and the Surviving Bank (1 Year After the Acquistion). #

234 ERIC J. NEUMAN ET AL.

9/1/2001 4/1/2004 5/14/2005

8/1/2001

10/1/1999 3/1/2000 12/31/2000

4/28/1998 7/23/1998 10/2/1998

3/31/1996 4/1/1996

MEAN MEDIAN STD. DEV

CHASE MANHATTAN CORP FIRST INTERSTATE BANCORPORATION CORESTATES FINANCIAL CORP BANKAMERICA CORP FIRST CHICAGO NBD CORPORATION BANKBOSTON PACIFIC BANK NA J.P. MORGAN & CO. INCORPORATED NATIONAL CITY BANCORPORATION WACHOVIA CORP FLEETBOSTON RIGGS NATIONAL CORPORATION 2.35 2 2.01

3 1 0

0

1 0 8

2 2 4

3 2

MEAN MEDIAN STD. DEV

MARSHALL & ISLEY CORPORATION FIRST UNION CORP BANK OF AMERICA CORP PNC FINANCIAL SERVICES GROUP

FLEET FINANCIAL GROUP CITY NATIONAL CORPORATION CHASE MANHATTAN CORP

FIRST UNION CORP NATIONSBANK CORP BANK ONE CORP

CHEMICAL BANKING CORP WELLS FARGO & CO

2.91 2 2.25

2 2 6

0

2 1 4

3 5 1

4 1

MEAN MEDIAN STD. DEV

MARSHALL & ISLEY CORPORATION WACHOVIA CORP BANK OF AMERICA CORP PNC FINANCIAL SERVICES GROUP

FLEETBOSTON FNCL CORP CITY NATIONAL CORPORATION JP MORGAN CHASE & CO

FIRST UNION CORP BANK OF AMERICA CORP BANK ONE CORP

CHASE MANHATTAN CORP WELLS FARGO & CO

3.26 3 2.24

4 1 5

1

2 1 6

3 3 2

7 3

Industry Consolidation and Network Evolution in U.S. Global Banking 235

236

ERIC J. NEUMAN ET AL.

strategies of U.S. banks. We theorized a link between the amount of international experience of a bank’s board members and the size of the foreign branch network operated by the bank. We anticipated that the widespread merger movement and subsequent industry consolidation that has taken place within American banking since 1986 would be a cause of the reconﬁguration of foreign branches operated by U.S. banks, and that it would do so in part through board of director networks. What we found instead was a growing disconnect between the characteristics of a bank’s board and its global strategy. We have documented several manifestations of this transformation. First, mergers did not lead global banks to increase the number of directors serving on their boards in a sustained manner. Some banks enlarged their boards slightly by adding select directors from the bank they acquired, while other banks constructed a post-acquisition board whose size was roughly equal to its pre-acquisition size and then populated it with an almost even split between former members of the merging banks’ boards. Those banks that did increase their number of directors following an acquisition quickly reduced the size of their boards to pre-acquisition levels through attrition, as illustrated in the cases of Bank America (formerly NCNB and NationsBank) and JP Morgan Chase (formerly Chemical Bank and Chase Manhattan). Even without a sustained increase in board size, banks still had an opportunity to change the composition of their boards based on which directors they kept from the board of the acquired bank. As the banking industry consolidated, we expected to see global banks increase the number of directors with international experience serving on their boards. Yet this did not occur; the number of directors with international experience serving on global banks’ boards did not increase. Even the largest global banks did not add multinational CEOs as they acquired increasingly more banks. Instead, they seem to have been operating according to an unwritten rule, or institutional logic, of board composition. Boards of post-acquisition banks appear to have been ﬁlled in an almost formulaic way: a group of top ofﬁcers from major corporations – most of them active but often one or more retired, and some with substantial international experience – plus one or two private investors, as well as public ﬁgures such as university presidents, foundation directors, and former public ofﬁcials. Concentration of an industry does not therefore translate into concentration of banks’ boards. Second, and following from the ﬁndings from the previous point, the strong cluster of banks found within the interﬁrm network in 1987 no longer

Industry Consolidation and Network Evolution in U.S. Global Banking

237

exists. By 2002 banks had become the sullen loaners of the corporate world. The network of director ties between banks and non-ﬁnancial corporations, which in 1987 saw most global banks connected to at least one other global bank through second-degree ties, had devolved into a series of discrete clusters, completely disconnected from one another. Given that this network fracturing occurred during unprecedented consolidation within the banking industry, it calls into question some of the conventional wisdom regarding the concentration of power among ﬁnancial ﬁrms. Writing about the major German and American banks near the end of World War I, Lenin (1916, p. 31) noted that ‘‘[t]he transformation of numerous modest intermediaries into a handful of monopolists represents one of the fundamental processes in the transformation of capitalism into capitalist imperialism.’’ As banks became ever more concentrated, the reasonable expectation would be that the resulting banks would occupy an ever-greater role in the American, and possibly global, economy, as well as an increasingly central position in the intercorporate network. Banks did remain highly central in the U.S. network into at least the early 1980s, although their centrality did not increase over time (Mizruchi, 1982). Since then, despite the rapid concentration of the U.S. banking industry, the trend in network centrality has, if anything, been the opposite of what Lenin predicted. Not only has the centrality of global banks not increased, but it has actually declined slightly (from an average of 4.86 in eigenvector centrality in 1987 and 1992 to 3.83 in 1997 and 2002), although this decline is not statistically signiﬁcant. Third, the regional banks that bought global banks did not retrench their newly acquired foreign networks. In several prominent instances, they kept them intact. In addition to the earlier example of Fleet Bank, NationsBank operated just ﬁve global branches upon acquiring Bank of America, but had 53 foreign branches in 1998, after the acquisition. Although NationsBank did close 14 of those branches over the next two years, this appears to have been done as a means of streamlining its operations, rather than as an attempt to abandon its foreign presence. As evidence of this, we note that despite its closure of individual branches, the bank withdrew completely from only one country during this retrenchment. One possible reason that these regional banks maintained the foreign presence of the banks they acquired may have been the changes in federal banking laws that were occurring during this period. The Riegle-Neal Act, passed in 1994, allowed full interstate banking (i.e., banks could own branches in different states) for the ﬁrst time. The lifting of this restriction opened numerous strategic possibilities to banks, and many responded by turning their attention from

238

ERIC J. NEUMAN ET AL.

the local economy to economies of the region, the United States as a whole, and eventually the rest of the world. Finally, despite the fact that the regional banks that acquired global banks maintained the latter’s foreign branches, the overall level of foreign branching has fallen to pre-1982 levels. Lenin’s concern about the concentration of ﬁnancial intermediaries would have led us to predict the opposite. In Lenin’s view, consolidation in the banking industry should have led to an increase in large banks’ export of capital, as they replace states and multilateral organizations as sources of capital for businesses and other states worldwide. One indicator of this, we have argued, would be an increase in the number of foreign branches operated by U.S. banks. Yet the bulk of the increase in foreign branching by American banks occurred in the 1960s and 1970s, and the level began to decline in the early 1980s. Although an increase did occur in the late 1990s, this was concentrated in one country – Argentina – and the banks subsequently scaled back their operations, both there and elsewhere.

CONCLUSION At its essence, this chapter is an attempt to answer a question about a curious phenomenon in U.S. banking: Why, given the increased global activity of American non-ﬁnancial corporations since 1980, did U.S. banks pull back from their involvement in foreign branching? One possible reason for this, we suggested, was the concentration of the banking industry and the concomitant acquisition of global banks by smaller banks with historically regional orientations. Our ﬁndings suggest, however, that with a few notable exceptions, the regional banks that acquired the global giants maintained the latter’s foreign branches. This means that the de-globalization of U.S. banking has to be a result of factors other than the increased concentration of the industry. One possibility that we mentioned earlier is that American banks in foreign nations were killed by their own success. In establishing a presence overseas, U.S. banks typically hired and trained bankers from the host country. These bankers, using the skills they had learned at the U.S. bank, in many cases chose to form banks of their own. The competitive advantage that they possessed as indigenous members of their societies may have made it more difﬁcult for the American banks to compete in those markets. We do not know how prevalent this phenomenon was, or whether it had the effect that we are suggesting. We believe that there are ways to test this

Industry Consolidation and Network Evolution in U.S. Global Banking

239

argument, however. If we are correct, we might expect to observe a correlation between the prevalence of startup banks and the closure of U.S. branches, net of other factors. We could also examine the extent to which the new banks in these nations are headed by former employees of American banks located in their nations. As we consider alternative reasons for why American banks have retrenched their foreign operations since 1980, we should not lose sight of the fact that so many of the banks that acquired foreign branches via the merger wave in the U.S. banking industry did not. Especially interesting is why regional banks with little experience in global branch banking, such as NationsBank and Fleet Bank, maintained almost all of their acquired foreign branching networks. Yet rather than only ask why certain banks kept their foreign branches and others did not, we can consider what this overall trend suggests about the relationship between a bank’s network of directors and its strategies. Our assumption going into this study was that a bank’s board would help shape its acquisition strategy and also its globalization strategy. In particular, the more international experience its directors have, the more extensive a bank’s foreign branch network, and the less international experience its directors have, the less likely a bank’s acquisition targets would have extensive foreign branch networks. In the latter’s case, any foreign branches that were part of an acquisition would ﬁt neither the bank’s (and the board’s) experience nor its identity and would be quickly jettisoned. The ﬁndings here suggest that we need to rethink these assumptions. It appears instead that banks’ strategies may have been independent of the international experience of their board members and that their acquisition strategies helped shape their globalization strategies or may have even been one and the same. A regional bank may have initially viewed an acquired bank’s foreign branches as unwanted baggage in the deal but then come to realize their value once it took ownership. Alternatively, a bank may have always viewed a target bank’s foreign branches as an integral part of an acquisition package as it looked to expand its geographic areas of coverage. Future work is necessary to explore these alternatives, perhaps by examining written documents and speeches given by the bank’s top executives about their strategic vision. Finally, the decisions that banks made in constructing their postacquisition boards leads us to speculate about the relationship between individual banks’ board composition strategies and the overall shape of the directorate network. Banks historically have taken different approaches to boards than have non-ﬁnancial ﬁrms, and these approaches have changed

240

ERIC J. NEUMAN ET AL.

over time. But rather than simply affecting the board characteristics and network position of individual banks, these individual-level strategies have consequences for the aggregate network to the point where we could say that banks have had particular network assembly rules. The literature on banks and their interlocks suggests that banks have used three network assembly rules that in turn have led to three distinct network signatures found in the U.S. directorate network. In the earliest period of public corporations, the U.S. directorate network was balkanized and centralized as ﬁnance capitalism dominated. Major bankers such as JP Morgan controlled groups of companies by placing themselves or their ofﬁcers on the boards of dozens of ﬁrms (Brandeis, 1914). As ﬁnance capitalism gave way to managerial capitalism in the 1920s, the directorate network became diffuse and centralized. Banks remained the most central actors in the intercorporate network (Mizruchi, 1982) but their network assembly rules shifted. Rather than controlling ﬁnancial ﬂows through the use of sent ties, banks sought access to information that would be available via the received ties of well-connected CEOs. These CEOs were more than willing to serve on bank boards not only because bank directorships were prestigious but because their ﬁrms relied on banks for ﬁnancing and the CEOs wanted to participate in decisions about capital ﬂows (Mintz & Schwartz, 1985). Beginning in the 1980s, however, technological and regulatory changes led ﬁrms to turn to other sources for ﬁnancing and allowed commercial banks to enter other lines of business. This weakened the mutual dependence between banks and ﬁrms, and once again banks changed how they assembled their board networks. In this new era of shareholder capitalism, banks began to look like every other ﬁrm oriented toward increasing shareholder value. The directorate network became diffuse and decentralized as banks reduced the size of their boards and their recruiting of centrally located directors. For the ﬁrst time since the rise of publicly traded companies in the U.S., banks were no longer the center of the directorate network (Davis & Mizruchi, 1999). Our observation of how post-acquisition banks constructed their boards is consistent with the transformation that took place between the second and third epochs. After acquiring another bank, the surviving bank either created a board with about the same number of members as it previously had and populated it with an approximately equal mix of directors from the boards of the acquiring and acquired banks, or it kept all of its own directors and added a select few – if any – directors from the acquired bank’s board. Thus the dissolution of strong clustering in the interﬁrm network between 1987 and 2002 can be seen as an aggregation of the decisions by

Industry Consolidation and Network Evolution in U.S. Global Banking

241

individual banks on how to compose their boards. This ﬁnding has parallels with that of Amburgey et al. (2008), who show that one basis of interﬁrm connection – preferential attachment – has a demonstrable effect on the structure of alliances of biotech ﬁrms. Unlike Amburgey et al., however, who found a core-periphery structure, our network reveals a group of loosely connected clusters with no clear core. This notion of network assembly rules suggests a re-orientation for how we think about network structure, network dynamics, and the role of strategy as these may be more intertwined than typical models of interﬁrm networks suggest. For instance, although networks are often modeled as exogenous entities whose properties inﬂuence the decisions of individual actors, in reality network properties are often endogenous with respect to the actors, who strategically construct their ties depending on their own characteristics and their partners’ characteristics, including, importantly, to whom their partners are already connected. We also need to pay attention to constraints on networking. Especially in social networks that require face-to-face communication such as boards of directors, there are social, organizational, and even physical constraints on how large a network should be. As Doreian (2008) suggests, there are liabilities to an actor of having too many ties. This point can be illustrated by the case of First RepublicBank, which thought that 46 directors would make a ﬁne board until it discovered that the directors could not all ﬁt in the board room (Apcar, 1988). Constraints may also involve ﬁlling speciﬁc roles, as we saw banks do when ﬁlling their post-acquisition boards. Even though banks already had begun to appoint fewer centrally located directors, acquisitions were an easy opportunity to bring aboard these ‘‘celebrity CEOs.’’ For the most part, though, banks chose not to. A thorough understanding of the properties and consequences of the macro network structure thus results from considering both the decisions by and properties of individual-level actors and the interactions between individual actors and the environment (Schelling, 1978). Acknowledging these complexities and maintaining focus on both individual and network levels simultaneously would be a step in this direction.

NOTES 1. Because we believe that all mergers involve an acquiring and acquired entity, we use the terms ‘‘merger’’ and ‘‘acquisition’’ interchangeably. Although some arrangements, such as when First Chicago and NBD became First Chicago NBD, are announced as a ‘‘merger of equals,’’ we believe that these statements are

242

ERIC J. NEUMAN ET AL.

face-saving tactics used to placate the employees, customers, and shareholders of the acquired entity. By examining characteristics of the surviving ﬁrm such as its CEO, board, and headquarters, one can almost always determine the role of each party in an arrangement that was announced as a merger. 2. Eurodollars are U.S. currency held in banks outside the United States, including, for example, foreign proﬁts of American multinationals. 3. We originally obtained identifying information on every U.S.-operated foreign branch from the Federal Financial Institutions Examination Council’s National Information Center (NIC) (http://www.fﬁec.gov/nicpubweb/nicweb/NicHome.aspx) using a web crawling computer program written by the ﬁrst author in October 2005. This search did not include identifying information on the branch’s parent bank, which necessitated the use of the Freedom of Information Act request. Unfortunately for researchers interested in this topic, the NIC has redesigned its web site since 2005 and made it impossible to access historical data on foreign branches of U.S. banks.

ACKNOWLEDGMENTS This chapter was written while the ﬁrst author was at the University of Michigan. Research was supported by the Rackham Graduate School, the Stephen M. Ross School of Business, and the College of Literature, Science, and the Arts, all at the University of Michigan. The chapter beneﬁted from comments by participants at the Economic Sociology Seminar at the University of Michigan and the Network Strategy Conference held at the Rotman School of Management, University of Toronto.

REFERENCES Amburgey, T. L., Al-Laham, A., Tzabbar, D., & Aharonson, B. (2008). The structural evolution of multiplex organizational networks: Research and commerce in biotechnology. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 171–209). Bingley, UK: JAI/Emerald Group. Apcar, L. M. (1988). First RepublicBank to cut board size, oust prominent members of old guard. Wall Street Journal, May 2, p. 9. Brandeis, L. (1914). Other peoples’ money: And how the bankers use it. New York: Frederick A. Stokes. Conyon, M. J., & Muldoon, M. R. (2008). Ownership and control: A small-world analysis. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advanced in strategic management (Vol. 25, pp. 31–65). Bingley, UK: JAI/Emerald Group. Davis, G. F. (1991). Agents without principles? The spread of the poison pill through the intercorporate network. Administrative Science Quarterly, 36, 583–613. Davis, G. F., & Mizruchi, M. S. (1999). The money center cannot hold: Commercial banks in the U.S. system of corporate governance. Administrative Science Quarterly, 44, 215–239.

Industry Consolidation and Network Evolution in U.S. Global Banking

243

Doreian, P. (2008). Actor utilities, strategic action, and network evolution. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 247–271). Bingley, UK: JAI/Emerald Group. Hallow, M. J. S. (1993). Over there: American banks abroad. New York: Garland Publishing. Haunschild, P. R. (1993). Interorganizational imitation: The impact of interlocks on corporate acquisition activity. Administrative Science Quarterly, 38, 564–592. Henisz, W. J., & Delios, A. (2001). Uncertainty, imitation, and plant location: Japanese multinational corporations, 1990–1996. Administrative Science Quarterly, 46, 443–475. Hite, J. M. (2008). The dynamic evolution of network ties: Navigating multi-dimensionality, dyadic contexts, bounded agency and strategic action. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in Strategic management (Vol. 25, pp. 133–170). Bingley, UK: JAI/Emerald Group. Larosiere, J. de. (2005). From Mexico to Argentina: What have we learned from two decades of debt crises? Princeton Institute for International and Regional Studies, Monograph Series. Number 3. Princeton University. Lenin, V. I. (1939; orig. 1916). Imperialism: The highest stage of capitalism. New York: International Publishers. Marquis, C., & Lounsbury, M. (2007). Vive la Re´sistance: Competing logics and the consolidation of U.S. community banking. Academy of Management Journal, 50, 799–820. Martin, X., Swaminathan, A., & Mitchell, W. (1998). Organizational evolution in the interorganizational environment: Incentives and constraints on international expansion strategy. Administrative Science Quarterly, 43, 566–601. Mintz, B., & Schwartz, M. (1985). The power structure of American business. Chicago: University of Chicago Press. Mizruchi, M. S. (1982). The American corporate network, 1904–1974. Beverly Hills: Sage Publications. Mizruchi, M. S. (1996). What do interlocks do? An analysis, critique, and assessment of research on interlocking directorates. Annual Review of Sociology, 22, 271–298. Mizruchi, M. S., & Davis, G. F. (2004). The globalization of American banking, 1962 to 1981. In: F. Dobbin (Ed.), The sociology of the economy (pp. 95–126). New York: Russell Sage Foundation. Rhoades, S. A. (2000). Bank mergers and banking structure in the United States, 1980–98. Board of Governors of the Federal Reserve System Staff Study 174. Schelling, T. C. (1978). Micromotives and macrobehavior. New York: W. W. Norton. Useem, M. (1984). The inner circle. New York: Oxford University Press.

First RepublicBank Corp Rainier Bancorporation

Irving Bank Corporation First Pennsylvania Corp Bank of New England Manufacturers Hanover Corp Security Paciﬁc Bancorporation Boston Co

7/30/1988 1/1/1989

10/6/1989 3/5/1990 7/14/1991 12/31/1991 4/22/1992

5/21/1993

Crocker National Corporation InterFirst Corporation

Acquired Bank

5/31/1986 6/6/1987

Merger Date

Mellon Bank Corporation

NCNB Corporation Security Paciﬁc Bancorporation Bank of New York Co CoreStates Financial Corp Fleet/Norstar Financial Group Chemical Banking Corp BankAmerica Corp

Wells Fargo & Co RepublicBank Corporation

Acquiring Bank

Mellon Bank Corporation

Wells Fargo & Co First RepublicBank Corporation NCNB Corporation Security Paciﬁc Bancorporation Bank of New York Co CoreStates Financial Corp Fleet/Norstar Financial Group Chemical Banking Corp BankAmerica Corp

Surviving Bank

Mergers between banks where the acquired bank operated at least one foreign branch (excluding the Bahamas and the Cayman Islands) at the time of acquisition, 1986–2004.

APPENDIX A

244 ERIC J. NEUMAN ET AL.

First Fidelity Bancorporation Chase Manhattan Corp First Interstate Bancorporation CoreStates Financial Corp BankAmerica Corp First Chicago NBD Corporation BankBoston Paciﬁc Bank NA JP Morgan & Co. Incorporated National City Bancorporation Wachovia Corp FleetBoston Riggs National Corporation

1/2/1996 3/31/1996 4/1/1996

8/1/2001 9/1/2001 4/1/2004 5/14/2005

10/1/1999 3/1/2000 12/31/2000

4/28/1998 9/30/1998 10/2/1998

Continental Bank Corp NBD Bancorp

9/1/1994 12/1/1995

Marshall & Isley Corporation First Union Corp Bank of America Corp PNC Financial Services Group

Fleet Financial Group City National Corporation Chase Manhattan Corp

First Union Corp NationsBank Corp Bank One Corp

First Union Corp Chemical Banking Corp Wells Fargo & Co

BankAmerica Corp First Chicago Corp

Marshall & Isley Corporation Wachovia Corp Bank of America Corp PNC Financial Services Group

FleetBoston Financial Corp City National Corporation JP Morgan Chase & Co

First Union Corp Bank of America Corp Bank One Corp

BankAmerica Corp First Chicago NBD Corporation First Union Corp Chase Manhattan Corp Wells Fargo & Co

Industry Consolidation and Network Evolution in U.S. Global Banking 245

ACTOR UTILITIES, STRATEGIC ACTION AND NETWORK EVOLUTION Patrick Doreian ABSTRACT The arguments in this chapter address all three of the questions motivating this volume on network strategy. First, they focus on the issue of network evolution and show how networks can emerge and change over time. Second, the chapter tackles the issue of endogeneity and shows that, under certain conditions, some structural advantages do precede rather than follow network positions. Networks evolve over trajectories and the trajectories matter. Third, the arguments respond to the core question of network entrepreneurship: does the awareness of structural advantages available to network positions inspire managers, acting on behalf of organizations, to seek these advantages? In responding, this chapter challenges the idea that ﬁlling structural holes necessarily confers advantages on the actors ﬁlling them. It follows that the advantages of bridging are dependent not only on the network structure, when decisions regarding tie formation or deletion are made, but also on the costs of forming and maintaining ties relative to the beneﬁts obtained from doing so.

Network Strategy Advances in Strategic Management, Volume 25, 247–271 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25007-2

247

248

PATRICK DOREIAN

1. INTRODUCTION The primary motivating example used here is one of organizations located in inter-organizational networks. A secondary example features individuals in interpersonal networks. Regardless of which example is used, and assuming the portability of network ideas (Burt, 2008), two items are axiomatic for most network analysts when considering actors in networks: (i) outcomes at the actor level are conditioned by the location of the actor in the network and (ii) the collective outcome for the network as a whole is conditioned by its structure. Put in these terms, the network and its structure have primacy. Yet it is reasonable to ask how networks form, how their structures arise and to ask about evolutionary principles for network change. For Doreian and Stokman (1997), ‘‘network processes are series of events that create, sustain and dissolve social structures’’ (p. 3), an emphasis endorsed by Hite (2008). These events include actors forming and dissolving network ties and so form generating mechanisms for structural change. In a list of evolutionary network principles, Stokman and Doreian (1997) allow that, for some networks, ‘‘the underlying process for network change is assumed to be located in the network structure (emphasis added)’’, an idea picked up by Amburgey, Al-Laham, Tzabbar, and Aharonson (2008) and extended fruitfully. Actors located in network structures make decisions about forming or eliminating ties with other actors. One reason for doing so it to derive greater beneﬁts from the network(s) that result. Hite (2008), drawing upon Stinchcombe (1965), contrasts agency (proactive choice) with path dependence (full constraint) and argues in favor of a position between these extremes (Evans, 2002). The examples considered below justify this claim fully: some actors have many options for productive agency while others have few depending on the network structure within which they are located. Social theorists have long had a concern with self-interest driving human action (Monge & Contractor, 1987). This applies also to considering actors in networks, especially organizations in inter-organizational networks. Thinking of ties forming and ending according to rational calculations, based on self-interest, provides a direct approach to answering the question ‘‘Where do network forms come from?’’ Jackson and Wolinsky (1996), hereafter JW, provided one set of answers by considering the costs, to an actor, of forming network ties in relation to the direct beneﬁts derived from these ties and indirect beneﬁts of being connected to other actors over paths in the network. Given different parameterizations of these costs and beneﬁts, different network structures are created as stable structures.

Actor Utilities, Strategic Action and Network Evolution

249

Hummon (2000) extended this line of inquiry by performing an agent-based simulation study of ‘‘JW actors’’, capable of adding and removing edges so as to improve their returns, to determine the stable structures generated by actors responding to different regimes of costs and beneﬁts. He showed that there were stable equilibrium structures not anticipated by the theorems provided by JW. Doreian (2006) examined the transitions between edge graphs for n ¼ 3, 4, 5 and established theorems for the evolved stable structures that were consistent with the Hummon simulation-based results while contradicting the JW results. Xie and Cui (2008a, 2008b) provided a correction to one part of Doreian’s (2006) result for n ¼ 5 edge networks. This chapter builds on this line of work. However, instead of seeking only results about which network structures are generated, the approach taken here considers the strategic choices facing actors as they elect to form (or not form) network ties. By shifting the emphasis from ‘‘which networks are stable?’’ to ‘‘what constraints and opportunities exists for actors in different locations in a network?’’ it is possible to address issues of strategic structure as well as network evolution. The JW actors are retained together with the JW speciﬁcation of utilities. We consider the lattice of edge graphs where the link between two graphs is established by adding or removing single edges. These are realized by choices based on the costs and beneﬁts of making these choices. The results giving the stable equilibrium structures (Doreian, 2006; Xie & Cui, 2008b) are presented. The concept of automorphic equivalence is used to deﬁne the idea of direct structural competitors in networks and the strategic choices facing actors in deciding to form, or drop, edges are considered in some detail.

2. ACTOR UTILITIES AND NETWORK STRUCTURE An edge graph (network), without loops, is denoted by G ¼ ðV; EÞ where V is a set of vertices and E is the set of edges in the network where E V V. The initial formulation of JW allowed edge-speciﬁc costs and beneﬁts but, in order to obtain general derived results, this was restricted so that the costs and beneﬁts associated with all edges were the same. Notationally, d represents the direct beneﬁt obtained by a pair of actors when they form a tie and g denotes the costs of maintaining this tie. We assume1 0odr1 and 0ogr1. In addition to these direct costs and beneﬁts, there are indirect beneﬁts for a pair of actors, denoted by i and j, by being linked to each other over geodesics of length tij. This beneﬁt is dtij.

250

PATRICK DOREIAN

The (simpliﬁed) utility to i in G is given by2: X X ui ðGÞ ¼ w þ ðd gÞ þ ðdtij Þ i; j2G

tij 41

The total utility of the network is simply the sum of the utilities for P the actors in the network. Denoting this by UT, where the nw term (from i(w)) is ignored because it not generated from the network structure, X UT ¼ ui ðGÞ i2G n

The set of all edge networks, G for a given size, n, forms a lattice which identiﬁes the possible transitions between graphs when a single edge is added (or removed) from a network. A graph, Gp, is a proper subgraph of Gq if GpCGq. If Ep and Eq are the edge sets of Gp and Gq, respectively, when considering all edge networks on n vertices, jVp j ¼ jVq j ¼ n. Gp is an immediate lower neighbor of Gq if Ep ¼ Eq/lg where lg is an edge of Gq. Similarly, if Eq ¼ Ep,lg then Gq is an immediate upper neighbor of Gp. Transitions occur between immediate neighbors either by adding or removing an edge. Hummon’s (2000) simulation approach brought a subtle difference to the JW formulation in so far as JW were concerned only with stable outcome networks that were strongly efﬁcient.3 Hummon focused attention on when adding and removing edges occur with simulated actors making choices in these terms. Focusing also in detail on the strategic options for actors is fully consistent with Hummon’s image of JW actors who change the network structures in which they are located.

2.1. Equivalences in Networks Partitioning networks in terms of different notions of equivalence deﬁned over the relations of a network is the foundation of blockmodeling (Lorrain & White, 1971; Breiger, Boorman, & Arabie, 1975) and generalized blockmodeling (Doreian, Batagelj, & Ferligoj, 2005). The idea of automorphic equivalence for studying social networks has been used by Winship (1988), Mandel (1983), Winship and Mandel (1983), Pattison (1988) and Borgatti and Everett (1992). A permutation j : V ! V is an automorphism of the relation R if and only if 8a; b 2 V : ðaRb ) jðaÞRjðbÞÞ. Automorphic equivalence is based on the idea that equivalent units occupy indistinguishable structural locations in the

Actor Utilities, Strategic Action and Network Evolution

251

network. Automorphic equivalence is deﬁned in terms of graph isomorphisms with j as a one-to-one mapping of G to itself with the same preserving features of an isomorphism. Two vertices a, bAV belong to the same orbit of G if and only if there exists an automorphism p ¼ (ab) such that p(a) ¼ b (Everett & Borgatti, 1988). Orbit membership deﬁnes an equivalence relation that partitions the vertices of G. Following Lorrain and White (1971), two units are structurally equivalent if they are connected to the rest of the network in identical ways. The units a and b are structurally equivalent, denoted as a b, if and only if the permutation (transposition) p ¼ (ab) is an automorphism of the relation R (Everett & Borgatti, 1988). An equivalence relation E on V is a regular equivalence on graph G ¼ ðV; RÞ if and only if for all a; b; c 2 V, aEb implies both (i) aRc ) 9d 2 V : ðbRd ^ d cÞ and (ii) cRa ) 9d 2 V : ðdRb ^ d cÞ. The following relation holds for these three types of equivalences (where, for a relation R, Str(R) is the set of structural equivalences of R, Aut(R) is the set of automorphic equivalences of R and Reg(R) is the set of regular equivalences of R: Str(R)DAut(R)DReg(R) (Doreian et al., 2005, p. 175). Structural equivalence is a special case of automorphic equivalence which, in turn, is a special case of regular equivalence. Graphs in Fig. 1 can be used to illustrate these equivalences. First, we consider automorphic equivalence. Trivially, there is only one orbit in G1 containing all the vertices. The two orbits in G2 are {(a,b),(c,d,e)}. There are four orbits in G9: {(a,c),(b),(d),(e)} while in G11 there are two orbits {(b), (a,c,d,e)}. G20 is another graph with a single orbit containing all ﬁve vertices. Examples of structurally equivalent vertices are: {a,c} in G3; {a,c,d} in G6 and {a,c,d,e} in G11. Theorem 1. Given a graph, G ¼ ðV; EÞ, and the utilities deﬁned according to the JW formulation, members of the same orbit of G have the same utility functions. Proof. Let P a and b belong to the same orbit of G. Given P tij u ðGÞ ¼ w þ ðd gÞ þ is trivial: uP i a ðGÞ ¼ w þ i;j2G tij 41 ðd Þ, the proof P P P taj tbk ðd gÞ þ ðd Þ and u ðGÞ ¼ w þ ðd gÞ þ b a;j2G taj 41 b;k2G tbk 41 ðd Þ. Necessarily, a and b have the same number of edges incident to them. P Hence, number of the (dg) terms are the same and ðd gÞ ¼ a;j2G P ðd gÞ. Second, the distributions, by length, of the geodesics b;k2G P P taj tbk for a and b are the same so taj41 ðd Þ ¼ tbk41 ðd Þ. It follows that ua(G) ¼ ub(G).

252

PATRICK DOREIAN

G1

G6

c

d

c

d

e b

e

a

G17

d

e

c G21

e G22

a c

d G26 a

b

e

G23

a

d b G27 a

b

e

c

G28

c

c G31

a

G32

b e

e

d

c

d

a

b

d

c

c a b

e d G25

e

c c G30 b e

d

c

d b a

c

d

a

G33

b e

a e

b

c a

b

G29

a G34

e

a

c d

b

b

G20

d

c e

a

d

a

a d

c

e

b

d c

b d

e

G24

a b

d

e

c

d

b

G15

G19

e d b

e

e d

b

c

e

a e

c

a

G18

c

b

a

a

c

d

c a

e

G14 b

c

d

d

a

e

d b

c

d G16

G13 a

c

b

c

b

e

G10

b d

c

G12

a

G9

a

b d

c

e

G5

a e

a

G8 b

e

G11

d

a

b

b

e

c

G7

a

e

b

d

G4

a

G3

e

b

e d

a

G2

a

b d

c

e

b d

c

Fig. 1. All Edge Graph Types for n ¼ 5.

Two simple implications follow from this result: (i) structurally equivalent vertices have the same utilities and (ii) regularly equivalent vertices need not have the same utilities. Also, G25 in Fig. 1 provides an example that actors (vertices) having the same utilities need not be in the same orbit.4 When the parameterization of the beneﬁts and costs of ties are such that it pays actors to form ties, all actors are competitors for forming ties.

Actor Utilities, Strategic Action and Network Evolution

253

As shown below, actors are competitors also for not forming ties when letting other actors bear the costs of maintaining edges is to their advantage. In particular, actors represented by automorphically equivalent vertices are deﬁned as direct structural competitors because they are structurally identical. We turn now to consider some of the strategic choices involved in forming edges (or not).

3. TRANSITIONS BETWEEN EDGE GRAPHS FOR n ¼ 5 The set of all the (isomorphism classes of) edge graphs for n ¼ 5 are given in Fig. 1 where 34 edge graphs are shown. The possible transitions between adjacent neighboring graphs, by the addition of an edge, are shown in Fig. 2. The numbers in diamonds on the left are counts of the number of edges in the graphs on the same level as these numbers. Fig. 2 is drawn so that the transitions represent evolutionary additions of edges. Reversing these transitions, by considering the removal of edges, gives a devolutionary set of transitions. By comparing the utilities for actors in one graph and those in another graph reached by the addition of an edge, it is possible to determine if the transition will be made. Similarly, given a graph and comparing the utilities to actors and the utilities obtained after the removal of an edge, the conditions for that change can be determined. Together, these comparisons allow the determination of structures that will be stable given the parameterizations of g relative to b. The equilibrium graphs for this lattice of networks with transitions between graphs taking the form of adding or removing edges are given by the following theorem from Xie and Cui (2008b) who corrected5 parts of a corresponding result in Doreian (2006). Theorem 2. For graphs with n ¼ 5 vertices, the equilibrium structures for JW actors are: pﬃﬃﬃ 1. For d in the range 0od ð 5 1Þ=2 (a) For gWdþd2d3d4, the null graph (G1) only; (b) For dþd2d3d4WgWd, the null graph (G1) and the cycle (G20); (c) For dWgWdd3, the star (G11), the near-star (G12) and the cycle (G20);

254

PATRICK DOREIAN

0

G1

1

G2

G3

2

G6

3

G4

G5

G7

4

G11

G10

5

G15

G16

G19

G17

G18

G20

6

G21

G22

G26

G23

G24

G25

G27

G28

G29

7

G9

G12

G8

G31

8

G14

G13

G30

G32

G33

9

10

Fig. 2.

G34

Lattice of Transitions for Edge Networks, n ¼ 5.

Actor Utilities, Strategic Action and Network Evolution

255

(d) For dd3WgWdd2, the star (G11), the cycle (G20) and a nearcycle (G26) and (e) For godd2, the complete graph (G34). pﬃﬃﬃ 2. For d in the range 0oð 5 1Þ=2odo1 (a) For gWd, the null graph (G1) only; (b) For dWgWdþd2d3d4, the star (G11), the near-star (G12) and the line (G14); (c) For dþd2d3d4WgWdd3, the star (G11), the near-star (G12) and the cycle (G20); (d) For dd3WgWdd2, the star (G11), the cycle (G20) and a nearcycle (G26) and (e) For godd2, the complete graph (G34). All of these equilibrium structures were identiﬁed by Hummon (2000) as equilibrium structures. However, two structures, the ‘‘near-null’’ and the ‘‘near-complete’’, provisionally identiﬁed by Hummon as equilibrium structures do not appear to be genuine equilibria. This result provides ranges of the parameterization of g relative to d and the equilibrium structures that result for those ranges. As Xie and Cui (2008a) note, the structures that persist over the greatest range of g are the star (G11) and the cycle (G20), suggesting something fundamental about these structures. It is worth noting that the notion of an equilibrium has been modiﬁed from the strongly efﬁcient networks deﬁned by JW to structures that persist because there are no ways of moving from them via the creation or deletion of edges. Actors can get locked into sub-optimal structures in terms of the utilities returned to them and to the network as a whole. Theorem 2 rests only on considering whether or not a transition could be made between a pair of graphs by considering the utilities in Table 1 and whether pairs of actors beneﬁt by forming a tie. When only one transition from one graph is possible this may be unproblematic, although strategic considerations for actors could still be relevant. When more than one transition is possible from a given graph it is no longer clear which transition will be made, nor how that transition to another network structure occurs, because the strategic choices of actors come into play. To the extent that actors seek to maximize utility, one future structure may be preferable to another from the point of view of an actor. This will depend on the utility preferences of the actors, the structure of the network at the time a decision is made (endogeneity), the amount of information that an actor has and the choices made by other actors.

256

PATRICK DOREIAN

Table 1.

Actor Utilities for n ¼ 5 Edge Graphs.

Graph

G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13 G14 G15 G16 G17 G18 G19 G20 G21 G22 G23 G24 G25 G26 G27 G28 G29 G30 G31 G32 G33 G34

Actor a

b

c

d

e

0 (dg) (dg)þd2 (dg) 2(dg) (dg)þ2d2 (dg)þd2þd3 (dg)þd2 2(dg)þd2 2(dg)þd2 (dg)þ3d2 (dg)þ2d2þd3 2(dg) 2(dg)þ2d2 3(dg) 2(dg)þ2d2 2(dg)þ2d2 2(dg)þd2þd3 3(dg)þd2 2(dg)þ2d2 3(dg) 2(dg)þ2d2 3(dg)þd2 4(dg) 2(dg)þ2d2 3(dg)þd2 4(dg) 4(dg) 4(dg) 3(dg)þd2 3(dg)þd2 4(dg) 4(dg) 4(dg)

0 (dg) 2(dg) (dg) 2(dg) 3(dg) 2(dg)þd2 2(dg) 3(dg) 2(dg)þd2 4(dg) 3(dg)þd2 2(dg) 2(dg)þd2þd3 2(dg)þd2 4(dg) 3(dg)þd2 3(dg)þd2 2(dg)þ2d2 2(dg)þ2d2 3(dg) 4(dg) 3(dg)þd2 2(dg)þ2d2 3(dg)þd2 2(dg)þ2d2 3(dg)þd2 2(dg)þ2d2 2(dg)þ2d2 3(dg)þd2 3(dg)þd2 3(dg)þd2 4(dg) 4(dg)

0 0 (dg)þd2 0 2(dg) (dg)þ2d2 2(dg)þd2 (dg)þd2 2(dg)þd2 2(dg)þd2 (dg)þ3d2 (dg)þ2d2þd3 2(dg) 2(dg)þd2þd3 3(dg) 2(dg)þ2d2 3(dg)þd2 2(dg)þd2þd3 2(dg)þd2þd3 2(dg)þ2d2 3(dg) 3(dg)þd2 3(dg)þd2 2(dg)þ2d2 3(dg)þd2 3(dg)þd2 3(dg)þd2 4(dg) 3(dg)þd2 3(dg)þd2 4(dg) 3(dg)þd2 4(dg) 4(dg)

0 0 0 (dg) 0 (dg)þ2d2 (dg)þd2þd3 (dg) (dg)þ2d2 2(dg)þd2 (dg)þ3d2 2(dg)þ2d2 (dg) (dg)þd2þd3þd4 2(dg)þd2 (dg)þ3d2 (dg)þ2d2þd3 2(dg)þ2d2 2(dg)þ2d2 2(dg)þ2d2 3(dg) 2(dg)þ2d2 2(dg)þd2þd3 2(dg)þ2d2 2(dg)þ2d2 2(dg)þ2d2 3(dg)þd2 2(dg)þ2d2 3(dg)þd2 2(dg)þ2d2 2(dg)þ2d2 3(dg)þd2 3(dg)þd2 4(dg)

0 0 0 (dg) 0 0 0 (dg) 0 0 (dg)þ3d2 (dg)þd2þ2d3 (dg) (dg)þd2þd3þd4 0 (dg)þ3d2 (dg)þ2d2þd3 (dg)þd2þ2d3 (dg)þ2d2þd3 2(dg)þ2d2 0 (dg)þ3d2 (dg)þ2d2þd3 2(dg)þ2d2 2(dg)þ2d2 2(dg)þ2d2 (dg)þ3d2 2(dg)þ2d2 2(dg)þ2d2 3(dg)þd2 4(dg) 3(dg)þd2 4(dg) 4(dg)

4. STRATEGIC CONSIDERATIONS FOR ACTORS When considering the formation (or deletion) of edges, by pairs of actors, in terms of their utilities given the regimes of the beneﬁts from edges and costs of maintaining them, there are different views that actors can take. Let i and j be two actors and consider the views of i (or j) regarding the

Actor Utilities, Strategic Action and Network Evolution

257

formation of ties. Three broad views appear relevant when considering adding6 edges: 1. Given the set of edges, does i gain by forming the (ij) edge? 2. Given the current network structure, which additional tie gives i the highest realized utility: when all possible edge formations involving i are considered? when all possible edge formations regardless of whether they involve i ? 3. Does i get the highest utility: compared to all other actors in the destination structure? in comparison with j when a tie is formed? We consider these views in turn.

4.1. The Local Tie Formation View In essence, the ﬁrst view that an actor can take is the one reﬂected by the approach underlying Theorem 2. Given b and g, all that is considered is whether there will be a gain for both actors from adding an edge between them. As such, this view leads to considering only the possible edges that can be formed for one transition and the sequences of possible edge formations. Actors adopting this view do not ask which of these edge additions are the best for them: all that matters is there is an improvement in utility following the change.

4.2. Seeking the Best Edge Addition View However, if actors do ask which changes in edges are best for them they adopt the second view. This can be explored by considering the possible transitions between network structures (in Fig. 1) as shown in Fig. 2. 4.2.1. Transitions from G2 Consider the transition from G2 to G3 versus the transition from G2 to G4 from the vantage point of actor, b. With regard to ﬁlling a structural hole (Burt, 1992) it would appear obvious that the best option is for b (or a) to form an edge with c (or d or e). However, if b examines all of the possible options, it is not clear that this is the case. The options for b are: (i) form an edge, bc, with c (or bd or be); (ii) do nothing and watch an edge, ac, form

258

PATRICK DOREIAN

between a and c; (iii) do nothing and watch ed form between e and d (or watch cd or ed form). Option (i) creates G3 (with b ‘‘in the middle’’); option (ii) creates a graph isomorphic with G3 (with a ‘‘in the middle’’) and option (iii) creates G4. The ﬁrst option is the only active option for b with the others being passive. Denote the utilities to b for these three outcomes as ub1, ub2 and ub3, respectively. From Table 1 we have: (i) ub1 ¼ 2(dg), ub2 ¼ (dg)þd2 and ub3 ¼ (dg). Given that dW0 and god, ub3 is the least desired. However, the ranking of the other two possible utilities depends on d and g: ub2>ub1.(dd2)og. However, ub1>ub2.(dd2)Wg. The implication of these two inequalities is clear: If the costs of tie formation are high (dd2og) then b prefers to see the edge ac (or ad or ae) form and beneﬁt more by being at the end of chain through a rather than ﬁll a structural hole by forming the edge bc (or bd or be). If the costs are low enough (dd2>g) then b will ﬁll the structural hole by forming the edge bc (or bd or be). Put differently, ﬁlling a structural hole by being ‘‘in between’’ two other actors is preferred only when the costs of tie formation are low enough. (As shown below, this kind of general conclusion does not depend on the speciﬁc network form which the transitions can be made, although the details do in terms of functions of d and g.) When the costs are too high, ﬁlling the structural hole, by being in the middle, is not the best option. Exactly the same arguments apply for a. The strategic problem is acute because a and b belong to the same orbit in G2: they are direct structural competitors with diametrically opposed interests. When one gains by forming an edge the other loses and vice versa. Further, when one gains by not forming an edge, the other loses when an edge is formed and vice versa. It appears that much of the discussion of structural holes and the beneﬁts of ﬁlling them rests on an implicit assumption that tie formation and tie maintenance are costless. Things look different for the actors, c, d and e, excluded from ties in G2. These three actors are also direct structural competitors. The options available to, say, c are: (i) form ca (or cb); (ii) form a tie with d (or with e) and (iii) do nothing and watch bd (or ad or be or ae) form. The utilities for c for these three outcomes are denoted by uc1, uc2 and uc3, respectively. Their values are uc1 ¼ (dg)þd2, uc2 ¼ (dg) and uc3 ¼ 0. For dW0 and god, there is a clear preference order: uc1Wuc2>uc3 with the worst option being excluded from tie formation. Identical arguments hold for d and e, the direct

Actor Utilities, Strategic Action and Network Evolution

259

structural competitors to c. These suggests that with full information about the state of G2, an isolate in this graph prefers to attach to an already connected actor (to beneﬁt both from the formed edge and being connected indirectly to another actor). In a way, this is similar to the discussion of Amburgey et al. (2008) concerning ‘‘preferential attachment’’ – only the preferences are driven by utility considerations given a network structure. The following question becomes obvious: Is G4 impossible because no actor, other than the potentially left out isolates, in G2 prefers G4 to G3? Under full information, it would seem so. However, actors make decisions in real time and the sequencing of decisions matter (Hummon, 2000; Doreian, 2002) and the possibility of being left out completely, in G4, could induce a pair from {c,d,e} to form a tie with each other if that opportunity occurs ﬁrst. Further, if the information available to them is limited and they are not aware of the ab link, then forming an edge like cd becomes a desirable outcome7. It would also form under a ‘‘satisﬁcing’’ framework (Simon, 1976) where the goal is not to be left out. 4.2.2. Transitions from G3 We next examine the possible transitions from G3 and consider ﬁlling structural holes through the addition of an edge, by looking at the four options of b: (i) form bd (or be) to create G6 (or a graph isomorphic with it); (ii) watch cd form (or ad, ce, ae) which leads to G7; (iii) watch ed form to create G8 and (iv) watch ac form to create G5. Denoting the utilities for each of these options as ub1, ub2, ub3 and ub4, we have (i) ub1 ¼ 3(dg), ub2 ¼ 2(dg)þd2 and ub3 ¼ 2(dg) ¼ ub4. Clearly, for dWg and do1, the worst outcome for b is ub3 (and ub4 even though it comes from a different graph) making G5 and G8 both undesirable. Having ub1Wub2 requires dd2>g and for ub1oub2 it follows that dd2og. This leads to the same conclusion as before: if dd2og then b prefers to not ﬁll the structural hole and will do so only if the cost of tie formation is low enough (i.e. dd2>g). The argument is exactly the same concerning the transitions from G6 and G11 is preferred by b only for low enough g. 4.2.3. Transitions from G8 The transitions from G8, a disconnected graph with connected components, also have some interest value. The orbits of G8 are {(a,c),(b),(d,e)} so that the pair (a,c) are direct structural competitors as are (d,e). Actor, d (or e), has ﬁve options. Fig. 3 contains G8 and the ﬁve graphs that can be reached

d

H1

e

c

a

b

e

G12

d

c

Fig. 3.

b

a

e

e

d

d

a

a

c

c

b

b

d

H3

e

Lattice of Edge Transitions from G8.

H2

G8

b

c

a

G13

e

d

a

c

b

260 PATRICK DOREIAN

Actor Utilities, Strategic Action and Network Evolution

261

from it by single edge additions. Two are active and three are passive: (i) attach to c (or to a) to form a graph, H1, isomorphic with G14; (ii) attach to b to form G12; (iii) do nothing and watch ea form (or ac) to create H2 which is isomorphic with G14; (iv) do nothing and watch eb form to create a graph, H3, isomorphic to G12 and (v) do nothing and watch ac form to create G13. Denoting the utilities to d for these ﬁve alternatives by {udq} for 1rqr5, these utilities are ud1 ¼ 2(dg)þd2þd3, ud2 ¼ 2(dg)þ2d2, ud3 ¼ (dg)þd2þd3þd4, ud4 ¼ (dg)þd2þ2d3 and ud5 ¼ (dg). For the two active options for d, do1 implies that ud2Wud1 so that attaching to the middle actor b is the preferred option for d. This implies that b would be in a position of ﬁlling a structural hole (as would d). Among the passive alternatives there is a clear ranking of the utilities: ud4>ud3>ud5 so that d prefers e to attach to the central actor, b, rather than to either a or c. Again, under this preference order, b ﬁlls a structural hole. The worst outcome for d is the ﬁfth option, where d and e remain in the smallest component of G13. From d ’s vantage point, G13 is the least desirable outcome. Making a comparison between the active and passive options for d requires a little more work. Having ud2>ud4 implies 2(dg)þ2d2> (dg)þd2þ2d3. In turn, this implies dþd2–2d3>g is the condition for d preferring to attach to b. So, if the condition for the structural hole to be ﬁlled by the bd edge is that g is low enough, this requirement is dþd2–2d3>g, a little more complicated that the condition for ﬁlling a structural hole established earlier. Conversely, d prefers the passive option of letting be form when dþd22d3og and the costs are too high for d to ﬁll the structural hole (with b). Because d and e are direct structural competitors their preferences with regard to the active and passive actions are the reverse of each other. Note that this holds only when d is in the range pﬃﬃﬃ ð 5 1Þ=2odo1 (Theorem 2.2(b)). However, for d in the range pﬃﬃﬃ 0odoð 5 1Þ=2 (Theorem 2.1(a)), gWdþd22d3.gWdþd2d3d4 and the null graph is the only stable equilibrium. The vantage point of b is considered next. This vertex has only three options, of which one is active: (i) form the edge bd (or be) to create a graph isomorphic with G12; (ii) do nothing and watch cd (or ad, ce or ae) form creating G14 and (iii) do nothing and watch ac form. Denoting the utilities to b by {ubq} for 1rqr3, these utilities are: (i) ub1 ¼ 3(dg)þd2; (ii) ub2 ¼ 2(dg)þd2þd3 and (iii) ub2 ¼ 2(dg). Clearly, the formation of G13 under option (iii) is the least desirable for b. Having ub1Wub2 implies 3(dg)þd2>2(dg)þd2þd3. This implies dd3>g as the conditions under which b ﬁlls the structural hole by forming an edge with d (or e).

262

PATRICK DOREIAN

Alternatively, b will prefer the passive option and let c from an edge with d (or e) when dd3og and makes the costs of bridging too high. This is also the conditions under which the star (G11) or the cycle (G20) are stable structures. However, the star cannot be reached from G8 via edge additions making the cycle the equilibrium structure that will develop under this (higher) set of values for g. This suggests that the cycle (G20) is stable because the costs of maintaining edges are shared by all of the vertices equally. The vertices a and c are direct structural competitors in G8. From the vantage point of a (with the same arguments holding for c), there are four options with two of them active: (i) form ad (or ae) to create a graph isomorphic with G14 in which a is the bridging vertex; (ii) form ac to create G13; (iii) do nothing and watch b form bd (or be) to create G12 and (iv) do nothing and watch c, its direct structural competitor, form cd (or ce) to create G14. Denoting the utilities to a by {uaq} for 1rqr4, these utilities are: (i) ua1 ¼ 2(dg)þ2d2; (ii) (i) ub2 ¼ 2(dg); (iii) ua3 ¼ (dg)þ2d2þd3 and (iv) ua4 ¼ (dg)þd2þd3þd4. Some implications are immediate. Option (i) is always the better active option for a because ua1>ua2 for non-zero d. Between the two passive options, option (iii) is always better because ua3ua4 ¼ d2d4>0 for 0odr1. This means that if a is to remain passive, watching b form the edge to the other component is always preferable to watching its direct structural competitor form such a link. Comparing ua1 and ua3 reveals that ua1>ua2.dd3>g. So a prefers to bridge the structural hole rather than b doing so only when g is low enough, i.e. when dd3>g. When dd3og, a’s preference is to let b form the link to the other component and so beneﬁt from the indirect path. Comparing options (i) and (iv), ua1>ua4-dþd2d3d4>g. Thus a prefers to pre-empt its structural rival, c, by forming the bridge to the other component only for dþd2d3d4>g. When dþd2d3d4og, a’s preference is to let c bear the burden of making the link to the other component. Of course, c’s preferences are exactly the reverse. The argument of the previous paragraph shows the zero-sum nature of the direct structural competitors stance towards each other. In the short term, when one gains the other loses. Whether this makes any difference in the longer term depends on other considerations pﬃﬃﬃ regarding d. According the Theorem 2.1(b), for d in the range 0od ð 5 1Þ=2 when dþd2d3d4>g the stable equilibrium structures are the null graph (G1) and the cycle (G20). If there is devolution to G1, all actors get zero utility and if there is evolution the G20, then all actors get the same non-zero utility. The shortterm advantage to a under this parameterization of g relative to d disappears

Actor Utilities, Strategic Action and Network Evolution

263

pﬃﬃﬃ in the longer run. But for d in the range 0oð 5 1Þ=2odo1, from Theorem 2.2(b), the stable structures are the star (G11), the near-star (G12) and the line (G14). The star cannot be reached from G8 and G12 is reached only if b forms the bridge. So, if the line is reached, then a will have secured a longer-term advantage over its direct structural competitor, c. The same arguments hold for c relative to a. As such, this is an example of the endogeneity issue: some advantages confer longterm advantages stemming from occupying a location in a network while others do not. Some of the comparisons made regarding strategic options for actors imply that it is sometimes beneﬁcial for actors to do nothing regarding forming ties and beneﬁt indirectly from the edges formed between other actors. Some examples include: (a) The transition from G3 to G7 by adding the edge cd increases the utility of a by d3 and that of b by d2; (b) if de is then added to form a graph isomorphic with G14, the further increases in utility for a and b are, respectively, d4 and d3; (c) the transition from G8 to G12 by adding de increases the utility of e by d2þ2d3; (d) the transition from G12 to graph isomorphic to G16 by adding ad increases the utility of e by d2d3 and (e) the addition of ed to G19 increases the utility of c by d2d3. These indirect increases in utility are all in the form of general sums of powers of d where the closer the third party is to the added edge, the higher the increase in utility for that actor. For all of these examples, removing edges from the destination graph will lower the utility of actors not involved in the edge removal.

4.3. Getting the Best Return View The third view that actors can take is to ask if they are getting the highest utility in a given structure. To examine this we consider the star (G11) reached by decisions made under the ﬁrst view that actors can take. It is one of the stable structures according to Theorem 2 and merits further attention because it can be reached by agreed additions of edges. Making the argument general (Hummon, 2000), for any n, we denote the utility of the central actor (e.g. b in G11) as uc. The remaining actors all have the same utility which we denote by up. These utilities are uc ¼ (n1)(dg) and up ¼ (dg)þ(n2)d2. The usual interpretation of such center-periphery structures, regardless of whether the central actor is an individual or an organization, is that the central actor beneﬁts the most by enjoying being in the middle and ﬁlling many structural holes. So it is

264

PATRICK DOREIAN

reasonable to ask about the conditions under which the actor ‘‘in the middle’’ does the best. If uc>up then it follows that (n1)(dg)W (dg)þ(n2)d2 which implies dd2>g. From Theorem 2, this is the condition also under which the complete graph forms as a stable structure. If the star is stable then ucoup. Having uc ¼ up requires side-payments from one of more of the peripheral actors to the central actor. If each of the peripheral actors pays the same side-payment, this is ((n2)/2)(d2(dg)). With this set of side-payments, uc ¼ up ¼ ((n1)/n)[2(dg)þ(n2)d2]. So, for the central actor to have the greatest utility requires that a side-payment, q, be paid where q>((n2)/n) (d2(dg)). This sequence of arguments seems counter-intuitive from the perspective of dominant central actors ﬁlling structural holes. However, it is fully consistent with the conditions under which G11 forms. For individuals as actors, an alternative realization, in terms of the utilities to the actors is that the central actor is a parent with the peripheral actors the parents’ children. This is, perhaps, less counter-intuitive where parents are compelled to ﬁll structural holes without enjoying beneﬁts in the short term. All of the transitions that can be made through edge additions or deletions as shown in Fig. 2 can be examined in the same fashion. For the comparisons made by doing this, some preliminary conclusions emerge: 1. Having an actor (vertex) ﬁll a structural hole, i.e. form a bridging relation, is not an automatic ﬁrst preference of that actor. The bridge will be formed only when the costs, g, are low enough. 2. The exact meaning of ‘‘low enough’’ in the ﬁrst conclusion depends on the structure from which the transition could be made. This range for g in terms of general sums of powers of d can get complicated. 3. Direct structural competitors, deﬁned as being members of the same orbit of the graph, have interests that are diametrically opposed (except when the link is formed between them – which is often one of the poorest outcomes available to them separately). This implies that even if the possible destination graphs are isomorphic (and indistinguishable according to Theorem 2), which realization is reached is critically important to these actors. Gains to one imply losses to the other and vice versa. This is consistent with Rowley and Baum’s (2008) observation that bridging positions are contested resources. 4. If there are multiple disconnected components, actors in the smaller component(s) fare less well and their preference is to be linked to a larger structure. (Intuitively, this is reasonable because there are fewer options

Actor Utilities, Strategic Action and Network Evolution

265

for direct ties and access is to shorter geodesics in the small components.) But this preference is conditioned by the ﬁrst two provisional conclusions: being in the bridge need not be the ﬁrst preference of an actor. 5. Actors can derive ‘‘remote’’ increases in utility (third party beneﬁts) for changes made to the network structure that do not involve them directly. This also suggests considering the use of side-payments to induce others to form ties from which they receive beneﬁts as remote actors. The most salient of these conclusions are (i) ﬁlling structural holes need not be in the best interests of an actor and (ii) direct structural competitors have opposing interests. Also, not all structural holes are the same and the exact structure of the network within which structural holes appear has to be considered. It is clear, also, that the micro-dynamics of strategic calculations by actors are complicated and studying them in detail with regard to long-term dynamics, as done here, may be an intractable problem. One implication is that actors cannot keep track of all of the options to them under full information. Put differently, using the language of van Liere, Koppius, and Vervest (2008), having full information requires having a network horizon that encompasses the entire network with regard to both completeness and accuracy. The results of van Liere et al. (2008) suggest that this is highly unlikely. If managers, as network actors, cannot have such a wide horizon, or information available to them is limited for some other reason, stable optimal structures for organizations seem either unlikely or will occur only fortuitously. Further, if the d and/or g regimes change then these stable structures can become unstable.

5. FURTHER CONSIDERATIONS While the provisional conclusions reached above have interest value, it is clear the discussion leading to them is less than general. Some additional issues can be considered when thinking about the evolutionary trajectories for both the location of actors and network structures in terms of JW utilities. All amount to generalizations of the approach considered here and include: 1. Having all actors with the same beneﬁts, d, from ties and costs, g, for maintaining ties is overly restrictive. Actors can vary in their costs for

266

2.

3.

4.

5.

6.

7.

PATRICK DOREIAN

maintaining ties as well as the beneﬁts they derive from the networks in which they are located. One modest generalization is to consider classes of actors that differ in terms of their costs and beneﬁts. This would accommodate Rowley and Baum’s (2008) claim that not all network actors are capable of pursuing network strategies. Of course, the most general situation is that actors, i, have unique values of di and gi (which appears to be intractable analytically). Actors can be motivated to be the ones getting the most from a structure even though the structure(s) that result are sub-optimal for them and for the network as a whole. (This could be another reason why network structures persist even if they are sub-optimal.) They also may be motivated primarily by not getting the worst outcome for themselves which makes the star (G11) problematic in strategic terms unless there are side-payments. There can be actors, i,j, who are opposed to each other and take actions to block each other’s gains rather than seek the best returns for themselves. All that matters to, say i, is making sure that the opposing actor, j, does not do well or, failing that, does not do as well as i. Structurally, this is generated by being in the same orbit and being direct competitors. Having d and g ﬁxed through time is overly restrictive and allowing them to vary over time seems useful. They could even vary depending on the structure of a network at a given point in time and change as the structure changes. Including actor mortality is another generalization as well as is allowing actors to join the network over time (Amburgey et al., 2008). An organization unable to get resources from other organizations via ties to them is unlikely to survive (Pfeffer & Salancik, 1978) and as organizations die, or if joining an extant network seems appealing, other organizations could join. Even so, joining with a limited set of network ties, would be to enter in a disadvantaged fashion and could be a part of the liability of newness (Stinchcombe, 1965). In the current formulation, nothing is speciﬁed about the resources available to actors. Organizations with more resources can sustain the costs of maintaining more ties than organizations with limited resources. And as resources change over time, it seems reasonable that both di and gi change with these organizational resources. Actors, in general, cannot support an unlimited number of ties. There are at least two ways of addressing this. One is to impose limits on the number of ties that an actor can have, a crude alternative,

Actor Utilities, Strategic Action and Network Evolution

267

and a second is to have some declining marginal utility for the addition of more ties. In an inter-organizational context, a third approach is to allow different organizations with different amounts of resources and specify those with more resources as capable of sustaining more ties. 8. It may be necessary to consider the simultaneous creation of multiple edges rather than focus solely on transitions that add, or remove, a single edge. This utility framework, in general, and the strategic considerations of actors, in particular, has the potential to help understand the micro-dynamics of actor choices that generate evolutionary trajectories of network structures to stable equilibria given the parameterization of beneﬁts, d, and costs, g. Analytically, derivations seem immensely difﬁcult as the sizes of networks increase and computer methods seem necessary. Empirically, the examination of the utility dynamics requires the measurement of beneﬁts and costs as well as the amounts of information that actors have regarding the networks structure(s) within which they are embedded. Both are daunting tasks.

6. FURTHER THOUGHTS Even though the approach taken here seems rather simplistic, it is still possible to address the three central concerns motivating this volume on managerial strategic action. It is clear that inter-organizational networks are dynamic and structural advantages, should they exist, need not persist. Given this, there are additional connections that need to be made to other strands of research on network agency to make the approach taken here more useful. The assumption of known and quantiﬁed beneﬁts (d) and costs (g), as alluded to above, is rather glib. Yet, if costs and beneﬁts are not considered explicitly, there will be severe limits on what can be accomplished when studying strategic management. Rowley and Baum (2008), building on Burt (1992), regard bridging ties as especially advantageous in information intensive environments. Their discussion of the Canadian banking industry in terms of a tension between forming ties within a set of known ﬁrms and forging ties with quite different organizations is insightful. It links to the discussion of Amburgey et al. (2008) who consider bridging ties, ties forming new components, creating pendant ties and forming ties within a (dense) component. This can be combined with the utility approach taken

268

PATRICK DOREIAN

here to operationalize the costs and beneﬁts of forging these ties. When Rowley and Baum (2008) report that their results ‘‘do not provide evidence of banks’ managers explicitly pursuing structural hole advantages’’, it is tempting to wonder, given the results established herein, if the costs of forming those bridging ties were too high, relative to the beneﬁts, for these managers to do this proﬁtably. In general, an approach to network evolution that considers the coevolution of actor attributes and network structure seems the most fruitful. Consistent with this, Hite (2008) considers a coevolution of actor agency and network structure. It would be useful to couple this line of thought to the actor utility models where there costs and beneﬁts are not uniform across ties and organizations. Some ties will be more costly than others to maintain, depending on organizational attributes of the partners, and both costs (gi) and beneﬁts (di) can change over time (as actor attributes). In turn, this can be coupled to the concerns of van Liere et al. (2008) concerning the information that actors have. In the approach taken here, the networks were tiny so the assumption, frequently made implicitly (Baum & Rowley, 2008), that actors know their networks is tenable. For larger networks it is not and utility calculations will be unreliable if both accuracy in perceptions is lacking and the network horizon (van Liere et al., 2008) is too local. Clearly, full information of actors regarding both their networks and their location in their networks is an assumption that will have to be abandoned, even for large organizations with plenty of resources. Finally, Neuman et al.’s (2008) consideration of two-mode actors suggests an intriguing extension for actor utility and network evolution models. For the banks they considered, it appears that the costs and beneﬁts of adding or removing directors from their boards were highly relevant for managerial decision-making. Thinking of such two-mode networks in terms of actor utility models explicitly in terms of costs and beneﬁts is a potentially fruitful extension of these models. There is much to de done – but, oh my, what fun it will be when exploring the social mechanisms of network agency and the evolution of social networks.

NOTES 1. This implies two things: (i) there are no edges that are costless to maintain and (ii) there are no edges that yield zero direct beneﬁts.

Actor Utilities, Strategic Action and Network Evolution

269

2. When edge-speciﬁc weights are speciﬁed, the equation giving ui (G) is: ui ðGÞ ¼ P P wii þ i;j2G ðdij wij gij Þ þ tij41 ðdtij wij Þ where wii denotes self-worth, wij the strength of the ij edge, dij the edge-speciﬁc beneﬁt of the ij edge to i and gij the edge-speciﬁc cost for i of maintaining the ij edge with j. With dij ¼ d8ðijÞ 2 E; gij ¼ g8ðijÞ 2 E; wii ¼ w8i 2 V and wij ¼ 18ðijÞ 2 E we get the simpliﬁed utility equation. 3. Let Gn be the set of all edge graphs on n vertices. Let G, GuAGn be two graphs in n G , and let values (e.g. utilities) be assigned to graphs in Gn. Let v(G) and v(Gu) be the values of G and Gu. Then G is strongly efﬁcient if vðGÞ vðG0 Þ8G0 2 Gn . 4. The orbits in G25 are {(a),(b,c)(e,d)} yet a, d and e all have the same utility (2(dg)þ2d2) even though a does not belong in the same orbit as d and e. 5. The critical correction that they provide stems from being attentive to the deletion of edges as well as their addition (Doreian, 2008). 6. This is stated in terms of adding edges but the same holds for the removal of edges. 7. In the approach of Amburgey et al. (2008), such a decision creates a new component in the network.

ACKNOWLEDGMENT I appreciate greatly the comments of Joel Baum, Tim Rowley and Natasˇ a Kejzˇar that helped me improve this version over an earlier draft of this chapter. Even more, I am grateful to Joel and Tim for organizing this volume together with the thoroughly entertaining and informative conference in Toronto in 2007 that preceded it.

REFERENCES Amburgey, T. L., Al-Laham, A., Tzabbar, D., & Aharonson, B. (2008). The structural evolution of multiple organizational networks: Research and commerce in biotechnology. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 171–212). Bingley, UK: JAI/Emerald Group. Borgatti, S. P., & Everett, M. G. (1992). Notions of positions in social network analysis. In: P. V. Marsden (Ed.), Sociological methodology (pp. 1–35). San Francisco: Jossey-Bass. Breiger, R. L., Boorman, S. A., & Arabie, P. (1975). An algorithm for clustering relational data with applications to social network analysis and comparison to multidimensional scaling. Journal of Mathematical Psychology, 12, 328–383. Burt, R. S. (1992). Structural holes. Cambridge, MA: Harvard University Press. Burt, R. S. (2008). Returns to secondhand brokerage in industry networks: Spillover effects on price-cost margins in American manufacturing. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 315–360). Bingley, UK: JAI/Emerald Group. Doreian, P. (2002). Event sequences as generators of social network evolution. Social Networks, 24, 93–119.

270

PATRICK DOREIAN

Doreian, P. (2006). Actor network utilities and network evolution. Social Networks, 28, 137–164. Doreian, P. (2008). A note on actor network utilities and network evolution. Social Networks, 30, 104–106. Doreian, P., Batagelj, V., & Ferligoj, A. (2005). Generalized blockmodeling. New York: Cambridge University Press. Doreian, P., & Stokman, F. N. (1997). The dynamics and evolution of social networks. In: P. Doreian & F. N. Stokman (Eds), Evolution of social networks (pp. 1–17). New York: Gordon and Breach. Evans, K. (2002). Taking control of their lives? Agency in young adult transitions in England and the New Germany. Journal of Youth Studies, 5, 245–269. Everett, M. G., & Borgatti, S. P. (1988). Calculating role similarities: An algorithm that helps determine the orbits of a graph. Social Networks, 10, 71–91. Hite, J. M. (2008). The evolution of strategic dyadic network ties: Strategically navigating bounded agency within multidimensional and dynamic dyadic relationships. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 133–170). Bingley, UK: JAI/Emerald Group. Hummon, N. P. (2000). Utility and dynamic networks. Social Networks, 22, 221–249. Jackson, M. O., & Wolinsky (1996). A strategic model of social and economic networks. Journal of Economic Theory, 71, 44–74. Lorrain, F., & White, H. C. (1971). Structural equivalence of individuals in social networks. Journal of Mathematical Sociology, 1, 49–80. Mandel, M. J. (1983). Local roles and social networks. American Sociological Review, 48, 376–386. Monge, P. R., & Contractor, N. S. (1987). Theories of communication networks. New York: Oxford University Press. Neuman, E. J., Davis, G. F., & Mizruchi, M. S. (2008). Networks and industry consolidation in U.S. global banking: 1986–2004. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 213–247). Bingley, UK: JAI/Emerald Group. Pattison, P. E. (1988). Network models; some comments on papers in this special issue. Social Networks, 10, 383–411. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations. New York: Harper and Row. Rowley, T. J., & Baum, J. A. C. (2008) The dynamics of network strategies and positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 641–671). Bingley, UK: JAI/Emerald Group. Simon, H. A. (1976). Administrative behavior. New York: Macmillan. Stinchcombe, A. L. (1965). Social structure and organizations. In: J. G. March (Ed.), Handbook of organizations (pp. 142–193). Chicago: Rand McNally and Company. Stokman, F. N., & Doreian, P. (1997). Evolution of social networks: Processes and principles. In: P. Doreian & F. N. Stokman (Eds), Evolution of social networks (pp. 233–250). New York: Gordon and Breach. van Liere, D. W., Koppius, O. R., & Vervest, P. H. M. (2008). Network horizon: An informationbased view on the dynamics of bridging positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 595–639). Bingley, UK: JAI/Emerald Group.

Actor Utilities, Strategic Action and Network Evolution

271

Winship, C. (1988). Thoughts about roles and relations: An old document revised. Social Networks, 10, 209–231. Winship, C., & Mandel, M. (1983). Roles and positions: A critique and extension of the blockmodeling approach. In: S. Leinhardt (Ed.), Sociological methodology 1983–1984 (pp. 314–344). San Francisco: Jossey-Bass. Xie, F., & Cui, W. (2008a). Cost range and network structure. Social Networks, 30, 100–101. Xie, F., & Cui, W. (2008b). A note on the paper ‘Cost range and Network Structure’. Social Networks, 30, 100–101.

CONTRADICTORY OR COMPATIBLE? RECONSIDERING THE ‘‘TRADE-OFF’’ BETWEEN BROKERAGE AND CLOSURE ON KNOWLEDGE SHARING Ray Reagans and Bill McEvily ABSTRACT Knowledge sharing is a fundamental source of competitive advantage. Social networks are thought to play an important role in knowledge sharing, but are presumed to create a trade-off such that a network can be optimized to promote either knowledge seeking or knowledge transfer, but not both. The trade-off, however, is premised on, and representative of a broader tendency to treat, brokerage and closure as contradictory network forms. We challenge this assertion and propose a theory of knowledge sharing with brokerage and closure as compatible and complementary. Evidence from a contract research and development ﬁrm broadly supports our theory. We also report the results of a simulation analysis, which illustrate that only in the extremely rare case when a network is characterized by nearly complete balance do brokerage and closure begin to create a trade-off.

Network Strategy Advances in Strategic Management, Volume 25, 275–313 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25008-4

275

276

RAY REAGANS AND BILL MCEVILY

A logic of opposition pervades the social sciences. The logic, which casts the world in stark ‘‘either-or’’ alternatives, extends to the study of strategy and organizations. Evident across a broad range of theories from motivation (theory X versus theory Y – McGregor, 1960), to learning (exploration versus exploitation – March, 1991), to competition (generalists versus specialists – Hannan & Freeman, 1977), the logic advances understanding by identifying and explaining sources and consequences of observed contrasts. While an emphasis on opposites can illuminate, it can also obscure. Trade-offs are often presumed, where none exist. A prime example is the contrast between brokerage and closure, two network forms of social capital. Bridging structural holes in a network (i.e., brokerage) provides an actor with access to and control of more diverse knowledge and information (Burt, 1992), while the absence of structural holes in a network (i.e., closure) increases the reputation costs of uncooperative behavior (Coleman, 1988). A network is optimized for brokerage-beneﬁts when all of the contacts are disconnected from each other, while a network is optimized for closurebeneﬁts when all of the contacts are connected through a dense web of interactions. Thus, brokerage and closure do appear to represent opposite network forms. The surface contrast, however, obscures how the two network features can and often do complement each other. Being successful often requires meeting more than one performance objective. Teams are an example. Being productive as a team is in part a function of being able to creatively solve problems and to quickly and efﬁciently deliver proposed solutions (Reagans & Zuckerman, 2001; Reagans, Zuckerman, & McEvily, 2004). The absence of structural holes between team members (i.e., closure) facilitates delivery, while the presence of structural holes between contacts outside the team (i.e., brokerage) promotes creative problem solving. Product markets provide another example. Strong ties between producers in a market allow producers to credibly commit to serving a market niche (Ingram & Roberts, 2000). Structural holes between suppliers reduce manufacturing costs, while structural holes between consumers increase proﬁt margins (Burt, 2000). In general, brokerage provides a ‘‘vision’’ advantage which is valuable for idea generation and innovation, while the beneﬁts associated with closure are critical for the execution and implementation of innovations (Burt, 2005; Cowan & Jonard, 2008; Obstfeld, 2005). Brokerage and closure are also complementary in the sense that the value of each depends upon the temporal context. Brokerage-beneﬁts tend to be short-lived and immediate, while closure creates value that is longer-lived and more enduring (Soda, Usai, & Zaheer, 2004; Baum, McEvily, & Rowley, 2008).

277

Contradictory or Compatible?

The observed compatibilities between brokerage and closure have been accounted for in two ways. First, from a contingency perspective, harmony between brokerage and closure can be achieved by emphasizing the contextual relevance of each theory. Second, the causal mechanisms underlying brokerage and closure while different, are not opposite, suggesting that the two network features can operate in different parts of a network structure and at different points in time (Burt, 2000, 2005). To these two accounts, we add a third – the structural conﬁgurations representative of brokerage and closure are distinct features of networks that can vary jointly or independently. We maintain that while it is true that brokerage and closure can be in opposition in an individual’s network, they need not be and can even be complementary. After all, the network that surrounds an individual is a collection of relationships, and dynamics in one relationship have implications for another relationship in the network. Consider the ﬁveperson network illustrated in Fig. 1. The nodes in the network are academics and the ties represent previous and on-going research projects. Xavier is at the center of the network. Consider Xavier’s relationship with Jim. Jim has collaborated with Joe and Jerry who have also collaborated with Xavier. Those ties can be expected to affect how much time and effort Xavier contributes to any project he has with Jim. Xavier also does research with

Jim

Jerry

Joe

Xavier

Rafe

Fig. 1.

Rachel

Structural Compatibility of Brokerage and Closure.

278

RAY REAGANS AND BILL MCEVILY

Rachel and Rafe. Rachel and Rafe are disconnected from each other and from the cluster containing Jim, Jerry and Joe. The projects that Xavier maintains with Rachel and Rafe make Xavier a more valuable member of the cluster containing Jim, Jerry and Joe. Xavier is in a position to bring fresh ideas and a new perspective to his collaboration with Jim. So if we focus on the relationship between Xavier and Jim, brokerage outside the focal relationship complements closure around the focal relationship. But if Rachel and Rafe start to collaborate, over time Xavier will add less to the relationship with Jim, not only in terms of perspective but also in terms of effort. The new tie between Rachel and Rafe will put them in a position to demand more from Xavier. Xavier now has conﬂicting afﬁliations. So just as brokerage in non-focal relationships complements closure in a focal relationship, closure in non-focal relations can offset the beneﬁts of closure in a focal relationship. If everyone collaborates with everyone else, Xavier will no longer feel any conﬂict and closure around each tie will be reinforcing. The ﬁctitious network described above illustrates the compatibilities between brokerage and closure and further suggests that the two structural conﬁgurations are in opposition only under certain conditions. We discuss three such conditions – level of analysis, nodal network capacity and global structural balance – which may impose a trade-off if empirically observed, but may not be observed with regularity. Level of Analysis. What is often overlooked in discussions emphasizing a logic of opposition in theories of social capital is that brokerage and closure operate at different levels of analysis. Brokerage is deﬁned at the nodal level (e.g., it is a property of Xavier’s network), whereas closure is deﬁned at the dyadic level (e.g., it is a property of the relationship between Xavier and Jim). Though subtle, the distinction is crucial. Since brokerage and closure exist at different levels of analysis, if we focus on dyads in a network, the two network conﬁgurations can represent different features of networks that need not be inversely related. Indeed, the two features of network structure may even vary independently (e.g., the addition of a new co-author by Xavier would change the level of brokerage in his network, but the degree of closure around his relationship with Jerry, Jim and Joe would remain the same). Nodal Network Capacity. A second condition required for brokerage and closure structures to be in opposition is that the focal node be operating at the limit of network capacity. For instance, if an individual is maintaining the maximum number of ties possible, then any increase in the level of one network feature results in a decrease of the other network feature.

Contradictory or Compatible?

279

Returning to Fig. 1, if Xavier cannot manage any additional co-author relationships, then increasing closure around his relationship with Jim (by, for example, including Rachel as a co-author on a paper with Jim) would decrease the degree of brokerage in Xavier’s network. If a node has excess capacity, however, and can afford to add new contacts to its network, then the level of brokerage and closure can vary independently. Thus, the degree of excess capacity in a network is one important condition for determining whether brokerage and closure are in opposition, and is a condition that is likely to vary over time and across settings. Global Structural Balance. Treating brokerage and closure structural conﬁguration as opposites also makes an implicit assumption about the degree of balance in the network. Structural balance refers to the tendency for strong ties to be shared among a set of contacts. According to balance theory (Heider, 1958), if an individual is strongly connected to two contacts, those contacts are likely to be strongly connected to each other. When a network is balanced, brokerage and closure are opposed because a strong tie with any one group member results in strong ties with every group member. The end result is a social network that is organized into internally cohesive cliques. All ties occur within a group. No ties connect individuals in different groups.1 If weak ties are not subject to the same balance constraints as strong ties, weak ties can act as bridges between individuals in different groups (Granovetter, 1973). Therefore, at the extreme of complete balance, the broader social network contains internally cohesive cliques that are completely disconnected. Complete closure in the group results in no ties between groups. If the tendency towards balance is less prevalent in weaker network connections, weak ties can act as bridges between groups. The opposition between brokerage and closure is less dramatic in this scenario, but this situation still suggests that the two network features are in conﬂict because the weak tie that provides brokerage comes at the expense of a strong tie that could have increased closure inside the group even more. While it is true that brokerage and closure are in conﬂict if a network is balanced, the degree to which a particular network actually is characterized by balance is an empirical question (Moody, 2004, p. 237). Moreover, when the broader social network is not balanced, the network pattern that surrounds any particular individual can be characterized by brokerage and closure, and each network feature can have independent effects (Burt, 2000, pp. 53–58; Reagans & Zuckerman, 2001; Reagans & McEvily, 2003; Reagans et al., 2004). Compatible Logics of Social Capital. Taken together, these arguments suggest that the network conﬁgurations representative of brokerage and

280

RAY REAGANS AND BILL MCEVILY

closure can exist simultaneously and vary independently. Nevertheless, the overwhelming tendency is treat the two theories of social capital as contradictory. The proliferation of a logic of opposition between brokerage and closure, however, has obscured understanding of the causes and consequences of social structure. To ground our arguments and illustrate how the logic of opposition can bias understanding of brokerage and closure, we focus on an application of social capital theory that has received considerable attention – knowledge sharing in organizations. As we discuss below, the case of knowledge sharing is particularly germane because the bias towards a logic of opposition in theories of social capital has been mirrored in (and possibly ampliﬁed by) research on knowledge sharing. We then go on to illustrate how treating brokerage and closure as compatible, rather than contradictory, can provide a more complete understanding of knowledge sharing. In particular, we develop an integrative theory that predicts complementary effects of brokerage and closure, and we test the validity of our theory with an empirical analysis of networks and knowledge sharing in a contract research and development ﬁrm. The empirical results support our assertion that the causal mechanisms of brokerage and closure are distinct, yet complementary. The empirical results are consistent with the idea that brokerage and closure are best understood as two unique features of social networks, rather than opposite ends of a single continuum.

NETWORKS AND KNOWLEDGE SHARING IN ORGANIZATIONS Knowledge sharing is a fundamental source of competitive advantage. The dissemination of knowledge within and across organizational boundaries provides individuals and organizations with the opportunity to improve upon existing work practices and routines, discard inefﬁcient work procedures, and develop new ideas and innovation. As research on knowledge sharing in organizations has cumulated, the literature has increasingly emphasized the importance of social structure (Argote, McEvily, & Reagans, 2003). A growing body of work has substantiated the idea that the overall conﬁguration of social ties in an organization, as well as the content of ties, inﬂuences knowledge sharing in important ways. At the same time, however, research in this literature differs considerably in how it applies brokerage and closure theories of social capital, both

281

Contradictory or Compatible?

Table 1.

Network Effects on Knowledge Sharing. Network Effects Structural

Knowledge Sharing Activities

Relational Seeking Transfer Combined

Brokerage Closure Tsai (2001) Cross and Cummings (2004) Hansen et al. (2005) Reagans and McEvily (2003) Hansen (1999) Borgatti and Cross (2003) Cross and Sproul (2004) Levin and Cross (2004)

X X X X

X

X X X X X X X

X X X

X X X

X X X

conceptually and structurally. Table 1 summarizes recent research on networks and knowledge sharing. The table catalogues a diversity of approaches to the study of network effects. The logic of opposition between brokerage and closure is evident in this research in two ways. First, and perhaps most obvious, is that few studies explicitly consider the simultaneous effects of brokerage and closure on knowledge sharing. Of those studies examining structural effects (identiﬁed in the ﬁrst two columns), all but one treat network structure as unidimensional. As a result, these studies are forced to focus on one theory of social capital while overlooking the other, or, more commonly, to treat a single feature of network structure as indicative of the opposing effects of both brokerage and closure. The second way that the logic of opposition has surfaced in research on knowledge sharing is less obvious, but arguably more pervasive and more dominant than the uni-dimensional approach to network structure. Rather than focusing on one structural effect or the other, several studies indirectly infer the effects of network structure from relational properties (i.e., tie strength). Most notably, Hansen’s (1999) highly inﬂuential study on knowledge sharing uses tie strength as a surrogate for the effects of network structure. On the one hand, weak ties are argued to promote knowledge seeking behavior because weak ties are more likely to bridge structural holes and, therefore, provide access to non-redundant information sources (Burt, 1992). The focus is on weak ties, but the condition responsible for the effect is brokerage, a feature of network structure. On the other hand, strong ties are argued to be critical for knowledge transfer because strong ties are more

282

RAY REAGANS AND BILL MCEVILY

likely to be embedded in a dense web of strong third party ties, and such closure leads to the development of cooperative norms and commitment (Coleman, 1988). Again, the focus is on strong ties, but the mechanism responsible for the observed effect is closure, a characteristic of network structure. The reliance on a single relational property (tie strength) to represent two different effects of network structure forces brokerage and closure to be treated as opposites and produces an apparent trade-off between knowledge seeking and knowledge transfer activities. The proliferation of a logic of opposition between brokerage and closure in research on networks and knowledge sharing has hindered the advancement of a cumulative body of knowledge. Given the inconsistencies in how network effects have been studied, it is difﬁcult to compare results across studies and integrate ﬁndings. More importantly, existing research provides incomplete and at time contradictory ﬁndings. For instance, although Hansen’s (1999) research supports the view that brokerage and closure produce a trade-off between knowledge seeking and knowledge transfer, Reagans and McEvily (2003) found that both brokerage and closure ease the transfer of knowledge.2 Moreover, because brokerage and closure are presumed to be in opposition, past research has overlooked some possible effects of networks on knowledge sharing. In particular, while the effect of closure on knowledge transfer has received considerable attention, research has yet to consider the potential effect of closure on knowledge seeking. Similarly, the effect of brokerage on knowledge seeking has been the focus of much research, but Reagans and McEvily’s (2003) ﬁnding for an effect of brokerage on knowledge transfer has yet to be corroborated in a model that simultaneously examines knowledge seeking. Thus, the presumed brokerage–closure trade-off for knowledge seeking and knowledge transfer has yet to be fully explored. And, the tendency to treat brokerage and closure as opposites has precluded consideration of the extent to which the two structural features are related to each knowledge sharing activity and to tie strength.

Theory and Hypotheses To address the limitations associated with applying a logic of opposition between brokerage and closure to the study of knowledge sharing, we develop an integrative framework that distinguishes the effects of brokerage and closure from each other, and that distinguishes these structural effects from relational effects of networks. We also identify how each of these

Contradictory or Compatible?

283

network effects inﬂuences the two distinct phases of knowledge sharing identiﬁed in the literature: knowledge seeking and knowledge transfer. During the knowledge seeking phase, the key objective is to identify knowledge. Knowledge seeking is problem oriented; the process is initiated after a problem arises, usually after performance drops below some threshold value (Cyert & March, 1963; Nelson & Winter, 1973). Resolving the problem requires identifying a more effective strategy, manufacturing technique or routine, or in general a piece of knowledge.3 Knowledge seeking ends when the relevant piece of knowledge has been identiﬁed. If the objective during the knowledge seeking phase is identifying knowledge, the goal in the transfer phase is integrating that knowledge into the ﬁrm’s existing routines and operations. Relational Effects on Knowledge Sharing Relational properties are features of dyads and, as such, they describe different attributes of individual relationships between a focal actor and a contact. From this perspective, tie strength is clearly a relational property since it characterizes the nature of the relationship between two parties. Tie strength is also among the most widely studied relational properties, but one that ‘‘has been clouded with ambiguity and inconsistency’’ since the initial underlying rationale for the causal force of tie strength, interpersonal attachment, is seldom emphasized (Krackhardt, 1992, p. 218). Yet, it is precisely this interpersonal attachment that is at the heart of theories linking strong ties to knowledge sharing. In particular, actors who are strongly tied to each other are likely to develop shared understandings and trust (Gulati, 1998; Krackhardt, 1992), both of which feature prominently in explanations of knowledge transfer. Less recognized, however, is the idea that shared understandings and trust are also important to understanding knowledge seeking. Who individuals turn to for knowledge is determined, in part, by who they know and how well they know them. Strong ties develop over time and through repeated interactions (Granovetter, 1973; Krackhardt, 1992). As individuals become more strongly connected, they develop shared understandings, common language and a better understanding each other’s expertise (Uzzi, 1997). Thus, the stronger the relationship, the better informed the recipient is likely to be about what a colleague knows. And, this familiarity is likely to shape the direction of knowledge seeking. According to theories of associative learning and absorptive capacity, one of the primary means through which people learn new ideas is by associating new knowledge to what they already know (Cohen & Levinthal, 1990).

284

RAY REAGANS AND BILL MCEVILY

To the extent that two people have a shared understanding of each other’s knowledge it will be easier for them to explain to each other a problem they are trying to solve, an idea they are attempting to develop or an issue they are trying to resolve. This suggests that individuals are more likely to seek knowledge from those they are strongly, rather than weakly, tied to because of familiarity. A second reason that strong ties shape the direction of knowledge seeking is that turning to others for knowledge involves risk. In particular, seeking knowledge entails revealing one’s limited, or entire lack of, expertise and requires reliance on another person to provide sound advice and guidance in that area. As a result, seeking knowledge involves making oneself vulnerable to the discretion and expertise of another. Over time and through repeated interactions strong ties also facilitate the formation of trust (Blau, 1964; Krackhardt, 1992), which can help overcome the vulnerability associated with knowledge seeking. In particular, trust affects the perceived veracity of knowledge, meaning that ‘‘the receiver is less likely to verify the knowledge for accuracy and is more inclined to accept the knowledge at face value’’ (McEvily, Perrone, & Zaheer, 2003, p. 97). In addition, to the extent that trusted contacts are viewed as acting in the best interests of the receiver, they are likely to be seen as a source of important and relevant knowledge. Taken together, these arguments lead to our ﬁrst hypothesis, H1. Tie strength will be positively related to knowledge seeking; the stronger the social tie, the greater the likelihood of a knowledge seeking relationship. Although the effect of tie strength on knowledge seeking has been largely overlooked, far more attention has been devoted to the relationship between tie strength and knowledge transfer (Hansen, 1999; Levin & Cross, 2004; Reagans & McEvily, 2003; Uzzi, 1997). Strong ties are characterized by frequent contact and communication, which can lead to the development of shared understandings in the form of relationship-speciﬁc heuristics that ease the transfer of knowledge (Uzzi, 1997; Uzzi & Lancaster, 2004). The relationship-speciﬁc ‘‘know how’’ associated with strong ties also make it easier for units and organizations to obtain knowledge from outside sources (Almeida, Dokko, & Rosenkopf, 2003; Rosenkopf & Almeida, 2003; Gomes-Casseres, Hagedoorn, & Jaffe, 2006). Moreover, the interpersonal attachment characteristic of strong ties affects the motivation to provide assistance and support (Granovetter, 1973), including transferring knowledge (Hansen, 1999; Reagans & McEvily, 2003). Relative to weak ties, individuals involved in strong ties are more willing to dedicate time to their

Contradictory or Compatible?

285

interaction with, and exert more effort on behalf of, each other. And, the trust that emerges from interpersonal attachment creates a sense of conﬁdence that knowledge shared will not be used inappropriately. Based on these arguments we predict, H2. Tie strength will be positively related to knowledge transfer; the stronger the social tie, the greater the ease of transferring knowledge. Structural Effects on Knowledge Sharing Whereas relational properties refer to dyadic characteristics, structural properties represent features of social systems that describe the conﬁguration of relationships among a set of actors. As discussed above, past research on knowledge sharing has considered brokerage and closure as two key structural properties, but has tended to treat them as having unilateral effects. Speciﬁcally, closure is thought to promote knowledge transfer and brokerage is believed to facilitate knowledge seeking. This emphasis has, however, overshadowed attention to the potential dual effect of network structure on knowledge sharing; namely, that closure promotes not only knowledge transfer, but also knowledge seeking, and similarly that brokerage facilitates not only knowledge seeking, but also knowledge transfer. Closure and Knowledge Sharing. As documented by previous research (e.g., Reagans & McEvily, 2003), the link between network cohesion and ease of knowledge transfer is rooted in cooperative norms. From a reputational standpoint, individuals are more likely to assist contacts with whom they share mutual ties because news of their (un)cooperative behavior will quickly spread to those third parties and inﬂuence the individual’s ability to obtain favors and support from the entire network of mutual contacts (Coleman, 1988). Closure also facilitates the internalization of group norms, such as cooperation, by creating greater understanding of the shared values and acceptable behaviors among a set of mutually connected individuals (Granovetter, 1992). H3. Closure will be positively related to knowledge transfer; the more that a social relationship is surrounded by mutual third party ties, the greater the ease of transferring knowledge. While its effect on knowledge transfer is fairly well established, how closure inﬂuences knowledge seeking is less well understood. Since closure increases the rate at which knowledge circulates within a clique, members

286

RAY REAGANS AND BILL MCEVILY

should have a clearer sense of who knows what inside, as opposed to outside, the group and who to turn to for speciﬁc expertise and knowledge. Closure makes it more likely that group members will not only have a better sense of who knows what, but also more likely that those group members will share frameworks with each other. Combined with the increased rate at which those frameworks diffuse across group members, closure makes it more likely that individuals will have similar mental models and frameworks. And shared mental models and frameworks have been shown to make it easier for team members to learn from each other (Weber & Camerer, 2003). The cooperative norms characteristic of closure are also likely to shape who individuals turn to for knowledge. Whether it be lending their own expertise or helping to identify and locate others in the broader network with relevant expertise, contacts that provide assistance with knowledge seeking are engaging in a form of cooperative behavior. And, when contacts share mutual third party ties with an individual, they are more likely to see cooperation with others in the group as the right, and prudent, thing to do. Based on these arguments we expect that closure will not only ease the transfer of knowledge, but also inﬂuence who individuals turn to for knowledge. Formally we predict, H4. Network cohesion will be positively related to knowledge seeking; the more that a social relationship is surrounded by mutual third party ties, the greater the likelihood of a knowledge seeking relationship. Brokerage and Knowledge Sharing. Whereas closure relates to ‘‘local’’ knowledge sharing in the immediate network neighborhood, brokerage relates to ‘‘global’’ knowledge sharing in more distant regions of the broader network. Thus far, research has primarily emphasized the effect of brokerage on knowledge seeking. In particular, since ties that bridge disconnects in social structure provide access to diverse and non-redundant pockets of knowledge they are particularly valuable for solving the problem of ﬁnding relevant knowledge (Hansen, 1999; Hansen, Mors, & Løvas, 2005). The more diverse and non-redundant the sources of knowledge, the broader the scope of knowledge accessed and the greater the likelihood of discovering novel and distinct ideas that may be relevant to a particular problem or issue. H5. Brokerage will be positively related to knowledge seeking; the more non-redundancy among an individual’s social ties, the greater the likelihood of a knowledge seeking relationship.

Contradictory or Compatible?

287

In addition to knowledge seeking, recent research has also begun to examine the effect of brokerage on knowledge transfer. The explanation for the effect emphasizes structurally induced capabilities (Reagans & McEvily, 2003). In particular, research on absorptive capacity has emphasized the beneﬁts of being exposed to diverse knowledge. Exposure to knowledge developed outside of the ﬁrm, increases the ﬁrm’s capacity for absorbing external knowledge and for transferring that knowledge from the environment to the ﬁrm. Brokerage is thought to lead to the development of the same capacity. Exposure to diverse information inside the ﬁrm increases an individual’s capacity for not only absorbing, but also for framing, translating and transferring that knowledge to others. Successful transfer requires that knowledge be conveyed to recipients in a language they comprehend. And, individuals accustomed to considering multiple perspectives and communicating in different languages should be effective at framing knowledge. The more an individual’s network is characterized by brokerage, the greater the capacity for translating knowledge because they are accustomed to considering multiple perspectives and communicating in different languages. By virtue of interacting with individuals across distinct bodies of knowledge, individuals with networks high in brokerage learn how to convey complex ideas to a diverse set of contacts. Therefore, brokerage should have a positive effect on the ease of knowledge transfer. Taken together, these arguments suggest that brokerage is conducive to not only knowledge seeking, but also knowledge transfer. This leads to our ﬁnal hypothesis, H6. Brokerage will be positively related to knowledge transfer; the more non-redundancy among an individual’s social ties, the greater the ease of transferring knowledge.

THE RESEARCH CONTEXT The setting is a small contract research and development ﬁrm located in a medium-size city in the American Midwest. During the time of the study, the ﬁrm employed 83 individuals. The vast majority of the employees are scientist and engineers, holding master’s degrees and doctorates. The ﬁrm distinguishes itself from other technical consultants in the area by providing interdisciplinary scientiﬁc consulting over the entire life cycle of customers’ products. The ﬁrm provides six distinct services, including assisting clients with designing products and selecting materials, developing

288

RAY REAGANS AND BILL MCEVILY

and improving manufacturing processes, performing scientiﬁc analyses and assessing product performance and quality. Given the interdisciplinary nature of work, project teams draw members from relevant areas inside the ﬁrm (e.g., analytic services, applied science, engineering, materials, etc.). The research setting has several characteristics that make it attractive for addressing the questions at hand. First, the organization is small, so employees have many opportunities to develop network connections and to share knowledge with each other. Second, given the interdisciplinary nature of projects and assignments, success turns on the ability of individuals to share knowledge, so the ability to effectively search for and transfer knowledge is important. Finally, project members are drawn from multiple functions and have distinct areas of expertise. Therefore, we are able to analyze how different network features affect search and transfer, while controlling for the amount of knowledge two individuals have in common.

Methods and Measures Data Collection During the summer of 2004 we administered a survey onsite and we received completed surveys from 78 individuals, for a response rate of 94%. The survey asked employees to report information about their knowledge sharing relationships and their social relationships. Knowledge Sharing Measures Knowledge sharing data was collected using two name generator questions. In particular, employees were asked to list the names of colleagues who had been a source of knowledge (‘‘people you contacted when you needed assistance with one of your projects’’) during the past year. And they were also asked to list the names of colleagues for whom they had been a signiﬁcant source of knowledge (‘‘people who contacted you when they needed assistance with one of their projects’’). In response to each question, an individual could cite as many of his or her colleagues as he or she liked. On average, respondents named 18 colleagues. The distribution of citations is skewed. 70% of the respondents cited fewer than 20 individuals. 85% named fewer than 25 contacts, while one individual named approximately three quarters of the people working at the ﬁrm. The list of cited colleagues deﬁnes our measure of knowledge seeking, which is the ﬁrst dependent variable. Knowledge seeking is a binary variable that equals one if the

289

Contradictory or Compatible?

respondent indicated that a knowledge sharing relationship existed with the focal colleague. After compiling the list of knowledge sharing relationships, each employee was then asked to respond to a series of six questions describing the ease with which he or she could transfer knowledge to each colleague named (see Table 2 for details). The items indicate the ease of knowledge transfer. Cronbach’s alpha for the six items is .87. And a single factor can explain approximately 62% of the covariance among the items. Our second dependent variable is the ease of knowledge transfer, which is the mean of the six items listed in Table 2. Social Network Data Social network data was collected using a standard sociometric instrument.4 Each respondent was given a roster with the names of all employees and asked to describe the strength of his or her relationship with each person in terms of emotional closeness and communication frequency (Burt, 1984; Marsden & Campbell, 1984; Marsden, 1987). With respect to emotional closeness, each respondent was asked to indicate if he or she felt ‘‘especially close’’, ‘‘close’’, ‘‘less than close’’ or ‘‘distant’’ to the focal colleague. With respect to communication frequency, he or she was asked to indicate if he or she spoke to the focal colleague ‘‘daily’’, ‘‘weekly’’, ‘‘monthly’’ or ‘‘less often’’. Emotional closeness and communication frequency are orthogonal in the general population but are more likely to be correlated inside ﬁrms. Table 2.

Ease of Knowledge Transfer.

Item It is easy for me to explain to this person a key idea, concept, or theory in my area of expertise. It is easy for me to explain to this person new developments in my area of expertise. It is easy for me to understand a key idea, concept, or theory explained by this person. It is easy for me to understand a new development explained by this person in his/her area of expertise. Multiple attempts are necessary for me to explain a key idea, concept, or theory to this individual.a Multiple attempts are necessary for me to understand a key idea, concept, or theory explained by this individual.a

Mean

S.D.

Loading

3.84

.93

.79

3.74

.94

.80

3.93

.85

.84

3.85

.84

.79

3.61

1.00

.80

3.73

.92

.68

Unless indicated otherwise, items measured with a 5-point Likert scale, ranging from strongly disagree to strongly agree. a Reverse coded.

290

RAY REAGANS AND BILL MCEVILY

The association between communication frequency and emotional closeness is positive (w2 ¼ 4206, df ¼ 9). Firm members tend to feel especially close to individuals with whom they communicate frequently and feel distant from those with whom they communicate less often (Hansen, 1999; Reagans & McEvily, 2003). The categorical responses to the emotional closeness and communication frequency questions are given quantitative scores using a multiplicative log-linear model (Goodman, 1984; Reagans & McEvily, 2003). Since previous research has emphasized the importance of communication for knowledge sharing, we use communication frequency as our indicator of tie strength. Results based on emotional closeness lead to the same substantive conclusions. Social Network Measures The social network data is used to construct three measures. Our ﬁrst variable is an indicator of a strong tie. The variable measures the intensity of the connection from person i to contact j: strong tie ¼ mij ¼

zij þ zji ; max ðziq þ zqi Þ

where zij indicates how much person i communicates with person j. The strength of a network connection is deﬁned relative to the strongest connection that a respondent has. The second variable, strong third party ties, is an indicator of closure and measures the strength of the indirect connection from person i to person j. The strength of their indirect connection is deﬁned by their tendency to have strong connections to the same third parties q. The variable measures the average strength of the connection from person i to contact j through mutual contacts q: X strong third party ties ¼ miq mjq =ðni 1Þ, q

where ni is the size of person i’s network. The more person i and person j allocate their strongest network connections to the same third parties q, the stronger the indirect connection between them. The focus on strong third party ties is consistent with the theory of social closure. The beneﬁts created by social closure are thought to increase with the presence of third party connections, especially when the third party ties are strong (Coleman, 1988, pp. S105–S107). Our third social network variable measures the absence of range as an indicator of brokerage. Network constraint measures the absence of

291

Contradictory or Compatible?

structural holes (Burt, 1992): Ci ¼

X X ð pij þ piq pqj Þ2 ; j

P

q

where pij ¼ ðzij þ zji Þ= p ðziq þ zqi Þ. Network constraint is generally a negative function of network size (Burt, 1992, p. 57). Descriptive statistics indicate that the correlation between network constraint and network size is .91. One explanation for the magnitude of the correlation is the tendency for egocentric networks in the ﬁrm to be large.5 Network size ranges from a minimum of 20 to a maximum of 80. With respect to communication frequency, a network connection exists whenever an individual speaks with a colleague at least once a month and nearly every relationship in the ﬁrm meets that threshold. Or the tendency for egocentric networks to be large could be an artifact of how the network data was collected. Individuals have a tendency to overestimate their status in social networks. Giving subjects a roster of names and asking them to describe their relationship with everyone else provides them with the opportunity to inﬂate their social standing. When network size is large, the network proportions (pij) that are used to calculate network constraint are small and as a result, the magnitude of the correlation between network size and network constraint increases. The large correlation between network constraint and network size raises a concern about multicollinearity, which would affect the size of the standard errors for these variables and therefore signiﬁcance tests. Given the large correlation between network size and network constraint, we estimate effects for the components of network constraint, instead of estimating an effect for network constraint itself. Network constraint is a function of network size, network density and network hierarchy (Burt, 1998). Network density measures the average strength of the relationship between members PP of an individual’s network: nd i ¼ j q mjq =ni ðni 1Þ. Network density is distinct from strong third party ties. The variable strong third party ties measures the average strength of the indirect connection between the respondent and a focal colleague through third party contacts. Network density, however, measures the average strength of the direct connections between contacts in the respondent’s network. Thus, whereas strong third party ties is a property of a dyad, network density is a property of the respondent’s network. Network hierarchy measures the tendency for the members of aPrespondent’s network to be strongly connected to a single contact: hi ¼ j yij lnðyij Þ=ni lnðni Þ; where yij equals cij divided by the mean level of constraint in the respondent’s network. Whereas network density

292

RAY REAGANS AND BILL MCEVILY

measures the average strength of the relationships in a respondent’s network, network hierarchy measures the tendency for strong ties to be concentrated on a single network member. Since both variables indicate the absence of range in the respondent’s network, support for the predicted positive relationships in hypotheses H5 and H6 would be provided by negative coefﬁcients for network density and network hierarchy. By using the components of network constraint we are able to attenuate the potential for multicollinearity since the correlations between network size and network density/hierarchy are more modest than the correlation between network size and constraint. The magnitude of the correlation between network size and network density equals .70 and the magnitude of the correlation between network size and network hierarchy equals .65. The lower correlations indicate that multicollinearity is less likely to be an issue. At the same time, while the correlations are smaller they are still large. Therefore, after estimating our equations, we check our estimates for the presence of multicollinearity. A variance inﬂation factor (VIF) indicates how much a variable in a model can be explained by other variables already included in the model and is therefore an indicator of multicollinearity. For each variable, the VIF is calculated and compared to the widely accepted threshold value of ten (Belsey, Kuh, & Welsch, 1980). Apart from reducing the potential for multicollinearity, a further advantage of focusing on the components of network constraint is that we can more clearly distinguish the effect of network structure from the effect of network size.6 Control Variables In addition to the variables of theoretical interest, we include several variables to control for potentially confounding effects. The knowledge sharing process varies with the amount of knowledge that two people have in common. The source of common knowledge can differ. To measure common knowledge with respect to formal training and expertise, respondents were asked to describe their areas of expertise and the responses were coded by the individual charged with managing knowledge in the ﬁrm. Each employee could have multiple expertise areas. The number of areas that two people have in common is an indicator of knowledge overlap and the variable is expertise overlap. Functional expertise is an additional source of common knowledge. The functional areas in the ﬁrm are analytical services, applied science, business services, engineering, materials and product-life prediction. Same functional area equals one if the respondent and the focal colleague work in the same functional area. There are informal sources of common knowledge as well. In particular, individuals who occupy similar positions in the communication

293

Contradictory or Compatible?

network are exposed to similar ideas and information. Structural equivalence measures the tendency for two individuals to be involved in the same communication relationships. The Euclidean distance between two relationship patterns measures the absence of structural equivalence. Nonequivalence is the Euclidean distance between the two focal individuals in the communication network. Nonequivalent individuals have access to different information and are therefore expected to have less information in common. The knowledge sharing process is also affected by the type of knowledge an individual possesses. One factor is the amount of knowledge the source has at his or her disposal. More knowledgeable people could ﬁnd it easy to transfer knowledge or may not feel the need to share knowledge with others. To control for either of these possible effects, we include the variable knowledge breadth, which is the number of expertise areas of the focal respondent. A second factor is the extent to which the knowledge being transferred is tacit or codiﬁed because codiﬁed knowledge is easier to transfer. Codiﬁed knowledge measures the degree to which the respondent’s knowledge can be documented or encoded (see Reagans & McEvily, 2003, p. 18). As noted above, we also control for network size in terms of the number of contacts in the respondent’s network. Network size can affect a person’s ability to manage the knowledge sharing process. Demographic characteristics such as race, gender and education can affect the knowledge sharing process. Data from the human resource management department allowed us to measure how different individuals were with respect to race, sex, education level and tenure within the ﬁrm. Different race equals one if the respondent and the focal colleague are not of the same racial heritage. Different sex equals one if the respondent and the focal colleague are not the same sex. Tenure was measured as the length of time (in years and months) that the ﬁrm had employed an individual. The variable tenure dissimilarity equals the absolute value of the difference in tenure between a respondent and the focal colleague. The ﬁrm has experienced some turnover in the past three years but very little before. Both older hires is a dummy variable that equals one if both individuals were employed with the ﬁrm before 2001. The variable education dissimilarity equals the absolute value of the difference in educational level between a respondent and the focal colleague.

ANALYSES Our analysis examines how different factors affect knowledge seeking and the ease of knowledge transfer between individuals in the ﬁrm.

294

RAY REAGANS AND BILL MCEVILY

The observations are dyads. Since individuals are involved in many relationships there are multiple observations for each individual. Such clustering violates the assumption of independence in regression analysis. To adjust our standard errors for clustering, we introduce a ﬁxed effect for every individual in the ﬁrm (Mizruchi & Koenig, 1988; Mizruchi, 1989). Within a particular dyad, the dummy variables for the focal respondent and the focal colleague are set equal to one and all other indicator variables are set equal to zero. The ﬁxed effects estimation also serves as an important control for unobserved differences between ﬁrm members that can affect the knowledge sharing process. Descriptive statistics and correlations among variables are reported in Table 3. These data suggest that the social network at the ﬁrm is not governed by balance. The correlation between tie strength and strong third party ties is .45. The correlation is signiﬁcant but the magnitude of the correlation is moderate. There is a tendency for strong ties to be embedded in strong third party ties, but a strong tie can and frequently does occur outside of this condition. The correlations in Table 3 also indicate that brokerage and closure do not appear to be in conﬂict. The correlation between strong third party ties (the indicator for closure) and network density (the ﬁrst component of brokerage) is .009 and the correlation between strong third party ties and network hierarchy (the second component of brokerage) is .12. Since network density and network hierarchy measure the absence of brokerage, the correlations indicate that any relationship in the ﬁrm is more likely to be embedded in strong third party ties if the broader social network around the respondent is expansive. Finally, the correlation between tie strength and network density is .21 and the correlation between tie strength and network hierarchy is .22. Hence, those individuals with more expansive networks (i.e., brokerage) tend to have strong relationships with other members of the ﬁrm. Combined, the pattern of correlations suggests that a large network full of strong ties is more likely to include contacts that are disconnected from each other. At the same time, each relationship in the network is more likely to be embedded in some (but not necessarily the same) strong third party ties. Overall, the correlations suggest that in the current organization the different network features are not in conﬂict.

Knowledge Seeking Next we consider how tie strength, brokerage and closure affect the likelihood that a transfer relationship exists between a respondent and a

Variables

Mean

S.D.

Knowledge transfer .21 .41 tie Network size 64.8 13.2 Knowledge breadth 1.44 .272 Knowledge 3.80 1.39 codiﬁability Different race .07 .32 Different sex .32 .46 Education .88 .86 dissimilarity Tenure dissimilarity 5.10 3.78 Expertise overlap .31 .50 Nonequivalence 4.59 .99 Same functional area .21 .40 Both older hires .58 .49 Strong tie .36 .28 Strong third party .18 .05 ties Network density .41 .02 Network hierarchy .05 .10

1

2

.13 .08 .19 .06 .09 .03 .03 .05

.01 .08 .07

Descriptive Statistics and Correlations. 3

4

5

6

.07 .10

.03

7

9

10

11

.32 .07 .11 .23

.01 .15 .11

12

13

.10 .15

.45

14

15

.01 .01 .12 .09 .05 .09 .03

.05 .06 .09 .07 .01 .03 .08 .28 .07 .29 .002 .04 .15 .06 .31 .007 .02 .04 .04 .009 .03 .23 .01 .04 .001 .003 .08 .08 .19 .14 .14 .00 .09 .06 .14 .22 .29 .02 .02 .04 .06 .009 .19 .05 .07 .01 .06 .12 .07 .16 .70 .05 .65

8

Contradictory or Compatible?

Table 3.

.18 .03

.14 .04

.02 .04

.03 .07

.06 .06

.01 .08 .21 .07 .29 .28 .12 .05 .12 .11 .07

.03 .07 .006 .001 .17 .21 .009 .02 .001 .05 .008 .03 .22 .12 .24

295

296

RAY REAGANS AND BILL MCEVILY

focal colleague.7 The results from the analysis are in Table 4. Control variables are entered in model 1. The variables of interests are entered after the control variables and added sequentially across Table 4 to illustrate their independent and joint contributions. Model 8 is the full model. The estimates in model 8 indicate that all three network features have a positive effect on knowledge seeking, providing support for hypotheses H1–H3.8 Speciﬁcally, the strength of the connection to another individual and the extent to which that tie is surrounded by strong third party ties increase the likelihood of seeking knowledge from that individual. The positive coefﬁcients are consistent with existing theoretical frameworks that emphasize the importance of local search for knowledge sharing. The coefﬁcients for network density and network hierarchy are negative, indicating that brokerage has a positive effect on knowledge seeking. The effects for brokerage are consistent with theoretical frameworks that emphasize the importance of broad information access for search. As noted previously, we examined the variance inﬂation factors for each predictor to assess the potential for multicollinearity. The VIF for network density is 4.95 and the VIF for network hierarchy is 4.49. Multicollinearity does not appear to be a problem for the network structure variables. However, the VIF for network size is 11.43, which is above the critical threshold, so multicollinearity could be a problem for our estimation of the network size effect. Since multicollinearity tends to inﬂate the standard errors of the affected variables (Schroeder, Sjoquist, & Stephan, 1986, p. 27), the signiﬁcance level for network size could be larger than the value reported in Table 4.

Ease of Knowledge Transfer We now turn to an analysis of how the properties of network structure affect the ease of transferring knowledge. Since this analysis focuses on observed knowledge sharing relationships and since knowledge sharing does not occur in every social relationship observed, the number of observations for the analysis of knowledge transfer is smaller than the number of observations in the preceding analysis. For this reason, we include a second set of descriptive statistics (Table 5). Results of the analysis using OLS regression are reported in Table 6 and ease of knowledge transfer is the dependent variable. Control variables are in model 1. Once again, the variables of interests are entered after the control variables and added sequentially across Table 6 to illustrate their independent and joint

Networks Effects on Knowledge Seekinga.

Variable

Intercept Network size Knowledge breadth Knowledge codiﬁability Different race Different sex Education dissimilarity Tenure dissimilarity Expertise overlap Nonequivalence Same functional area Both older hires Strong tie

Knowledge Seeking (2)

(3)

2.498 (.626) 004 (.005) .111 (.046) .022 (.039) .724 (.575) .079 (.116) .215 (.060) .004 (.015) .853 (.088) .673 (.046) 1.014 (.097) .407 (.305)

2.043 (.631) .006 (.005) .110 (.046) .019 (.039) .850 (.595) .071 (.116) .201 (.061) .008 (.015) .843 (.089) .660 (.046) .903 (.098) .468 (.307) 1.260 (.165)

1.948 (1.079) .033 (.007) .111 (.046) .018 (.039) .733 (.581) .091 (.116) .209 (.060) .003 (.015) .846 (.088) .556 (.052) .973 (.097) .420 (.306) –

(4) 14.22 (2.20) .037 (.009) .077 (.047) .015 (.039) .724 (.576) .080 (.117) .214 (.060) .004 (.015) .859 (.088) .680 (.046) 1.023 (.097) .417 (.306) –

(5) 1.504 (1.085) .029 (.007) .110 (.046) .017 (.039) .851 (.600) .081 (.116) .198 (.061) .007 (.015) .839 (.089) .568 (.052) .880 (.098) .469 (.307) 1.168 (.168)

(6) 13.52 (2.21) .034 (.009) .011 (.039) .012 (.039) .849 (.597) .073 (.117) .200 (.061) .007 (.015) .849 (.089) .667 (.046) .913 (.098) .480 (.308) 1.245 (.166)

(7) 10.05 (2.33) .007 (.010) .079 (.047) .012 (.039) .733 (.583) .093 (.117) .208 (.060) .003 (.015) .852 (.088) .559 (.052) .980 (.097) .432 (.307) –

(8) 10.19 (2.34) .010 (.010) .079 (.047) .008 (.039) .850 (.601) .084 (.117) .197 (.061) .006 (.015) .845 (.089) .570 (.052) .888 (.098) .483 (.308) 1.149 (.169)

297

(1)

Contradictory or Compatible?

Table 4.

298

Table 4. (Continued ) Variable

Knowledge Seeking (1)

(2)

Strong third party ties

(3)

(4)

(5)

10.052 (1.96)

–

8.148 (2.00) –

Network density Network hierarchy Model ﬁt Number of observations Log likelihood

6,314 2,256.82

a

Standard errors in parentheses.

6,314 2,243.40

6,314 2,240.27

–

6,314 2,219.21

(7)

(8)

19.78 (3.88) 17.57 (4.79)

10.48 (1.98) 21.19 (3.86) 17.84 (4.77)

8.63 (2.02) 20.79 (3.89) 17.04 (4.80)

6,314 2,212.01

6,314 2,225.91

6,314 2,202.77

–

RAY REAGANS AND BILL MCEVILY

po.10; po.05; po.01.

6,314 2,227.56

20.06 (3.84) 18.54 (4.77)

(6)

Variables

Mean S.D.

Ease of transfer 3.84 .74 Network size 68.2 10.0 Knowledge breadth 1.61 1.16 Knowledge 3.64 1.39 codiﬁability Different race .05 .32 Different sex .29 .46 Education .79 .86 dissimilarity Tenure dissimilarity 4.70 3.78 Expertise overlap .58 .50 Nonequivalence 4.00 .99 Same functional area .39 .40 Both older hires .77 .49 Strong tie .48 .28 Strong third .20 .05 party ties Network density .40 .02 Network hierarchy .04 .10

1

2

.09 .006 .13 .08 .05

Descriptive Statistics and Correlations. 3

4

5

7

8

9

10

11

12

13

14

15

.08

.04 .07 .04

.001 .02 .09 .12 .16 .003 .06 .08 .02 .06 .04

.04 .08 .10 .11 .04 .04 .01

.04 .05 .09 .008 .29 .05 .09 .05 .03 .07 .11 .07 .13 .09 .03 .22 .03 .05 .03 .06 .17

.004 .69 .06 .57

6

Contradictory or Compatible?

Table 5.

.09 .14

.07 .08

.10

.03 .07 .02 .03 .26 .09 .01 .02 .02 .04 .02 .11 .11 .21 .07 .04 .26 .29 .10 .03 .13 .28 .08 .05 .05 .08 .03 .02 .02 .08 .14 .18 .04 .10 .06 .01 .01 .23 .10 .05 .05 .08 .02 .13 .09

.09 .01 .01 .01 .02 .07

.03 .001

.10 .21 .003 .07

.45 .10 .007 .08 .14 .25

299

300

Table 6. Networks Effects on the Ease of Knowledge Transfera. Variable

Intercept

Knowledge breadth Knowledge codiﬁability Different race Different sex Education dissimilarity Tenure dissimilarity Expertise overlap

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

4.372 (.308) .007 (.002) .016 (.022) .013 (.019) .134 (.252) .062 (.054) .042 (.029) .005 (.007) .109 (.039)

4.270 (.310) .007 (.002) .014 (.022) .012 (.019) .115 (.252) .058 (.054) .038 (.029) .005 (.007) .106 (.039)

3.379 (.581) .001 (.004) .016 (.022) .012 (.019) .140 (.252) .058 (.054) .040 (.029) .005 (.007) .109 (.039)

7.478 (1.01) .018 (.004) .010 (.022) .015 (.019) .130 (.251) .060 (.054) .040 (.029) .005 (.007) .109 (.039)

3.455 (.581) .002 (.004) .014 (.022) .012 (.019) .122 (.252) .055 (.054) .037 (.029) .005 (.007) .107 (.039)

7.364 (1.106) .017 (.004) .009 (.022) .014 (.019) .111 (.251) .056 (.054) .037 (.029) .005 (.007) .106 (.039)

6.571 (1.175) .011 (.005) .011 (.022) .013 (.019) .136 (.251) .053 (.054) .038 (.029) .005 (.007) .109 (.039)

6.600 (1.173) .012 (.005) .009 (.022) .013 (.019) .118 (.251) .053 (.054) .035 (.029) .005 (.007) .107 (.039)

RAY REAGANS AND BILL MCEVILY

Network size

Knowledge Seeking

Same functional area Both older hires

.060 (.026) .183 (.046) .316 (.169)

Strong tie

.054 (.026) .166 (.046) .339 (.169) .216 (.082)

Strong third party ties

.034 (.029) .177 (.046) .325 (.169) –

.062 (.026) .187 (.046) .300 (.169) –

2.111 (1.049)

–

Network density Network hierarchy Model ﬁt Number of observations R2 Adj R2

1384 .2924 .2413

1384 .2961 .2448

1384 .2946 .2431

5.492 (1.974) 3.825 (2.321) 1384 .2971 .2452

.033 (.029) .163 (.046) .344 (.169) .196 (.083) 1.754 (1.058) – – 1384 .2976 .2458

.056 (.026) .170 (.046) .323 (.169) .214 (.082) – 5.541 (1.969) 3.434 (2.321) 1384 .3007 .2486

.033 (.029) .181 (.046) .310 (.169) – 2.391 (1.051) 5.925 (1.980) 3.835 (2.317) 1384 .2999 .2477

.032 (.029) .166 (.046) .329 (.168) .191 (.083) 2.046 (1.060) 5.906 (1.976) 3.484 (2.319) 1384 .3027 .2501

Contradictory or Compatible?

Nonequivalence

po.10; po.05; po.01. a

Standard errors in parentheses.

301

302

RAY REAGANS AND BILL MCEVILY

contributions. The coefﬁcients indicate that tie strength9 has a positive effect on the ease of knowledge transfer as does the presence of strong third party ties. The coefﬁcients for the components of brokerage (network density and network hierarchy) are both negative, but the coefﬁcient for network hierarchy is not signiﬁcant. When an individual is surrounded by a network where contacts are strongly connected to each other, he or she ﬁnds it more difﬁcult to share knowledge with others.10 The tendency for the network to be focused on a single individual does not have an effect on the ease of knowledge transfer. Overall, the observed pattern of effects provides strongest support for H4 and H5 and weaker support for H6.11 All three network features have a positive effect on the ease of knowledge. Analysis of the variance inﬂation factors for this model indicates that multicollinearity is not a concern. The VIF for network size is 7.57. The VIF for network density is 4.71 and the VIF for network hierarchy is 3.95.

DISCUSSION How do the observed empirical effects inform our understanding of the effects of social networks on knowledge sharing? Recall in particular our observations that the literature on networks and knowledge sharing has proliferated a logic of opposition between brokerage and closure and has been inconsistent in how network effects have been studied. The observations suggest an inability to compare results across studies, to develop a cumulative body of knowledge, and ultimately to derive a more precise understanding of the underlying causal mechanisms. This study addresses these observations by developing and testing an integrated framework that simultaneously examines the direct effects of a consistent set of network properties on a common set of knowledge sharing activities. And, as the results of our analyses make clear, adopting such an approach is particularly important for evaluating whether networks present a trade-off between knowledge sharing activities. Based on the evidence reported here, we ﬁnd little support for the notion that the different network properties are in opposition. Instead, the evidence reported here is more consistent with the view that the different network properties are mutually reinforcing in the sense that they each promote knowledge seeking and knowledge transfer, albeit in different ways and to differing degrees. Starting with the relationships among the different network properties themselves, the descriptive statistics suggest that the three properties are related, but distinct. Tie strength tends to co-occur with, but is also distinct

Contradictory or Compatible?

303

from, closure. Similarly, and perhaps less intuitively, tie strength and brokerage are positively related, reﬂecting the idea that a network of expansive ties is compatible with strong ties. And brokerage and closure are not in opposition. Moving on to our multivariate models, we also found that each network feature had a positive effect on knowledge seeking and knowledge transfer. The same network features that promote knowledge seeking also ease knowledge transfer. Therefore, the observed effects suggest that the presumed trade-off between knowledge seeking and knowledge transfer is subject to question.12 At the extreme it may be the case that social networks produce a trade-off between knowledge seeking and transfer, particularly where an individual is forced to choose whether to invest in closure around a particular relationship or invest in brokerage in the broader network. Yet, focusing on the extreme obscures the fact that a particular network pattern can be characterized by closure and brokerage, and that each network feature can have a positive effect on the outcome of interest. Therefore, instead of thinking in terms of trade-offs on investments, our ﬁndings suggest that it could be more valuable to think in terms of returns on those investments. In particular, the return to investing in either a strong tie or closure is limited to a small set of relationships. An investment in a strong tie with a focal colleague has a positive effect on knowledge sharing in the relationship with that focal colleague and indirectly if the strong tie promotes closure around other relationships. However, the beneﬁt created by brokerage accumulates across every relationship in a person’s network, which can make the investment in brokerage more valuable. This illustrates a fundamental difference between brokerage and closure. Both network forms of social capital are a function of social relationships. But closure (and tie strength) is a property of a particular relationship, while brokerage is a property of a network position. The difference has important implications for the debate over whether one kind of social capital is more valuable than the other kind. In addition to advancing our understanding of how social networks affect knowledge seeking and knowledge transfer, the empirical results provide insight into the underlying process of seeking knowledge at the ﬁrm. All three network features have a positive effect on the likelihood of observing a knowledge sharing relationship. The brokerage effect describes the extent that an individual is willing to share knowledge with a colleague based on information gathered through indirect sources. Even if two people are not connected by a strong tie or embedded in a dense web of strong third party ties, brokerage increases the likelihood that a knowledge sharing

304

RAY REAGANS AND BILL MCEVILY

relationship exists. Put differently, if an individual lacks brokerage, he or she is only willing to share knowledge with colleagues with whom he or she has a strong tie directly or indirectly through closure. This suggests that individuals with less brokerage in their networks seek knowledge locally in the immediate network neighborhood deﬁned by tie strength and closure. However, since all three network features have a positive effect on knowledge seeking and since the features are not in conﬂict, social networks can facilitate local and global knowledge seeking. But since the coefﬁcients for brokerage are larger than the coefﬁcients for tie strength and closure, our ﬁndings suggest that broad information access has more of an affect on knowledge sharing outcomes. So even if an individual can seek knowledge locally and globally, in the ﬁrm that we studied global knowledge seeking is the more frequent mode and is more likely to result in success.

Directions for Future Research This research is not without its limitations. We discuss three limitations that affect our ability to generalize our ﬁndings to other empirical contexts and represent important areas for future research. Knowledge Seeking Process First, we treat the existence of a knowledge sharing relationship as the successful culmination of knowledge seeking activities. As a result, we do not know how much time individuals spent seeking knowledge, which is a clear drawback to our approach and limits our ability to generalize our ﬁndings. Moreover, our approach to studying knowledge seeking also prevents us from observing attempts where the relevant piece of knowledge was identiﬁed but the person was unable to access the knowledge. We were, however, able to estimate the tendency for knowledge seeking to end unsuccessfully because an individual was unable to establish a knowledge sharing relationship. Speciﬁcally, we asked respondents to provide the names of colleagues with whom they currently do not, but would like to, share knowledge. The 78 respondents reported a total of 60 contacts, or less than 1 contact per person on average. To put this in perspective, on average respondents identiﬁed 18 colleagues, with whom they currently share knowledge. If the list of contacts also includes desired but unsuccessful knowledge sharing relationships, at most unsuccessful knowledge seeking represents about 5% (i.e., 1/18) of successful efforts. Based on this, it seems unlikely that knowledge seeking ends unsuccessfully because sources were

Contradictory or Compatible?

305

unwilling to share what they know. While we think focusing on an outcome of the knowledge seeking process and inferring aspects of the process from the pattern of the network effects has merits, we recognize the limitations of our approach and see the value of analyzing process as well. The different network features could have different effects on the knowledge seeking process. For example, it is possible that it takes less time to seek knowledge when the conduit is a strong relationship. Therefore, a more thorough analysis would consider how the different network features affect different dimensions of the knowledge seeking processes as well as multiple knowledge sharing outcomes. Structural Balance Second, our analysis is based on data from a single ﬁrm at a single point in time and ﬁrm characteristics affect the evidence that we observe. In particular, the ﬁrm is small. And since the ﬁrm is small, everyone has an opportunity to communicate with everyone else and almost everyone does so at least once a month. As a result, the communication network is dense. The observed density level affects how the different network features vary and co-vary. With so many connections in the ﬁrm, variation with respect to brokerage is restricted. For example, network density (an indicator of the absence of brokerage) has a minimum of .36 and a maximum value of .51, while the variable strong third party ties (an indicator of closure) has a minimum value of .01 and a maximum value of .44. Brokerage varies more in an organization that is full of structural holes. It is possible that the small negative correlation that we observe between brokerage and closure is a function of the observed density level in the ﬁrm. Since almost everyone is connected to everyone else, brokerage varies little, which can attenuate the correlation between brokerage and closure (Bollen & Barb, 1981). In a situation where brokerage could vary more, for example, a ﬁrm that is organized into a number of internally cohesive groups, the correlation between brokerage and closure could be more negative. This outcome would be consistent with our earlier observation that the correlation between brokerage and closure is not automatic but instead results from the overall pattern of connections in the broader organization – i.e., the degree of structural balance. To examine the correlation between brokerage and closure in a context characterized by a higher degree of structural balance, we generated connections among 100 nodes with the likelihood of two nodes being connected varying as a function of a systematic component (proportion of third party ties in common) and a random component. Network size was set equal to 20 to allow for structural holes in the network and the focus on

306

RAY REAGANS AND BILL MCEVILY

strong third party ties increased the tendency for network connections to be balanced. The simulation was run 10 times to generate a large number of dyadic relationships. On average, network connections are balanced. The correlation between being connected and the strength of third party ties around a relationship was .75. However, even in this empirical context, the presence of strong third party ties around one relationship did not guarantee strong third party ties around all of the relationships in ego’s network. If ego and alter are connected, the correlation between strong third party ties around the focal relationship (balance in the focal relationship) and network density (balance in all of ego’s relationships) equals .60. The correlation is large but not so large to indicate that the presence of balance in one of ego’s relationships means that all of ego’s relationships will be balanced as well. Therefore, even when the broader social network is characterized by a high degree of structural balance, the trade-off between brokerage and closure is not automatic. As the broader social network becomes more balanced, the magnitude of the trade-off between brokerage and closure will increase. It seems unlikely, however, that most organizational/social networks will approach the degree of structural balance that we simulated and since it is difﬁcult to predict the degree of structural balance a priori, it seems advisable to treat the possible trade-off between brokerage and closure as an empirical question that is to be examined instead of being assumed. Along these lines, then, a fruitful area for future research is to consider the issues that we have addressed in this research in a ﬁrm that is full of structural holes. The fact that we observed brokerage effects in a context that is characterized by few structural holes, however, suggests that the brokerage effects would be even stronger in a ﬁrm with more structural holes. The extent to which the beneﬁts provided by brokerage come at the expense of the beneﬁts provided by closure is an empirical question, which depends on the overall organization of ties in the ﬁrm. Information versus Control Finally, brokerage models in social capital research emphasize the information and control beneﬁts provided by ties that bridge structural holes. The level of density in the ﬁrm suggests that the brokerage effects that we observed have very little to do with control and more to do with exposure to diverse information. Since everyone is connected to everyone else, either directly or indirectly, there are few structural holes in the ﬁrm. The large number of indirect connections means that individuals have access to information through multiple colleagues, reducing the control that any one individual has

307

Contradictory or Compatible?

over any piece of information. This suggests that the brokerage effects we observed are not simply a function of being exposed to diverse information but being exposed to diverse information more frequently.

CONCLUSION Despite these limitations, this study has important implications for research on knowledge sharing and for organizational strategy deﬁned more broadly. The fact that sharing knowledge is a kind of cooperative behavior provides an additional explanation for why strong ties and closure increase the likelihood of observing a knowledge sharing relationship. Strong ties are more likely to be characterized by emotional commitment and motivation to provide support and assistance. And closure promotes cooperation by making uncooperative behavior more costly. The information beneﬁts provided by strong third party ties combine with the emphasis on cooperation to increase the likelihood that relevant knowledge will be identiﬁed and that the source will share it when it is requested. In addition to highlighting the importance of cooperation for knowledge sharing, the ﬁndings can extend our understanding of cooperation. Direct and indirect strong ties provide a form of social insurance that generates cooperation by reducing risk. While it is certainly the case that individuals cooperate when it is safe to do so, our ﬁndings remind us that another reason that individuals cooperate is because it can be beneﬁcial to do so (Coleman, 1990, Chapter 5). Our ﬁndings suggest that individuals with more brokerage in their communication network have a better sense of the beneﬁts of cooperation. Broad information access increases the likelihood that an individual will cooperate with others, even when cooperating is risky. That is, brokerage has a positive effect on transfer, even when the two individuals are not connected by a strong tie directly or indirectly through mutual third parties. The exclusive focus on strong ties for cooperation has resulted in a lack of appreciation for how broad information access facilitates cooperation. To obtain a better appreciation of the different ways that social networks generate cooperation, future research should explore this issue in greater depth. Previous research has highlighted the importance of the social guarantee provided by strong ties. The decision to cooperate is not only driven by the risks associated with cooperation, it is also driven by potential gains. Our ﬁndings suggest that social networks provide information that is critical for ﬁguring out not only who is worthy of cooperation, but also what cooperation is worth.

308

RAY REAGANS AND BILL MCEVILY

NOTES 1. Instead of being organized into multiple cliques, the social network could be organized into a single clique. Since all members of such a network would be strongly connected to each other, information would be completely redundant and brokerage will be absent. 2. A recent study by Tortoriello and Krackhardt (2008) similarly treats brokerage and closure as compatible, but in the context of innovation rather than knowledge sharing, and focuses on the co-occurrence of brokerage and closure around a particular dyad, known as Simmelian ties, rather than examining their independent effects. 3. We do not distinguish among different types of knowledge sought. Whether individuals are searching for new ideas or seeking to corroborate existing knowledge that is not uniformly understood, the fact the individuals are striving to solve a problem is consistent. The extent to which such problem solving occurs locally in the immediate network, or more globally across relatively disconnected social spheres is the question of interest. 4. Subjects completed separate survey booklets for social network and knowledge transfer data and the survey booklets were completed one at a time to prevent crossreferencing of responses. 5. The tendency for the egocentric networks to be large could be a function of the time when the network data was collected. The survey was administered at the end of an economic downturn in the consulting market, so ﬁrm members were more focused on revenue generating activities. The increased focused on revenue could have increased the tendency for ﬁrm members to communicate with each other. 6. Results to be presented lead to the same substantive conclusions, if we estimate an effect for network constraint, instead of estimating effects for density and hierarchy. 7. Consistent with previous social network research we assume that dynamics in a particular relationship are a by-product of existing network connections (Granovetter, 1985; Coleman, 1988). That is, we treat networks as endogenous (Gould, 2002; Burris, 2004). The endogenous nature of network connections has been analyzed by previous researchers in two distinct ways. Some researchers use structural characteristics of a network to predict dyadic relationships in the same network either at a single point in time (Sorenson & Stuart, 2001) or over time (Gulati & Gargiulo, 1999; Podolny, 2001). An example would be predicting the likelihood that two individuals are friends or will become friends as a function of the number of friends they have in common. Other researchers use characteristics of one network to predict outcomes in a network deﬁned by a different kind of relationship. An example would be predicting business relationships from family connections (Peng, 2004). Another example would be using the strength of direct and indirect connections in the social network to predict the likelihood that two individuals are connected in the trust network (Burt & Knez, 1995). 8. We conducted a bootstrap analysis to examine the robustness of the empirical results (Efron, 1992). The ﬁnal model in Table 4 was estimated 1,000 times after drawing 1,000 sub-samples of the original data set. Each sub-sample is based on random draws of the original dyads with replacement, so any particular dyad is at

Contradictory or Compatible?

309

risk to be sampled multiple times. The bias corrected 95% conﬁdence interval describes how much the model coefﬁcients vary across the different sub-samples. The 95% conﬁdence interval for strong ties ranges from .734 to 1.448. The interval for strong third party ties ranges from 4.197 to 12.596. The interval for network density ranges from 29.007 to 12.663 and the interval for network hierarchy ranges from 28.419 to 8.451. The magnitude of the coefﬁcients varies across the sub-samples but lead to the same substantive conclusions as the coefﬁcients in Table 4. The robustness check indicates that our ﬁndings are not sensitive to the inclusion or exclusion of particular dyadic relationships. 9. Previous research suggests that strong ties are even more valuable for the transfer of tacit knowledge, which has lead many researchers to conclude that there should be a signiﬁcant interaction between type of knowledge and tie strength. If we add an interaction between codiﬁed knowledge and tie strength to our ease of transfer equation, the coefﬁcient is not signiﬁcant. However, if we remove the individual ﬁxed effects from the model, the interaction is positive and signiﬁcant. This suggests that any tendency for strong ties to make it easier to share knowledge is a function of unobserved differences between the source and the recipient. If we add the interaction between codiﬁed knowledge and tie strength to the knowledge seeking equation, the predictor is never signiﬁcant, even if the individual ﬁxed effects are not included in the model. 10. We also checked for possible interactions between the different network variables. The interaction terms involving the different network variables are entered in blocks. We started by adding the interaction terms between tie strength and closure. Next we entered the interactions between tie strength and the brokerage variables. Finally, we added the interaction terms between closure and the brokerage variables. None of the interactions terms are signiﬁcant when the individual ﬁxed effects are included in the model. 11. As with the ﬁrst stage, we conducted a bootstrap analysis to examine the robustness of our ﬁndings (Efron, 1992). The ﬁnal model in Table 6 was estimated 1,000 times after drawing 1,000 sub-samples of the original data set. The bias corrected 95% conﬁdence interval describes how much the model coefﬁcients vary across the different sub-samples. The conﬁdence interval for strong ties ranges from .018 to .235. The interval for strong third party ties ranges from .242 to 2.892. The interval for network density ranges from 6.574 to 1.022. And the interval for network hierarchy ranges from 2.679 to 3.027. The magnitude of the coefﬁcients vary across the different sub-samples but lead to the same substantive conclusions as the coefﬁcients in Table 6. 12. We have focused on the potential trade-off between knowledge seeking and knowledge transfer created by the different network features. It is possible that there is an inherent trade-off between the two activities. Time spent seeking knowledge could come at the expense of time spent transferring knowledge. This would suggest that the level of knowledge seeking activity has a negative effect on the ease with which an individual can transfer knowledge with colleagues. Or individuals could have limited capacity for knowledge seeking such that as the level of this activity increases, an individual is less likely to engage in further knowledge seeking. To explore these possibilities, we re-estimated the knowledge seeking and ease of transfer equations and included a predictor for the number of knowledge transfer

310

RAY REAGANS AND BILL MCEVILY

relationships in which the focal individual was involved, not including the focal relationship. The number of knowledge transfer relationships indicates the level of knowledge seeking activity. The results indicate that the level of knowledge seeking has a positive effect on observing a transfer relationship, or further knowledge seeking. This either suggests that the level of knowledge seeking does not have a negative effect on further knowledge seeking or the association is not estimated at a high enough level to observe the limit. If we add the number of transfer relationships to the ease of transfer equation, the coefﬁcient is not signiﬁcant, which suggests that the level of knowledge seeking does not have a negative effect on the ability to transfer knowledge. So there does not appear to be an inherent trade-off between the two kinds of activities. Moreover, when the level of knowledge seeking is included in the knowledge seeking and ease of transfer equations, the effects for the network variables are essentially the same.

ACKNOWLEDGMENTS We would like to thank Joel Brockner, Ron Burt, Martine Haas, Olav Sorenson and Marco Tortoriello for helpful comments and suggestions on earlier drafts of this chapter. The chapter also beneﬁtted from insightful feedback from the participants in the Knowledge Management conference, held at Imperial College, London, in November 2006 and the Network Strategy conference, held at the University of Toronto in May 2007.

REFERENCES Almeida, P., Dokko, G., & Rosenkopf, L. (2003). Startup size and the mechanisms of external learning: Increasing opportunity but declining opportunity? Research Policy, 23(2), 301–315. Argote, L., McEvily, B., & Reagans, R. (2003). Managing knowledge in organizations: An integrative framework and review of emerging themes. Management Science, 49(4), 571–582. Baum, J.A.C., McEvily, B., & Rowley, T.J. (2008). Better with age? Tie longevity and the performance implications of bridging and closure. Working paper, University of Toronto, Toronto, Canada. Belsey, David A., Kuh, E., & Welsch, Roy E. (1980). Regression diagnostics. New York: Wiley Press. Blau, P. M. (1964). Exchange and power in social life. New York: Wiley. Bollen, K., & Barb, KH. (1981). Pearson’s R and coarsely categorized measures. American Sociological Review, 46, 232–239. Borgatti, S. P., & Cross, R. (2003). A relational view of information seeking and learning in social networks. Management Science, 49(4), 432–445.

Contradictory or Compatible?

311

Burris, V. (2004). The academic case system: Prestige hierarchies in PhD exchange networks. American Sociological Review, 69, 239–264. Burt, R. S. (1984). Network items and the general social survey. Social Networks, 6, 293–339. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. (1998). The gender of social capital. Rationality and Society, 10, 5–46. Burt, R. S. (2000). The network structure of social capital. In: R. I. Sutton & B. M. Staw (Eds), Research in organizational behavior (Vol. 22). Greenwich, CT: JAI Press. Burt, R. S. (2005). Brokerage and closure: An introduction to social capital. New York: Oxford University Press. Burt, R. S., & Knez, M. (1995). Kinds of third-party effects on trust. Rationality and Society, 7, 255–292. Cohen, W., & Levinthal, D. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35, 128–152. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Harvard University Press. Cowan, T., & Jonard, T. (2008). If the alliance ﬁts y : Innovation and network dynamics. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in Strategic Management (Vol. 25, pp. 427–455). Bingley, UK: JAI/Emerald Group. Cross, R., & Cummings, J. N. (2004). Tie and network correlates of individual performance in knowledge-intensive work. Academy of Management Journal, 47(6), 928–937. Cross, R., & Sproul, L. (2004). More than an answer: Information relationships for actionable knowledge. Organization Science, 15(4), 446–462. Cyert, R. M., & March, J. G. (1963). A behavioral theory of the ﬁrm. Englewood Cliffs, NJ: Prentice-Hall. Efron, B. (1992). Jackknife-after-bootstrap standard errors and inﬂuence functions. Journal of Royal Statistical Society. Series B (Methodological), 54, 83–127. Gomes-Casseres, B., Hagedoorn, J., & Jaffe, A. (2006). Do alliances promote knowledge ﬂows? Journal of Financial Economics, 80, 5–33. Goodman, L. (1984). The analysis of cross-classiﬁed data having ordered categories. Cambridge, MA: Harvard University Press. Gould, R. (2002). The origins of status hierarchies: A formal theory and empirical test. American Journal of Sociology, 107, 1143–1178. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91(3), 481–510. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380. Granovetter, M. S. (1992). Problems of explanation in economic sociology. In: N. Nohria & R. G. Eccles (Eds), Networks and Organizations (pp. 25–56). Cambridge, MA: Harvard Business School Press. Gulati, R. (1998). Alliances and networks. Strategic Management Journal, 19, 293–317. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 177–231. Hannan, M., & Freeman, J. (1977). The population ecology of organizations. American Journal of Sociology, 82, 29–964. Hansen, M. T. (1999). The search-transfer problem: The role of weak ties in sharing knowledge across organization subteams. Administrative Science Quarterly, 44, 82–111.

312

RAY REAGANS AND BILL MCEVILY

Hansen, M. T., Mors, M. L., & Løvas, B. (2005). Knowledge sharing in organizations: Multiple networks, multiple phases. Academy of Management Journal, 48(5), 776–793. Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley. Ingram, P., & Roberts, P. (2000). Friendship among competitors in the Sydney Hotel Industry. American Journal of Sociology, 106(2), 387–423. Krackhardt, D. (1992). The strength of strong ties: The importance of philos in organizations. In: N. Nohria & R. G. Eccles (Eds), Networks and Organizations (pp. 216–239). Cambridge, MA: Harvard Business School Press. Levin, D. Z., & Cross, R. (2004). The strength of weak ties you can trust: The mediating role of trust in effective knowledge transfer. Management Science, 50(11), 1477–1490. March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), 71–87. Marsden, P. V. (1987). Core discussion networks of Americans. American Sociological Review, 52, 122–131. Marsden, P. V., & Campbell, K. E. (1984). Measuring tie strength. Social Forces, 63, 482–501. McEvily, B., Perrone, V., & Zaheer, A. (2003). Trust as an organizing principle. Organization Science, 14(1), 91–103. McGregor, D. (1960). The human side of enterprise. New York: McGraw-Hill. Mizruchi, M. S. (1989). Similarity of political behavior among large American corporations. American Journal of Sociology, 95, 401–424. Mizruchi, M. S., & Koenig, T. (1988). Economic concentration and corporate political behavior: A cross-industry comparison. Social Science Research, 17, 287–305. Moody, J. (2004). The structure of a social science collaboration network: Disciplinary cohesion from 1963 to 1999. American Sociological Review, 69, 213–238. Nelson, R. R., & Winter, S. G. (1973). Toward an evolutionary theory of economic capabilities. American Economic Review, 63, 440–449. Obstfeld, D. (2005). Social networks, the tertius iungens orientation, and involvement in innovation. Administrative Science Quarterly, 50, 100–130. Peng, Y. (2004). Kinship networks and entrepreneurs in China’s transitional economy. American Journal of Sociology, 109, 1045–1074. Podolny, J. (2001). Networks as pipes and prisms of the market. American Journal of Sociology, 107, 33–60. Reagans, R., Zuckerman, E. W., & McEvily, B. (2004). How to make the team: Social networks vs. demography as criteria for designing effective teams. Administrative Science Quarterly, 49, 101–133. Reagans, R. E., & McEvily, B. (2003). Network structure and knowledge transfer: The effects of cohesion and range. Administrative Science Quarterly, 48, 240–267. Reagans, R. E., & Zuckerman, E. W. (2001). Networks, diversity and performance: The social capital of R&D units. Organization Science, 12, 502–517. Rosenkopf, L., & Almeida, P. (2003). Overcoming local search through alliances and mobility. Management Science, 49, 751–766. Schroeder, L. D., Sjoquist, D. L., & Stephan, P. E. (1986). Understanding regression analysis: An introductory guide. Newbury Park, CA: Sage. Soda, G., Usai, A., & Zaheer, A. (2004). Network memory: The inﬂuence of past and current networks on performance. Academy of Management Journal, 47(6), 893–906. Sorenson, O., & Stuart, T. (2001). Syndication networks and the spatial distribution of venture capital investments. American Journal of Sociology, 106, 1546–1588.

Contradictory or Compatible?

313

Tortoriello, M., & Krackhardt, D. (2008). The role of strong ties, simmelian ties, and bridges in the generation of innovations. Working paper, H.E.C. Tsai, W. (2001). Knowledge transfer in intraorganizational networks: Effects of network position and absorptive capacity on business unit innovation and performance. Academy of Management Journal, 44(5), 996–1004. Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Uzzi, B., & Lancaster, R. (2004). Embeddedness and price formation in the corporate law market. American Sociological Review, 69, 319–344. Weber, R. A., & Camerer, C. F. (2003). Cultural conﬂict and merger failure: An experimental approach. Management Science, 49, 400–415.

INDUSTRY PERFORMANCE AND INDIRECT ACCESS TO STRUCTURAL HOLES Ronald S. Burt ABSTRACT What is the scope of brokerage network to be considered in thinking strategically? Given the value of bridging structural holes, is there value to being afﬁliated with people or organizations that bridge structural holes? The answer is ‘‘no’’ according to performance associations with manager networks, which raises a question about the consistency of network theory across micro to macro levels of analysis. The purpose here is to align manager evidence with corresponding macro evidence on the supplier and customer networks around four-digit manufacturing industries in the 1987 and 1992 benchmark input–output tables. In contrast to the manager evidence, about 24% of the industry-structure effect on industry performance can be attributed to structure beyond the industry’s own buying and selling, to networks around the industry’s suppliers and customers. However, the industry evidence is not qualitatively distinct from the manager evidence so much as it describes a more extreme business environment.

There is a lively literature on the advantages associated with networks that bridge the structural holes in social networks. Given a disconnect between Network Strategy Advances in Strategic Management, Volume 25, 315–360 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25009-6

315

316

RONALD S. BURT

two groups – a gap in the ﬂow of information between the groups – the groups can be expected to develop their own language, beliefs and business practice, such that information becomes sticky within the groups, creating potential advantage to a network that coordinates across the groups. Diverse evidence shows higher performance in networks that bridge structural holes. I will present illustrative evidence in a moment. Given the accumulating evidence, what are the implications for strategic action intended to improve performance? Business magazines publish practical guidance on bridging structural holes (e.g., Hargadon & Sutton, 2000; Maletz & Nohria, 2001; Uzzi & Dunlap, 2005). In the natural evolution of academic work, research has matured from questions about the average value of bridging structural holes into questions about contingent value. For example, returns to networks rich in structural holes increase from negligible to substantial from junior to senior job rank, as work becomes more ambiguous and political (Burt, 1997, 2004, p. 371, 2005, pp. 156–162). In this volume, Venkatraman, Lee, and Iyer (2008) show that revenue growth is greater for software ﬁrms in alliance networks rich in structural holes (software ﬁrm is allied with companies not allied with one another), but particularly if the ﬁrm has a broad mix of products in a broad mix of markets. Also in this volume, van Liere, Koppius, and Vervest (2008) report on a series of inventive laboratory experiments with middle managers and M.B.A. students that show how building a network rich in structural holes is contingent on a subject’s ‘network horizon’ (see van Liere, 2007, for more detail and corroborating evidence). Subjects able to see more of the forming and dissolving connections among others in the business simulation are more successful in building their own rewarding network of relations bridging structural holes. This chapter too is a study of contingent returns to bridging structural holes, on a question also concerning scope: What is the scope of the brokerage network to be considered in thinking strategically? There are my contacts, their contacts, and their contacts’ contacts. How far out should strategic thinking extend? The answer with respect to manager networks turns out to be attractively simple: You only need to strategize about your immediate contacts (Burt, 2007). The answer is simple, greatly simpliﬁes the study of strategic behavior in networks, and is surprisingly robust, but it raises a question about the consistency of network theory across micro to macro levels of analysis. My goal in this chapter is to re-establish consistency, using analogous evidence on industry networks. I begin with illustrative evidence on manager networks, to establish a baseline and to explain why direct and indirect access to structural holes can

Industry Performance and Indirect Access to Structural Holes

317

be an advantage. Direct access refers to structural holes in the immediate network of a manager’s colleagues, or an industry’s suppliers and customers. Indirect access refers to structural holes between friends of friends, in the networks around colleagues, or around suppliers and customers. We know there are returns to direct access, in fact very similar returns at micro and macro levels of analysis. If there is advantage to afﬁliation with the well-connected, there should also be returns to indirect access. However, the returns are negligible in manager networks. Second, I describe the analogous industry network model, introducing the industry data (two years of benchmark performance and network data on detailed American manufacturing industries), and highlighting complementarities between the manager and industry evidence (consistency across levels of analysis, greater variety in manager networks, less endogeneity in the industry networks). Third, I present the evidence on industry performance and indirect access to structural holes.

MANAGER ADVANTAGE AND ACCESS TO STRUCTURAL HOLES A cluster of related network concepts emerged in the 1970s developing the general idea that there is advantage in having connections to multiple, otherwise disconnected, groups and individuals. At the center of the concept cluster are Granovetter (1973, 1983) on weak ties as bridges between groups, Freeman (1977, 1979) on network centrality as a function of being between contacts, (Cook & Emerson, 1978; Cook, Emerson, Gillmore, & Yamagishi, 1983) on the advantage of having alternative exchange partners, Burt (1980, 1983) on the advantage of disconnected contacts, later discussed as access to structural holes (Burt, 1992, 2005), and Lin, Ensel, and Vaughn (1981) on the advantage of distant, prestigious contacts, later elaborated in terms of having contacts in diverse status groups (Lin, 2002). Two facts – from a stream of research beginning around World War II on inﬂuence and social networks (e.g., Festinger, Schachter, & Back 1950; Lazarsfeld, Berelson, & Gaudet 1944) – provided foundation for the network concepts: (1) People are clustered into groups by factors (later discussed as social foci, Feld, 1981) deﬁned by the places where people meet; the neighborhoods in which they live, the organizations with which they afﬁliate, the ofﬁces where they work, the projects in which they are involved. (2) Communication, and thus socialization, is more frequent within than between these groups such that

318

RONALD S. BURT

people in the same group develop similar views of the history that led to today, similar views of proper opinion and practice, and similar views of how to move into the future. People tire of repeating arguments and stories explaining why they believe and behave the way they do. They make up short-hand phrases to reference whole paragraphs of text with which colleagues are familiar. Jargon ﬂourishes. What was once explicit knowledge interpretable by anyone becomes tacit knowledge meaningful only to insiders. With continued time together, the group deepens its tacit knowledge as new combinations and nuances emerge. Much of what we know is not readily understood beyond the colleagues around us. Inside the tribe, one only needs to say the punch line of a popular joke to elicit bonding recollection of the whole story. Explicit knowledge converted into local, tacit knowledge makes information sticky (von Hippel, 1994) such that holes tear open in the ﬂow of information between groups. These holes in the social structure of communication, or more simply ‘‘structural holes,’’ are missing relationships that inhibit information ﬂow (‘‘like an insulator in an electric circuit,’’ Burt, 1992, p. 18).

Direct Access to Structural Holes The network image of groups separated by structural holes, with the image’s implications for sticky information within groups and heterogeneity more likely between than within groups, is foundation for network models of competitive advantage. Structural holes are a source of efﬁciency at the same time that they are a source of growth. As a source of efﬁciency, structural holes are boundary markers in the division of labor. By not having to attend to the interpretations of people beyond the boundary around my specialty, I can focus on deepening my knowledge of what I already know pretty well. Without structural holes, we would be overwhelmed with the diversity of knowledge out there. I expect that we would quickly establish structural holes to re-establish a sense of control over our lives. Structural holes are simultaneously a source of growth from the hardy souls among us who reach out to broker connections across the holes to create new combinations of existing opinion and practice (see Burt, 2005; Chapters 1 and 2, for review). Brokerage opportunities are measured in terms of the structural holes between contacts. When contacts are all connected with one another, there are no structural holes to broker. The more disconnected a manager’s contacts, the more likely her network spans holes in the surrounding organization and market. People who connect

Industry Performance and Indirect Access to Structural Holes

319

across structural holes – call them network brokers, connectors, or entrepreneurs – have a vision advantage in detecting and developing productive opportunities. As described in the previous paragraph, people disconnected from one another often employ different problem-solving and practices in their work. Because network brokers are more exposed to the diversity of these diverse opinions and practices, brokers have a vision advantage in selecting early between alternative ways to go, synthesizing new ways to go, framing a proposal to be attractive to needed supporters, and detecting likely supporters/opponents to implementing a proposed way to go. For reasons of information breadth, timing, and arbitrage, people with strong relations to otherwise disconnected groups have a competitive advantage in detecting and developing productive opportunities. The advantage expected from manager access to structural holes is manifest in standard performance metrics. The graph in Fig. 1 contains illustrative results pooled across ﬁve populations of managers listed in the graph (see Burt, 2005, p. 56, for a similar graph based on eight study populations): human resource managers in a commercial bank (Burt, Jannotta, & Mahoney 1998), investment analysts and bankers in a ﬁnancial services organization (Burt, 2007), managers in the Asia-Paciﬁc launch of a new software product (Burt, 2008), and supply-chain managers in a large electronics company (Burt, 2004, 2007). Various performance metrics were obtained from company archives on each manager as described in the research cited in the previous sentence (job evaluations, compensation, recognition in external professional awards), then regressed across job rank, job function, education, seniority, geographic location, and other background variables obtained from company archives to remove performance variance associated with the background variables. Take the prediction residual scores, standardize them, and you have a measure of performance relative to peers – which is the vertical axis in Fig. 1. A score of zero means that a manager is performing at a level expected for someone with his or her background. A score of zero in the population of supply-chain managers, for example, means that you received compensation and evaluation typical for someone at your job rank, with your seniority, in your area, and with your background. A score of one means that you are one standard deviation ahead of what is typical for people like you, and so on. The horizontal axis in Fig. 1 measures manager access to structural holes. The measure, network constraint, is an index of the extent to which a manager’s time and energy are concentrated in a single group of interconnected colleagues – which means no access to structural holes. As described in the research cited above for each of the ﬁve study populations

Fig. 1.

Performance and Direct Access to Structural Holes.

320 RONALD S. BURT

Industry Performance and Indirect Access to Structural Holes

321

in Fig. 1, the discussion network around each manager was constructed such that the following measure of network constraint could be computed (equations are presented here for later analogy to corresponding industry measures). X Ci ¼ w ; iaj (1) j ij where Ci is network constraint on manager i, and wij is a measure of i’s dependence on colleague j. X p p Þ2 ; iaqaj, (2) wij ¼ ðpij þ q iq jq where pij is the proportion of manager i’s network time and energy spent on colleague j, so dependence weight wij varies from 0 to 1 with theP extent to which i’s network time and energy is directly (pij) or indirectly ( q piqpqj) spent on colleague j. Network constraint, as the sum of dependence weights, measures the extent to which the manager’s network of colleagues is like a straightjacket around the manager, limiting his or her vision of alternative ideas and sources of support. Ideographs below the horizontal axis in Fig. 1 illustrate colleague networks posing low constraint (to the left) and high constraint (to the right). The low-constraint network has the manager at the center of the network working with disconnected colleagues. Disconnections, holes, between the manager’s colleagues provide opportunities to broker connections. The high-constraint network to the right has the manager working with connected colleagues. There are no opportunities for brokerage. I multiply the constraint scores by 100 to discuss points of constraint. Networks around managers in the ﬁve populations varied widely, from two points of constraint up to 100 points, around a mean of 33 points. The graph in Fig. 1 illustrates an empirical result that has become familiar over the last 15 years: managers with access to structural holes have an advantage in detecting and developing productive opportunities. There is a strong association between performance and network constraint in each population (t-tests of 4.4 to 7.3), and the regression line in the graph shows performance decreasing as a manager’s colleagues become more interconnected.

Indirect Access to Structural Holes Managers also vary in their indirect access to structural holes. Around each of a manager’s contacts is a network of the contact’s contacts.

322

RONALD S. BURT

Direct contacts are the people with whom a manager has personal contact. Indirect contacts are friends of friends reached through direct contacts as intermediaries. Given the value of direct access to structural holes through contacts in diverse groups, it is reasonable to expect that value is enhanced if those contacts themselves are connected to diverse groups. Networks are jointly owned by the people in them (not equally, but jointly), so it is not difﬁcult to imagine that advantage spills over between adjacent networks such that it is an advantage to be afﬁliated with well-connected people. For one thing, well-connected colleagues can be a source of opportunity and resources. If you discover an opportunity for which you do not have the time or energy, you pass it on to a friend. In the course of working with a colleague you learn about new gossip and ideas of interest to the colleague. Colleagues are also a signal. Well-connected colleagues signal to observers that you have standing among the right people. These commonsense expectations are nicely illustrated by a pair of quotes that Rowley and Baum (2004, p. 122) cite from their interviews with investment bankers: ‘‘information and access to it are king y being close to the source is the name of the game. y I don’t have time to know everyone, but I need to be close to those that have the best contacts.’’ ‘‘The best players in the industry build reputations by getting the biggest clients and controlling information, and carefully passing it out to others. It makes you a hot commodity, like a hot concert ticket or restaurant – everybody wants some.’’ Common sense has a formal analogue. The imagery of advantage spilling over between adjacent networks is foundation for the idea of ‘‘increasing returns to networks’’ and ‘‘Metcalf ’s Law’’ in which the value of a network increases with the square of the people in it. As Spence (2002, p. 453) referenced the imagery in his lecture on the occasion of receiving a Nobel Prize for his work on information in markets: ‘‘Metcalfe’s law states that the value of a network to the entities attached to it is proportional to the square of the number of connected entities. In economic terms this probably means that the value and hence the speed of connecting accelerates as the number increase. This is sometimes referred to as the network effect.’’ Indirect access to structural holes turns out to provide none of the advantage associated with direct access. The evidence is documented elsewhere (Burt, 2007, 2008), but illustrated in Fig. 2 using the ﬁve manager populations pooled in Fig. 1. Indirect access to structural holes is measured on the horizontal axis in Fig. 2. The network around each of a manager’s direct contacts poses a level

323

Fig. 2.

Performance and Indirect Access to Structural Holes.

Industry Performance and Indirect Access to Structural Holes

324

RONALD S. BURT

of constraint and opportunity, on the contact directly and on the manager indirectly through the contact. Let ICi be network constraint scores Cj pooled across the manager i’s contacts j.1 Where the network constraint index C in Eq. (1) measures the lack of structural holes in a manager’s immediate network of direct contacts, IC measures the lack of structural holes in the networks around the manager’s direct contacts. There is low indirect constraint on a manager connected to brokers (low Cj scores average to a low ICi score). A manager subject to low indirect constraint is connected to colleagues whose networks are rich in brokerage opportunities. Through those colleagues, the manager has indirect access to structural holes. There is a strong performance correlation with indirect access to structural holes. In the graph to the left in Fig. 2, performance is standardized within population and year. There are no controls for job rank or background variables. Consider the population of investment bankers pooled with other populations in Fig. 1. Banker performance was measured by annual salary and bonus compensation. When I regress z-score compensation (ZP) across indirect network constraint, I get the following result (standard error in parentheses). ZP ¼ :93 :45 lnðICÞ. ð:13Þ The statistically signiﬁcant negative association (–3.46 t-test) shows that bankers afﬁliated with colleagues in small, dense networks tend to receive compensation below their peers. When the bankers look around the ofﬁce, they see that peers doing well are afﬁliated with well-connected colleagues (well-connected in the sense of having low-constraint networks rich in brokerage opportunities). The graph to the left in Fig. 2 shows the result across the ﬁve manager populations. There is a strong correlation between manager performance and indirect network constraint (–7.66 t-test). In fact, the nonlinear, downward-sloping association between performance and indirect network constraint to the left in Fig. 2 looks very similar to the association in Fig. 1 with direct network constraint. The performance correlation with indirect access is spurious. Wellconnected colleagues have their own interests. Why should they sustain a connection with you if you are not attractive in your own right? When I hold constant manager job rank and direct network constraint – measures of manager attractiveness as a productive contact – the association between performance and indirect network constraint disappears. Again using the bankers for a speciﬁc example, compensation does not vary with the

Industry Performance and Indirect Access to Structural Holes

325

networks around a banker’s colleagues so much as it varies with the banker’s own network. ZP ¼ 1:30

:47 lnðCÞ :14 lnðICÞ: ð:13Þ

ð:14Þ

Compensation is strongly associated with a banker’s own network (3.62 t-test). It is not associated with the networks around his or her colleagues (1.00 t-test). The graph to the right in Fig. 2 illustrates the point across the ﬁve manager populations. Residual performance in the graph is the same as residual performance in Fig. 1 except I have also held constant the level of network constraint in the manager’s immediate network (horizontal axis in Fig. 1). There is no performance association with indirect access to structural holes once a manager’s attractiveness is held constant (1.26 t-test). The lack of returns led me to discuss the brokerage opportunities of indirect access as ‘‘secondhand’’ brokerage (Burt, 2007), to distinguish it sharply from the performance-related brokerage opportunities of direct access.

Network Brokerage a Forcing Function for Human Capital? These results emphasize the importance of agency in networks. People who do not build their own brokerage networks do not show the beneﬁts of brokerage. It is not enough to afﬁliate with known brokers. But there should be returns to secondhand brokerage if brokerage creates advantage by providing quick, early access to distant, novel information. Consistently negligible returns to secondhand brokerage in diverse populations lead me to conclude that the advantage of network brokerage is not about quick, early access to distant, novel information so much as it is about what happens to a person who has to manage communication across a network full of structural holes. Either way, ego has a vision advantage in detecting and developing rewarding opportunities. The question is whether the vision advantage comes from better glasses or better eyes. A network that spans structural holes could provide a manager with better information access and control, which would be an advantage, or it could, by exercising one’s ability to manage heterogeneous information, make the managers better able than less ‘‘exercised’’ peers to see opportunities, which would amount to the same advantage. Brokerage exposes ego to diverse opinion and practice in other groups. In the course of managing contradictory relationships, ego develops

326

RONALD S. BURT

cognitive skills of analogy and synthesis, and emotional skills for reading, engaging, and motivating colleagues. One is perhaps less troubled by sharp differences in opinion or practice. One becomes, perhaps, more skilled in analogy and metaphor in order to communicate across differences. Whatever speciﬁc skills are involved, brokerage is not valuable for the information it provides so much as it is valuable as a forcing function for cognitive and emotional skills required to manage communication between colleagues with divergent belief and practice. It is the cognitive and emotional skills produced as by-product in managing brokerage networks that are the proximate source of competitive advantage. In a phrase, brokerage could be a forcing function for human capital (the theme in Coleman’s, 1988, initial description of social capital). The case is made and discussed in detail elsewhere (Burt, 2008), but the above results are sufﬁcient illustration for the purposes of this chapter.

MICRO–MACRO CONNECTION DISRUPTED A central tenet in network theory, if not the central tenet, is that causal spark is released by the pattern in which relations intersect. Something about the pattern of relations intersecting in a network node encourages, facilitates, or inhibits. Speciﬁc models focus on the spark released by a speciﬁc pattern. Whatever the causal spark, it is expected from the relational pattern regardless of where the pattern occurs; in a person, a team, an organization, a geographic region. For example, the network status model that Podolny (1993) uses to explain why certain investment banks are able to obtain capital at more attractive rates is the same eigenvector model used by Kadushin (1995) to describe the status of individuals in the French ﬁnancial elite. The network brokerage model that Burt (1992, Chapter 3) uses to explain why proﬁt margins are high in certain markets is the same model used in the subsequent chapter of the same book to explain why certain managers are promoted more quickly to senior job rank in a large organization. The network brokerage model that Freeman (1977) uses to explain why certain people are more satisﬁed in a laboratory task is the same model used by Owen-Smith and Powell (2004) to explain why certain companies are more likely to ﬁle successful patent applications. The simple embedding model used to describe mutual contacts increasing the persistence of relationships and reputations (e.g., Burt, 2005, Chapter 4; Feld, 1997; Krackhardt, 1998) is the same model that Gulati and Gargiulo (1999) use to describe the higher odds of repeated alliances between ﬁrms

Industry Performance and Indirect Access to Structural Holes

327

with mutual alliance partners, that Ingram and Roberts (2000) use to describe mutual friendships enhancing the survival of hotels, that Rowley, Greve, Rao, Baum, and Shipilov (2005) use to describe mutual contacts lowering the probability of exit from investment-bank cliques, and that Løva˚s & Sorenson (2008) in this volume use to describe mutual contacts decreasing the risk otherwise associated with sharing scarce resources between consultants. Consistent network theory across levels of analysis is attractive because the consistency is a bridge for analogies between otherwise disparate research results, which is all the more powerful because disparate research results are likely to have complementary strengths if the results can be compared in a meaningful way. As illustrated by the examples cited in the previous paragraph, network explanations for performance differences between people can be used to draw inferences about performance differences between macro units of analysis such as organizations or industries or regions – just as network explanations for macro performance differences can be used to draw inferences about performance differences between people. I will be more speciﬁc about network brokerage models in the next section. For the moment, I can say that the integrative and crossfertilizing potential of network theory consistent across levels of analysis has contributed in some part to the widespread use of network models in studies of competitive advantage. Now the problem: The fact that mangers do not beneﬁt from indirect access to structural holes raises a question about consistency across levels of analysis. The role of cognition and emotion in network brokerage makes sense for sentient individuals. It is not obvious how the image of sentient individuals applies at the macro level. Organizations, and the industries and regions in which they operate, are assemblies of people who individually think and feel. To attribute thinking and feeling to macro units such as organizations, industries, or regions, requires an unattractively anthropomorphic metaphor. To continue the ‘‘better glasses or better eyes’’ metaphor in the discussion of network brokerage as a forcing function for human capital, the ‘‘better glasses’’ explanation generalizes readily to the macro level of organizations and markets. The ‘‘better eyes’’ story, with its emphasis on enhanced cognitive and emotional skills, does not. It would be useful to see macro-level evidence on performance and indirect access to structural holes. I begin with the macro advantages of direct access to illustrate what has been a consistent micro–macro connection for network models of brokerage. I then turn to performance and indirect access.

328

RONALD S. BURT

CORPORATE ADVANTAGE AND DIRECT ACCESS TO STRUCTURAL HOLES As manager networks rich in structural holes provide an advantage in detecting and developing opportunities by exposing managers to diverse business opinion and practice, advantage comes twice at the macro level to producer organizations with hole-rich networks of suppliers and customers. (1) Within supplier and customer industries, structural holes mean more likely variation in business practice so there is something for producers to learn from the industry (groups separated by structural holes are more likely to evolve on separate paths), and competing organizations within the industry mean that producers can play them against one another to negotiate attractive prices. (2) Between supplier and customer industries, structural holes mean that large ﬁrms are unlikely to have integrated operations across the industries, so the producer advantages of withinindustry holes occur between industries: more likely exposure to variation in business practice, and more likely independent competitors that can be played against one another. Using gross proﬁt margins to measure performance, producer margins on average should increase with the structural holes in their immediate network of suppliers and customers. This was the initial intuition for returns to network brokerage at the macro level, modeled as structural autonomy, here stated in a multiplicative form (Burt, 1980, 1983, 1992, Chap. 3; Burt, Guilarte, Raider, & Yasuda, 2002). A ¼ aðk OÞb C g ,

(3)

where A is producer structural autonomy, an advantage provided by an industry’s network position in the economy, a is an intercept term, O is a measure of producer coordination within an industry, k is a constant just above the upper limit of O so (k-O) measures the lack of coordination between industry producers, b measures the corrosive effect of disorganized producers, C is a network constraint measure of producer dependence on well-organized suppliers and customers, and g measures the corrosive effect of organized suppliers or customers. Network constraint at the macro level of an industry is deﬁned by the dependence weights that deﬁne network constraint at the micro level on individual people (wij in Eq. 2), but there is now a question of whether supplier and customer establishments are organized to exploit producer dependence on them. Producer dependence on another industry is not a problem if businesses in the other industry can be played against one another. Dependence is a problem when there are few

Industry Performance and Indirect Access to Structural Holes

329

alternatives within a key supplier or customer industry. Network constraint on industry i is a weighted sum of dependence on supplier–customer industries j in which business is concentrated in a few dominant companies: X Ci ¼ w O ; iaj (4) j ij j where Oj is the coordination of businesses in market j, measured as it is measured for the producer market in Eq. (3). The product wijOj in Eq. (4) is a network measure of the condition that Pfeffer and Salancik (1978, p. 51) so productively explored as resource dependence: ‘‘Dependence can then be deﬁned as the product of the importance of a given input or output to the organization and the extent to which it is controlled by a relative few organizations.’’ The network constraint index in Eq. (4) is the sum of such dependencies, measuring the aggregate extent to which producers are dependent on coordinated suppliers or customers. With respect to Porter’s (1980, p. 4) inﬂuential ﬁve-forces metaphor – grounded in the economics of industrial organization (e.g., Caves, 1992) and a close relative in time and content to Pfeffer and Salancik’s resource dependence metaphor – b measures the negative effect on industry proﬁts from producer ‘‘rivalry’’ within the industry and g measures negative effects from ‘‘supplier power’’ and ‘‘buyer power.’’ In sum, Eq. (3) is a baseline industry network model for which estimates of b and g should be negative. The estimates have been signiﬁcantly negative in the American economy since the 1960s and in other economies where estimates are available (Burt et al., 2002). The results merely express empirically the old idea that monopolists do well exploiting disorganized partners. The optimum industry network for proﬁts combines coordination inside the industry with brokerage outside the industry (high O combined with low C in Eq. (3); see Burt, 2005, pp. 139–146 for a surface plot of industry data, c.f. Baum, van Liere, & Rowley, 2007, Fig. 3a, for a similar graph describing company data within an industry).

Network Data on Industry Dependencies (wij) Much of the data needed to estimate effects in Eq. (3) can be obtained at high quality in the U. S. Department of Commerce benchmark input– output tables. Each table is a network of dollar ﬂows between sectors of economic activity: cell (i,j) is dollars of goods purchased by organizations in sector j from organizations in sector i. In theory, organizations assigned to the same input–output sector, or industry, draw supplies in similar

330

RONALD S. BURT 2.0

High-Concentration Industries [n = 164, ZPCM = 2.01 - .63 ln(C)]

(PCM standardized within year)

Z-Score Industry Performance

1.5

1.0

0.5

0.0

-0.5

Low-Concentration Industries [n = 156, ZPCM = .13 - .14 ln(C)]

Vertical axis indicates relative performance and horizontal indicates network constraint. Graph to left shows how price-cost margins in American manufacturing industries change with increasing network constraint on producers from coordinated suppliers and customers. Graph below shows how performance metrics for managers in Figure 1 change with increasing connections among a manager’s key contacts. Thin lines describe returns when peer competition is intense (low concentration, many peer managers). Bold lines describe returns when peer competition is less intense (high concentration, few peer managers).

-1.0 1

6

11 16 21 26 31 36 41 46

2.0

1.5 (evaluation, compensation, promotion)

Z-Score Manager Residual Performance

t-test

People in senior job ranks [n = 440, Z = 1.94 - .49 ln(C)]

1.0

Population

-6.2

Human Resource Managers

-4.4

Investment Analysts

-7.3

Investment Bankers

-4.5

Product Launch Managers

-7.4

Supply Chain Managers

0.5

0.0

-0.5

People in lower job ranks [n = 1044, Z = .55 - .23 ln(C)]

-1.0 1

6

11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96

Network Constraint (C) many ————— Structural Holes ————— few

Fig. 3.

Baseline: Micro-Macro Connection for Direct Access to Structural Holes.

proportions from the same supplier industries and sell product in similar proportions to the same customer industries. Thus, an input–output table is a summary network, like a density table, describing patterns of buying and selling between structurally equivalent organizations (Burt, 1988; Burt &

Industry Performance and Indirect Access to Structural Holes

331

Carleton, 1989), and an input–output industry composed of structurally equivalent organizations corresponds to a market in that the industry contains organizations competing for the same supplier and customer business (Burt, 1992, pp. 208–215). Regional markets, government regulations, business practice, and data limitations must create data deviations from theory, but the industry concept remains in theory a concept of industry organizations using similar processes, to produce similar goods, available to customers according to customer input requirements. Treating the input–output dollar ﬂows as cells in a network density table, the dependence weight wij in Eq. (2) can be computed with pij deﬁned as the proportion of industry i buying and selling across industries that is conducted with establishments in industry j,2 ðzij þ zji Þ P pij ¼ P ð k zik þ k zki zii Þ

(5)

where zij is dollars of sales from industry i to j in the input–output table, and k ranges across all product categories in the table (i.e., everything excluding government and ﬁnal demand). Weight wij varies from 0 to 1 with P the extent to which producer buying and selling is directly (pij) or indirectly ( q piqpqj) with establishments in market j (see Burt, 1992, pp. 54–62, for other speciﬁcations and connections with laboratory results on exchange networks). In this chapter, I estimate effects for detailed manufacturing industries in 1987 and 1992. I use the most detailed input–output categories to preserve the highest level of structural equivalence available between producers treated as competitors in the same industry. I focus on the years 1987 and 1992 for consistent, reliable sector deﬁnitions. The U.S. Department of Commerce expanded distinctions between service sectors in the 1987 benchmark input–output table, then shifted from Standard Industrial Classiﬁcation (SIC) categories to the North American Industry Classiﬁcation System (NAICS) for 1997 and later benchmark input–output tables (Lawson, Bersani, Fahim-Nader, & Guo, 2002). Sector deﬁnitions in the 1987 and 1992 panels are similarly expanded from earlier benchmark tables, but still based on SIC categories familiar to the operations people at the Department of Commerce before they changed over to the substantially different NAICS categories. Dollar ﬂows between industries can be downloaded from the U.S. Department of Commerce, Bureau of Economic Analysis website (www.bea.gov/industry/io_benchmark.htm). Excluding government and ﬁnal demand, the 1987 benchmark input–output table

332

RONALD S. BURT

distinguishes 469 production sectors, of which 362 are manufacturing (Lawson & Teske, 1994). Respective numbers for the 1992 table are 485 and 361 (Lawson, 1997). There is almost no difference between manufacturing industries in the two tables. The one difference is that chewing gum and a portion of candy manufacturing are separate sectors in 1987, but combined in 1992 (sectors 142001 and 142003 in 1987 are combined as sector 142005 in 1992). For consistency across the tables, I combined the two 1987 candy categories to correspond to their combined category in the 1992 table. Thus, I have 361 manufacturing industries in 1987 and 1992. Each industry is subject to some level of network constraint in its buying and selling with suppliers and customers in the other 402 industries. Producers in manufacturing industry i are dependent on industry j, wij, as deﬁned in Eq. (2), with proportional buying and selling deﬁned in Eq. (5). Producer dependence is combined with the data on organization within other industries to compute measures of direct and indirect network constraint. To compute the network constraint scores for an industry, I need a measure of coordination within each of the 402 potential supplier or customer industries.

Industry Concentration (O) I follow standard practice in using market shares to measure the extent to which producers are coordinated within an industry. The four-ﬁrm concentration ratio for an industry varies from 0 to 100 as the percent of industry output that comes from the four ﬁrms producing the largest volumes of industry output. Higher concentration is presumed to indicate more coordination, less rivalry, so producers can price for higher proﬁt margins. The four-ﬁrm concentration ratio of 91% in the 1987 ‘‘Tire and Cord Fabric’’ industry indicates that almost all industry output came from establishments operated by one of the four leading ﬁrms in the industry. In contrast, concentration in the 1987 ‘‘Sheet Metal Work’’ industry indicates that only 10% of industry output came from the four leading ﬁrms, so there must be numerous other competitors within the industry. Concentration ratios for manufacturing industries in 1987 and 1992 are available from the U.S. Census Bureau website for four-digit SIC categories (www.census.gov/epcd/www/concentration.html). The input–output tables are published with a list of SIC categories that map into each input-output category. Of the 361 manufacturing industries on which I have input–output data, 320 correspond to a unique four-digit SIC category. The other 41

Industry Performance and Indirect Access to Structural Holes

333

correspond to multiple four-digit SIC categories. For example, the input– output ‘‘Sugar’’ industry (141900) is composed of three four-digit SIC categories (2061 ‘‘Cane Sugar,’’ 2062 ‘‘Cane Sugar Reﬁning,’’ and 2063 ‘‘Beet Sugar’’). For the 41 manufacturing industries that correspond to multiple four-digit SIC categories, concentration is averaged across component SIC categories, weighting by the volume of business in each component category: Sk CRk*(Sk/[Sk Sk]), where CRk is the four-ﬁrm concentration ratio in component SIC category k, and Sk is dollars of sales by establishments in SIC category k. Buying and selling with 42 aggregate industries beyond manufacturing is included in the network measures. The industries are taken from a network analysis of boundaries between detailed input–output categories of agriculture, mining, construction, distribution, and services. The 42 nonmanufacturing industries are described in the Data Appendix and listed with concentration scores for 1987 and 1992. There are no authoritative concentration scores in these industries. Input–output tables provide dollar-ﬂow data beyond manufacturing, but there are no measures of producer organization comparable to the concentration data on manufacturing. Concentration in non-manufacturing can be estimated using data on the relative size of companies (e.g., Burt, 1992, pp. 89–91), but the practice is disconcerting because companies often operate in multiple industries and competition in non-manufacturing industries is often more local and regulated than competition in manufacturing industries (e.g., Burt et al., 2002). In the Data Appendix, I report tests with alternative approximations to concentration, but the approximations based on company size provide the clearest results. Effect estimates in this chapter are based on network constraint computed from size-based approximations to concentration in non-manufacturing. I now have a measure of concentration (O) in each of the 403 manufacturing and non-manufacturing industries in 1987 and 1992. I can compute network constraint in the baseline model (C in Eq. (4)) and measures to be presented of indirect constraint. I focus on predicting performance in certain industries because the concentration scores are not equally valid across industries. Scores in the 42 non-manufacturing industries are approximations correlated with the effective level of competition in non-manufacturing (Burt et al., 2002). The scores in which I have the most conﬁdence are those for the 320 manufacturing industries that correspond to a unique four-ﬁrm SIC category. These are the industries in which producer concentration is deﬁned by the same industry boundaries that deﬁne producers buying and selling

334

RONALD S. BURT

with suppliers and customers. Concentration scores in the other 41 manufacturing industries are an average of scores within segments of the industry so it is impossible to know the extent to which the four leading producers within industry segments account for total industry output. I compute network constraint scores for all 361 manufacturing industries and test for selection bias from my focus on the 320 that correspond to unique four-digit SIC categories. I obtain similar results for the 320 and the 361 industries. Effect estimates based on all 361 manufacturing industries differ slightly in metric, and are statistically stronger since they are based on 82 additional observations across the two panels, however, I focus where I have the most authoritative industry-structure data: the 320 industries for which transaction data and concentration data are deﬁned by the same industry boundaries.

Baseline Effects on Industry Performance (PCM) The input–output data provide a measure of industry performance. Pricecost margins (PCM) are a performance measure of net income to sales introduced by Collins and Preston (1969) and widely used in market structure research: PCM as originally computed from Census of Manufactures data equals net income (dollars of value added minus labor costs) divided by sales. Computed from input–output data, PCM equals net income (dollars of ‘‘other value added’’ plus indirect business taxes) divided by volume of business. The input–output data could be argued to provide a better measure of performance because more production and distribution costs such as advertising and entertainment are removed from value added, but the ﬁnal result is that the two data sources provide pricecost margins similarly associated with industry structure (Burt, 1988, pp. 372–378). The average price-cost margin is .162 across manufacturing industries in 1987 and 1992, showing a price-cost proﬁt of 16.2b on the average dollar of sales. As a concrete example, the 1987 input–output table shows $1,047.3 million in business by establishments in the ‘‘Tire Cord and Fabric’’ industry. Of that sum, $742.8 million were production and distribution costs, leaving $304.5 million in value added, of which $134.8 million was labor cost (input–output category 880000), $3.5 million went to indirect business taxes (category 890000), and $166.2 million was other value added not attributed to speciﬁc costs (category 900000). Removing labor costs from the value added, dividing by volume of business, and multiplying by

335

Industry Performance and Indirect Access to Structural Holes

100 yields a price-cost margin of 16.2b, the average across all manufacturing. The margin seems modest given the high 91% four-ﬁrm concentration in the industry, however, it is well known that industry margins have only a weak correlation with industry concentration (Schmalensee, 1989, pp. 973–976; Weiss, 1989). Relative industry performance in 1987 continued by and large into 1992, but margins were slightly higher on average in 1987, and nine industries operated at a loss in one or the other year. No industry operated at a loss in both years. Given that the nine negative price-cost margins are year speciﬁc (each is positive in the other panel), and would have disproportionate inﬂuence on estimated effects because they are at the extreme edge of the data distributions, I put the nine aside as intrusive outliers. This turns out not to affect conclusions about the statistical signiﬁcance of effects, but it does make effects stand out more clearly since the nine temporary outliers do not have to be ﬁt into the aggregate performance associations with industry structure. Detailed discussion is in the Data Appendix. As quick illustration, here are estimates for the baseline model (Eq. (3)) ﬁt across all 722 observations of the 361 manufacturing industries, including adjustment for the slightly higher margins in 1987.3 PCM ¼ 41:37

4:07 lnð100 OÞ 3:99 lnðCÞ ð1:48Þ ð:81Þ

þ2:45D87; ð:41Þ

where standard errors are given in parentheses (adjusted for autocorrelation across repeated observations with the ‘‘cluster’’ option in STATA). There is a statistically signiﬁcant –2.75 t-test for the negative effect of producer rivalry, and a –4.92 t-test for the negative effect of supplier–customer network constraint. Here are estimates for all 640 observations of the 320 industries that correspond to unique four-digit SIC categories. PCM ¼ 42:31

4:14 lnð100 OÞ 4:18 lnðCÞ ð1:52Þ ð:87Þ

þ2:51D87; ð:41Þ

which deﬁne t-tests of –2.70 and –4.81 respectively for producer rivalry and network constraint. And here are estimates for the baseline model ﬁt across the further subset of 632 observations in which price-cost margins were nonnegative. PCM ¼ 48:41

5:42 lnð100 OÞ 4:39 lnðCÞ

þ2:38D87;

ð1:41Þ

ð:41Þ

ð:80Þ

336

RONALD S. BURT

which deﬁne t-tests of 3.83 and 5.47 respectively for producer rivalry and network constraint. Three points are illustrated: First, the two industrystructure effects are, as expected, negative and statistically signiﬁcant. Second, estimates do not differ much between the equation estimated across all 361 manufacturing industries and the one estimated across the 320 manufacturing industries that correspond to unique four-digit SIC categories. Third, effects are more clear – stronger magnitudes and smaller standard errors – in the equation for which I put aside the nine negative price-cost margins as temporary outliers. Micro–Macro Connection The two graphs in Fig. 3 illustrate micro–macro consistency for performance as a function of direct access to structural holes. The graph at the top in Fig. 3 describes the industry performance-structure association. I standardized price-cost margins within years to have a measure of relative industry performance comparable to the z-score performance metrics on managers in Fig. 1. The z-score performance measure, ZPCM, is then a function of the two industry-structure variables, O and C, in the baseline model (Eq. (3)): ZPCM ¼ 3:41

:56 lnð100 OÞ :46 lnðCÞ ð:15Þ

ð:08Þ

where estimation is across the 632 non-negative margins in the 320 manufacturing industries that correspond to unique four-digit SIC categories, standard errors are given in parentheses, and the two network effects are clearly negative. Lines in the graph at the top in Fig. 3 show how z-score price-cost margins vary with decreasing brokerage opportunities among suppliers and customers. The bold line shows the negative effect of coordinated suppliers or customers on industries in which producer rivalry is low (concentration is in the top quartile of manufacturing). The thin line shows the negative effect on industries in which producer rivalry is high (concentration is in the bottom quartile). The graph illustrates two characteristics of the macro performancestructure association: First, the bold and thin lines both decrease, showing how producer margins are eroded by increasing dependence on supplier and customer industries in which rivalry is low. Second, producers in concentrated industries lose more. Dependence on coordinated suppliers and customers can erase the advantage of producer coordination. The bold

Industry Performance and Indirect Access to Structural Holes

337

line lies well above the thin line in the graph, showing the higher margins enjoyed by producers in concentrated industries. Where suppliers and customers are completely disorganized (far left in the graph), the difference between the bold and thin lines is almost two standard deviations (.13 z-score price-cost margin for the thin line, 2.01 for the bold line). The gap corresponds to 18b extra proﬁt on a dollar of sales.4 As producers become more dependent on supplier and customer industries in which rivalry is low (far right in the graph), the bold line decreases more quickly than the corresponding thin line, narrowing the gap between the lines (.42 for the thin line at the far right in the graph, versus .45 for the bold line, a difference that corresponds to a mere .3f proﬁt advantage to industries in which concentration is high). Similarity between the graphs in Fig. 3 illustrate network-effect consistency across levels of analysis. The network model of brokerage applied to markets is a bit more complicated than the model applied to managers, but it is the same model. The difference is that applications to managers usually assume that each manager is equally able to act in his or her own interest. Consider the implications of making that assumption about producers in markets. If it could be assumed that producers were equally coordinated within each market, then O would be a constant, so Eq. (4) would reduce to a sum of dependence weights as in Eq. (1), the producer organization term in Eq. (3) would be absorbed into the intercept and Eq. (3) would reduce to aCg (where a is the intercept in Eq. (3), a, plus an adjustment for constant O), which is the form of the log constraint predictions illustrated for managers in Fig. 1. In fact, managers are not equally able to act in their own interest. When the assumption of equal ability to act is relaxed, returns to manager brokerage resemble the returns to market producer brokerage. Ceteris paribus, managers doing a job in which they have many peers are less able to act in their own interest. Numerous peers increase competitive pressure on each manager. Jobs in which there are many peers are more subject to company processes. Individuals are less the author of their own jobs, more a reﬂection of company prescriptions. Returns to brokerage decrease as the number of peers increases (Burt, 2005, pp. 156–162). In the graph at the bottom of Fig. 3, I use job rank as a crude surrogate for number of peers and re-estimate the prediction in Fig. 1 for managers in senior job ranks separate from managers in lower job ranks. The bold line in the graph at the bottom of Fig. 3 describes for senior managers the rate at which performance erodes with decreasing access to structural holes. The thin line describes the same for managers in lower ranks.

338

RONALD S. BURT

The two points made about the industry graph at the top in Fig. 3 can be made equally well about the manager graph at the bottom in Fig. 3. In both graphs, the regression lines decrease showing the corrosive effect on performance of increasing network constraint. The bold line is higher than the thin line, showing the advantage of being a producer in a concentrated industry or a manager in a senior rank. Second, the advantaged lose more. The bold line decreases more quickly in both graphs as suppliers, customers, or colleagues, become more coordinated with one another. The micro and macro effects are also complements in their differences. For one thing, there is a difference in network variability. Managers are more varied in network constraint. Once a manager ﬁnds a secure niche in a large organization, he or she can work it to personal advantage. Unproductive managers are not removed from organizations with the same ruthlessness by which competition removes unproductive organizations from markets. The two graphs in Fig. 1 are drawn to scale. They are the same height, but the manager graph is wider. The industry graph is less wide because surviving industries rarely exist at the upper extremes of network constraint. Only 2% of industry observations used to estimate industry effects for Fig. 3 lie above 40 points of network constraint on the horizontal axis. Only 1% of the observations lie above 50 points. The managers exist in more varied circumstances. A third of the manager observations used to estimate network effects for Fig. 3 lie above 40 points of network constraint. A ﬁfth of the observations lie above 50 points, and many managers are embedded in completely closed networks, networks that pose 100 points of constraint. The industry data have their own strength: they provide a stronger foundation for claims that network structure affects performance. The stronger foundation is due to network data that are more authoritative, and network relations that are more exogenous to performance. With respect to more authoritative, the benchmark input–output tables deﬁning industry networks are based on a census of business establishments. Anyone who studies industry networks deﬁned by the tables begins with the same dollarﬂow relations. Results are directly comparable across research projects. Manager network data, in contrast, are always open to questions about how networks have been sampled and measured, and whether the measured relations are real or a reﬂection of passing interests. With respect to more exogenous, the dollar-ﬂow relations are not discretionary. They are deﬁned by production technology, which makes them more exogenous to performance than is usually the case in network analysis. Car producers, for example, can purchase steel from one or another company, but they must

Industry Performance and Indirect Access to Structural Holes

339

purchase steel somewhere. Producers are dependent on another industry to the extent that existing production technology has them transacting a large portion of their business with the other industry. In contrast, relations in manager networks are typically cited and maintained at the discretion of individuals. Who I select as my ‘‘friend’’ is my choice, as is naming ‘‘frequent’’ or ‘‘valued’’ contacts. Where I have discretion in selecting friends, I can select for reasons other than friendship, which creates an endogeneity problem: a relationship can appear, or be obscured, because the person naming contacts is reacting to performance. Whatever the performance advantage provided by access to structural holes, for example, there must be some effect in the opposite direction. People seek out successful colleagues. Successful people will attract relations from colleagues from other groups such that a network measured after a manager has achieved success is likely to span structural holes. Input–output relations are more exogenous to performance. The relations are deﬁned by production technology and performance results from how producers execute the technology. This is not to say that industry performance and production technology do not have mutual effects over time. Both evolve and are subject to exogenous shocks (e.g., McGahan, Argyres, & Baum, 2004). However, relative to the networks around managers, industry networks are more exogenous to performance. In short, what managers do not provide in authoritative network data as a research site, they provide in variety. What industries lack in variety, they provide in authoritative data. Industries and managers are together a more powerful platform for network studies of competitive advantage than either would be alone.

CORPORATE ADVANTAGE AND INDIRECT ACCESS TO STRUCTURAL HOLES That is, unless something disrupts the ability to draw research inferences between manager and industry networks, which is the central issue for this chapter: Advantage does not spill over between adjacent manager networks. Is the same true of industry networks? Expected Advantage: Maybe, Yes, and No A priori, the performance association with indirect access could be almost anything; negligible, positive, or negative. Argument can be made for each

340

RONALD S. BURT

of the three possibilities. Indirect access to structural holes in manager networks corresponds to industry networks: Organizations with which producers buy and sell deﬁne the producer industry’s direct suppliers and customers. Organizations with which those suppliers and customers do business are the industry’s indirect suppliers and customers. The effect on producers of structural holes among indirect suppliers and customers follows from the effect of holes among direct suppliers and customers. A priori, my prediction would have been a negligible association in industry networks because the association in manager networks is so obviously negligible. Given the similar micro and macro performance associations with direct access to structural holes (Fig. 3), and given no performance association with indirect access for managers (Fig. 2), my default prediction would have been to assume similar micro and macro associations with indirect access, and so predict a negligible industry performance association with indirect access. The storyline would be that supplier and customers advantage is irrelevant to producer advantage. All that matters is whether producers are in a position to beneﬁt from supplier or customer diversity and disunion. A person unaware of the manager results could be expected to predict a correlation between producer margins and supplier–customer advantage – for much the same reason that correlation with manager performance was expected before the results in Fig. 2 were known: Given the known advantage of direct access to structural holes, and the fact that networks are jointly owned (producers have nothing without customers and customers have nothing without suppliers), an advantage enjoyed by suppliers and customers must affect producer margins. The performance effect could be positive. We know that direct access to structural holes is an advantage. Producers with direct access to structural holes among suppliers and customers are more exposed to variation in business practice and have more opportunities to play competing organizations against one another. Extend the immediate network one step to predict the performance association with having suppliers and customers advantaged by direct access to structural holes. Advantaged industries are more likely to have budget to experiment with new business practice so producers with advantaged industries as suppliers and customers are more likely to see new business practice and alternative ways to implement the practice. Advantaged industries in this view would be hubs in the spread of new business practice (e.g., Davis, 1991) and the abandoning of old practice (e.g., Greve, 1995). There is a precedent for this possibility in Baum et al.’s (2007) analysis of U.K. investment banks predicting the value of a bank’s bond

Industry Performance and Indirect Access to Structural Holes

341

deals from bridges in the bank’s network and bridges in the networks around partners in the bank’s bond deals (where network ties are deﬁned by bank participation in the same syndicates). They report positive associations with the number of a bank’s own bridges and the number of bridges that its partners have. Beyond information and access, the morelikely slack resources available to advantaged suppliers and customers (illustrated in the graph at the top of Fig. 3), can make them more lucrative customers and suppliers. The summary story would be that advantaged suppliers and customers offer lucrative business opportunities and an enhanced portal into new business practices, so advantaged suppliers and customers have a positive association with producer performance. The performance effect could equally well be negative. The performance advantage of direct access to structural holes is anchored on the assumption that producers gain advantage from supplier and customer disadvantage. The corollary is that producers lose advantage when dealing with advantaged suppliers and customers. Laboratory experiments with exchange networks clearly show that people with multiple exchange opportunities exploit their partners who have few opportunities (Cook & Emerson, 1978; Cook et al., 1983). Outside the lab, Fernandez-Mateo (2007) reports disadvantage to contingency workers from continued afﬁliation with one placement ﬁrm that brokers access to jobs. Speciﬁcally, Bidwell and Fernandez-Mateo (2007) show that contingency workers receive a decreasing share of their earnings the longer they stay with the same placement ﬁrm. With respect to industry networks, the story would be that advantaged suppliers and customers extract a disproportionate share of proﬁt from their business, so advantaged suppliers and customers have a negative association with producer performance.

Tire Cord Industry To illustrate the arguments, consider the tire cord industry network displayed in Fig. 4. Fig. 4 is a sociogram of the network around the tire cord industry in 1987 (‘‘Tire Cord and Fabrics,’’ input–output industry 170700, SIC code 2296). Lines in Fig. 4 indicate volumes of business. Dots indicate industries. The tire cord industry is indicated by the square ‘‘dot’’ in the sociogram. The tire cord industry is a useful example because of its simplicity. There is one primary supplier and one primary customer. The bulk of tire cord supplies are purchased from the manmade ﬁbers industry (‘‘Manmade Organic Fibers,’’ input–output category 280400). The bulk of tire cord output is sold to tire manufacturers (‘‘Tires and Inner Tubes,’’

Fig. 4.

Network around Tire Cord and Fabrics Industry in 1987.

342 RONALD S. BURT

Industry Performance and Indirect Access to Structural Holes

343

input–output category 320100). The two primary supplier and customer relations are indicated by the two solid lines in Fig. 4. Together, the two relations account for 86.7% of tire cord buying and selling with other production industries (the pij deﬁned in Eq. (5), are given in Fig. 4 as 52.8% with tire manufacturers, 33.9% with manmade ﬁbers). I have further simpliﬁed Fig. 4 by presenting only relations that constitute more than 5% of an industry’s buying and selling (all pij greater than .05). I am using a broader 2% criterion to deﬁne suppliers and customers in the analysis, but a 5% criterion is better for the purposes of the example in Fig. 4. The lack of a solid line in Fig. 4 between tire manufacturers and manmade ﬁbers means that each does less than 5% of its business with the other. There is little more to report on the immediate network around tire cord producers. After the 33.9% of business with manmade ﬁbers, the next largest volume of tire cord business is 3.1% with advertising, followed by 1.7% with the local electric utility, followed by still smaller percentages spread across 44 other industries with many relations constituting less than .01% of tire cord business. In short, the tire cord industry is little more than a way station in the ﬂow of product from manmade ﬁbers to tire manufacturers. The immediate network helps explain why tire cord proﬁts are low despite the high level of industry concentration. The tire cord price-cost margin of 16.2b equals the average margin across all manufacturing industries, yet the concentration ratio of 91% is well above the 40% average in manufacturing (2.45 z-score). The price-cost margin should be higher in such a concentrated industry. However, the sociogram in Fig. 4 shows that concentration within the industry is counterbalanced by severe network constraint from suppliers and customers. Tire cord manufacturers are dependent on one primary supplier and one primary customer. The industries on which they are dependent are highly concentrated. Concentration is color coded in Fig. 4 as high (black), above average (grey), below average (light grey), and low (white) distinguished by the median and interquartile range of 1987 scores. The text box shows that concentration is high in the direct supplier and customer industries: 76% in manmade ﬁbers and 69% in tires and inner tubes. Dependence on concentrated supplier– customer industries deﬁnes a high level of direct network constraint on the industry (C equals 37 for tire cord and fabrics, well above the average of 15 for manufacturing, 2.36 z-score). Under strong pressure from suppliers and customers, tire cord proﬁts should be lower than would be otherwise expected from high concentration in the industry – as they are. But tire cord proﬁts are even lower than predicted by industry concentration and direct network constraint in the baseline model. The

344

RONALD S. BURT

text box in Fig. 4 shows a 23.0b price-cost margin predicted for the tire cord industry in 1987, which is well above the observed margin of 16.2b (z-score difference is .72).5 Explanation can be found in the broader network of indirect suppliers and customers. Dashed lines in Fig. 4 indicate buying and selling beyond the immediate network around tire cord producers. Network constraint computed within the immediate network around an industry – the solid lines in Fig. 4 – measures the extent to which industry producers have direct access to structural holes from which they could beneﬁt. Network constraint computed within the broader network of suppliers and customers to the industry’s direct suppliers and customers – the dashed lines in Fig. 4 – measures the extent to which industry producers have indirect access through their suppliers and customers to structural holes in the network structure around their suppliers and customers. In predicting tire cord proﬁts from the Eq. (3) baseline network model, I held constant supplier and customer concentration as a component in direct network constraint (C in Eq. (4)). However, the supplier and customer industries for tire cord producers have a further advantage: they are subject to low network constraint from their own networks of suppliers and customers. Fig. 4 shows that suppliers in the ‘‘manmade organic ﬁbers’’ industry do business with many supplier and customer industries, few of which are especially concentrated – so tire cord suppliers face much less direct network constraint than tire cord producers (C for manmade ﬁbers is 13 versus 37 for tire cords, a 2.52 z-score difference). The lower direct network constraint on suppliers means that they enjoy a higher proﬁt margin (PCM is 21.6b in manmade ﬁbers versus 16.2 in tire cord), which could affect on tire cord producers. Tire manufactures, the primary customer industry for tire cord, are similar subject to lower network constraint (C ¼ 13). In this case, having advantaged suppliers and customers seems to have a negative effect on tire cord margins. Advantaged suppliers and customers enjoy proﬁts at a level expected from direct access to structural holes (PCM hat is about the same as PCM in the text box) while producer proﬁts are well below expected.6

Returns to Indirect Access In contrast to the tire cord example, the aggregate effect is positive: producers derive advantage from business with advantaged suppliers and customers. Results with alternative measures are presented in Table 1.7

345

Industry Performance and Indirect Access to Structural Holes

Table 1.

Price-Cost Margins and Industry Network Structure. (1)

(2)

(3)

(4)

(5)

(6)

Rivalry within industry (1005.42 5.71 5.77 5.95 5.94 5.41 (1.41) (1.37) (1.37) (1.40) (1.37) (1.38) industry four-ﬁrm concentration ratio) Direct network constraint from 4.39 3.11 3.28 2.63 3.11 suppliers–customers (C in Eq. (.80) (.82) (.80) (.80) (1.02) (4)) Indirect network constraint from their suppliers–customers Unweighted average constraint on 6.81 5.09 industry suppliers–customers (1.30) (1.32) Weighted average constraint on 3.32 industry suppliers–customers (.95) Percent industry business with 1.48 low-constraint suppliers– (.69) customers Percent industry business with .74 high-constraint suppliers– (.37) customers Total constraint across indirect 3.92 supplier–customers (1.93) 1987 Intercept R2

2.38 2.06 2.16 2.26 2.20 2.28 (.41) (.40) (.41) (.41) (.41) (.41) 48.41 53.98 58.14 54.96 41.86 49.19 .15 .16 .19 .18 .19 .16

Percent industry-structure effect 56% from: Rivalry within industry Direct network constraint from 44% suppliers–customers Indirect network constraint from – their suppliers–customers

40% – 60%

48%

53%

49%

52%

25%

29%

21%

29%

27%

18%

30%

19%

Note: These are ordinary least-square regression equations predicting nonnegative price-cost margins in manufacturing industries corresponding to unique four-digit SIC categories in 1987 and 1992 (N ¼ 632). Criterion to be a supplier-customer is 2% of industry business. All predictors are measured as log scores except the dummy variable for 1987. Means, standard deviations, and correlations are given in the Data Appendix (see acknowledgement note). Standard errors (in parentheses) are corrected for autocorrelation across repeated observations of same industry using ‘cluster’ option in STATA. p o.05; p o.001.

346

RONALD S. BURT

Suppliers and customers in Table 1 are the industries with which producers transact two or more percent of their business.8 As a point of reference, Model (1) in Table 1 provides estimates for the baseline model in Eq. (3). The estimates, discussed in the text on pages 335–336, show the negative performance effect of rivalry within the industry (reversed industry concentration in the ﬁrst row of the table) and the negative effect of dependence on supplier–customer industries in which there is little rivalry (network constraint C in second row of the table). Zero-Order Correlation As a further point of reference, Model (2) is the same as Model (1), but with indirect network constraint replacing direct constraint. Recall the correlation between manager performance and indirect access to structural holes (graph to the left in Fig. 2). Similarly, Model (2) in Table 1 shows a strong positive association between industry margins and indirect access to structural holes. The measure of indirect network constraint is average direct network constraint on supplier and customer industries, which is an exact analogue to the measure of indirect network constraint in Fig. 2 for managers.9 In Fig. 4, for example, network constraint on suppliers in the ‘‘Manmade Organic Fibers’’ industry (C ¼ 13) would be averaged with network constraint on customers in the ‘‘Tires and Inner Tubes’ industry (C ¼ 14), which together deﬁne 13.5 points of indirect network constraint on tire cord producers. Model (2) in Table 1 shows that producer margins increase with decreasing direct constraint on suppliers and customers (–5.21 t-test). Returns to Average Indirect Network Constraint Models (3) and (4) test direct and indirect network constraint as alternative effects on producer margins. The measure of indirect constraint in Model (3) is the average across suppliers and customer industries used in Model (2). No consideration is given to the relative volume of producer business with different industries. Any supplier or customer industry over the criterion volume of business is equally a source of indirect network constraint on producers. This is a crude measure, but it is sufﬁcient to show that producer margins increase with indirect access to structural holes in the networks of suppliers and customers, above and beyond the effect of direct access within their immediate network. The –5.09 coefﬁcient for indirect network constraint in Model (3) generates a strong –3.84 t-test (c.f. Baum et al., 2007; Table 2, for association between investment bank performance and

Industry Performance and Indirect Access to Structural Holes

347

the average number of bridging ties in the networks around the bank’s syndicate partners). Model (4) differs from (3) in weighting supplier and customer industries for volume of business.10 In Fig. 4, for example, indirect network constraint on tire cord producers is more deﬁned by the network constraint on tire manufacturers than the constraint on manmade ﬁbers because tire cord producers do more business with tire manufacturers. Speciﬁcally, the weight for tire manufacturers is .61, which is 52.8/(52.8þ33.9), and the weight for manmade ﬁbers is the complement, .39. The two weights together deﬁne 14 points of weighted indirect network constraint on tire cord producers: 13.61 ¼ .61(14)þ.39(13). Weighting in Model (4) offers no improvement over the count of indirect suppliers and customers in Model (3). The –3.32 coefﬁcient in Model (4) for indirect network constraint generates a –3.48 t-test, which is about the same as the corresponding t-test in Model (3). Returns to High versus Low Indirect Network Constraint In Model (5), I disaggregate indirect network constraint into positive and negative elements to see whether either extreme makes disproportionate contribution to the spillover. Models (3) and (4) show that indirect network constraint erodes producer performance, but the effect is some mix of negative effect from indirect network constraint and positive effect from the lack of indirect network constraint. I suspected that the negative effect might be less negotiable in industry buying and selling, and so more likely to spill over between adjacent networks. Measuring positive spillover potential in Model (5), ‘‘Percent Industry Business with Low-Constraint Suppliers-Customers’’ is the percent of industry business transacted with suppliers or customers that are advantaged by their own networks of suppliers and customers, which could be an indirect advantage to producers. The measure is pij for producer industry i (in Eq. (2)), summed across supplier–customer industries j, where industry j is under ‘‘low’’ network constraint from its own suppliers and customers, and ‘‘low’’ refers to the bottom quartile of network constraint scores (C less than 7.72 points). In Fig. 4, tire cord producers score zero on this measure. The 13 points of network constraint on manmade ﬁbers is above the 7.72 criterion for a low-constraint industry, and the 14 points of network constraint on tire manufactures is above the criterion. Measuring negative potential, ‘‘Percent Industry Business with HighConstraint Suppliers-Customers’’ is the percent of industry business transacted with suppliers and customers weakened by severe network constraint from their own suppliers or customers. The measure is pij for

348

RONALD S. BURT

producer industry i summed across supplier–customer industries j where industry j is under ‘‘high’’ network constraint from its own suppliers and customers, and ‘‘high’’ refers to the top quartile of network constraint scores (C greater than 17.43 points). In Fig. 4, tire cord producers score zero on this measure. Network constraint on supplier and customer industries falls below the 17.43 criterion for a high-constraint industry. The results for Model (5) show that producer performance is affected by both the positive and negative effects of indirect network constraint (t-tests of 2.14 for the positive and –2.04 for the negative). Returns to Constraint from the Whole Network of Indirect Suppliers and Customers Model (6) in Table 1 measures indirect network constraint for the whole extended network that does business with producer suppliers and customers. The measures of indirect network constraint in Models (2), (3), (4), and (5) average network constraint in the networks around each supplier–customer industry. Business relations between networks are ignored. The measure of indirect network constraint in Model (6) deﬁnes constraint within and across the networks around an industry’s suppliers and customers. The measure is created as follows: Deﬁne the immediate network around a focal industry by identifying every other industry where focal-industry producers do more than 2% of their business. Second, deﬁne in the same way the immediate network around each industry supplier and customer in the immediate network. The M industries identiﬁed in the second step, but not the ﬁrst, are indirect suppliers or customers for the focal industry. In Fig. 4, for example, M equals 19. There are 19 indirect supplier–customer industries for tire cord producers. For the lower 2% criterion used in Table 1, the number of indirect supplier–customer industries increases to 26. Third, assemble from the input–output table buying and selling among the M industries to deﬁne the extended network of indirect suppliers and customers. By deﬁnition, the focal industry has no direct buying or selling (above the criterion) with its M indirect supplier–customer industries. I operationalized indirect relations with the strongest two-step connection through a direct supplier or customer.11 Fourth, and ﬁnally, compute constraint C in Eq. (4) from the network of buying and selling among indirect supplier–customer industries, and concentration data on producer rivalry within the industries. This measure of network constraint across indirect supplier–customer industries has a statistically signiﬁcant effect on industry performance (2.03 t-test), but the effect is less clear than the corresponding effects for the more

Industry Performance and Indirect Access to Structural Holes

349

narrowly deﬁned indirect constraint measures in Models (2), (3), (4), and (5). The implication is that what matters most for indirect network constraint is the immediate network around each supplier and customer industry, not the whole network of business relations within and between the immediate networks.

SUMMARY I opened this chapter with a question: What is the scope of brokerage network to be considered in thinking strategically? Given the value of bridging structural holes, is there value to being afﬁliated with people or organizations that bridge structural holes? The answer is ‘‘no’’ according to evidence on performance associations with manager networks. Indirect access to structural holes through colleagues, deemed ‘‘secondhand brokerage,’’ shows no performance advantage. Advantage depends on people building their own bridge relations across structural holes. The answer ‘‘no’’ is simple, greatly simpliﬁes the study of strategic behavior in networks, and is surprisingly robust, but its interpretation in terms of enhanced cognitive and emotional skills raises a question about network theory generalized across levels of analysis. Cognitive and emotional skills are more obviously qualities of people than they are qualities of organizations or industries. My goal in this chapter has been to re-establish micro–macro consistency, using evidence on industry networks analogous to the evidence on manager networks. I began with illustrative evidence on performance and manager networks, to establish a baseline and to explain why direct and indirect access to structural holes can be an advantage. Direct access refers to structural holes in the immediate network of a manager’s colleagues, or an industry’s suppliers and customers. Indirect access refers to structural holes between friends of friends, in the networks around colleagues, or around suppliers and customers. We know there are returns to direct access (Fig. 1), in fact very similar returns at micro and macro levels of analysis (Fig. 3). If there is advantage to afﬁliation with the well-connected, there should be returns to indirect access. The returns are negligible in manager networks, as illustrated in Fig. 2, showing no advantage to afﬁliation with well-connected colleagues despite the fact that a manager’s own network is strongly associated with performance. I then described the analogous industry network model (Eq. (3)), introducing the industry data (two years of benchmark performance and

350

RONALD S. BURT

network data on detailed American manufacturing industries), and highlighting in Fig. 3 complementarities between manager and industry evidence (consistency across levels of analysis, greater variety in manager networks, less endogeneity in the industry networks). The micro–macro consistency in industry and manager associations with direct access to structural holes breaks down with respect to indirect access. I analyzed industry performance in terms of three industry-structure effects: rivalry within the industry, direct network constraint from industry suppliers and customers, and indirect network constraint spilling over from the networks around suppliers and customers. In contrast to the manager evidence showing no performance association with indirect access, there is clear evidence of a positive association at the industry level of analysis. The bottom rows in Table 1 show that about 24% of the industry-structure effect on price-cost margins can be attributed to structure beyond the industry’s own buying and selling, to networks around the industry’s suppliers and customers.12

CONCLUSION The industry results could be interpreted as evidence that the network theory used here is not consistent across micro and macro levels of analysis. However, there is also much that is consistent across the levels. I conclude that the industry evidence is not qualitatively distinct from the manager evidence so much as it describes a more extreme business environment. Consider the disaggregate evidence in Table 2. Performance is reported for six study populations – the ﬁve manager populations in Fig. 1 plus the population of industries in Table 1 – broken down into four network categories distinguished by high versus low direct and indirect network constraint. Each network category is a row panel in Table 2 illustrated by a sociogram to the left. There are several industry measures of indirect network constraint in Table 1. For Table 2, I use the measure in Model (4) of Table 1 – the industry measure most similar to the manager measure. The manager and industry results are similar for extreme networks, at the top and bottom panels in Table 2. Performance in Table 2 is a z-score residual holding constant job rank and year for the managers, concentration and year for the industries. At the top of the table are the networks around broker of brokers. These networks provide direct and indirect access to structural holes. These are managers and industries with many disconnected contacts, themselves in networks of many disconnected contacts.

351

Industry Performance and Indirect Access to Structural Holes

Table 2. Manager and Industry Returns to Direct and Indirect Access to Structural Holes. Network Categorya Network Constraintb Direct, Indirect

Study Population (N)

Residual Z-Score Test Performancec Statisticd

Low, Low

Product Launch (68) Supply Chain (134) HR (46) Bankers (134) Analysts (122) Industries (200)

.35 .44 .54 .46 .39 .34

3.78 6.33 3.79 5.09 3.76 4.69

Low, High

Product Launch (78) Supply Chain (97) HR (49) Bankers (88) Analysts (55) Industries (115)

.07 .10 .26 .26 .01 .04

2.27 3.18 2.43 4.44 2.53 2.17

High, Low

Product Launch (55) Supply Chain (95) HR (55) Bankers (102) Analysts (123) Industries (129)

.20 .29 .09 .28 .21 .05

.62 .12 .66 1.37 1.38 2.22

High, High

Product Launch (57) Supply Chain (129) HR (133) Bankers (145) Analysts (54) Industries (188)

.32 .3 .21 .39 .40 .30

2.45 3.68 2.10 6.25 3.35 3.26

po.05; po.001. a

Focal manager or industry is dot in the center. Dashed lines are relations beyond immediate network. b Constraint is dichotomized at its median in each population, except in the HR organization, where it is split to distinguish lowest 33% of scores. c This is performance holding constant year and job rank for individuals, year and concentration for industries. d These are test statistics for effects when z-score residual performance is regressed across the rows in each study population (analyst, banker, and industry results are adjusted for autocorrelation using ‘cluster’ option in STATA). ‘Closed Network’ is reference category.

352

RONALD S. BURT

Performance scores at the top of Table 2 are the highest in the table. At the bottom of the table are the closed networks providing neither direct nor indirect access to structural holes. These are managers and industries with densely interconnected contacts which are themselves in networks of densely interconnected contacts. Performance scores at the bottom of Table 2 are the lowest in the table. The critical results for this chapter are in the middle of the table, describing networks that provide direct or indirect access to structural holes, but not both. In the second panel of Table 2, managers and industries are similarly advantaged by ‘‘only direct access’’ to structural holes. These are producers relatively free from the constraint of dependence on concentrated supplier or customer industries, but beyond those suppliers and customers are concentrated industries that pose severe indirect network constraint. The –.04 average residual price-cost margin reported for industries in the second panel of Table 2 is lower than the .34 residual margin enjoyed by producers free from direct and indirect network constraint, but signiﬁcantly higher than the –.30 residual margin observed in industries oppressed by high direct and indirect network constraint (2.17 t-test). Test statistics in the second panel are sufﬁcient to reject the null hypothesis – a magnitude of two or three – but are for managers and industries similarly weaker than the test statistics for the broker-of-brokers networks in the ﬁrst panel of the table. The manager and industry results disagree in the third panel of Table 2, for networks that provide ‘‘only indirect access’’ to structural holes. Average manager performance in the third panel is no better than the low performance observed in closed networks (t-tests of .12 to 1.38). In contrast, price-cost margins for industries in the third panel of Table 2 are signiﬁcantly higher than the margins in closed-network industries (2.22 t-test). The panel-three industries contain producers dependent on concentrated supplier–customer industries that are themselves relatively free from constraint. The network in Fig. 4 is illustrative. Tire cord producers face severe direct network constraint. They are dependent on a concentrated supplier industry and a concentrated customer industry. Both the supplier industry, and customer industry, do business in a wide variety of their own supplier–customer industries (dotted lines in Fig. 4), which would put tire cord producers in the third panel of Table 2. However, the indirect supplier and customer industries are sufﬁciently concentrated to put tire cord producers in the ‘‘closed network’’ panel at the bottom of Table 2 (indirect network constraint score of 13.32 is higher than the median score of 8.53). In other words, industries in the third panel of Table 2 are less constrained than the example in Fig. 4 in the sense that their indirect

Industry Performance and Indirect Access to Structural Holes

353

supplier–customer industries are more numerous, more disconnected, or more riddled with internal rivalry. That relative freedom from indirect network constraint is an advantage manifest in higher margins despite severe direct network constraint. In fact, margins in the third panel are about as high as the margins observed in the industries just above them with more attractive network structures (mean residual price-cost margins of –.04 and –.05 for industries in the second and third panels of Table 2, versus –.30 for ‘‘closed networks’’ at the bottom of the table).13 The disagreement between manager and industry results in the third panel of Table 2 is not a qualitative jump from managers to industries so much as it is a matter of degree. The manager effects are not equally negligible. There is an order to effects in the panel: statistically signiﬁcant for industries (2.22 t-test; Po.05), not signiﬁcant for the bankers and analysts (respective t-tests of 1.37 and 1.38, Po.10), zero for the product-launch, supply-chain, and HR managers (t-tests of .62, .12, and .66 respectively, P W .50). What do analysts and bankers have in common with industries that distinguish all three populations from the other, more bureaucratic, populations in Table 2? Of the many dimensions of competition that would put industry markets at one extreme of a continuum with bureaucratic organizations at the other extreme, two dimensions stand out as likely candidates for productive network research in future: information and inhibition. The information dimension I have in mind is the familiar contrast between Austrian and neoclassical markets (e.g., with respect to network models, Birner, 1996; Burt, 2005, pp. 227–244; 2007). At the Austrian end of the continuum lie networks in which information is tacit and complex so it moves between groups slowly and inaccurately, if it moves at all. Here, the product-launch and supply-chain managers work in networks balkanized by geography, technology, and legacy culture. Indirect connections beyond the immediate network have limited value, or, judging from Table 2, no value. At the neoclassical end of the continuum lie networks in which information moves rapidly and accurately. Here are the mature capital markets in which I would have thought the investment bankers and analysts work. There must always be an element of local interpretation, but capital markets are mature in the sense that news about investments and company developments in distant locations routinely ﬂashes around the globe to affect plans and share price in London, New York, and Tokyo. The more easily that meaningful information moves quickly between distant places, the more advantage there is to the diverse information provided by indirect access to structural holes. There is a severe scope condition to this advantage.

354

RONALD S. BURT

Indirect access shows negligible advantage for the investment bankers and analysts in Table 2. Advantage is only visible at the extreme of industry buying and selling, where information can be codiﬁed into routines and apparently moved with impact through indirect connections. The inhibition dimension I have in mind is social norms of proper behavior. The more personal and local the business, the more likely that people feel obligation to support friends and return favors. Among the six populations discussed in this chapter, HR, product-launch, and supplychain managers show no returns to brokerage beyond their immediate network. In conﬁdential talk with these managers, I would expect to hear stories about actions justiﬁed by personal loyalty and favors that people owe one another. In contrast, no one ‘‘owes’’ their industry. No one counts their industry among their friends. You can drive a business into bankruptcy, but it would be poor form to hammer a friend insensible. There must always be some element of inhibition to corporate behavior. If you think corporations are wild based on what you know about their behavior, imagine what was ruled out as improper. The analysts and bankers in Table 2 show a negligible but nonzero advantage from indirect access to structural holes, so I put them somewhere between the extremes, distinct from the impersonal market behavior of industry buying and selling, but not quite the personal work environments of the HR, product-launch, or supply-chain managers. Protection from market forces can be discussed in terms of human decency or corrupt bureaucracy. Either way, it is an intriguing question for the next round of empirical research. One thing is clear: a wide range of business environments – from corporate bureaucracies up through the mature capital markets in which investment bankers and analysts work – show no performance advantage to brokerage beyond the immediate network of direct contacts. There is a detectable performance advantage at the extreme of industry market relations; but short of that extreme, advantage is limited to the immediate network of direct contacts.

NOTES 1. Alternative aggregations yield similar results. Indirect network constraint on manager i is measured by aggregating the networks around each of the manager’s P contacts, ICi ¼ j dijCj, i6¼j, where Cj is network constraint on contact j within his or her own network, and dij is a weight for pooling contact networks. I tried measuring indirect network constraint as the arithmetic average across a manager’s contacts (dij ¼ 1/N, where N is the number of the manager’s contacts). This is the measure on

Industry Performance and Indirect Access to Structural Holes

355

the horizontal axis of Fig. 2. I also tried the constraint on the manager’s boss, under the assumption that the chain of command is the primary source for opportunities (dij ¼ 1 for manager’s boss, 0 for all other contacts), and constraint on the manager’s best-connected colleague, under the assumption that every contact need not be a source of opportunity, but you need at least one (dij ¼ 1 for the contact with the lowest network constraint, which means the largest, least redundant, network; 0 for all other contacts). These three aggregations yield the same result: strong zero-order association with performance and negligible partial association. 2. After suppliers and customers in an industry’s immediate network are identiﬁed, proportions are normalized within the immediate network to compute network constraint. The raw proportions deﬁned by Eq. (5) are normalized to sum to one across all production industries in the economy. As reported in Table A3 in the Data Appendix, I get stronger network constraint effects if I compute constraint P from pij normalized within the immediate network around an industry: pij ¼ pij/ k pik, i 6¼ j, where the sum is across all industries k in the immediate network excluding industry i itself. This assumes that the connections most relevant to the focal industry are the connections within its immediate network, not connections across the economy. Normalizing within the immediate network is what is done with manager networks when relations beyond the immediate network are unknown (as is often the case in survey network data), so I am comfortable using the same operationalization with industry networks to obtain stronger network effects. 3. Throughout this chapter, I use the structural autonomy score deﬁned by industry structure (A in the baseline model, Eq. (3)) to predict exponential industry performance (ePCM), rather than raw performance (PCM), where PCM is the industry price-cost margin. The exponential of performance yielded clearer results in Burt et al. (2002), and I ﬁnd the same for the more narrowly deﬁned industries analyzed here. Thus, natural logs of industry-structure variables predict price-cost margins in the text. 4. The statement is based on regressing observed price-cost margins across z-score margins, holding constant the slightly higher margins in 1987, which shows a 9.6b average increase in price-cost margin for a unit increase in z-score margin. 5. The expected price cost margin is predicted using the estimates presented below for Model (1) in Table 1. 6. The performance link with industry structure is all the more impressive because buying and selling this constrained is rarely left exposed to the vicissitudes of market price. Such buying and selling is typically embedded in a corporate hierarchy to manage the risk (e.g., Pfeffer & Salancik, 1978; Burt, 1983). The typical pattern is borne out here. For example, one of the leading ﬁrms in the tire cord industry is Firestone Fibers and Textiles. Firestone is owned by BFS Diversiﬁed Products, which also runs establishments in Firestone’s primary supplier industry, manmade ﬁbers. BFS Diversiﬁed Products is owned by the Japanese tire company, Bridgestone, the American operations of which are a major tire supplier for automobiles produced in the United States. In other words, Bridgestone has embedded its American tire production in a corporate hierarchy. Bridgestone tire production can draw on Firestone tire cord, which can draw on BFS ﬁber output. Nevertheless, market advantage emerges in the transfer prices negotiated between business units. Tire cord production is anchored on three industries: itself, a supplier industry, and a

356

RONALD S. BURT

customer industry. All three are highly concentrated. However, the supplier and customer industries are less subject to network constraint from their own suppliers and customers, which is manifest in them enjoying their expected level of proﬁts while tire cord producers report margins well below expected. 7. The ﬁve effect estimates in Table 1 for measures of indirect network constraint are statistically signiﬁcant if I include observations on the manufacturing industries that correspond to multiple SIC categories, increasing the number of observations from 632 to 713. Here are the coefﬁcients in Table 1 and their corresponding t-tests (in parentheses): 5.09 (3.84), 3.32 (3.48), 1.48 (2.14), .74 (2.04), and 3.92 (2.03). When estimated across all manufacturing industries with nonnegative pricecost margins, the results are similar, but stronger because of the additional observations: 5.33 (4.41), 3.49 (4.09), 1.38 (2.43), .76 (2.33), and 4.22 (2.52). 8. The 2% criterion is based on tests of higher and lower criteria reported in Table A3 in the Data Appendix. The 2% criterion keeps the immediate network to a minimum size without losing predictive power in the baseline model, and leaves more of the economy available as potential indirect suppliers and customers. I estimated the models adding as a predictor the percentage of industry business that was over the 2% criterion (74% on average, see Table A2 in the Data Appendix). There is no zero-order association with performance (1.37 t-test) and no partial associations in the ﬁve models in Table 1 (t-tests of .81, .63, .42, .95, and .66 respectively for the ﬁve models in Table 1). 9. See measure IC in footnote 1, where dij is here equal to 1/N if producers in industry i do a criterion volume of business with industry j, and N is the number of industries with which producers do more than the criterion amount of business. For example, 5% is the criterion in Fig. 4, and N equalsP 2, so the dij for each is 1/2. 10. Weight dij in the preceding footnote equals pij / k pik for Model (4), volume of producer business with industry j divided by the sum of business relations that exceed the criterion for inclusion in the network. 11. The relation pij from focal-industry i to indirect supplier–customer industry j is set equal to the square-root of the maximum pikpkj across industries k in the immediate network around industry i, where pik is the proportion of industry i business conducted with industry k and pkj is the proportion of industry k business conducted with industry j (where industry j is not in the immediate network around focal industry i). 12. The 24% Figure in this sentence is the average across the four percentages for indirect network constraint in Table 1. The speciﬁc averages across Models (3) through (6) are 50.5% for rivalry within the industry, 26.0% for direct network constraint, and 23.5% for indirect network constraint. 13. The network cross-classiﬁcation in Table 2 almost always elicits a workshop question about interaction effects. Do direct and indirect network constraint affect one another’s effect on performance? They do not. To determine this, I multiplied log direct network constraint times log indirect constraint, and entered the interaction term to the performance prediction in each study population. The interaction term is a negligible addition: .45 t-test for compensation in the product launch, .36 for supply-chain manager salary, 1.71 for HR salary, .40 for investment banker compensation, .12 for analyst election to the All-America Research Team,

Industry Performance and Indirect Access to Structural Holes

357

and .26 for industry price-cost margins (Model (3) in Table 1). The concentration of effect in panel three of Table 2 is a heuristic. It is true that the disagreement between manager and industry results is most apparent in panel three of Table 2, but the industry performance association with indirect access to structural holes exists in the other three panels as well. If I delete the 139 industry observations in panel three of Table 2 from the estimation of Model (3) in Table 1, there is still a 3.36 t-test for the performance association with indirect network constraint. Binary distinctions in Table 2 are a useful heuristic. They do not fully capture the continuous-variable results in Table 1.

ACKNOWLEDGMENTS I am grateful to the University of Chicago Graduate School of Business for ﬁnancial support of work on this chapter, to Edward Smith for comment and replicating the analysis with the download data, and to workshop participants at the Rotman School of Management, University of Toronto, and Nufﬁeld College, Oxford University. This work began in reaction to a question from Henrich Greve in a presentation at the Academy of Management meetings in 2007: ‘‘What are the implications of secondhand brokerage for macro applications of network brokerage?’’ It was a good question. It took me unawares. I was surprised to realize that I had not thought about the question. I wish I knew then what I can report now. Source data and a Data Appendix are available online (www.chicagogsb. edu/fac/ronald.burt/research).

REFERENCES Baum, J. A. C., van Liere, D. W., & Rowley, T. J. (2007). Between closure and holes: Hybrid network positions and the performance of U. K. investment banks. Paper presented at annual meetings of the Academy of Management, Philadelphia, PA. Bidwell, M., & Fernandez-Mateo, I. (2007). Relationship duration and returns to brokerage in the stafﬁng sector. Unpublished manuscript, INSEAD, Singapore. Birner, J. (1996). Mind, market and society: Network structures in the work of F. A. Hayek. Unpublished manuscript, Computable and Experimental Economics Laboratory, University of Trento. Burt, R. S. (1980). Autonomy in a social topology. American Journal of Sociology, 85, 892–925. Burt, R. S. (1983). Corporate proﬁts and cooptation. New York: Academic Press. Burt, R. S. (1988). The stability of American markets. American Journal of Sociology, 94, 356–395. Burt, R. S. (1992). Structural holes. Cambridge: Harvard University Press.

358

RONALD S. BURT

Burt, R. S. (1997). The contingent value of social capital. Administrative Science Quarterly, 42, 339–365. Burt, R. S. (2004). Structural holes and good ideas. American Journal of Sociology, 110, 349–399. Burt, R. S. (2005). Brokerage and closure. New York: Oxford University Press. Burt, R. S. (2007). Secondhand brokerage: Evidence on the importance of local structure for managers, bankers, and analysts. Academy of Management Journal, 90, 119–148. Burt, R. S. (2008). Secondhand brokerage. Unpublished manuscript, University of Chicago Graduate School of Business. Burt, R. S., & Carleton, D. S. (1989). Another look at the network boundaries of American markets. American Journal of Sociology, 95, 723–753. Burt, R. S., Guilarte, M., Raider, H. J., & Yasuda, Y. (2002). Competition, contingency, and the external structure of markets. In: P. Ingram & B. S. Silverman (Eds), Advances in Strategic Management (Vol. 19, pp. 165–215). New York: Elsevier. Burt, R. S., Jannotta, J. E., & Mahoney, J. T. (1998). Personality correlates of structural holes. Social Networks, 20, 63–87. Caves, R. (1992). American industry: Structure, conduct, performance. Englewood Cliffs, NJ: Prentice-Hall. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Collins, N. R., & Preston, L. E. (1969). Price-cost margins and industry structure. Review of Economics and Statistics, 51, 271–286. Cook, K. S., & Emerson, R. M. (1978). Power, equity and commitment in exchange networks. American Sociological Review, 43, 712–739. Cook, K. S., Emerson, R. M., Gillmore, M. R., & Yamagishi, T. (1983). The distribution of power in exchange networks: Theory and experimental results. American Journal of Sociology, 89, 275–305. Davis, G. F. (1991). Agents with principles? The spread of the poison pill through the intercorporate network. Administrative Science Quarterly, 36, 583–613. Feld, S. L. (1981). The focused organization of social ties. American Journal of Sociology, 86, 1015–1035. Feld, S. L. (1997). Structural embeddedness and stability of interpersonal relations. Social Networks, 19, 91–95. Fernandez-Mateo, I. (2007). Who pays the price of brokerage? Transferring constraint through price setting in the stafﬁng sector. American Sociological Review, 72, 291–317. Festinger, L., Schachter, S., & Back, K. W. (1950). Social pressures in informal groups. Stanford, CA: Stanford University Press. Freeman, L. C. (1977). A set of measures of centrality based on betweenness. Sociometry, 40, 35–41. Freeman, L. C. (1979). Centrality in social networks: Conceptual clariﬁcation. Social Networks, 1, 215–239. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380. Granovetter, M. S. (1983). The strength of weak ties: A network theory revisited. In: R. Collins (Ed.), Sociological Theory 1983 (pp. 201–233). San Francisco, CA: Jossey-Bass. Greve, H. R. (1995). Jumping ship: The diffusion of strategy abandonment. Administrative Science Quarterly, 40, 444–473.

Industry Performance and Indirect Access to Structural Holes

359

Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Hargadon, A., & Sutton, R. I. (2000). Building an innovation factory. Harvard Business Review, 78, 170–176. Ingram, P., & Roberts, P. W. (2000). Friendships among competitors in the Sydney hotel industry. American Journal of Sociology, 106, 387–423. Kadushin, C. (1995). Friendship among the French Financial Elite. American Sociological Review, 60, 202–221. Krackhardt, D. (1998). Simmelian tie: Super strong and sticky. In: R. M. Kramer & M. A. Neale (Eds), Power and Inﬂuence in Organizations (pp. 21–38). Thousand Oaks, CA: Sage. Lawson, A. M. (1997). Benchmark input-output accounts for the U.S. economy, 1992. Survey of Current Business, 77, 36–82. Lawson, A. M., Bersani, K. S., Fahim-Nader, M., & Guo, J. (2002). Benchmark input-output accounts of the United States, 1997. Survey of Current Business, 82, 19–109. Lawson, A. M., & Teske, D. A. (1994). Benchmark input-output accounts for the U.S. Economy, 1987. Survey of Current Business, 74, 73–115. Lazarsfeld, P. F., Berelson, B. R., & Gaudet, H. (1944). The people’s choice. New York: Duell, Sloan and Pierce. Lin, N. (2002). Social capital. New York: Cambridge University Press. Lin, N., Ensel, W., & Vaughn, J. (1981). Social resources and strength of ties: Structural factors in occupational status attainment. American Sociological Review, 46, 393–405. Løva˚s, B., & Sorenson, O. (2008). The mobilization of scarce resources. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in Strategic Management (pp. 361–390). Oxford, UK: JAI/Elsevier. Maletz, M. C., & Nohria, N. (2001). Managing in the Whitespace. Harvard Business Review, 79, 102–111. McGahan, A. M., Argyres, N., & Baum, J. A. C. (2004). Context, technology and strategy: Forging new perspectives on the industry life cycle. In: J. A. C. Baum & A. M. McGahan (Eds), Advances in Strategic Management (Vol. 21, pp. 1–21). Oxford, UK: JAI/Elsevier. Owen-Smith, J., & Powell, W. W. (2004). Knowledge networks as channels and conduits: The effects of spillovers in the boston biotechnology community. Organization Science, 15, 5–21. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations. New York: Harper & Row. Podolny, J. M. (1993). A status-based model of market competition. American Journal of Sociology, 98, 829–872. Porter, M. E. (1980). Competitive strategy. New York: Free Press. Rowley, T. J., & Baum, J. A. C. (2004). Sophistication of interﬁrm network strategies in the Canadian investment banking industry. Scandinavian Journal of Management, 20, 103–124. Rowley, T. J., Greve, H. R., Rao, H., Baum, J. A. C., & Shipilov, A. V. (2005). Time to break up: Social and instrumental antecedents of ﬁrm exits from exchange cliques. Academy of Management Journal, 48, 499–520. Schmalensee, R. (1989). Inter-industry studies of structure and performance. In: R. Schamalensee & R. Willig (Eds), Handbook of industrial organization (pp. 951–1009). New York: North-Holland.

360

RONALD S. BURT

Spence, M. (2002). Signaling in retrospect and the informational structure of markets. American Economic Review, 92, 434–459. Uzzi, B., & Dunlap, S. (2005). How to build your network. Harvard Business Review, 83, 53–60. van Liere, D. W. (2007). Network horizon and the dynamics of network positions. Unpublished doctoral dissertation, Erasmus University, Rotterdam. van Liere, D. W., Koppius, O. R., & Vervest, P. H. M. (2008). Network horizon: An information-based view on the dynamics of bridging positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in strategic management (pp. 597–642). Oxford, UK: JAI/Elsevier. Venkatraman, N., Lee, C. H., & Iyer, B. (2008). Interconnect to win: The joint effects of business strategy and network positions on the performance of software ﬁrms. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in strategic management (pp. 391–426). Oxford, UK: JAI/Elsevier. von Hippel, E. (1994). Sticky information and the locus of problem solving: Implications for innovation. Management Science, 40, 429–439. Weiss, L. W. (Ed.) (1989). Concentration and price. Cambridge, MA: MIT Press.

THE MOBILIZATION OF SCARCE RESOURCES Bjørn Løva˚s and Olav Sorenson ABSTRACT We examine how the ability of one actor to gain access to resources controlled by another depends on two factors: (i) the number of mutual acquaintances connecting the prospective lender and borrower and (ii) the scarcity of the resources in question. We argue that the incentives to renege on an agreement grow as the resources being traded become increasingly scarce. Mutual acquaintances, however, dampen these incentives, and therefore become more important to facilitating exchange as demand for the good of interest rises. Our analysis of qualitative and quantitative evidence from a study of senior partners at an international consultancy supports these propositions.

Would you ‘‘loan’’ a reliable and diligent assistant to a colleague? What about to someone elsewhere in the organization, maybe in a different department? Would it matter whether you had anything for the assistant to do? Though responses to these questions vary, an informal poll suggests typical answers of: (i) maybe, (ii) probably not, and (iii) absolutely. Why? Though less extensively studied than spot markets and formal contracts, much of the movement of goods and services both within and across organizations occurs via informal transactions in which one party receives a Network Strategy Advances in Strategic Management, Volume 25, 361–389 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25010-2

361

362

BJØRN LØVA˚S AND OLAV SORENSON

good or service in return only for a promise of repayment at some later date. Let us refer to transfers of this type as resource mobilization. For example, a manager may request a rush order from a supplier who might comply thinking that doing so could lead to future sales (Uzzi, 1996). Or, he could refer a client to another ﬁrm under the belief that that ﬁrm will reciprocate (Ingram & Roberts, 2000). Resource mobilization probably plays an even larger role, however, within ﬁrms than across organizational boundaries. Employees, teams and other sub-units must frequently gain access to ‘‘organizational’’ assets – equipment, expertise, manpower, relationships – held by some other member of, or group within, the ﬁrm. Though managerial authority can and does sometimes adjudicate the conﬂicts of interest that arise between these prospective lenders and borrowers, the escalation of action to higher levels of the organization is not always feasible and, even when possible, carries costs. Consequently, resource mobilization serves as an important mechanism for reallocating assets within organizations, and employees routinely negotiate such transfers amongst one another. Though common, resource mobilization is by no means unproblematic. Not only do lenders provide resources on the mere promise of repayment, but also the exact terms of this repayment – where, when, and how – typically remain unspeciﬁed (Blau, 1964). Should the recipient refuse to honor his debt, the absence of a speciﬁc agreement, either oral or written, leaves the lender with few routes for recourse. Given the risks involved, one might then wonder how such transactions ever transpire. As the opening questions illustrate, intuitively we understand that successful resource mobilization depends not only on the nature of our relationship to the other party but also on the demand for the resource being requested. As the resources of interest become increasingly scarce, the cost to the borrower of repaying the loan or returning favors at some later date rises (as do the opportunity costs to the lender of loaning them). Lenders recognizing these incentives for the borrower to renege therefore refuse to grant access to the resources they control unless something reduces this risk. We argue that mutual acquaintances – third parties connected to both the borrower and lender – dampen the desire of borrowers to disavow their commitments through three mechanisms: (i) they improve the lender’s ability to monitor the borrower, thereby increasing the odds of him being punished for failing to repay the loan; (ii) they allow the lender to coordinate with others, intensifying the expected severity of punishment; and (iii) they may also allow potential lenders to assess better other factors that might affect the likelihood of repayment. As the scarcity of a resource

The Mobilization of Scarce Resources

363

rises, individuals therefore rely increasingly on mobilization through relations buttressed by a set of mutual acquaintances. Our empirical analysis focuses on senior partners at ABC Consulting.1 These managers need not worry about promotions, as they have already reached the highest echelon of the ﬁrm. Their jobs (and consequently their bonuses) nevertheless depend on their ability to mobilize resources both inside and outside the ﬁrm. We draw on two types of data. We began by gathering qualitative information through interviews. Using the insights garnered from these interviews, we then developed a survey relevant to this context and administered it to a randomly selected set of senior partners. Analyzing the effects of mutual acquaintances at the level of the dyad, our results conﬁrm that relations supported by a set of mutual acquaintances become more and more important to effective individual performance as the resources mobilized through them become increasingly scarce. Our research extends the literature on social relations and resource mobilization in at least three ways. First, we identify resource scarcity as an important modiﬁer of – and perhaps even scope condition to – the value of social relations to exchange. Existing research tends to treat positions as uniformly useful for speciﬁc outcomes (e.g., Burt, 1992; Uzzi, 1996; Podolny & Baron, 1997). Our research demonstrates, however, that situational properties – that can vary not only from one dyad to another for the same actor, but also from period to period even within a dyad – importantly mediate the value of social relations. Second, we highlight the role of mutual acquaintances in facilitating resource mobilization within a dyad and describe the mechanisms that could account for this effect. Prior research, by comparison, has been more attentive to the effects of relationship strength (e.g., Granovetter, 1973; Uzzi, 1996; Hansen, 1999), and has given less empirical attention to the effects of common connections to others (for an excellent exception, see Reagans & McEvily, 2003). Finally, whereas most studies investigate the consequences either of connections to contacts outside the ﬁrm or of positions within the ﬁrm, by examining both in the same setting, we show that a common set of forces govern resource mobilization in both cases.

INFORMAL EXCHANGE AT A CONSULTANCY ABC Consulting, one of the world’s largest and most diversiﬁed professional service ﬁrms, employs more than 55,000 professionals in over 100 countries. Our research focuses on the 1,100 senior partners that own

364

BJØRN LØVA˚S AND OLAV SORENSON

and manage the ﬁrm. We began our study by interviewing 32 of these senior partners to understand better their roles and learn how they thought interpersonal relations facilitated their jobs (see the Appendix and Table 7 for information about these interviews and our methodology for conducting them.) Partners produce proﬁts for the ﬁrm and earn their salaries and bonuses through two complementary activities: selling projects and then fulﬁlling them to the clients’ satisfaction. Revenue targets require each senior partner at ABC Consulting to generate roughly $10 million per year in revenue. Once sold, partners must assemble the expertise and resources necessary to deliver on these contracts. Throughout this chapter we will refer to the former activity as business generation and the latter as project execution. When discussing the challenges of business generation and project execution, all partners stressed, without prompting, the importance of their informal relationships – generally described as ‘‘business acquaintances’’ rather than as ‘‘friendships’’ (i.e. these relationships appeared to have limited emotional content). The partners discussed several ways in which these contacts helped them to gain access to resources both inside and outside the ﬁrm. We clustered their examples into four categories described below.2

Business Generation Relationships external to ABC Consulting helped partners generate business in at least two ways. On the one hand, relationships could allow them to claim credibly that they would staff a project with particular personnel of interest to the client. Partners reported that potential clients became especially sensitive to these stafﬁng issues during periods of high demand for consulting services, when consulting ﬁrms would frequently expand their capacity by hiring less experienced personnel. One London-based partner, for example, describing the sale of a project, noted: I have certain situations lately y where I say to a client ‘‘And the good news is team members three, four and ﬁve y I can get Joe, Fred and Tom onto the job,’’ and the guy says: ‘‘If you do that it’s yours, because I know them and I have worked with them before y that is it, it is sold.’’

Because ABC Consulting does not allow their contracts with clients to specify who will work on a project, such assurances can only occur through personal promises. Partners nonetheless considered them critical to project sales.

The Mobilization of Scarce Resources

365

At other times, partners’ contacts at clients help them to sell projects to those ﬁrms; partners referred to it as having an ‘‘internal salesman.’’ Though these external contacts might simply provide the consultant with valuable information, internal salesmen would sometimes even ‘‘call in favors’’ from their colleagues to garner support for the consultant. Consider a German partner’s description of the process of bidding for and winning a large consulting project: The sales process of this project took 18 months, was highly competitive, was going back and forth y [My contact got] us going [through] the relations [he] had internally in the [client] organization, moving forward information, triggering those people, conducting meetings, providing the right information in the right way, and the right shaping to the client.

These internal salesmen put their reputations at risk and partners discussed the fact that they typically expected some sort of quid pro quo, often in terms of the project’s recommendation supporting the champion’s point of view or examining some side issue of interest to him.

Project Execution Within the ﬁrm, social relations facilitated the recruitment and retention of personnel to execute projects. Experienced personnel played many important roles. Especially on large projects (more than 100 consultants), senior staff helped partners to manage the client relationship. Even on smaller projects, associate partners and project managers would coordinate and monitor the activities of 5 to 25 less experienced consultants. Moreover, to the extent that projects required expert knowledge, it generally resided within experienced managers and junior partners. At the time of the interviews, many partners revealed that ﬁnding staff for their projects had become their main challenge. Demand in many segments of consulting had expanded rapidly, with growth rates of more than 20% per year common. Consultants in these areas found it easy to generate business but difﬁcult to fulﬁll contracts once sold. As a London-based partner in a particularly popular practice observed, ‘‘Now that we can pick and choose [business opportunities] the real challenge is to ﬁnd the resources that you need to deploy on those opportunities.’’ Another observed: What is more difﬁcult today is attracting the right talent, and retaining that talent, and growing that talent professionally. So there is almost a behavioural shift trying to get partners to think about people and our knowledge y as the most important assets we have.

366

BJØRN LØVA˚S AND OLAV SORENSON

Under the strain of these shortages, formal stafﬁng procedures failed. As one managing partner, who helped to develop and implement the resource allocation system, explained: When you are resource constrained it isn’t much about balancing, it’s about managing scarcity, which isn’t well supported by the best scheduling system – even if I designed the perfect one. Managing scarcity is about taking away from somebody to give to somebody else, or not doing something [i.e. foregoing a sale].

As a result, partners had resorted to mobilizing human resources directly from the people holding them. An Italian partner expressed views representative of many: ‘‘If one [experienced employee] is free I have 200 requests claiming, ‘this guy is mine’ y when you are in permanent lack of resources you are in permanent conﬂict.’’ In these requests, established relationships within the ﬁrm helped partners gain access to personnel. For example, one London-based partner noted,’’ There’s a big black market right now in [informal] stafﬁng y You need to focus on the internal network to try to get the resources.’’ Partners described two typical approaches to seeking a loan of experienced employees. In some cases, one would ask another to release a staff member for a limited period of time (e.g., 4 to 12 weeks). Such timebound borrowing usually involved senior managers who had knowledge or technological expertise that the requester needed to execute a project. In other cases, however, these personnel loans would have an indeﬁnite duration. The partners at ABC Consulting thought relatively explicitly about this process as one of favor trading. In the words of one Italian partner: ‘‘You [lend staff] to make a favor. You do it because you’re a friend of this partner y you do it because you expect return.’’ Partners would also use their internal relationships to retain personnel. When valued staff expressed concerns, partners would typically not address those concerns immediately. Moving an individual off of an assignment, or deploying reinforcements, requires time to locate and train personnel. Promotions and raises meanwhile must follow ABC Consulting’s review cycle. Partners would therefore bargain with at-risk employees in terms of future assignments, future promotions, and future bonuses. Though one can tell from these descriptions that the categories used to classify the ways in which social relations operate tend to overlap and occur in conjunction, these interviews nevertheless inspired and informed our thinking about the process of resource mobilization. Let us then turn to a more precise discussion of this process.

The Mobilization of Scarce Resources

367

INCENTIVES IN INFORMAL EXCHANGE To motivate our empirical analysis, let us describe a simple model of the incentives governing resource mobilization (for illustrative purposes only; for a sophisticated analysis, see Kreps & Wilson, 1982). The model could easily apply to a variety of situations. Consider the following cases, garnered from the examples offered in our interviews: (1) One partner has sold more projects than he can staff and requests a loan of personnel from another partner; (2) One partner has excess staff not currently engaged on any projects and asks another partner for assistance in selling projects to support them; (3) A senior partner offers a junior partner an attractive assignment if she will ﬁrst assist him on a less glamourous one; (4) An employee at a client offers to help sell a project to his superiors if the consultant later supports his pet project. Despite the model’s consistency with all of these situations, for the purpose of exposition, we will describe it as an exchange of personnel between two partners. Imagine that two partners, Bob and Lisa, labeled B and L, contemplate a transaction. Each of these partners requires two complementary resources: projects and personnel. Each resource comes in discrete units of one, and the combination of one unit of each produces one unit of value.3 Allow X to represent the cardinality of the set of projects and Y the cardinality of the set of personnel. Although the model is symmetric with respect to these resources, assuming that Xb W Yb (i.e. that Bob has more projects than people to complete them) simpliﬁes our discussion. For example, if Bob sold projects that required 10 staff, then Xb ¼ 10. If he had ﬁve people available to execute those projects (i.e. Yb ¼ 5), then he could only create ﬁve units of value – the scarcity of personnel is a binding constraint on the value that Bob can produce. Because resource mobilization includes a (possibly implicit) promise of repayment, it has a time dimension. We capture this ordering by considering the effects of exchange across two periods. In the ﬁrst period, Bob, B, wishes to borrow a employees from Lisa, L, the potential lender. In the second period, the roles reverse as Lisa calls in her chit.4 Given these assumptions, we can deﬁne the incentives facing Bob, the borrower. The value, VB, that Bob can create across the two periods is: V B ¼ minðX B1 ; ðY B1 þ aÞÞ þ minðX B2 ; ðY B2 ð1 bÞaÞÞ bac;

(1)

where (b denotes the probability of reneging in the second period and c captures the costs of not repaying one unit of the loan (discussed in greater detail below). Let us assume that aoXB2YB2.5

BJØRN LØVA˚S AND OLAV SORENSON

368

By differentiating with respect to a (the number of units of Y, employees, transferred from the lender, Lisa, to the borrower, Bob, in the ﬁrst period), we can see Bob’s incentives to borrow: ( if X B2 4Y B2 a; bð1 cÞ DV B ¼ (2) if X B2 Y B2 a; 1 Da Though Bob always wants to borrow personnel – recall that we assumed that he had more projects than people in period 1 – Eq. (2) demonstrates that he only has an incentive not to repay his debt in period 2 if his demand for personnel remains high when Lisa asks him to return the favor (i.e. XB2WYB2a). In that situation, Bob faces an opportunity cost to loaning employees back to Lisa because doing so requires him to forgo the completion of some of his own projects, and consequently the value they would allow him to produce (and presumably the bonuses associated with that value). Lisa confronts a similar calculus in her decision to lend, but where a reduces her production in the ﬁrst period and increases it (if repaid) in the second: V L ¼ minðX L1 ; ðY L1 aÞÞ þ minðX L2 ; ðY L2 þ ð1 bÞaÞÞ

(3)

Again differentiating with respect to a reveals Lisa’s incentives to lend: 8 if > > > < if DV L ¼ > if Da > > : if

X L1 4YL1 a and X L2 4Y L2 ; X L1 4Y L1 a and X L2 Y L2 ;

b 1

X L1 Y L1 a and X L2 4Y L2 ; 1 b X L1 Y L1 a and X L2 Y L2 ; 0

(4)

Because we did not constrain the relative scarcity of projects or personnel in either period for Lisa, we must consider four possibilities. If Lisa always has more personnel than she requires to execute her own projects – the fourth case in Eq. (4) – she has no incentive not to lend freely to Bob. On the other hand, if she expects to only need people in the ﬁrst period but not the second (the second case), she has no incentive to lend because she cannot beneﬁt by calling in the favor. The other two cases – the ﬁrst and third in (4) – seem more interesting. In each of these, whether or not Lisa would agree to transfer personnel to Bob depends on her expectation of his likelihood of repaying her.

The Mobilization of Scarce Resources

369

Likelihood of Reneging The odds of the borrower failing to repay – and consequently the probability that any transaction transpires – depends on the costs to him of doing so. Our focus on the costs of reneging stems from the fact that our theory concerns factors governing the assurance of reciprocation. Our central contention is that these costs increase as a function of the degree to which mutual acquaintances surround the relation. That is, c ¼ f ðSÞ

(5)

where f’(S) W 0 and S represents the number of third parties, such as Mike in Fig. 1, connected to both Bob and Lisa. In our interviews, partners frequently discussed the importance of mutual acquaintances to both business generation and project execution. A British partner, for instance, discussing a difﬁcult sale related that, ‘‘[The client] was a bit hostile to start with y basically what I started doing was talking about all the people at his company that I have a good relationship with, to kind of get approval.’’ He sold the project. Similar processes appeared to govern the internal lending of employees. For example, a partner based in Italy described the process of requesting an employee: You introduce [your request] and within a week the person has made ﬁve phone calls y to understand if he can trust you or not. He will talk to other partners in ABC Consulting to actually hear the rumors y some feedback is always activated.

Three mechanisms contribute to the importance of mutual acquaintances in this evaluation.6 On the one hand, these connections facilitate monitoring. By improving information ﬂow and with it the ability of the lender to observe whether the borrower fails to hold their side of the (possibly tacit) agreement, mutual acquaintances increase the odds of being punished – and therefore also the overall expected cost – of reneging. Assessing whether the other party has

Fig. 1.

A Simple Example of a Dyad with a Mutual Acquaintance.

370

BJØRN LØVA˚S AND OLAV SORENSON

failed to repay could require ﬁne-grained information for at least two reasons: First, the repayment requested from the borrower may elude veriﬁcation. An employee at the client organization might ﬁnd it difﬁcult, for example, to determine whether the consultant has supported his perspective without thirdparty veriﬁcation. Second, when the repayment involves a different type of good – imagine repaying a loan of personnel with equipment – lenders require substantial information about the relative value of the goods to both the borrower and other parties to determine the fairness of the repayment. On the other hand, multiple mutual relations also increase the expected magnitude of punishment, conditional on punishment occurring. Under this mechanism, widely discussed in the existing literature on rational trust, mutual acquaintances allow the spurned actor to coordinate with many others when sanctioning a failure to repay (Bott, 1957; Kreps & Wilson, 1982; Granovetter, 1985; Coleman, 1990; Raub & Weesie, 1990; Burt & Knez, 1995). Punishment may occur either passively or actively. Passive forms of punishment generally amount to no more than exclusion from future trades. A London-based partner, for example, commented on a partner who had been excluded from personnel trading: ‘‘The people who manipulate new networks totally for their own gain, people y see through that because it is a bit about give and take. You have to put in as well as take out.’’ More active forms of punishment at ABC Consulting might include manipulating a compensation committee to prevent the offending partner from receiving an annual bonus. Finally, having multiple mutual acquaintances with a potential trading partner allows one to assess other costs that may inﬂuence that actor’s likelihood of reneging on an agreement. These costs might stem from a wide variety of factors: personal guilt or regret (possibly induced by community norms; see Durkheim, 1957), reputation loss among those on whom the actor would not rely for resources (e.g. the borrower’s church congregation), or direct costs involved with hoarding resources (e.g., ﬁnding ofﬁce space to house personnel). Much of the information required to assess these costs remains private, and mutual acquaintances therefore provide the potential lender with better information on the borrower’s costs of non-repayment.

PROPOSITIONS Together, Eqs (2), (4), and (5) lead us to two propositions: Proposition 1. When alters face relative scarcity in the availability of resources required to execute projects, the value of relationships used to

The Mobilization of Scarce Resources

371

mobilize those resources rises with the number of mutual acquaintances shared by ego and alter. The intuition here is relatively straightforward. If the alter, in this case, Lisa, faces scarcity in personnel (most clearly when XL1 W YL1a and XL2 W YL2), then she only lends if she believes that Bob will transfer personnel to her at some later date. Bob, however, will only return the favor if he faces sufﬁciently high costs (c) related to not doing so to dissuade him from reneging. For the reasons noted above, we expect these costs to increase with the number of mutual acquaintances the two share. Proposition 2. When alters face relative scarcity in the demand for consulting services, the value of relationships used to sell projects rises with the number of mutual acquaintances shared by ego and alter. Notice that the model above is completely symmetric with respect to resources. In other words, we can just as easily allow X to represent personnel and Y projects. Imagine then that Bob instead has an excess of personnel relative to projects and that he has asked Lisa for assistance in selling additional projects. If Lisa also faces scarcity in business (in the sense that she has more people than projects sold for them to staff), Lisa then will only help Bob to sell a project to one of her clients if she expects him to return the favor. Bob again only reciprocates if the costs of non-repayment – again dependent on the number of mutual acquaintances – exceed the potential beneﬁts.

SURVEY CORROBORATION We used a survey to collect additional data. Our survey, administered in face-to-face sessions, polled 102 senior partners (for more details, see the Appendix). Table 1 provides some descriptive information about these individuals. To test our propositions, we adopted the dyad as a unit of analysis, estimating the importance of each contact to a respondent’s job performance. In the situation described above, for example, we would estimate the correlates of Lisa’s importance to Bob. Dyadic analysis has one main advantage relative to aggregating the ego network into one or more summary measures (e.g., centrality or constraint): It estimates the model at the level of the theory. Recall that our propositions concern how the value of a single relation changes as a function of resource scarcity and

BJØRN LØVA˚S AND OLAV SORENSON

372

Table 1.

Age Gender Tenure in ﬁrm Tenure as partner Direct reports Partners Associate partners Senior managers Total

Characteristics of Partners Surveyed. Mean

SD

44 years 92% male 16 years 6 years

4.7 5.1 3.9

4.5 9.0 24.5 38.0

11.8 18.7 65.8 96.3

Table 2. Name Generator Questions. Question

Who are your most reliable sources of valuable information in terms of identifying attractive business opportunities? Who are your most valuable contacts in terms of gaining new business (i.e. closing deals)? Who do you consider your most important sources of valuable knowledge and expertise (e.g. industry, competency, functional)? Who are the associate partners or managers on whom you rely to get things done? On whom do you rely to help you develop skills and knowledge in your area of expertise? On whom do you rely to sponsor and support your projects and activities? Please list any other individuals who are an important part of your network and do not ﬁt into the previous categories.

Number of Names Generated Mean

SD

Minimum Maximum

5.1

1.26

2

6

4.4

1.51

1

6

4.7

1.54

0

6

4.8

1.37

1

6

3.8

1.75

0

6

4.2

1.52

1

6

3.3

2.1

0

6

mutual acquaintances, rather than the value of any particular set of relations. Testing dyadic-level theories at a higher level requires one to assume that effects cumulate in some speciﬁc way (usually linearly), an assumption that may mask the underlying processes. Our information on respondents’ alters came from a set of name generator questions (see Table 2). We customized these questions to the

The Mobilization of Scarce Resources

373

partners’ activities so that they would ﬁnd them meaningful and so that we could distinguish whether they used them to help to generate business or to execute projects.7 In particular, we used the ﬁrst two name generators in Table 2 to identify alters that assisted respondents in selling projects. In total, respondents nominated 333 alters on these two questions (hence we have 333 dyads used for mobilizing resources related to business generation). The third and fourth name generators, yielding a total of 766 dyads, meanwhile allowed us to determine those ties used to gain access to resources required for the execution of projects. We use the last three generators – and the 881 dyads they identify – later in our analysis as a validity check.

Dependent Variable Analyzing relationship importance at the dyadic level required us to rely on respondent self-reports. We used responses to the following question: ‘‘How critical is each person listed for your success at ABC consulting? (1 ¼ less important, 5 ¼ essential).’’ Our assumption is that respondents’ ratings reﬂect the degree to which alters assist them in creating value for the ﬁrm and for themselves. Debrieﬁng from the pilot surveys suggested that partners would generally consider someone important to the extent that they either helped them to sell projects or loaned them key personnel. We nonetheless recognize that using this measure opens our results to an alternative account: Perceptual inaccuracies could color our ﬁndings. In particular, respondents might view as most important those they depend on most heavily. If so, mutual acquaintances might simply reﬂect dependence. We further address this possibility below.

Mutual acquaintances The survey gathered information on the number of mutual acquaintances connecting ego and alter by asking respondents to report whether each pair of their alters shared a direct relationship of some sort, and whether they believed these relations weak or strong (an approach pioneered by Burt, 1992). To assess the accuracy of these reports of alter–alter relations, we compared them to the names reported as alters by those respondents who appeared as alters on others’ surveys (56 cases). For example, if we surveyed both Bob and Lisa in the example above, we compared Bob’s report of

374

BJØRN LØVA˚S AND OLAV SORENSON

whether Lisa had a relationship to Mike with her own self-report (and vice versa). Out of the 149 alter–alter ties respondents rated as strong (and 284 alter–alter ties they rated as weak), 76% (or 26% for weak ties) appeared as important alters on those alters’ self-reports. Because of their greater reliability and closer correspondence to our theory, we use only strong alter– alter ties in our measure of mutual acquaintances. As respondents could name as many as 24 alters, the total number of possible mutual acquaintances ranged from 0 to 23.8 Partners, however, maintain quite diffuse networks, and the average dyad had only 2.8 mutual acquaintances. We logged this count (plus one to remove zeros) to accommodate decreasing returns to additional mutual acquaintances.9

Scarcity Since we lack survey responses for many alters, we constructed our measure of scarcity from the ego responses to the question: ‘‘Think about your efforts to build the consulting practice in the last 12 months. What has been your greatest challenge? (1 ¼ demand: identifying and negotiating opportunities much more challenging, 5 ¼ supply: identifying and managing resources much more challenging).’’ A low value on this scale says that partners in a particular segment of the ﬁrm face scarcity in terms of business generation, while a high value indicates that they face scarcity in terms of the resources required for project execution. We calculated our measure of scarcity for alters in two steps. First, we regressed egos’ responses to this question on egos’ positions within the ﬁrms on three dimensions: countries (8 categories), industry groups (5 categories), and competencies (4 categories). Together these factors accounted for roughly 30% of the variance in the ego-level responses. Second, we used these regression estimates to produce predicted scarcity values for each alter as a function of his or her ofﬁce, industry, and competency. Because our propositions claim that the value of mutual acquaintances increases with the scarcity of the resource being mobilized, our tests focus on the effects of the interaction of scarcity with the number of mutual acquaintances (we mean deviated both variables before creating interaction terms).10 Even if data had been available for all alters, we believe that our method has advantages over using respondents’ own answers. Most notably, our approach purges individual-level variation in response patterns from the data. When using survey data, researchers worry that their results might stem not from real correlations between two factors, but rather from

375

The Mobilization of Scarce Resources

individual-level variation in how people interpret the questions (sometimes referred to as common method bias). The key subjective independent variable in our models, however, depends not on individuals’ own responses, but rather on the responses of similar others in the sample, thereby circumventing common method bias and many endogeneity problems.

Controls The models included several controls (see Table 3 for descriptive statistics). Multiplex relation records whether the respondent attributed more than one kind of relation to the contact. For example, a dyad would receive a score of two if the respondent named an alter on three name generators. Padgett and Ansell (1993), among others, have argued that more complex relations may confer an advantage in resource mobilization. Closeness measures the strength of the direct relationship between the ego and each contact, measured using the question: ‘‘How close are you with each person? (1 ¼ distant, 5 ¼ especially close).’’ Emotional affect and the trust engendered by it may substitute for indirect relations in facilitating resource mobilization (Uzzi, 1996; Sorenson & Waguespack, 2006). Same industry is a dummy variable with a value of one when both members of the dyad focus on the same primary industry. Same region received a coding of one when both work in the same country.

Estimation Ordinal dependent variables, such as ours, potentially cause problems for linear regression, which frequently yields inefﬁcient estimates in these Table 3.

1. 2. 3. 4. 5. 6. 7.

Importance Multiplex Closeness Same industry Same region ln (mutual acquaint.) Alter resource scarcity

Dyad-Level Descriptive Statistics.

Mean

SD

1

2

3

4

5

6

3.27 0.40 3.69 0.70 0.85 1.12 2.40

1.20 0.49 1.05 0.46 0.35 0.69 0.44

.26 .31 .08 .01 .24 .04

.20 .05 .04 .13 .06

.01 .09 .27 .02

.04 .14 .06

.03 .05

.05

376

BJØRN LØVA˚S AND OLAV SORENSON

situations (McCullagh, 1980). We therefore estimated the data using ordered logit regression. By including multiple intercepts, this procedure accommodates the possibility that respondents may not view hierarchically ordered categories as equally distant (e.g., they may perceive the difference between a 3 and a 4 on the scale as larger or smaller than the distance between a 4 and a 5). Our data also potentially exhibit network autocorrelation. Prior research has dealt with this issue in at least two ways: QAP regression permutes the afﬁliation matrix to generate efﬁcient standard errors (Krackhardt, 1988). This procedure, however, requires information on the complete network rather than egocentric data. A second approach samples the data to arrive at a set of sparsely connected cases (Sorenson & Stuart, 2001). The limited size of the full sample, however, prevents us from doing so here. Since we could not resort to one of these established strategies, we generated conﬁdence intervals using bootstrapping as a means of limiting the potential effects of autocorrelation.11 The logic of bootstrapping corresponds closely to that of QAP regression (see Fernandez, Castilla, & Moore, 2000). The bootstrap procedure samples from the data set to generate a distribution of possible coefﬁcient estimates (Efron, 1982); essentially, it asks, assuming that the sample represents the population, how likely are these particular point estimates? Its attractiveness stems from its minimal assumptions; for example, it does not assume independence across cases. Our conﬁdence intervals came from 200 draws – the upper end of the range required for efﬁcient intervals (Efron, 1982). Each draw consisted of random selection with replacement, within egos, of a set of alters equal to the number reported by ego. For example, if an ego nominated 10 alters, we drew 10 names from this set of 10. Though drawn from the same data, these samples differ because selecting with replacement allows some cases to appear multiple times and others not at all in any particular sample.

RESULTS To test our ﬁrst proposition (regarding project execution), we examined dyads in which the alter provided knowledge or talent (name generators 3 and 4 in Table 2). In debrieﬁng those involved in the pilot survey, partners explained that these resources moved primarily through the lending of personnel. Table 4 reports the ordered logit estimates of the correlates of relationship importance for these dyads. The ﬁrst model estimated the

377

The Mobilization of Scarce Resources

Table 4.

Ordered Logit Estimates of Dyad Importance. P1: Project Execution (1)

ln (mutual acquaint.) Alter resource scarcity ln (mutual acquaint.) resource scarcity Multiplex

Same region Same industry

Intercept 2 Intercept 3 Intercept 4 Log-likelihood Dyads

(3)

.595 (.068) .748 (.073) .093 (.178) .153 (.128) .232 (.315) 2.32 (.311) 3.73 (.319) 5.29 (.331) 1,046.9 766

.406 (.104) .103 (.118) .585 (.218) .595 (.060) .651 (.084) .076 (.134) .061 (.134) .050 (.420) 2.07 (.403) 3.52 (.408) 5.11 (.412) 1,033.8 766

.803 (.091) .147 (.101) .406 (.182)

Closeness

Intercept 1

(2)

1.96 (.187) .007 (.133) 1.25 (.147) 2.61 (.161) 1,117.9 766

P2: Business Generation (4)

(5)

(6)

.454 (.113) .627 (.095) .013 (.343) .122 (.228) .703 (.464) .811 (.423) 2.20 (.430) 3.70 (.466) 462.4 333

.248 (.136) 1.07 (.410) 1.07 (.503) .444 (.110) .644 (.112) .084 (.312) .135 (.223) 3.08 (1.01) 1.55 (1.01) .146 (1.03) 1.37 (1.02) 458.8 333

.460 (.162) .764 (.428) .557 (.364)

4.31 (1.05) 2.88 (1.03) 1.61 (1.02) .238 (1.02) 482.9 333

Notes: Bootstrapped standard errors in parentheses; coefﬁcients signiﬁcantly different from zero at po .05 in bold.

effects with only the variables relevant to our proposition. Consistent with Proposition 1, mutual acquaintances prove most valuable to project execution when consultants ﬁnd it difﬁcult to assemble the resources necessary to fulﬁll contracts. In the second model, we estimate the effects of only the control variables. Multiplex and close ties corresponded to higher importance scores (consistent with Padgett & Ansell, 1993; Uzzi, 1996). Model 3 tests the robustness of the hypothesized effects to the inclusion of control variables. The addition of these controls actually strengthens the results. To explore the second proposition (business generation), we analyze those dyads that partners identiﬁed as helping to close deals (name generators

378

BJØRN LØVA˚S AND OLAV SORENSON

1 and 2 in Table 2). Model 4 regresses relationship importance on the number of mutual acquaintances, alter resource scarcity and the interaction of those two variables. Though the coefﬁcients are of similar size to those in model 1, they are not signiﬁcant because of larger standard errors (possibly because these estimates come from fewer dyads). Model 5 estimates the effects of the controls, and model 6 the effects of both the variables of interest and the controls. With the controls included, the results support our proposition: the interaction has a positive and signiﬁcant effect – mutual acquaintances prove most valuable to the generation of business in situations when consultants face a low demand for their services (and consequently client contacts face a high opportunity cost to assisting the consultant). Further exploration revealed that closeness is the critical control for ﬁnding signiﬁcant effects on the interaction term.

Robustness We ran several robustness checks. First, we estimated the models with ego ﬁxed effects to purge the estimates of factors that do not vary across dyads related to the same respondent (Johnson, 2004). As one can see in models 7 and 9, the results remain robust to the inclusion of these separate intercepts. Second, we estimated the effects within a more restrictive set of dyads, those with only one kind of relation. One might worry that multiplex relations add substantial measurement error to our dependent variable because we only have one measure of relationship importance for each dyad. We therefore eliminated dyads involving multiplex relations from the data and estimated our effects on the resulting set (models 8 and 10). Our results remain robust within this set of simpler relations; in fact, they become even stronger. Third, in unreported models, we calculated a measure of the number of nominations that each alter received from all respondents – in essence, the degree centrality of each alter. One might, for example, worry that respondents’ relationship importance scores merely reﬂect the status of alters. Alter centrality nevertheless has no signiﬁcant effect on relationship importance (Table 5).

Validity As a validity check, we regressed the same variables against alter importance for dyads not used to mobilize resources (name generators 5 through 7 in

379

The Mobilization of Scarce Resources

Table 5. Ordered Logit Estimates of Dyad Importance (Bootstrapped Standard Errors). P1: Project Execution

ln (mutual acquaint.) Alter resource scarcity ln (mutual acquaint.) resource scarcity Multiplex Closeness Same region Same industry Intercept 1 Intercept 2 Intercept 3 Intercept 4 Ego ﬁxed effects Log-likelihood Dyads

P2: Business Generation

(7)

(8)

(9)

.397 (.131) .092 (.142) .726 (.245) .662 (.071) .730 (.095) .093 (.216) .003 (.171) .588 (.687) 2.94 (.687) 4.53 (.699) 6.27 (.715) Yes po .01 970.0 766

.381 (.151) .012 (.164) .741 (.275)

.547 (.172) 1.17 (.491) 1.22 (.597) .403 (.149) .513 (.140) .036 (.032) .026 (.338) 3.68 (1.14) 2.05 (1.13) .529 (1.12) 1.16 (1.12) Yes po .06 431.1 333

.782 (.098) .055 (.190) .124 (.172) .287 (.504) 2.55 (.486) 4.18 (.487) 5.81 (.513)

450.8 334

(10) .248 (.144) 1.04 (.403) 1.04 (.485)

.544 (.098) .084 (.327) .077 (.246) 3.35 (.961) 1.80 (.934) .520 (.932) 1.01 (.924)

421.5 294

Others (11)

(12)

.547 .658 (.129) (.126) .288 .225 (.143) .212 .021 .058 (.196) (.238) .907 .929 (.292) (.345) .422 .479 (.067) (.076) .497 .402 (.201) (.312) .198 .374 (.139) (.164) 1.16 .1.36 (.447) (.759) .231 .204 (.440) (.760) 1.44 1.57 (.434) (.749) 3.08 3.37 (.433) (.755) Yes po .01 874.6 458.8 589 589

Notes: Bootstrapped standard errors in parentheses; coefﬁcients signiﬁcantly different from zero at po.05 in bold. Models 8 and 10 have fewer cases because they drop all dyads involving multiplex relations.

Table 2). Most alternative explanations, such as personal affect, predict similar effects across dyads regardless of what ﬂows through them. By contrast, our propositions only predict interactions between the value of mutual acquaintances and resource scarcity for relations that actors use to mobilize resources. Consistent with our theory, we do not see signiﬁcant effects in dyads not used for resource mobilization (model 11). This result holds even when controlling for ego-level ﬁxed effects (model 12). To explain our observed results, then, any alternative story must not only

BJØRN LØVA˚S AND OLAV SORENSON

380

elucidate why we ﬁnd the positive effect between the value of a relationship and the interaction of mutual acquaintances and alter resource scarcity, but also explicate why we do not ﬁnd such an effect for other types of ties. One alternative account that does seem consistent with our dyad-level results is that mutual acquaintances facilitate resource mobilization through the conferral of status (Emerson, 1962; Flynn, 2003). Social systems frequently reward those that provide resources to needy parties without any direct expectation of reciprocity. Consider, for example, the honor that accrues to gift givers in Japan and other cultures (Benedict, 1946), or the ﬁerce competition in giving during the ‘‘potlatch’’ among Native American tribes in the Northwest (Mauss, 1925). Mutual acquaintances might enhance this effect by diffusing awareness of the givers’ actions more widely and more credibly. We nonetheless believe that the next validity check allows us to discount this possibility. Recall that we mentioned above that one shortcoming of the dyadic analysis is that it forces us to rely on a subjective dependent variable. One might therefore worry that perceptual biases inﬂuence our results. Though we cannot dismiss this concern directly, we do address it indirectly. We do so by regressing respondent performance – as reported by the people that evaluate our partners for bonuses – on a summary variable of mutual acquaintances and resource scarcity.12 We asked the managing partners heading each of the practices at ABC Consulting to rate respondents on a ﬁve-point scale as to their contribution to the proﬁtability of the practice. Because different managing partners provided information on disjoint sets of respondents and because they differed somewhat in their distributions of scores, we normalized their responses to z-scores (i.e. by subtracting out the mean and dividing by the standard deviation). We regressed the resulting score on a summary measure of mobilization capacity (mc): X X mc ¼ ln Sj spj þ ln Sj sbj (6) j2P

j2B

where P and B represent the sets of relations used respectively for project execution and business generation, S denotes the number of mutual acquaintances surrounding the relationship between ego and alter j, sp indicates the scarcity faced by j in resources for project execution, and sb denotes the scarcity faced by j on business generation. In essence, for each ego, it sums the dyad-level interaction terms across all project execution and business generation ties. The resulting variable has a mean of .01 and a standard deviation of .95.

381

The Mobilization of Scarce Resources

Table 6. OLS Estimates of Respondent Performance.

Tenure Mobilization capacity

(13)

(14)

.033 po.01 .105 po.06

.032 po.01 .127 po.03 .776 po.23 4.37 po .01 .15 76

Constraint F R2 N

5.76 po.01 .14 76

The results of this analysis appear in Table 6. Because we immediately noticed a strong relationship between tenure at the ﬁrm and these performance ratings, we included a control (tenure) for years at ABC Consulting. The estimates in model 13 suggest that consultants whose patterns of relationships should facilitate the mobilization of scarce resources perform better. Our results, moreover, do not stem simply from closure.13 Model 14 adds Burt’s (1992) constraint measure as a covariate. Not only does the effect of mobilization capacity remain robust, but also the standard constraint measure fails to show a signiﬁcant effect.

DISCUSSION We began by asking: What factors govern the decision to lend resources to another? Our answer is that resource mobilization – the ability to borrow on only the (possibly implicit) promise of reciprocation – depends jointly on two factors: the number of mutual acquaintances connecting the lender and borrower, and the scarcity of the resources (potentially) being exchanged between them. Intuitively, as scarcity rises, the borrower faces increasingly strong incentives not to repay. The resource holder, recognizing this temptation, therefore refuses to lend unless something mitigates these incentives. We contend that mutual acquaintances increase the cost of reneging (and therefore enable exchange) through three mechanisms: (i) they improve monitoring, thereby raising the odds of being punished; (ii) they raise the expected severity of punishment by allowing a coordinated response; and (iii) they allow prospective lenders to assess other factors that limit the borrower’s likelihood of not repaying. Qualitative and quantitative

382

BJØRN LØVA˚S AND OLAV SORENSON

data from a sample of senior partners at a large consultancy corroborate this thesis. Our ﬁndings echo much of the literature on network ‘‘closure’’ – where an actor’s contacts also have a high density of connections to each other (Coleman, 1990). Ingram and Roberts (2000), Reagans and Zuckerman (2001), and Reagans and McEvily (2003), for example, all ﬁnd beneﬁts to closure, primarily through the sharing of information. Our research nonetheless focuses on a distinct set of mechanisms. In much of the literature on closure, the advantages accrue to the sublimation of individual interests. Reagans and Zuckerman (2001, p. 503), for example, discuss how network closure facilitates ‘‘the joining of individual interests for the pursuit of common initiatives.’’ Resource mobilization, by contrast, only becomes problematic because actors remain self-interested (and therefore do not cooperate). Mutual acquaintances can nonetheless mitigate these incentives by imposing individual-level costs on behaving badly. Our research differs from earlier studies on at least two dimensions. On the one hand, our study moves in new directions by allowing the effects of relations to depend on their context. In this respect, we follow a similar line to Burt (this volume) and Venkatraman (this volume). Burt (this volume) argues that positions may have differing effects across levels of analysis, while Venkatraman (this volume) provides evidence that the consequences of positions may vary as a function of the strategies that generated them (see also Ryall & Sorenson, 2007). We meanwhile show that, even for a particular dyad, the value of a relationship depends on the local supply of and demand for the resources moving through the network. The value of a relationship consequently varies over time. Aggregation to the actor level can mask these differences. On the other hand, whereas much of the focus has been on the movement of information (e.g., Burt, 1992; Hansen, 1999; Reagans & McEvily, 2003), our study concerns the lending and borrowing of rival goods, resources with two properties: (i) the lender faces an opportunity cost to lending them (generally because she can then not use them herself) and (ii) the resources themselves can exist in insufﬁcient quantities. Information has neither of these properties – one can replicate it with little cost and pass it along without losing the ability to use it. As a result, studies of information ﬂow do not reveal the important moderating role of resource scarcity on optimal patterns of relationships. The fact that the same relational topologies may not maximize both information gathering and resource mobilization seems interesting with respect to at least two practical applications. Consider ﬁrst the nascent

The Mobilization of Scarce Resources

383

entrepreneur. Individuals beginning new ventures must convince investors to devote ﬁnancial capital, employees to invest time, and buyers and suppliers to exchange goods and services, all on the promise of return. Those best positioned to identify opportunities may ﬁnd it difﬁcult to assemble the resources necessary to exploit those opportunities, precisely because expansive and sparse sets of relations aide in the former while denser interconnections facilitate the latter. CEOs and other executives at the highest levels engage in similar chores – negotiating access to external resources, assembling political support within the ﬁrm, etc. To the extent that individuals with diverse relations climb the hierarchy fastest (Burt, 1992; Podolny & Baron, 1997), those that reach the top may then ﬁnd themselves with patterns of relations poorly suited to their responsibilities.

NOTES 1. We have assigned a pseudonym to the ﬁrm to protect the conﬁdentiality of the data. 2. One should not consider these categories exhaustive. Because of our theoretical interest in resource mobilization, we have excluded from our discussion some purposes served by these relations (e.g., information on internal politics). 3. Since the model can accommodate sales of different size by considering them aggregates of single unit projects, this assumption does not limit the generality of our setup. 4. In reality, partners neither own nor completely control the fates of those that report to them. In the interest of simplicity, however, our depiction of this process omits the cajoling and negotiation in which partners frequently engage to implement these loans. Of course, as the third example at the beginning of this section suggests, mutual acquaintances may also facilitate these promises between partners and their subordinates. 5. Though the results hold beyond this assumption, it nonetheless allows us to simplify the presentation of Bob’s incentives. Under aoXB2 YB2, XB2 WYB2 a implies XB2 W YB2 (1 b)a. 6. These mechanisms differ from the ‘‘third-party trust’’ discussed by Coleman (1990) because mutual acquaintances here do not act as guarantors of repayment ready to cover the debt should the borrower renege. 7. These questions reﬂect the language that partners used to discuss business generation and project execution in the qualitative interviews. They clearly had a less precise delineation of the kinds of exchange than assumed in our propositions. To the extent that these name generators also capture other types of relations, however, it should add noise to our estimates and therefore increase the conservativeness of our results. 8. The pilot survey suggested that 24 spaces would allow us to avoid truncation in the naming of relationships. Indeed, only ﬁve respondents ﬁlled the available spaces.

BJØRN LØVA˚S AND OLAV SORENSON

384

9. A linear spline estimating the relationship between the number of mutual acquaintances and relationship importance suggested a declining marginal effect. Experiments with different functional forms revealed that the log and square root transformations offered the best ﬁts to the data. 10. Because the measure increases with the scarcity of resources for project execution, we invert it (i.e. multiply it by 1) for tests that require a measure of the scarcity of business generation resources. 11. Some have suggested that actor ﬁxed effects can account for autocorrelation (e.g., Mizruchi, 1989), but separate intercepts do not correct for the fact that alters connected within the same ego might correlate on unobserved dimensions when connected to one another. 12. Gathering third-party evaluations of alter importance is not practical. Given the size and dispersion of partners’ networks as well as the conﬁdentiality surrounding many of these relations, no one other than the focal partner can assess the relative importance of each relation to his success. 13. We use the standard measure for binary relationships:

Constrainti

X j

rij þ

X

! rik rkj

k

where r denotes whether a relationship exists between two actors. Partners constraint levels ranged from .02 to .47 with a mean of .17 and a standard deviation of .09. 14. Though substantially larger than networks in most prior research, the size seems consistent with that of other high-level managers; Burt (2004), for example, ﬁnds that vice presidents have two to four times as many contacts as middle managers.

ACKNOWLEDGMENTS London Business School ﬁnancially supported this research. We thank Louise Mors and Michelle Rogan for their assistance with the data, and Matt Bothner, Ron Burt, Yves Doz, Constan¸ca Esteves, Martin Gargiulo, Susan Lynch, Bill McEvily, Mikolaj Piskorski, Michelle Rogan, Michael Ryall, and Brian Uzzi for their advice in developing this chapter.

REFERENCES Benedict, R. (1946). The Chrysanthemum and the sword: Patterns of Japanese culture. Boston: Houghton Mifﬂin. Blau, P. M. (1964). Exchange and power in social life. New York: Wiley.

The Mobilization of Scarce Resources

385

Bott, E. (1957). Family and social network: Roles, norms, and external relationships in ordinary urban families. London: Tavistock. Burt, R. S. (1992). Structural holes. Cambridge, MA: Harvard University Press. Burt, R. S. (2004). Structural holes and good ideas. American Journal of Sociology, 110, 349–399. Burt, R. S. (2008). Returns to secondhand brokerage in industry networks: Spillover effects on price cost margins in American manufacturing. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 315–360). Bingley, UK: JAI/Emerald Group. Burt, R. S., & Knez, M. (1995). Kinds of third-party effects on trust. Rationality and Society, 7, 255–292. Carroll, G. R., & Hannan, M. T. (2000). The demography of organizations and industries. Princeton, NJ: Princeton University Press. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Harvard University Press. Durkheim, E. (1957). Professional ethics and civil morals. London: Routledge. Efron, B. (1982). The jackknife, the bootstrap and other resampling plans. Philadelphia: Society for Industrial and Applied Mathematics. Emerson, R. M. (1962). Power-dependence relations. American Sociological Review, 27, 31–41. Fernandez, R. M., Castilla, E. J., & Moore, P. (2000). Social capital at work: Networks and employment at a phone center. American Journal of Sociology, 105, 1288–1356. Flynn, F. J. (2003). How much should I help and how often? The effects of generosity and frequency of favor exchange on social status and productivity. Academy of Management Journal, 46, 539–553. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380. Granovetter, M. S. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Hansen, M. T. (1999). The search-transfer problem: The role of weak ties in sharing knowledge across organization subunits. Administrative Science Quarterly, 44, 82–111. Ingram, P., & Roberts, P. W. (2000). Friendships among competitors in the Sydney hotel industry. American Journal of Sociology, 106, 387–423. Johnson, E. (2004). Ordered logit with ﬁxed effects. Working paper, Stanford University. Krackhardt, D. (1988). Predicting with networks: Non-parametric multiple regression analyses of dyadic data. Social Networks, 10, 359–382. Kreps, D. M., & Wilson, R. (1982). Reputation and imperfect information. Journal of Economic Theory, 27, 253–279. Maister, D. H. (1993). Managing the professional services ﬁrm. New York: Free Press. Mauss, M. (1925). The gift: Forms and functions of exchange in archaic societies. New York: W.W. Norton & Co. McCullagh, P. (1980). Regression models for ordinal data (with discussion). Journal of the Royal Statistical Society, (Series B), 42, 109–142 Mizruchi, M. S. (1989). Similarity of political behavior among large American corporations. American Journal of Sociology, 95, 401–424. Padgett, J. F., & Ansell, C. K. (1993). Robust action and the rise of the medici, 1400–1434. American Journal of Sociology, 98, 1259–1319. Podolny, J. M., & Baron, J. N. (1997). Resources and relationships: Social networks and mobility in the workplace. American Sociological Review, 62, 673–693.

386

BJØRN LØVA˚S AND OLAV SORENSON

Raub, W., & Weesie, J. (1990). Reputation and efﬁciency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Reagans, R., & McEvily, B. (2003). Network structure and knowledge transfer: The effects of cohesion and range. Administrative Science Quarterly, 48, 240–267. Reagans, R., & Zuckerman, E. (2001). Networks, diversity and performance: The social capital of R &D units. Organization Science, 12, 502–517. Ryall, M. D., & Sorenson, O. (2007). Brokers and competitive advantage. Management Science, 53, 566–583. Sorenson, O., & Stuart, T. E. (2001). Syndication networks and the spatial distribution of venture capital investments. American Journal of Sociology, 106, 1546–1588. Sorenson, O., & Waguespack, D. M. (2006). Social structure and exchange: Self-conﬁrming dynamics in Hollywood. Administrative Science Quarterly, 51, 560–589. Uzzi, B. (1996). The sources and consequences of embeddedness for economic performance of organizations. American Sociological Review, 61, 674–698.

APPENDIX Several aspects of ABC Consulting and the roles and responsibilities of its senior partners made it an ideal setting for investigating the theoretical arguments developed above: Prior research has identiﬁed the importance of social networks to the operation of professional service ﬁrms (Maister, 1993). By focusing on the senior partners at the top echelon of the ﬁrm, our study concentrated on individuals for whom resource mobilization accounted for a large portion of their job responsibilities. Finally, focusing on a single ﬁrm allowed us to control for many factors other than social networks that affect the variables of interest. Just as focusing on a single industry leads to more precise statistical estimates when studying ﬁrms (see Carroll & Hannan, 2000, pp. 85–99, for a discussion), studying individuals in the context of a single ﬁrm controls for ﬁrm-level heterogeneity.

Interviews The managing partners for strategy and for global operations, who sponsored the project, selected 38 individuals – representing a variety of industries and functional specialties – in ﬁve countries in Western Europe for our interviews. The ﬁrst author met with 32 of them, for a total of nearly 50 hours (see Table 7 for details); due to either illness or scheduling problems, six of them could not meet during the three months allocated to interviewing. To facilitate analysis, the interviewers taped and transcribed all except one of these sessions.

387

The Mobilization of Scarce Resources

Table 7. Partner

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

Overview of Interviews.

Ofﬁce

Industry Group

Competency

Firm Tenure (Years)

Interview Duration (Hours)

London London London London London London London London London London London London Frankfurt Frankfurt Frankfurt Madrid Madrid Madrid Madrid Milan Milan Milan Milan Milan Milan Paris Paris Paris Paris Paris Paris Paris

Consumer products Consumer products Financial services Financial services Financial services Financial services Government High tech High tech Natural resources Natural resources Natural resources Consumer products Financial services High tech Consumer products Financial services High tech Natural resources Consumer products Financial services Financial services Financial services High tech Natural resources Consumer products Financial services Financial services High tech High tech High tech Natural resources

Strategy Technology Change Process Strategy Strategy Process Process Strategy Change Strategy Technology Technology Change Technology Strategy Change Technology Strategy Process Process Strategy Technology Technology Technology Process Process Strategy Process Process Technology Process

24 23 N/A 20 21 N/A N/A 25 20 20 9 28 18 17 20 22 23 12 15 20 N/A 13 15 15 12 8 26 15 23 30 15 22

1:58 1:05 1:13 1:56 1:00 3:28 1:12 1:50 0:51 1:03 0:51 1:23 1:09 1:32 1:11 1:20 1:26 1:38 5:15 1:07 1:06 1:38 1:09 1:19 2:13 1:13 1:35 2:13 1:15 1:00 1:14 1:20

Each interview began with an unstructured discussion of the challenges that the partner faced and how he felt his social networks enabled and constrained him in meeting those challenges. Participants had clearly considered the issue consciously before the interviews, as many spoke at length, revealing considerable sophistication in their perceptions of how networks operated. Following this ﬁrst stage, the interview proceeded into a semi-structured discussion of how both internal and external networks

388

BJØRN LØVA˚S AND OLAV SORENSON

helped and hindered the interviewee in the performance of his job, how he built and maintained his networks, and ﬁnally how he felt the formal structure of the ﬁrm affected both of these issues.

Survey Because the interviews revealed that partners maintained relations with both fellow employees in other ofﬁces (i.e. other countries) and to actors outside the ﬁrm, we collected network information with an egocentric strategy, building on the methodology developed by Burt (for a detailed discussion, see 1992, pp. 121–125). Partners’ diverse relations made it impractical to develop a list of all possible contacts for each respondent to consider. The ﬁrm itself had 55,000 individuals that a respondent might identify as important contacts, and the inclusion of external parties made such a list essentially boundless. We conducted the ﬁrst pilot by sending questionnaires to six senior partners through ABC Consulting’s internal mail. This pilot revealed a problem: Partners’ large networks (18 ties on average) made answering the questions regarding alters time consuming.14 Because a second pilot suggested that having a researcher present would improve the response rate, we administered the surveys in face-to-face sessions. Even though researchers answered only clarifying questions, their presence and the prebooked slot in appointment calendars prevented respondents from being disturbed while completing the survey. Since surveying a random sample across ABC Consulting would have required ﬂying researchers to more than 50 locations, exceeding the project’s budget and available manpower, we adopted a cluster sampling design, with senior partners selected at random from ten major ofﬁces: San Francisco, Chicago, New York, London, Paris, Frankfurt, Milan, Madrid, Tokyo, and Sydney. The managing partner in the CEO’s ofﬁce sponsored the survey. Each potential respondent received three communications soliciting their participation. First, the global managing partner sent each an e-mail (i) explaining the survey’s objective (to learn more about how the ﬁrm worked), (ii) indicating that ABC Consulting would not have access to individual responses, and (iii) emphasizing the importance of their participation. Next, the managing partners of the 20 competency and market units called the sampled partners that reported to them soliciting their participation. Finally, the researchers scheduled one hour for the survey. Four partners did not respond because they left the ﬁrm between the

The Mobilization of Scarce Resources

389

sample selection and surveying stages. In total, the researchers scheduled meetings with 133 partners, resulting in 102 completed surveys (a 71% response rate). Tests comparing the 133 sampled to the 102 completed surveys did not reveal any signiﬁcant differences in means on the available demographic information (e.g., ofﬁce, practice, tenure with ﬁrm). Moreover, since variation comes from differences across contacts for a particular individual (the results remain robust to respondent ﬁxed effects), selection effects cannot account for our results.

INTERCONNECT TO WIN: THE JOINT EFFECTS OF BUSINESS STRATEGY AND NETWORK POSITIONS ON THE PERFORMANCE OF SOFTWARE FIRMS N. Venkatraman, Chi-Hyon Lee and Bala Iyer ABSTRACT We develop and test a model of how a software ﬁrm’s business strategy (product scope and market scope) interacts with the ﬁrm’s network position (alliance degree and structural holes) to impact performance. We test the joint-effects hypotheses on a sample 359 packaged software ﬁrms that have entered into 5,489 alliances involving 2,849 distinct ﬁrms during the time period, 1990–2002. While prior studies have demonstrated the importance of network positions as a determinant of ﬁrm strategy and performance, this chapter begins to examine the performance effects of how a ﬁrm’s business strategy and network positions interact. We ﬁnd support for three of the four hypotheses lending empirical support for our theoretical model. We develop implications for network-based perspectives of strategy and outline areas for further research. Network Strategy Advances in Strategic Management, Volume 25, 391–424 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25011-4

391

392

N. VENKATRAMAN ET AL.

INTRODUCTION Management research is concerned with the sources and drivers of competitive advantage from different perspectives. In a stylized fashion, these sources can be separated into those that exist within a ﬁrm (e.g., resources and capabilities) and those that are appropriated through a ﬁrm’s position in the industry’s local external environment (e.g., position relative to suppliers, buyers, and complementors). Studies emphasizing internal sources subscribe to a resource-based view of the ﬁrm (e.g., Barney, 1991; Conner & Prahalad, 1996). Asymmetrically distributed heterogeneous ﬁrm resources and capabilities may result in the creation and persistence of material ﬁrm differences and competitive advantage. Studies emphasizing external sources follow industrial organization economics (Porter, 1980). Industry structure (e.g., industry concentration and market power vis-a`-vis up- and down-stream industries) has a potential impact on industry and ﬁrm performance (Scherer & Ross, 1990). A recent stream of work focuses on how a ﬁrm’s position in a network of external relationships – that for a given ﬁrm forms the business ecosystem – impact ﬁrm performance (Gulati, 1999; Gulati, Nohria, & Zaheer, 2000). A business ecosystem consists of ﬁrms embedded in formalized, signiﬁcant, and enduring interﬁrm relationships (Gulati & Gargiulo, 1999). The business ecosystem permits a focal ﬁrm to access resources and capabilities external but local to the ﬁrm (Gulati et al., 2000). The asset, information, and status ﬂows in the ecosystem have the potential, when combined with the focal ﬁrm’s resources, to positively impact ﬁrm performance (Gulati, 1999; Gnyawali & Madhavan, 2001; Oliver, 1990; Podolny, 2001). For example, Rowley, Baum, Shipilov, Greve, and Rao (2004) found that a ﬁrm’s clique structure signiﬁcantly impacted ﬁrm performance for a sample of Canadian investment banks. Ahuja (2000) found that a ﬁrm’s direct and indirect ties, and structural holes signiﬁcantly impacted its innovative output for a sample of international chemical companies. Lee (2007) found that ecosystem structure impacted the time required for a telecommunications ﬁrm to enter a market. Interestingly, Madhavan, Caner, Prescott, and Koka (2008) argue that many of the studies that examine the relationship between a ﬁrm’s ecosystem and performance adopt a structuralist view – the structure of the business ecosystem and a ﬁrm’s position are dominant drivers of ﬁrm performance. In such a perspective, ﬁrm characteristics are assumed to be constant among ﬁrms (Koka, Prescott, & Canella, 2007; Madhavan et al., 2008). Recent studies are beginning to challenge a strictly structuralist

Interconnect to Win

393

emphasis by arguing that ﬁrm resources (i.e., internal view) and the business ecosystem (i.e., local external view) together inﬂuence ﬁrm performance. For example, Zaheer and Bell (2005) concluded that a ﬁrm’s performance can be explained by both internal innovative capabilities and network structure. Madhavan et al. (2008) and Tsai (2001) both concluded that a ﬁrm’s performance can be explained by its absorptive capacity and network position. Lavie (2007) demonstrated that a ﬁrm’s resources and network structure both inﬂuenced ﬁrm performance. Echols and Tsai (2005) found that network embeddedness moderated the niche – performance relationship for a sample of venture capital ﬁrms. Network structure is endogenous with respect to ﬁrm attributes – namely resources and capabilities. Our study is in the spirit of Madhavan et al.’s (2008) call to ‘‘‘bring the ﬁrm back in’ to network strategy (pp. 457–502)’’ not only for greater conceptual clarity but also because of its importance within the packaged software sector that we study here. First, the software industry is characterized by a shift from vertical integration to horizontal layers dominated by specialist ﬁrms that form relationships across these layers. These horizontal layers have been variously referred to as modular clusters (Baldwin & Clark, 2000) or stacks (Gao & Iyer, 2006) that represent divided technical leadership (Bresnahan, 1998). In the packaged software ecosystem, ‘‘there is no single vertically integrated ﬁrm with control over direction of a platform’’ (Bresnahan, 1998, p. 13). Second, the packaged software industry is prototypical of networked within-industry markets that call for simultaneous ﬁrm cooperation and competition. Thus, every software ﬁrm faces the challenge of ensuring product interoperability with another ﬁrms’ products. Interoperability is critical because no single ﬁrm provides all the software required by users. The value to end customers is based on coordinated product launches of complementary products that work seamlessly (Shapiro & Varian, 1998). Thus, this sector is prototypical of new modes of competition, where ‘‘co-opetition, in which ﬁrms selling products which are complements compete with one other, is prevalent’’ (Bresnahan, 1998, p. 2). Take the case of an alliance announced between SUN and Microsoft. They compete in the operating system market but have agreed to create interoperability between their products because of the perceived threat from the Linux operating system. Thus, specialized ﬁrms need to interconnect with ﬁrms to deliver superior value to end customers. Indeed, success in the software industry calls for different modes of strategy that recognizes intense and simultaneous competition and cooperation (Bresnahan, 1998; Bresnahan & Greenstein, 1999).

394

N. VENKATRAMAN ET AL.

One way to do this is to have common ‘‘agreements’’ and coordination requirements for bringing disparate systems to communicate and exchange data digitally with one another (Gates, 2002). These agreements serve to create interoperability among ﬁrms’ diverse and heterogeneous software products. Beyond such implicit contracts and agreements, ﬁrms enter into alliances and relationships to formally ensure that their products interoperate and are deployed with other products and services in customers’ (i.e., end users) IT operations. Several descriptions of software ecosystems exist (Campbell-Kelly, 2003; Cusumano, 2004; Messerschmitt & Szyperski, 2003) but there has been no systematic empirical test of whether ecosystems create distinct competitive advantage. Moreover, the software sector – despite its inherent need for dynamic ecosystems – has not yet been studied for the performance effects of networks of relationships. No study has looked at the interaction of ﬁrm business strategy and network positions on ﬁrm performance in general and the software sector in particular. To this end, we study 359 software ﬁrms over a 13-year period (1990–2002) to test if there are signiﬁcant performance effects of how business strategy and network positions interact. We believe that a study of software ecosystems with intense competition and cooperation could provide insights for other settings that are increasingly becoming digital and knowledge intensive such as ﬁnancial services, retail, entertainment, and healthcare.

OUR RESEARCH MODEL AND HYPOTHESES Our model is based on a key premise: success in the software sector depends on how a ﬁrm’s business strategy interacts with its network positions. We subscribe to the view that internal activities and interorganizational relationships are not substitutes but complements (e.g., Mowery & Rosenberg, 1989; Powell, Koput, & Smith-Doerr, 1996). A ﬁrm’s network can be a resource of inimitable and non-substitutable value (Gulati, 1999) and that such resources lie both inside the ﬁrm and also in the network (Afuah, 2000; Powell, White, Koput, & Owen-Smith, 2005). Such hybrid governance with intermediate degree of administrative control and coordinated adaptation sometimes combine the beneﬁts of pure forms of governance – namely: markets and hierarchies (Williamson, 1991). Our thesis is that a ﬁrm’s performance is inﬂuenced not just by the environmental characteristics (market structure) and ﬁrm characteristics (resource deployments) but also by how a ﬁrm is positioned in the business

395

Interconnect to Win

Fig. 1.

Research Model.

network to obtain complementary resources for strategy execution. We develop a stylized time-sequenced model of how the ﬁrm–network interactions drive performance. We argue that a ﬁrm’s business strategy at time t0 interacts with its network position at time period t1 to impact performance at time t2. Fig. 1 is a schematic representation of the research model. We discuss the rationale for our theoretical model and derive our hypotheses below.

Business Strategy A ﬁrm’s business strategy can be conceptualized in multiple ways using different typologies or dimensions. In this study, we focus on two dimensions that capture business strategy – product scope and market scope. Software ﬁrms often design and deploy multiple products for different market segments. The products and the markets served by a ﬁrm represent its business scope that evolves over time reﬂecting a dynamic view of business strategy. Product Scope Although results from the extensive research stream on ﬁrm diversiﬁcation are far from conclusive, evidence suggests a positive relationship between product relatedness and ﬁrm performance (Palich, Cardinal, & Miller, 2000). In fact, recent ‘‘within-industry’’ studies in particular suggest a positive association between relatedness and superior ﬁrm performance (Cottrell & Nault, 2004; Li & Greenwood, 2004; Stern & Henderson, 2004). The resource-based view of diversiﬁcation posits that ﬁrms enter into multiple product lines to exploit excess productive capacity of their critical resources (Farjoun, 1994; Markides & Williamson, 1994; Robins &

396

N. VENKATRAMAN ET AL.

Wiersema, 1995). Relatedness or the use of common resources across multiple product lines within a single business creates synergies in the form of economies of scope (Davis & Thomas, 1993). Related within-industry diversiﬁcation positively impacts ﬁrm survival (Cottrell & Nault, 2004; Stern & Henderson, 2004), product survival (Cottrell & Nault, 2004), return on assets (Li & Greenwood, 2004), and growth (Tanriverdi & Lee, 2008). Within the packaged software sector, product relatedness has special signiﬁcance. Software is an intangible knowledge-intensive experience good. Product relatedness is associated with software code (e.g., designs, routines and know-how, and components) reuse and portability (Tanriverdi & Lee, 2008). Software code reuse, speciﬁcally, across multiple products decreases new product development time by simplifying and managing product complexities (Ulrich, 1995), allows the ﬁrm to rapidly address new market opportunities (Meyer & Lehnerd, 1997), and increases product compatibility, interoperability, and quality (Boehm, 1981; Baldwin & Clark, 2000). Knowledge and capabilities gained in the process of developing one product can further be used as a catalyst to improve the quality of other products (Markides & Williamson, 1994). Recently, Tanriverdi and Lee (2008) found a positive relationship between a ﬁrm’s software product relatedness and ﬁrm performance. The concept of modular architectures has been used to explain outcomes such as software reuse, compatibility, interoperability, and quality (Baldwin & Clark, 2000; Simon, 1996; Ulrich, 1995; Ulrich & Eppinger, 1999). At the core of a modular architecture are the ﬁrm’s products or modules. Module independence (e.g., the design of a module can be partially decoupled from the design of a different module) and interdependence (e.g., modules interoperating) create synergistic value to both the ﬁrm and the ultimate users (Baldwin & Clark, 1997; Schilling, 2000; Sanchez & Mahoney, 1996). A high level of modularity is associated with a one-to-one mapping between functions and modules (Ulrich, 1995) or code reuse. The seemingly dual paradoxical objectives of independence and interdependence are possible because modular architectures partition module information into hidden and visible sets of interfaces (Baldwin & Clark, 2000). The visible information is often denoted as the application program interfaces (APIs). Because a module’s hidden information is opaque to other modules, they need only be concerned with the visible information thus decreasing the complexity of a ﬁrm’s software products. A decrease in the complexity of the products positively impacts performance. Moreover, modular architectures facilitate cooperation between ﬁrms (Schilling, 2000; Schilling & Steensma, 2001). Higher interoperability and quality, for example, within a ﬁrm’s

Interconnect to Win

397

products can extend to higher interoperability and quality among alliance partners’ software products. Modular architectures decrease the costs associated with making products interoperate. Market Scope While the ﬁrst dimension, product scope, captured the logic of the portfolio using the lens of within-industry relatedness, market scope captures the scope of markets targeted by the product portfolio in the competitive marketplace. If the ﬁrst dimension reﬂected the logic of resource allocations to products within a portfolio, this dimension reﬂects the realized market acceptance of the portfolio. We capture market scope in the software market through the collective installed base of software products in the portfolio. The role of installed base and its implications for business strategy have been convincingly established in extant research. The classic treatise on system-based competition and network effects is based on the central notions of installed base (Katz & Shapiro, 1994). When interoperability is important and no single ﬁrm is able to provide all the software required by users, the installed base can impact a ﬁrm’s business strategy and performance (Farrell & Saloner, 1985, 1986). Empirically, Brynjolfsson and Kemerer (1996) demonstrated that network externalities, as measured by the size of a product’s installed base signiﬁcantly increased the price of the spreadsheet product and subsequent ﬁrm performance. Kauffman, McAndrews, and Wang (2000) found that the size of a bank’s customers (i.e., local install base) decreased the banks time-to-entry into ATM alliances. Installed base is also central to ‘‘standards war’’ and battles for market dominance amongst a set of plausible competing software standards (Farrell & Saloner, 1986; Shapiro & Varian, 1998). The ultimate winner in such battles may be the one with a higher installed base (Schilling, 2002; Gallagher & Park, 2002). We extend the concept of installed base that has been most commonly used at the level of a single product line to the level of business strategy. Given interoperable and complementary products, the install bases of the products are positively interdependent. A single software product cannot fulﬁll all the needs of users. A single ﬁrm’s products are not present in all product categories (i.e., horizontal layers or modular clusters). Not only are the install bases of the products interdependent, this interdependency may be reinforced over time due to path dependencies (Tanriverdi & Lee, 2008; Venkatraman & Lee, 2004). Firms, that have products serving markets with small install bases, will be at a disadvantage to ﬁrms that have products serving markets with large install bases. By creating a weighted

398

N. VENKATRAMAN ET AL.

proﬁle of the different products across different market segments covered, this concept captures the richness in market scope choices exercised by ﬁrms over time.

Network Positions Strategy researchers have begun to examine a variety of network characteristics in terms of their likely impact on strategic choice and ﬁrm performance. We use two characteristics of the network that have been commonly examined in the network-perspectives of strategy: (1) alliance degree and (2) structural holes. Alliance Degree Alliance degree has been used to understand the degree of direct and indirect network ties as well as the attractiveness of the focal ﬁrm to link with a broad set of ﬁrms within a business ecosystem. Firms link with other ﬁrms for exchanging, sharing, or joint development of products, technologies, services, or other information and knowledge resources (Gulati, 1995). Such links are representations of ‘‘not only pipes carrying the stuff of the market (information about exchange opportunities as well as actual goods, services and payments); they are prisms, splitting out and inducing differentiation among actors y’’ (Podolny, 2001, p. 35). Presence or absence of a tie between two ﬁrms is an informational cue on which others rely to make inferences about the quality of one or more of these actors and their resources and capabilities (Venkatraman & Lee, 2004). Alliance degree is thus an important network positional characteristic. Alliances formed within the software sector act as channels and conduits for accessing product and informational resources through direct and indirect ties (Ahuja, 2000). A ﬁrm’s partners could potentially bring not only the resources that lie internally but also their knowledge and experience from their interactions with their partners in the network. Thus, selecting the right set of partners to link with is crucial to maximize access to the requisite resources while minimizing the cost of coordination of a wide array of alliances. Moreover, since each link represents mutual commitment by the ﬁrms to form the relationship, it is qualitatively different from links in other networks such as cross-citation of papers or patents or hyper-linking of documents, which could be done unilaterally. Thus, a ﬁrm’s set of relationships reﬂects its strategic intent to form purposive relationships and to create an advantageous position in a network of relationships.

Interconnect to Win

399

The alliance degree serves as ways to access complementary resources in the network to maximize performance.

Structural Holes In general, ﬁrms maximize their access to unique information and rare resources from complementary entities. They do so by coordinating their activities to make their products interoperate to increase the joint value of their products to customers. When ﬁrms form relationships to further these coordination efforts, a network of ties is formed because focal ﬁrms’ partners may also have ties among themselves giving rise to ties within networks. Firms occupy network positions that allow for bridging structural holes – gaps between ﬁrms not otherwise connected – to realize advantages due to superior resource access, information, and control (Burt, 1992). Structural holes represent gaps in resource ﬂows (e.g., information and products) between alters linked to the same focal ﬁrm. Not all ﬁrms have equal opportunities – ﬁrms with distinctive, valuable resources have clearly richer and far-reaching collaboration opportunities (Ahuja, 2000). When ﬁrms positioned in networks are able to access mutually unconnected partners in a complex network, they are likely to access distinct set of resources. Burt’s (1992) thesis that the existence of structural holes spanned is a central feature of an efﬁcient resource-rich network is supported by extensive empirical evidence in intra- and interorganizational networks (Walker, Kogut, & Shan, 1997; Gulati, 1995; McEvily & Zaheer, 1999; Uzzi, 1997). Lower constraint levels may also provide ﬁrms with more opportunities for forming new ties. Structural holes have additional relevance in software due to the inherent need for interoperability of underlying parts of the system that is composed of hardware, software and communication components. This has given rise to a new category of software termed as middleware – software that connects two otherwise separate applications (e.g., linking a database system to a Web server to allows users to request data from the database using forms displayed on a Web browser). In recent years, specialist ﬁrms such as BEA Systems, TIBCO software, Webmethods, Vitria, and others have emerged primarily to bridge otherwise disconnected parts of the software network. Software companies seek to form links with such ﬁrms and broaden the appeal of their products. Moreover, since middleware is used to interoperate across multiple operating systems such as Linux, Unix, and Windows, such companies are suited for bridging structural holes.

400

N. VENKATRAMAN ET AL.

Beyond middleware as a case, under fast-changing conditions, common linking partners serve as referral agents and relay expectations and obligations while bringing together parties to encourage cooperation and coordination of products and services. These ﬁrms positioned to span structural holes could also encourage cooperation and reciprocity that otherwise may limit effective interoperability of software with related components within a system-based competitive setting (Shapiro & Varian, 1998). In contrast, some empirical evidence suggests that network closure (Coleman, 1988) may lead to higher ﬁrm performance. A combination of direct and indirect ties gives rise to network closure or cohesion. Indirect ties – links between two existing alters of the ego – are key to network closure which facilitates trust and managerial sense-making, diminishes risks of opportunism, potentially enhances interﬁrm coordination (Coleman, 1988; Granovetter, 1985; Uzzi, 1997), allows richer information and knowledge exchanges, and impacts innovation (Ahuja, 2000) especially under fast-changing technological environments (Krackhardt & Stern, 1988; Rindﬂeisch & Moorman, 2001). Kogut and Walker (2001) found that the clustering coefﬁcient explained the rate of mergers and acquisitions amongst the ﬁrms. Thus, higher magnitude of embeddedness can potentially lower costs of coordination of interﬁrm linkages (Gulati & Singh, 1998) with potentially greater value realization (Kale, Dyer, & Singh, 2002). Dense and repeated links between two ﬁrms have the potential to increase the ‘‘capacity of the channel linking them’’ (Freeman, Borgatti, & White, 1991, p. 145) and relational capability (Dyer & Singh, 1998) which can result in more efﬁcient access to and leveraging of resources (Beckman & Haunschild, 2002; Koka & Prescott, 2002; Lee, 2007). Identiﬁcation of opportunities, development of complementary products, access to resources and competencies that they do not otherwise posses, and coordination of product development and marketing activities are all potential beneﬁts. Embedded alliances can increase the value of their joint products and services for end customers and lower their joint costs of production depending on the business landscape in which ﬁrms cooperate and compete (Baum & Singh, 1994).

The Joint Effects of Product Scope and Alliance Degree This hypothesis examines the performance impacts of the joint effects of product scope and alliance degree. While product scope or speciﬁcally,

Interconnect to Win

401

relatedness within a business enhances performance due to potential synergies in the form of economies of scope (Davis & Thomas, 1993), the number of alliances serves as the conduit for realizing the potential synergy. Relatedness reduces a ﬁrm’s product development and marketing costs through the redeployment of technical and marketing expertise, brands, and sales forces across multiple product lines (Capron & Hulland, 1999) and higher degree of alliances provide more opportunities to derive enhanced value from such expertise in knowledge-intensive settings (Powell et al., 2005). High levels of software product reuse among a set of products, also called platform strategies, system-based competition, or both, is a signiﬁcant source of resource-based synergies (Meyer & Seliger, 1998). Software businesses with higher levels of product relatedness can use alliances and linkages to cross-sell interoperable products to the same customer segments. More alliances allow for more product systems that are related within a tight-knit portfolio to be offered to more customers. For example, Microsoft’s success in the enterprise arena can be attributed to its ability to make available its suite of software to major companies in different industries through its impressive number of alliances and partnerships (Iyer, Lee, & Venkatraman, 2006) H1. The joint effect of product scope and alliance degree on software ﬁrm performance is positive and signiﬁcant.

The Joint Effects of Product Scope and Structural Holes This hypothesis examines the performance impacts of the interactions between product scope and structural holes reﬂecting a thesis that software ﬁrms with a related product portfolio will ﬁnd greater value when bridging structural holes because it expands the base of interoperability of the products. In general, ﬁrms maximize their access to rare resources from complementary entities. Burt (1997) argued that a network structure that is rich in structural holes positively inﬂuences ﬁrm performance because holes permits companies on either side to access unique resources (e.g., Rowley, Behrens, & Krackhardt, 2000). They do so by coordinating their activities to make their products interoperate to increase the joint value of their products to customers. A focal ﬁrm’s products, if they interoperate with another ﬁrm’s products, are more likely to fulﬁll the multiple needs of ultimate customers. An increase in the interoperability of ﬁrms’ products also helps to establish standards and protocols (Shapiro & Varian, 1998).

402

N. VENKATRAMAN ET AL.

An example from the software sector can be used to motivate this hypothesis. SAP sells products such as enterprise resource planning (ERP), supply chain management (SCM), customer relationship management (CRM), and product life cycle management to large and mid-size companies. Since customers would like to focus on the information aspects of the suppliers, customers and their own operations, they tend to use these products in unison. Independently, within the open-source community, ERP, SCM and CRM products were being developed around the MySQL database management system. When customers used SAP and its partners they were separated from the open-source ERP products. These two islands of products were not connected until 2003 when SAP decided to join the MySQL ecosystem by bridging the structural hole. SAP announced a collaboration relationship with MySQL AB to interoperate their database product with MySQL. With this move, customers will be able to mix and match open-source ERP products with SAP’s. If SAP had not announced this business linkage, customers will either abandon mixing and matching of ERP products or spend additional resources to make this integration. We argued previously and empirical evidence suggests that network closure (i.e., dense and cohesive network) facilitates trust and rich information ﬂows, diminishes risks of opportunism, potentially enhances interﬁrm coordination (Coleman, 1988; Granovetter, 1985; Uzzi, 1997), and impact innovation (Ahuja, 2000). In general, embeddedness could play constraining as well as enabling roles (e.g., White, 2002) as empirical research conﬁrms both positive and negative consequences in different contexts (Coleman, 1988; Granovetter, 1985; Uzzi, 1997). Soda, Zaheer, and Carlone (2008) found that high network closure resulted in more imitative ﬁrm behavior, under conditions of positive interdependence, with subsequent negative performance effects. In the software sector and for highly related (i.e., modular) software products in particular, the advantages associated with embeddedness decrease in salience. To reiterate, module information is partitioned into hidden and visible sets. First, only visible module information need be exchanged between ﬁrms thus decreasing the need for rich information ﬂows between ﬁrms. The visible information is only needed to create product interoperability (Baldwin & Clark, 2000). Second, the emphasis on visible information facilitates innovation and modular recombination (Sanchez & Mahoney, 1996; Schilling, 2000). The complexities associated with creating interoperability decrease. Third, a ﬁrm needs to only expose visible information to a potential rival therefore attenuating the likelihood of opportunism. The hidden module information is opaque to ﬁrms other than

Interconnect to Win

403

the focal ﬁrm. ‘‘Hiding’’ resources that are a basis for competitive advantage from potential rivals results in the products being more resistant to competitive erosion. Thus: H2. The joint effect of product scope and structural holes on software ﬁrm performance is positive and signiﬁcant.

The Joint Effects of Market Scope and Alliance Degree While the ﬁrst two hypotheses focused on the interaction effects of product scope and a ﬁrm’s network position, this hypothesis tests the joint effects of market scope and a ﬁrm’s network position – namely alliance degree. We subscribe to the views expressed by Shapiro and Varian (1998) that ‘‘the old industrial economy was driven by economies of scale; the new information economy is driven by economies of networks’’ (p. 14). We go further to argue that economies of networks is stronger for those companies that have wider, more broad-based market scope that attracts more companies willing and eager to form relationships within the web of interoperable products and services that make up the software sector. An increase in alliance degree signals the quality of a ﬁrm’s internal resources and capabilities. Barabasi (2002) found that a new web page will most likely link to popular web pages on the subject and a new scientiﬁc paper is likely to cite the more well-cited scientiﬁc papers in the ﬁeld creating ‘‘hubs.’’ A new node (i.e., focal) preferentially attaches to those nodes (i.e., alters) with higher ex ante links. Baum, Shipilov, and Rowley (2003) found that when a focal ﬁrm (from outside of a clique) seeks to link to a clique, it will do so with a ﬁrm (i.e., alter) within the clique that has the highest degree centrality and that ‘‘positions densely interconnected with partners y provide network advantage’’ (p. 721). The multiplicative effect is created due to the interaction of two powerful forces – market dominance due to the composite effect of the installed base of the product portfolio and alliance degree due to the preference of ﬁrms to be linked to dominant ﬁrms (e.g., Powell et al., 2005). Alliance degree is important to maintaining and increasing product install base. Moreover, when serving customers of one product through alliances, the ﬁrm has an opportunity to learn about the needs and preferences for other products. Thus, higher alliance degree also serves to obtain valuable information about product modiﬁcations and possible new product introductions that can be more readily actualized by a ﬁrm with more

404

N. VENKATRAMAN ET AL.

focused product portfolio. Customers gravitate to those product and markets with large network externalities given they are in the packaged software sector. The increase likelihood of encountering useful information (i.e., information from more sources) increases with an increase in the customer base and alliance degree. The associated learning is important for understanding and serving multiple product needs of the existing and evolving customer base (Farjoun, 1998). When SAP announced the launch of the Netweaver platform, they already had a large installed base of their products running within many enterprises across the world. With Netweaver, third parties could create products faster by reusing functions and development tool kits that were made available to them by SAP. Since they launched the platform with many application providers already in partnership with SAP, it made it very attractive for others to join in. SAP’s installed base and multitudes of products launched by their alliance partners, would make it very attractive for software ﬁrms like SmartOps with their inventory optimization solution to capitalize on it. Since most companies using the ERP solutions from SAP will have the transaction data to provide to the SmartOps product, they can market their solution to the installed base. The alliance partners on the Netweaver platform add value to the customers by providing related products. The higher the number of alliance partners, the more attractive is the platform to companies like SmartOps that have to be very careful with their allocation of resources. This hypothesis complements the ﬁrst hypothesis in the sense that alliances enhance the served-market by pooling complementary resources. Thus: H3. The joint effect of market scope and alliance degree on software ﬁrm performance is positive and signiﬁcant.

The Joint Effects of Market Scope and Structural Holes This hypothesis asserts that those software ﬁrms with a broader market scope are in a better position to leverage the structural holes and achieve superior performance. Software ﬁrms realize their market dominance through the degree of acceptance of their products by consumers. Within the stream on system-based competition, network externalities (Katz & Shapiro, 1985) leads to superior performance due to the interplay between direct and indirect network effects. As more customers use a software

Interconnect to Win

405

product, it will attract more complementors to create interoperable products, which will in turn make the focal software product more attractive to the consumers. While complementors can often create supporting products without any formal linkages, ﬁrms do use alliances and interﬁrm relationships to prime the network pump to design and deploy products and services that could accelerate market acceptance. Since not all ﬁrms can equally exploit the structural holes in the network, this hypothesis tests the value of the interaction between installed base of the product portfolio and structural holes. A ﬁrm occupying many structural holes will have an advantage over its competitors who rely on it to send and receive information and resources in the network of strategic alliances. Structural holes are particularly valuable under conditions of rapid change where ﬁrms strongly embedded with a small group of ﬁrms may ﬁnd themselves unable to adapt to the changing requirements. Not only did Lee (2007) ﬁnd that dense network may increasingly create lock-in effects, she found that spanning structural holes increases the quality of information which in turn, decreased ﬁrms’ time-to-entry into new markets. In the software sector, those ﬁrms spanning structural holes across diverse segments of the market will be in a better position to gather information about new developments and experiences in the use of different products and use such information to ﬁne-tune and adapt their product lines. Thus, structural holes will have a catalytic impact to those ﬁrms with high degree of product relatedness. Structural holes serve as referral agents and relay expectations and obligations while bringing together parties to encourage cooperation and coordination of products and services. Such network positions are particularly valuable to software ﬁrms with higher degree of product relatedness since they span structural holes to encourage cooperation and reciprocity that otherwise may limit effective interoperability of software with related components in settings characterized by system-based competition (Shapiro & Varian, 1998). Dynamic shifts in the software sector call for organizational routines that balance exploration and exploitation (March, 1991), which favor ﬁrms that occupy positions spanning structural holes. For example, Microsoft by virtue of its strong installed base in operating systems (Windows) and applications (Ofﬁce) is in a much better position to reap performance beneﬁts by closing the structural hole in the network than a ﬁrm such as Apple with a niche position in the network. When Microsoft

406

N. VENKATRAMAN ET AL.

announced that Microsoft’s Word will run on Apple’s operating system, they were trying to do two things. First, they wanted to join an ecosystem with a robust installed base that consisted of many independent software ﬁrms creating applications for the Windows and Ofﬁce environments. By being part of this ecosystem, Macintosh developers would be exposed to additional opportunities for application development and have opportunities to learn from these ﬁrms. The second beneﬁt to them was the ability to access customers that were already using Word on Windows and were not willing to switch to a Macintosh because they would have to use a new word processing package. Prior to this product announcement, these two user communities were using separate packages for word processing. With this announcement, Microsoft has bridged a structure hole between them. Thus: H4. The joint effect of market scope and structural holes on software ﬁrm performance is positive and signiﬁcant. Fig. 2 is a schematic representation of the four hypotheses.

Fig. 2.

Schematic of the Four Hypotheses.

407

Interconnect to Win

METHODS Data and Sample Frame We assembled a database of packaged software ﬁrms that integrates data from multiple sources (cf. Table 1). Our sample frame for the study is a set of software ﬁrms that design and deliver packaged software (SIC 7372). For these ﬁrms, we collected ﬁrms’ package software product sales data from International Data Corporation (IDC) and alliance data from Securities Data Company (SDC). The IDC database contains the revenues (in millions of dollars) of approximately 1,200 software ﬁrms (e.g., IBM and SAP) from approximately 90 within-industry markets (e.g., database, accounting, and logistics) over a period of 13 years. In total, the database contains over 16,000 unique company – product (sold in a single within-industry market) – year – revenue observations. The IDC database also captures ﬁrm entry or exit and new within-industry markets (e.g., middleware and XML software markets). The SDC database contains data on alliance, alliance participants, and the alliance date. We ﬁrst extracted from the SDC alliance database all alliances between 1988 and 2001, inclusive. For every focal software ﬁrm in the IDC database, we identiﬁed the same ﬁrm in the SDC database. We took extreme care in ‘‘mapping’’ the IDC database ﬁrms to the SDC alliance database ﬁrms. Alliances that did not contain at least one focal IDC ﬁrm where deleted Table 1. Sources of Data for the Variables Used. Variables/Database Source Growthi,tþ2 Degreei,tþ1 Structural holei,tþ1 Product scopei,t Market scopei,t Firm market diversityi,t Firm market agei,t Firm market growthi,t Firm market concentrationi,t Firm sharei,t Firm agei,t Firm sizei,t Growthi,tþ1

SDC

IDC

Mergent

Lexis-Nexis

P

Compustat V

P P

V V P P P P P P P P P P

V

V

Notes: P, primary database used and V, veriﬁcation where appropriate.

V V

408

N. VENKATRAMAN ET AL.

from the study sample. Moreover, we also only included an alliance if the focal packaged software ﬁrm partnered with another packaged software, hardware, or services ﬁrm. Most of the alliances were either technology or marketing agreements. The ﬁnal sample for this study contains 5,489 alliances and 2,849 distinct ﬁrms. The set of 2,849 ﬁrms contains 359 focal packaged software ﬁrms. The total number of ﬁrm – year observations is 1,672. We followed the procedure outlined by Anand and Khanna (2000) and Kale et al. (2002) to clean the data. We also consulted other sources such as Lexis/Nexis and Compustat to triangulate the alliance data. The SDC Thompson and IDC proprietary databases are the primary sources for this study. We used the other databases to verify the integrity of the merged database given that the data are collected from multiple sources. For example, we removed semantic variation in the names of companies. We also veriﬁed for a sample using both the text in the SDC database and actual alliance announcement to check for consistency in the data elements such as alliance date and that the alliance indeed contained a ‘‘software component.’’ We consulted the Compustat database to verify the SIC code for public companies. For recent alliances, we also consulted the company’s website to verify the existence of the alliance.

Network Speciﬁcation We used a symmetric weighted adjacency matrix to represent the alliance network in year tþ1. This matrix encoded the topological properties of the network and was the basis for computing the two network measures. Per Newman (2001a, 2001b), the value or weight is zero (wij ¼ 0) if no alliances existed between nodes (i.e., ﬁrm) i and j and greater than zero (wijW0) if alliances did exist. A weighted network speciﬁcation captures not only the existence of an interaction between two ﬁrms but also captures the intensity of the interactions (Freeman et al., 1991; Wasserman & Faust, 1994). Speciﬁcally, let 1 if firm i is a member of alliance k 2 K tþ1 dik;tþ1 ¼ 0 otherwise where Ktþ1 denotes the set of alliances that were formed between the years (t2, tþ1) and each alliance contained at least one focal software ﬁrm. To this end, the value of the edge between two companies i and j or adjacency

409

Interconnect to Win

P matrix element wij;tþ1 ¼ k2K tþ1 ðdik;tþ1 djk;tþ1 =ðnk 1ÞÞ for i6¼j and nkW1 and wii,tþ1 ¼ 0. The term nk denotes the number of distinct companies in alliance k and is used to incorporate the observation that the intensity of interactions decreases as the number of participants in an alliance increases (Newman, 2001b). This network speciﬁcation differentiates, for example, between a dyadic alliance (nk1 ¼ 21) and a triadic alliance (nk1 ¼ 31). Because of difﬁculties in determining the boundaries of the network, we used a fouryear moving window. This criterion for alliance network inclusion is consistent with prior research (e.g., Gulati & Gargiulo, 1999; Stuart, 2000). While alliance formations are announced and can be reasonably corroborated, terminations of relationships are not consistently reported. We remove an alliance from the network after four years unless new information was available to indicate otherwise.

Operationalization of Constructs Growthi,tþ2. We measured the dependent variable, ﬁrm Growthi,tþ2, as the two-year lagged natural logarithm of the total ﬁrm revenues in year tþ2 divided by total ﬁrm revenues in year tþ1 (Carroll & Hannan, 2000). Sales growth is critical to the success of software ﬁrms (Campbell-Kelly, 2003). The range for the time variable t is 1990–2000. Structural holei,tþ1. A key variable P in the calculation of network structural holes is pij;tþ1 ¼ ðwij;tþ1 þ wji;tþ1 =ð k ðwik;tþ1 þ wki;tþ1 ÞÞÞ or the proportion of ﬁrm i’s relations with ﬁrm j compared to all of ﬁrm i’s alliances. The structuralPhole for ﬁrm P i in year tþ1 is one minus the network constraint or 1:0 j ðpij;tþ1 þ k;kai;kaj þ pkj;tþ1 Þ2 (Burt, 1992). Alliance degreei,tþ1. We measured a focal ﬁrmPi’s degree in year tþ1 as the total weight of the focal ﬁrm edge weights or j wij;tþ1 . Product scopei,t. We computed a ﬁrm i’s product scope in year tþ1 in two steps. First, we followed recent operationalization of within-industry market relatedness (Li & Greenwood, 2004) that builds on Sohn’s (2001) similarity metric. We computed the complementarities between two within-industry software markets j and k by P P computing a market similarity metric ojk;t ¼ ð i2N firm xij;t minðxij;t ; xik;t Þ= i2N firm ðxij;t Þ2 Þ (Sohn, 2001) where xij,t denotes t t the sales of ﬁrm i in market j in year t and Ntﬁrm denotes the set of all ﬁrms in the IDC database in year t. This similarity metric changes yearly. We repeated this calculation for every market pair and every year that we examine. Per Lemelin (1982), the relatedness of markets can be inferred from the customers’ purchase patterns because ﬁrms tend to diversify into

410

N. VENKATRAMAN ET AL.

markets where customers of their principle products buy other products from them. Second, we computed the ﬁrm’s product scope by taking a sales weighted average of the similarity metrics across all pairs of ﬁrm i’s products or P product o j 1Þ. The set Ni,tproduct denotes ﬁrm i’s jkjjak jk;t ðpij;t þ pik;t Þ=ðjN i;t products in year t and pij,t and pik,t denote proportion of product sales generated from products j and k, respectively. We normalized the sum to remove double-counting bias (Robins & Wiersema, 1995). Market scopei,t. We measured ﬁrm i’s market scope in year t as the sum, multiplied by 100 to minimize rounding errors, of the product’s share of the packaged software sector multiplied by the proportion of the ﬁrm’s sales P from the product or j pij;t Shrj;t . The term Shrj,t is the total size of market j divided by the total size of the software sector in year t. The computation of install base of a technology has been used in prior studies (e.g., Farrell & Saloner, 1986; Kauffman et al., 2000).

Controls Firm market diversityi,t. We measured a ﬁrm i’s market diversity in year t using the extensive literature on corporate diversiﬁcation (Palepu, 1985). There is a growing body of within-industry diversiﬁcation studies suggesting that diversiﬁcation may have a signiﬁcant impact on ﬁrm performance (e.g., Cottrell & Nault, 2004; Li & Greenwood, 2004; Stern & Henderson, 2004). Let pij,t in year t denote the proportion of ﬁrm i’s sales from product j. P We deﬁned market diversityi,t ¼ j pij;t ln ðp1 ij;t Þ. Firm market agei,t. We computed ﬁrm i’s product age in year t as the sales weighted age of the ﬁrm’s products. We took the sum of the proportion of the ﬁrm’s P sales from the product j multiplied by the age of the market in year t or j pij;t MktAgej;t . The term MktAgek,j is the age of product j in year t and pij,t is the proportion of the ﬁrm’s sales from product j in year t. Firm market growthi,t. A ﬁrm’s position across growing markets can inﬂuence sales growth (Robins & Wiersema, 1995). We thus controlled for sales growth in the markets in which the ﬁrm competes. Firm market growthi,t for ﬁrm i in year t is the sum, over all products, of the growth of market j multiplied by the ﬁrm’s proportion of sales from the product in the P market or p MktGrowth j;t . The term MktGrowthj,t is computed as j ij;t the natural logarithm of the size of market j in year t divided by the size of the same market in year t1.

411

Interconnect to Win

Firm market concentrationi,t. We measured the ﬁrm’s market concentrationi,t as the sum of the proportion the ﬁrm’s sales from product j multiplied by the share of the four largest ﬁrms in the market (i.e., P p Top4 j;t ). Market concentration is a descriptor of industry structure j ij;t that can impact ﬁrm performance. Firm sharei,t. We measured ﬁrm i’s share in year t as the weighted sum of a ﬁrm’s share in each market. Speciﬁcally, we summed the ﬁrm’s share of product j in the corresponding market multiplied by the proportion of ﬁrm i’s sales from product j. Studies (e.g., Li & Greenwood, 2004) suggest that ﬁrm size inﬂuences ﬁnancial performance. ln(Firm sizei,t). We measured ﬁrm i’s size in year t as the natural logarithm of the ﬁrm’s total sales. Firm size may also impact ﬁrm growth (Sorenson & Stuart, 2000). ln(Firm agei,t). We measured ﬁrm is are in year t as the natural logarithm of the difference between the year t and the ﬁrst year in which the company generated sales or founding. We determined the ﬁrst year that a company generated sales from the full IDC database, which spans the years 1990–2002. For those companies that existed prior to 1990, we used the difference between year t and its year of incorporation as the ﬁrm’s age. Studies (e.g., Podolny, Stuart, & Hannan, 1996) suggest a signiﬁcant relationship between ﬁrm age and performance. Growthi,tþ1. We measured ﬁrm Growthi,tþ1 as the log of the total revenues in year t divided by total revenues in year tþ1 (Carroll & Hannan, 2000). It is customary to include in a panel design the prior value of the dependent variable. Yeart. Finally, we included a set of indictor variables that are contemporaneous with the independent variables to capture industry effects such as the total size of the software sector and the numbers of ﬁrms, markets, and other industry-level effects in that year. Table 2 presents the descriptive statistics of our sample. It summarizes the means, standard deviations, and zero-order correlations among the variables in our model.

Analysis We tested the hypotheses using a cross-sectional time series or panel design. The sample contains 359 distinct focal packaged software ﬁrms with approximately 4.7 years observations per ﬁrm or 1,672 ﬁrms – year observations. The design repeatedly measures ﬁrm performance and covariates, which includes a lagged performance measure. We estimated a

412

Sample Descriptives.

Table 2. Mean

1

2

3

4

5

6

7

8

9

10

11

Structural holei,tþ1 0.523 0.370 Degreei,tþ1 10.438 24.102 0.422 Product scopei,t 0.478 0.464 0.333 0.276 0.113 0.645 0.066 0.026 0.094 Market scopei,t Firm market diversityi,t 0.560 0.661 0.394 0.482 0.572 0.042 Firm market agei,t 6.034 2.700 0.099 0.068 –0.008 0.188 0.038 0.224 0.314 0.064 0.008 0.006 0.003 0.053 0.428 Firm market growthi,t Firm market concentrationi,t 0.600 0.121 0.190 0.222 0.083 0.046 0.133 0.307 0.088 Firm sharei,t 0.055 0.091 0.289 0.378 0.222 0.011 0.220 0.214 0.153 0.463 ln(ﬁrm age)i,t 1.616 1.095 0.308 0.356 0.411 0.154 0.491 0.045 0.103 0.084 0.193 3.928 1.934 0.539 0.535 0.563 0.093 0.562 0.056 0.136 0.232 0.502 0.506 ln(ﬁrm size)i,t Firm growthi,tþ1 0.271 0.376 0.009 0.031 0.121 0.058 0.134 0.189 0.352 0.005 0.043 0.287 0.315

Note: N ¼ 1,672; panels ¼ 359.

N. VENKATRAMAN ET AL.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

SD

413

Interconnect to Win

two-period lagged model. Although our database contains data for the year 1990–2002, we needed the last two years to estimate period tþ1 and tþ2 growth measures. Because of this constraint, the range of time t is 1990– 2000. This speciﬁcation helps to mitigate possible concerns of causality and temporal dependence. A ﬁrm’s product position (i.e., resources) at time t inﬂuences the alliance network at time tþ1. This plus other controls impacts performance at time tþ2. Under these conditions, ordinary least squares (OLS) may be inefﬁcient and result in biased estimates (Greene, 2003). We used the generalized estimating equations (GEE) approach (Liang & Zeger, 1986; Zeger, Liang, & Albert, 1988) that has been utilized in prior studies (e.g., Katila & Ahuja, 2002). GEE is a ﬂexible estimation procedure that incorporates ﬁrm correlation and heterogeneity and thus results in more efﬁcient and unbiased parameters than OLS. GEE estimators are asymptotically normal and consistent given an arbitrary correlation among observations (Liang & Zeger, 1986; Zeger et al., 1988). We used a Gaussian distribution for the dependent variable, an identity link function, and an unstructured working correlation matrix. We also performed estimations using an exchangeable working correlation matrix and also used a sandwich variance estimator for correcting standard errors. We estimated ﬁve separate models. Model 1 is the base model. It contains the controls. Model 2 adds the two network measures (i.e., structural holes and degree) and two product measures (i.e., product relatedness and network effects). Model 3 adds the four interactions between the network and product measures. Models 4 and 5 add independently the product and network measures, respectively. All models are estimated using Stata. We used Model 3 to test the hypotheses.

RESULTS Table 3 is a summary of the Stata estimations of our model estimations and all ﬁve models are signiﬁcant. The last row in the Table 3 indicates that the key changes in the model ﬁt statistics between the models are signiﬁcant. Speciﬁcally, the addition of the two network positions and two business strategy variables signiﬁcantly improved model ﬁt (i.e., Dw2 ¼ 20.82, df ¼ 4, po.001). Our hypotheses called for the addition of four interactions and the model ﬁt with these interactions also signiﬁcantly improved the model ﬁt (i.e., Dw2 ¼ 13.44, df ¼ 4, po.01) lending support to the thesis that the interactions between business strategy and network positions signiﬁcantly

Model 1

df Wald w2 Models Dw2

0.066 0.009 0.064 0.015 0.621 0.006 0.062 0.145 0.214 0.087w 0.116 0.004 0.043 0.062 0.025 0.081 0.254 0.273 0.287

GEE Regression Results. Model 2

0.018 0.007 0.045 0.101 0.296 0.011 0.013 0.032 0.078 0.054 0.058 0.050 0.052 0.059 0.058 0.061 0.069 0.072 0.083

18 484.64

Model 3

0.101 0.001w 0.070 0.022

0.037 0.001 0.033 0.013

0.070 0.010 0.057 0.028 0.631 0.006 0.077 0.121 0.211 0.088w 0.105w 0.002 0.038 0.053 0.036 0.064 0.265 0.292 0.315

0.024 0.007 0.046 0.101 0.309 0.011 0.014 0.033 0.078 0.053 0.059 0.051 0.053 0.059 0.060 0.063 0.070 0.073 0.089

22 505.46 (2)(1) 20.82

0.009 0.001 0.040 0.072 0.004 0.277 0.133 0.146 0.090 0.010 0.055 0.016 0.654 0.004 0.077 0.122 0.211 0.064 0.136 0.032 0.073 0.089 0.002 0.101 0.227 0.254 0.304

Model 4 0.048 0.001 0.039 0.033 0.002 0.099 0.068 0.045 0.025 0.007 0.046 0.101 0.307 0.011 0.014 0.033 0.079 0.052 0.060 0.051 0.053 0.060 0.059 0.063 0.069 0.071 0.089

26 518.90 (3)(2) 13.44

Notes: N ¼ 1,672; semi-robust standard errors; wo.10; o.05; o.01; and o.001.

Model 5

0.064 0.023

0.033 0.012

0.080 0.008 0.062 0.015 0.626 0.006 0.060 0.146 0.210 0.096w 0.101w 0.018 0.024 0.040 0.047 0.060 0.275 0.294 0.299

0.023 0.007 0.045 0.100 0.293 0.011 0.013 0.032 0.079 0.054 0.060 0.052 0.054 0.060 0.060 0.063 0.070 0.073 0.089

20 497.52 (2)(4) 7.94

0.100 0.001

0.037 0.001

0.060 0.010 0.060 0.002 0.621 0.006 0.079 0.118 0.214 0.081 0.119 0.012 0.056 0.075 0.014 0.085 0.244 0.271 0.297

0.019 0.007 0.045 0.101 0.312 0.011 0.014 0.033 0.078 0.053 0.058 0.050 0.051 0.058 0.058 0.061 0.069 0.072 0.083

20 491.92 (2)(5) 13.54

N. VENKATRAMAN ET AL.

Structural holei,tþ1 Degreei,tþ1 Product scopei,t Market scopei,t H1: Degreei,tþ1 product scopei,t H2: Degreei,tþ1 market scopei,t H3: Structural holei,tþ1 product scopei,t H4: Structural holei,tþ1 market scopei,t Firm market diversityi,t Firm market agei,t Firm market growthi,t Firm market concentrationi,t Firm sharei,t ln(ﬁrm age)i,t ln(ﬁrm size)i,t Firm growthi,tþ1 Year 1991 Year 1992 Year 1993 Year 1994 Year 1995 Year 1996 Year 1997 Year 1998 Year 1999 Year 2000 Constant

414

Table 3.

415

Interconnect to Win

impact sales growth in the software companies over and beyond ﬁrm level and industry controls. Thus, we use Model 3 as the best-ﬁt model for deriving the empirical support for our hypotheses. Finally, the independent addition of focal ﬁrm’s business strategy variables (Dw2 ¼ 7.94, df ¼ 2, po.05) and network positions (Dw2 ¼ 13.54, df ¼ 2, po.01) signiﬁcantly reduced model ﬁt vis-a`-vis the main effects only model. These results suggest that the set of business strategy and network position variables taken together are a better predictor of ﬁrm sales growth than either set independently. More importantly, the interactions effects that underlie our hypotheses emerged as signiﬁcant predictor of performance. The control variables have effects in the expected direction and magnitude in Model 3. Prior sales growth is positively associated with the dependent variable and is an important control to rule out spurious effects; market share is positively associated with sales growth; and the negative and signiﬁcant coefﬁcient for ﬁrm size suggests that larger ﬁrms ﬁnd growth more difﬁcult than smaller ﬁrms as expected. H1 is a test of the positive effect of the interaction between product scope and alliance degree on sales growth. The GEE coefﬁcient is 0.004 but is not signiﬁcant (pW.10). Thus, H1 is not supported. H2 is a test of the positive effect of the interaction between market scope and alliance degree on sales growth. The GEE coefﬁcient estimate is 0.277 and signiﬁcant (po.01), providing support for H2. H3 is a test of the positive effect of the interaction between product scope and structural holes on sales growth. The GEE coefﬁcient estimate is 0.133 and signiﬁcant (po.05), supporting the hypothesis. Finally, H4 is a test of the positive effect of the interaction between market scope and structural holes. The GEE coefﬁcient estimate is 0.146 and signiﬁcant (po.01), providing empirical support for this hypothesis. Collectively, three of the four hypotheses are supported.

DISCUSSION Networks have emerged as an important frame to understand not only how ﬁrms develop their strategies but also to explain the impact of strategies on performance. Recently, empirical evidence is forthcoming that network positions matter for performance (e.g., Rowley et al., 2000; Zaheer & Bell, 2005). Our empirical analyses demonstrated as a baseline that strategy and network positions have positive implications for ﬁrm performance after accounting for a battery of control variables. Comparison of Model 4 with Model 2 shows that network positions do add statistically to explaining ﬁrm

416

N. VENKATRAMAN ET AL.

performance. Thus, this adds to the empirical literature stream on the performance impacts of network positions. Going beyond such universal ﬁndings, our point of departure for this study was to examine the joint interaction effects of business strategy and network positions. Speciﬁcally, we focused on a central aspect of business strategy (business strategy captured through product scope and market scope) and network positions (alliance degree and structural holes). Using available prior evidence that could be marshaled to motivate the interaction effects and relying on our understanding of the importance of linkages within the software industry to shape the technical architecture, we derived a set of four general hypotheses. Our results – based on the fact that three of the four hypotheses were empirically supported – validate our thesis of looking at joint effects of business strategy and network positions. The results also raise a set of promising directions for further theorizing and empirical explorations to go deeper into the performance effects of interactions of strategic resource deployments and network characteristics. The results raise a set of promising directions for further theorizing and empirical explorations to go deeper into the performance effects of interactions of strategic resource deployments and network characteristics. Such examinations will pave way for bringing the ﬁrm back at the center of network strategy research. We believe that the performance of ﬁrms needs to be understood through joint interactions of ﬁrm strategy and network positions.

Role of Alliance Degree in Software Looking at alliance degree as a network characteristic, we found that its interaction with product scope was not signiﬁcant but its interaction with market scope emerged as being signiﬁcant. Based on an understanding of the key characteristics of the software industry, we conceptualized and operationalized market scope to reﬂect the collective installed base of the products. Given that the installed base of software product is a central variable in system-based competition (Shapiro & Varian, 1998), our results indicate that software ﬁrms enter into different types of alliances to further the acceptance of their products in the different target market segments. Since market acceptance of a software product depends on how they interoperate with related software products (i.e., as part of the technical architecture – not made in isolation) within different customer settings, this empirical ﬁnding reinforces an implicit, untested assertion within the information systems and technology ﬁeld. Furthermore, taking the results of H1 and H3 together, one

417

Interconnect to Win

could surmise that ﬁrms appear to rely on interﬁrm linkages to achieve interoperability closer to the point of use than in the design of software products upfront. This could very well be due to the fact that speciﬁcations for interoperating different software products may be made explicit by the software providers to potential complementors without necessarily relying on bilateral agreements. Our results for these two hypotheses point to a fruitful direction for extending this study. It suggests that we should further decompose the set of linkages within the software industry into distinct categories and evaluate the impact of different dimensions of business strategy on network positions under different types of linkages. Such ﬁne-grained conceptualization and empirical tests will go a long way towards understanding the logic of interactions between strategy and network positions.

Role of Structural Holes in Software Looking at structural holes as the other network characteristic, we ﬁnd that this emerged as signiﬁcant in interactions with product scope and market scope. Structural holes have been seen as a central idea underlying social capital in studies of individuals in different settings (e.g., Burt, 1992; Echols & Tsai, 2005) and more recently been used as a core concept to understand the formation and evolution of interorganizational relationships (e.g., Gulati, 1999; Lee 2007; Zaheer & Bell, 2005). This study is the ﬁrst to understand its effect in the software sector, where structural holes may represent an opportunity to interconnect disparate islands of technologies. Network effects within system-based competition is based on demarcating distinct boundaries of competition – for example: direct network effects enjoyed by Windows is based on the fact that customers can more easily share data and applications within the same operating system than across different operating systems. But, by entering into speciﬁc interﬁrm agreements that close structural holes, some software ﬁrms may be able to enhance their network effects. Thus, structural holes do have speciﬁc property in the software sector that is akin to social capital in human networks but inﬂuences the path (evolutionary direction) and pace of shifts in the dynamic software sector. Armed with the strong results that we found for both facets of business scope in this study, we believe that it is worthwhile to delve into explicating the role of bridging structural holes as part of the dynamic shifts in software architecture. Such forays could be guided by mapping design moves in Baldwin and Clark (2000) against the different interorganizational arrangements in the software sector.

418

N. VENKATRAMAN ET AL.

Going beyond using a simple use of scope of agreements (licensing, R&D agreements, marketing agreements, etc.), we could develop a map of how the different alliances enable the focal ﬁrm to leverage the key design moves (substituting, porting, augmenting, etc.). Such an approach has the potential to connect the core ideas of structural holes with the nuanced understanding of the evolution of the software architecture, which is the result of multiple independent and coordinated actions in this fast-changing sector. Time-Sequenced Interactions Our model is based on a time-sequenced logic of interaction effects between business strategy and network positions. Speciﬁcally, as shown in Fig. 1, we speciﬁed that strategy at time t0 and network positions at time t1 will impact performance during t1t2. In specifying such a model, we have implicitly recognized that network positions act in ways to support the execution of the chosen business strategy. Our empirical results do support this view. Armed with this baseline results, it may be worthwhile to delve into decomposing the network linkages into ones that are exploratory and those that are exploitative (March, 1991; Rothaermel & Deeds, 2004) to better understand how product and market scope impacts with network positions. Such extensions will go a long way in developing a dynamic perspective of how ﬁrms compete and navigate in networks. We carried out some tests to rule out a plausible alternative temporal sequence. We empirically tested the possibility that network position in time t0 (instead of time t1) inﬂuences business strategy in time t1 (instead of time t0) with subsequent impact on ﬁrm performance (time t1t2). In such a speciﬁcation, network positions dictate business strategy rather than vice versa but we found little empirical support for this temporal sequence (results available upon request). This set of results may be speciﬁc to the packaged software sector – where product and alliance preannouncements are commonplace to signal future moves and potentially locking-out rivals within these networks. Clearly, more systematic thinking and analysis is needed to develop plausible alternative time-sequenced models with a nuanced typology of linkages for different types of software products.

CONCLUSION Strategies for success within the software sector are inherently networkcentric: they call for complementors that design and deliver applications

419

Interconnect to Win

that enhance the value and acceptance of software products. Software ﬁrms also enter into a broad array of relationships to ensure that their products are part of the solutions in different settings. In this chapter, we present the ﬁndings of what we believe to be the ﬁrst set of results on the importance of network positions for ﬁnancial performance of the software ﬁrms. We not only demonstrated that the network positions have additional explanatory power by themselves but also that they interact with business strategy variables to inﬂuence ﬁrm growth.

ACKNOWLEDGMENTS We thank Joel Baum, Ron Burt, Pat Doeian, Tim Rowley, and Gordon Walker, and the other participates of the Advances in Strategic Management Conference, Toronto for their insightful comments and suggestions.

REFERENCES Afuah, A. (2000). How much do your co-opetitors’ capabilities matter in the face of technological change? Strategic Management Journal, 21, 387–404. Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Anand, B. N., & Khanna, T. (2000). Do ﬁrms learn to create value? The case of alliances. Strategic Management Journal, 21, 295–315. Baldwin, C., & Clark, K. (1997). Managing in an age of modularity. Harvard Business Review, 75, 84–93. Baldwin, C., & Clark, K. (2000). Design rules: The power of modularity. Cambridge: MIT Press. Barabasi, A. L. (2002). Linked: The new science of networks. Cambridge: Perseus Publishing. Barney, J. B. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17, 99–120. Baum, J. A. C., & Singh, J. V. (1994). Organizational niches and the dynamics of organizational mortality. American Journal of Sociology, 100, 346–380. Baum, J. A. C., Shipilov, A., & Rowley, T. (2003). Where do small worlds come from? Industrial and Corporate Change, 12, 697–725. Beckman, C. M., & Haunschild, P. R. (2002). Network learning: The effects of partners’ heterogeneity of experience on corporate acquisitions. Administrative Science Quarterly, 47, 92–124. Boehm, B. (1981). Software Engineering Economics. New Jersey: Prentice Hall. Bresnahan, T. (1998). New modes of competition: Implications for the future structure of the computer industry. Presented at the Competition, Convergence, and the Microsoft Monopoly. Bresnahan, T., & Greenstein, S. (1999). Technological competition and the structure of computer industry. Journal of Industrial Economics, 47, 541–558.

420

N. VENKATRAMAN ET AL.

Brynjolfsson, E., & Kemerer, C. (1996). Network externalities in microcomputer software: An econometric analysis of the spreadsheet market. Management Science, 42, 1627–1647. Burt, R. (1992). Structural Holes. Cambridge: Harvard University Press. Burt, R. (1997). The contingent value of social capital. Administrative Science Quarterly, 42, 339–365. Campbell-Kelly, M. (2003). From airline reservation to sonic the hedgehog: A history of the software industry. Cambridge: MIT Press. Capron, L., & Hulland, J. (1999). Redeployment of brands, sales forces, and general marketing management expertise following horizontal acquisitions: A resource-based view. Journal of Marketing, 63, 41–54. Carroll, G. R., & Hannan, M. T. (2000). The demography of corporations and industries. Princeton: Princeton University Press. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Conner, K. R., & Prahalad, C. K. (1996). A resource-based theory of the ﬁrm: Knowledge versus opportunism. Organization Science, 7, 477–501. Cottrell, T., & Nault, B. R. (2004). Product variety and ﬁrm survival in the microcomputer software industry. Strategic Management Journal, 25, 1005–1025. Cusumano, M. (2004). The business of software: What every manager, programmer, and entrepreneur must know to thrive and survive in good times and bad. New York: Free Press. Davis, R., & Thomas, L. G. (1993). Direct estimation of synergy: A new approach to the diversiﬁcation-performance debate. Management Science, 39, 1334–1346. Dyer, J., & Singh, H. (1998). The relational view: Cooperative strategy and sources of interorganizational competitive advantage. Academy of Management Review, 23, 660–679. Echols, A., & Tsai, W. (2005). Niche and performance: The moderating role of embeddedness. Strategic Management Journal, 26, 219–238. Farjoun, M. (1994). Beyond industry boundaries: Human expertise, diversiﬁcation and resource-related industry groups. Organization Science, 5, 185–199. Farjoun, M. (1998). The independent and joint effects of the skill and physical bases of relatedness in diversiﬁcation. Strategic Management Journal, 19, 611–630. Farrell, J., & Saloner, G. (1985). Standardization, compatibility, and innovation. Rand Journal of Economics, 16, 70–83. Farrell, J., & Saloner, G. (1986). Installed base and compatibility: Innovation, product preannouncements, and predation. American Economic Review, 76, 940–955. Freeman, L., Borgatti, S., & White, D. (1991). Centrality in valued graphs: A measure of betweenness based on network ﬂow. Social Networks, 13, 141–154. Gallagher, S., & Park, S. H. (2002). Innovation and competition in standard-based industries: A historical analysis of the U.S. home video game market. IEEE Transactions on Engineering Management, 49, 67–82. Gao, L. S., & Iyer, B. (2006). Analyzing complementarities using software stacks for software industry acquisitions. Journal of Management Information Systems, 23, 119–147. Gates, B. (2002). Available at http://www.microsoft.com/presspass/trial/mswitness/2002/ billgates/billgates.asp Gnyawali, D. R., & Madhavan, R. (2001). Cooperative networks and competitive dynamics: A structural embeddedness perspective. Academy of Management Journal, 26, 431–445.

Interconnect to Win

421

Granovetter, M. S. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Greene, W. (2003). Econometric analysis. New Jersey: Prentice Hall. Gulati, R. (1995). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38, 85–112. Gulati, R. (1999). Network location and learning: The inﬂuence of network resources and ﬁrm capabilities on alliance formation. Strategic Management Journal, 20, 397–420. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Gulati, R., Nohria, N., & Zaheer, A. (2000). Strategic networks. Strategic Management Journal, 21, 203–215. Gulati, R., & Singh, H. (1998). The architecture of cooperation: Managing coordination costs and appropriation concerns in strategic alliances. Administrative Science Quarterly, 43, 781–814. Iyer, B., Lee, C., & Venkatraman, N. (2006). Managing in a ‘‘Small World Ecosystem’’: Lessons from the software sector. California Management Review, 48, 28–47. Kale, P., Dyer, J., & Singh, H. (2002). Alliance capability, stock market response and long-term alliance success: The role of the alliance function. Strategic Management Journal, 23, 747–767. Katila, R., & Ahuja, G. (2002). Something old, something new: A longitudinal study of search behavior and new product introduction. Academy of Management Journal, 45, 1183–1194. Katz, M. L., & Shapiro, C. (1985). Network externalities, competition, and compatibility. American Economic Review, 75, 424–440. Katz, M. L., & Shapiro, C. (1994). Systems competition and network effects. Journal of Economic Perspectives, 8, 93–115. Kauffman, R., McAndrews, J., & Wang, Y. (2000). Opening the ‘‘black box’’ of network externalities in network adoption. Information Systems Research, 11, 61–82. Kogut, B., & Walker, G. (2001). The small world of Germany and the durability of national networks. American Sociological Review, 66, 317–335. Koka, B., & Prescott, J. (2002). Strategic alliances as social capital. Strategic Management Journal, 23, 795–816. Koka, B., Prescott, J., & Canella, B. (2007). Does network position really matter?: A metaanalysis of the network position-ﬁrm performance relationship in interﬁrm networks. Working paper. Krackhardt, D., & Stern, R. N. (1988). Informal networks and organizational crises – an experimental simulation. Social Psychology Quarterly, 51, 123–140. Lavie, D. (2007). Alliance portfolios and ﬁrm performance: A study of value creation and appropriation in the U.S. software industry. Strategic Management Journal, 28, 1187–1212. Lee, G. K. (2007). The signiﬁcance of network resources in the race to enter emerging product markets: The convergence of telephony communications and computer networking, 1989–2001. Strategic Management Journal, 28, 17–37. Lemelin, A. (1982). Relatedness in the patterns of interindustry diversiﬁcation. Review of Economics and Statistics, 64, 646–657. Li, X., & Greenwood, R. (2004). The effect of within-industry diversiﬁcation on ﬁrm performance: Synergy creation, multi-market contact and market structuration. Strategic Management Journal, 25, 1131–1153.

422

N. VENKATRAMAN ET AL.

Liang, K. Y., & Zeger, S. (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73, 13–22. Madhavan, R., Caner, T., Prescott, J., & Koka, B. (2008). Bringing the ﬁrm back in: Networking as antecedent to network structure. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 457–501). Bingley, UK: JAI/Emerald Group. March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2, 71–87. Markides, C. C., & Williamson, P. J. (1994). Related diversiﬁcation, core competences and corporate performance. Strategic Management Journal, 15, 149–165. McEvily, B., & Zaheer, A. (1999). Bridging ties: A source of ﬁrm heterogeneity in competitive capabilities. Strategic Management Journal, 20, 1133–1158. Messerschmitt, D. G., & Szyperski, C. (2003). Software ecosystem: Understanding an indispensable technology and industry. Cambridge: MIT Press. Meyer, M. H., & Lehnerd, A. P. (1997). The power of product platforms: Building value and cost leadership. New York: Free Press. Meyer, M. H., & Seliger, R. (1998). Product platforms in software development. Sloan Management Review, 40, 61–74. Mowery, D. C., & Rosenberg, N. (1989). New developments in U.S. technology policy: Implications for competitiveness and international trade policy. California Management Review, 32, 107–124. Newman, M. E. J. (2001a). Scientiﬁc collaboration networks II: Shortest paths, weighted networks, and centrality. Physical Review E, 64, 016131. Newman, M. E. J. (2001b). Scientiﬁc collaboration networks I: Network construction and fundamental results. Physical Review E, 64, 016132. Oliver, C. (1990). Determinants of interorganizational relationships. Academy of Management Review, 15, 241–265. Palepu, K. (1985). Diversiﬁcation strategy, proﬁt performance and the entropy measure. Strategic Management Journal, 6, 239–255. Palich, L. E., Cardinal, L. B., & Miller, C. C. (2000). Curvilinearity in the diversiﬁcationperformance linkage: An examination of over three decades of research. Strategic Management Journal, 21, 155–174. Podolny, J. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107, 33–60. Podolny, J., Stuart, T., & Hannan, M. (1996). Networks, knowledge, and niches: Competition in the worldwide semiconductor industry, 1984–1991. American Journal of Sociology, 102, 659–689. Porter, M. E. (1980). Competitive strategy: Techniques for analyzing industries and competitors. New York: Free Press. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational co-llaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41(1), 116–145. Powell, W., White, D., Koput, K., & Owen-Smith, J. (2005). Network dynamics and ﬁeld evolution: The growth of interorganizational collaboration in the life sciences. American Journal of Sociology, 110, 1132–1205. Rindﬂeisch, A., & Moorman, C. (2001). The acquisition and utilization of information in new product alliances: A strength-of-ties perspective. Journal of Marketing, 65, 1–18.

Interconnect to Win

423

Robins, J., & Wiersema, M. F. (1995). A resource-based approach to the multibusiness ﬁrm: Empirical analysis of portfolio interrelationships and corporate ﬁnancial performance. Strategic Management Journal, 16, 277–299. Rothaermel, F. T., & Deeds, D. L. (2004). Exploration and exploitation alliances in biotechnology: A system of new product development. Strategic Management Journal, 25, 201–221. Rowley, T. J., Baum, J. A. C., Shipilov, A. V., Greve, H. R., & Rao, H. (2004). Competing in groups. Managerial and Decision Economics, 25, 453–471. Rowley, T., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21, 369–386. Sanchez, R., & Mahoney, J. T. (1996). Modularity, ﬂexibility, and knowledge management in product and organizational design. Strategic Management Journal, 17, 63–76. Scherer, F. M., & Ross, D. (1990). Industrial market structure and economic performance. Boston: Houghton and Mifﬂin Company. Schilling, M. A. (2000). Toward a general modular systems theory and its application to interﬁrm modularity. Academy of Management Review, 25, 312–334. Schilling, M. A. (2002). Technology success and failure in winner-take-all markets: The impact of learning orientation, timing, and network externalities. Academy of Management Journal, 45, 387–398. Schilling, M. A., & Steensma, H. K. (2001). The use of modular organizational ﬁrms: An industry-level analysis. Academy of Management Journal, 44, 1149–1168. Shapiro, C., & Varian, H. R. (1998). Information rules: A strategic guide to the network economy. Boston: Harvard Business School Press. Simon, H. A. (1996). The sciences of the artiﬁcial. Cambridge: MIT Press. Soda, G., Zaheer, A., & Carlone, A. (2008). Imitative behavior: Network antecedents and performance consequences. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 531–560). Bingley, UK: JAI/Emerald Group. Sohn, M. W. (2001). Distance and cosine measures of niche overlap. Social Networks, 23, 141–165. Sorenson, J., & Stuart, T. (2000). Aging, obsolescence, and organizational innovation. Administrative Science Quarterly, 45, 81–112. Stern, I., & Henderson, A. D. (2004). Within-business diversiﬁcation in technology-intensive industries. Strategic Management Journal, 25, 487–505. Stuart, T. (2000). Interorganizational alliances and the performance of ﬁrms: A study of growth and innovation rates in a high technology industry. Strategic Management Journal, 21, 791–811. Tanriverdi, H., & Lee. C. (2008). Within-industry diversiﬁcation and ﬁrm performance in the presence of network externalities: Evidence from the software industry. Academy of Management Journal, 51, 381–397. Tsai, W. (2001). Knowledge transfer in interorganizational networks: Effects of network position and absorptive capacity on business unit innovation and performance. Academy of Management Journal, 44, 996–1004. Ulrich, K. (1995). The role of product architecture in the manufacturing ﬁrm. Research Policy, 25, 419–440. Ulrich, K., & Eppinger, S. D. (1999). Product design and development. New York: McGrawHill.

424

N. VENKATRAMAN ET AL.

Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Venkatraman, N., & Lee, C. (2004). Preferential linkage and network evolution: A conceptual model and empirical test in the US videogame sector. Academy of Management Journal, 47, 876–892. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8, 109–125. Wasserman, S., & Faust, K. (1994). Social network analysis. Cambridge: Cambridge University Press. White, H. (2002). Markets from networks: Socioeconomic models of production. Princeton: Princeton University Press. Williamson, O. E. (1991). Comparative economics organization: The analysis of discrete structural alternatives. Administrative Science Quarterly, 36, 269–296. Zaheer, A., & Bell, G. (2005). Beneﬁting from network position: Firm capabilities, structural holes, and performance. Strategic Management Journal, 26, 809–825. Zeger, S., Liang, K., & Albert, P. (1988). Models for longitudinal data: A generalized estimating equation approach. Biometrika, 44, 1049–1060.

IF THE ALLIANCE FITS y: INNOVATION AND NETWORK DYNAMICS Robin Cowan and Nicolas Jonard ABSTRACT Network formation is often said to be driven by social capital considerations. A typical pattern observed in the empirical data on strategic alliances is that of small-world networks: dense subgroups of ﬁrms interconnected by (few) clique-spanning ties. The typical argument is that there is social capital value both to being embedded in a dense cluster, and to bridging disconnected clusters. In this chapter we develop and analyze a simple model of joint innovation where we are able to reproduce these features, based solely on the assumption that successful partnering demands some intermediate amount of technological similarity between the partners.

1. INTRODUCTION In this chapter we are interested in strategic alliance networks. Two ﬁrms form an R&D alliance with the goal of innovating, and the expected proﬁtability of this alliance depends on the very nature of joint innovation.

Network Strategy Advances in Strategic Management, Volume 25, 427–455 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25012-6

427

428

ROBIN COWAN AND NICOLAS JONARD

We develop a simple model of joint innovation to which ﬁrms respond, forming partnerships, and thereby create an industry alliance network. We show that when ﬁrms make partnership choices based solely on the nature of innovation, the emerging network displays all the properties characteristic of observed strategic alliance networks. The networks that arise in the model are small worlds with skewed link distributions. This is the case in spite of the fact that in the model ﬁrms pay no attention to issues of social capital while making alliance decisions.

1.1. Alliance Networks Networks of strategic alliances have been studied in many industries.1 Several properties seem to be very common: the networks are sparse; they tend to be small worlds; and the link distributions are skewed to the right. Sparseness is easily explained by the fact that links are costly to create and maintain. Skewness is explainable through the fact that in most industries ﬁrm-size distribution is skewed. If links have costs, it is likely that a larger ﬁrm has more resources with which to create or maintain links than does a smaller ﬁrm. So the distribution of links should reﬂect the distribution of ﬁrm size. The small-world properties of alliance networks have been more challenging to explain. Typically the explanation invokes social capital. Clustering arises from the tendency of ﬁrms to partner with past partners and with partners of partners. The former creates inertia; the latter closes triangles in the network. The reasons given for these tendencies come from a form of social control. Partnering is risky, and information is a good way to reduce that risk. If a ﬁrm forms a link with a past partner, that link is said to be ‘‘relationally embedded’’; a link with partners of partners is ‘‘structurally embedded’’. In both cases embeddedness is a source of information about potential partners; in the ﬁrst case from our history together; in the second from our common partners. The information value of structurally embedded links serves as an incentive for ﬁrms to create closed triangles in their local networks. Structural embeddedness is also valuable as a source of social control. If a ﬁrm behaves badly towards one of its partners, that behaviour will be reported to, and presumably punished by the local community. If the local community is dense (which is the case when links are structurally embedded) this is an effective way of creating incentives to behave well. The social capital of embeddedness (see Coleman, 1988) works to create cliques of densely connected agents at the local level.2

Innovation and Network Dynamics

429

But small worlds are not only cliquish, they also have enough cliquespanning ties that the average length of the shortest path between two ﬁrms is low. To explain clique-spanning ties, a different sort of social capital is invoked (Burt, 1992). A ﬁrm with links outside its local neighbourhood is in a position to access information originating in distant parts of the network. It might also be in a position to act as an information broker between different parts of the network. Both provide incentives for ﬁrms to look for partners that are not embedded in their local networks, and not to be themselves too embedded in dense local networks. But of course, as more and more ﬁrms engage in these strategies, the value of the strategy falls. Thus we might expect to see some, but not too many ﬁrms forming clique-spanning ties.3

1.2. Innovation Joint innovation involves two agents combining forces to create new knowledge. If the agents are identical, there is little value in combining forces – they can only duplicate each others thoughts and actions. If the agents are extremely different, they will have difﬁculties exploiting each other’s competences. Thus we might expect that when a ﬁrm evaluates potential partners it would ﬁnd desirable ﬁrms that are somewhat similar but not too similar. There are many ways in which ﬁrms can be similar or dissimilar: in terms of product portfolios, production technologies, organizational structures, organizational culture, competences generally, and so on. For our purpose we can focus on whether or not ﬁrms are similar in their knowledge portfolios. Several studies have examined this question empirically.4 If we consider that ﬁrms are located in some underlying knowledge space, the general conclusion of these studies is that when ﬁrms ally (or merge) the success of their venture is an inverted-U in the distance in that space.5 This is the result on which we base our model of innovation. We show that when this is what drives success in joint innovation, the networks that result from strategic alliance formation can be small worlds with a skewed link distribution.

2. THE MODEL Consider a ﬁnite population of ﬁrms located in a single, highly innovative industry. We assume that innovation is necessary for survival and growth,

430

ROBIN COWAN AND NICOLAS JONARD

so the immediate goal of every ﬁrm is to accumulate knowledge. We model knowledge as a set W of discrete facts or ideas, each of which a ﬁrm either knows or does not know. The knowledge endowment of any ﬁrm i can be represented by a binary vector vi of length #W, in which viz ¼ 1 indicates that the ﬁrm knows fact z. The knowledge of a ﬁrm changes over time as the ﬁrm innovates. Innovation can take place in isolation, or in a partnership between two ﬁrms. Because our main concern is with joint innovation, we have a simple, reduced form model of autarchic innovation: autarchic innovations arrive at a rate l, independently of a ﬁrm’s alliance activities. When two ﬁrms form an alliance to innovate, the probability of success depends on how well their knowledge stocks complement each other. Following the empirical literature, we model this as an inverted-U relationship in the distance between the two ﬁrms in knowledge space. We measure distance by ‘‘overlap’’ – that is, the number of facts which both ﬁrms know.6 We assume a ﬁxed cost of alliance formation, so two ﬁrms will form an alliance if (and only if) the overlap of their knowledge stocks is not ‘‘too far’’ from the optimal overlap. Innovation success implies the discovery of a fact new to the innovator, or in the case of joint innovation, new to both of the innovation partners. The new fact is incorporated into the knowledge stocks of (each of) the innovator(s). Every period ﬁrms evaluate all possible alliances and form exactly those that have positive expected value, and within those alliances attempt to innovate. At the end of each period, all alliances dissolve; and in the subsequent period the process repeats, ﬁrms possibly having added knowledge to their portfolios. The network thus evolves; co-evolving with the changing knowledge portfolios of the ﬁrms, but the network structure itself at any point in time is determined entirely by the knowledge held by ﬁrms and by the nature of joint innovation. We have made the very simple assumption that ﬁrms form alliances purely on the basis of whether or not the knowledge portfolios of the prospective partners are complementary. Based only on this assumption we are able to derive several structural properties of the strategic alliance network. The network responds in predictable ways to changes in the amount of knowledge ﬁrms hold on average, to the size of the optimal overlap, and to how strictly the optimal overlap must be met.

2.1. Innovation and Equilibrium In this section we derive a description of the equilibrium network, as determined by our assumptions on joint innovation.

Innovation and Network Dynamics

431

2.1.1. Innovation Initially each ﬁrm knows each idea with probability p0. This probability is independent both of knowing other facts, and of the state of knowledge of other ﬁrms. Innovation is deﬁned as the discovery of a previously unknown (to the innovators) fact, and has, without loss of generality, a value of 1.7 Deﬁne vij ¼ {z:viz ¼ vjz ¼ 1} as the intersection, or overlap, of the knowledge portfolios of i and j, and deﬁne yij ¼ #vij to be the size of the overlap, or the number of facts known to both i and j. If the partnership ij forms, both i and j pay a ﬁxed cost c, and the partnership innovates with probability f(yij), independently of the other alliances formed by i and j.8 The function f(y) is the inverted-U discussed earlier, so f(y) is positive, symmetric about the unique optimal value, y ¼ d, with cof(d){1, increasing (decreasing) monotonically on the left (right) of d. If the partnership ij innovates, it discovers a new fact z which both partners receive, implying that their post-innovation knowledge portfolios become viþ{z} and vjþ{z}. As stated above, an innovation can only take place in a location in the knowledge vector where the innovator(s) is(are) currently ignorant. Thus some innovations will be ‘‘new to the innovator’’ but not new to the world: the innovation takes place within the knowledge frontier. We also allow the frontier to expand, that is, innovations can take place beyond the frontier. The innovation takes place at location z, 1rzrwþD, where D measures the extent to which an innovation can extend the frontier. If the innovation is within the frontier, i’s innovation has no effect on the knowledge stocks of other ﬁrms. However, if i’s innovation expands the frontier, we assume that this makes redundant older knowledge. We make the assumption that if an innovation takes place at location z, wþ1rzrwþD, then the ﬁrst zw knowledge elements become obsolete. The frontier is thus pushed forward by the discovery of ideas beyond w. As innovation consists in drawing the location of an ‘‘empty’’ slot uniformly at random in {1, y, wþD}, the frontier expansion is more likely to be caused by more knowledgeable ﬁrms, and the innovative potential of the industry is never exhausted. We can characterize this in the following way. In principle there are a countably inﬁnite number of facts in the world. At any time t there is a relevant set, which we denote Wt ¼ fwt ; . . . ; wt g with wt wt ¼ w for any t. This relevant set evolves: if at t an innovation takes place within the frontier (or no innovation takes place), Wtþ1 ¼ Wt . If an innovation takes place at z4wt then wtþ1 ¼ z and wtþ1 ¼ z w.

432

ROBIN COWAN AND NICOLAS JONARD

2.1.2. Equilibrium In any period, the independent alliance formation results in a network g. Deﬁne the neighbourhood of ﬁrm i as N gi ¼ f jai : ij 2 gg, that is, the set of the agents to whom i is directly connected. The degree of a ﬁrm is the size of its neighbourhood, which we denote ngi ¼ #N gi . The value of forming a link lies purely in its potential to produce an innovation. In this model, interﬁrm links are only useful for creating knowledge. Unlike, for example, the communication model of Jackson and Wolinsky (1996), links here are not conduits for knowledge spillovers from other (distant) ﬁrms, nor do they perform any other task (control, information brokerage, etc.). Thus the value of a link is simply the expected value of the potential innovation less the costs of the link. So given a network g, we can write the value to ﬁrm i of its links as X p gi ¼ f ðyij Þ cngi (1) j2N gi

Every ﬁrm faces the same problem when evaluating potential links, and a link only forms if both partners agree. Thus a network g is stable if and only if every existing alliance is beneﬁcial to both partners, and the creation of a (currently) non-existent link would reduce the net proﬁts of at least one of the partners. In the present model, the simple form of ﬁrms’ proﬁts implies that both existing and non-existing links have their potential value determined in exactly the same way, the value of ij being f(yij)c to both i and j. As a consequence, the stable network g is simply {ij : f(yij)Zc}. The equilibrium network structure is determined by the interaction of c with f(y). The effect of c is clear: if c>f(d), no partnership forms and the network is empty; if comin{f(0); f(w)} then the network is complete; but for intermediate values of c (min{f(0); f(w)}ocof(d)), all partnerships between ﬁrms having overlap yij such that f(yij)Zc form. In each case , we observe a unique equilibrium network. This can be seen relatively easily graphically, by recalling that f(y) is an inverted-U. Deﬁne rZ0 by f(d7r) ¼ c. Two ﬁrms i and j, with an overlap that differs from d by less than r ﬁnd it proﬁtable to form an alliance since f(yij)oc. Fig. 1 above illustrates this result and Proposition 1 below states it. Proposition 1. For any cZ0, there exists a unique equilibrium network g. When c>f(d) the empty network is stable; when comin{f(0); f(w)} the complete network is stable; when min{f(0); f(w)}ocof(d) the stable network is g ¼ {ij:|yijd|rr}.

433

Innovation and Network Dynamics Expected benefit of an alliance f(yij)

f(yij)

Cost c

−

δ

+

Distance between i and j, yij

Fig. 1.

Costs and Beneﬁts of an Alliance as a Function of Distance in Knowledge Space.

2.2. Digression on a Special Case Consider for the moment a special case of this model, in which the invertedU is actually a spike: joint innovation is possible if and only if partners’ knowledge vectors overlap in exactly d positions. In addition, assume that ﬁrms knowledge vectors are independent both internally and with respect to those of other ﬁrms. In other words, ﬁrm i knows fact k with some ﬁxed probability p, where p is independent of what other facts the ﬁrm knows and what other ﬁrms know. In this case, p is both the probability that a ﬁrm knows some particular fact, and a measure of the prevalence of knowledge in the industry – if p is small, the typical ﬁrm is ignorant of many potentially relevant facts; if p is large, most ﬁrms know most things. Cowan and Jonard (2008) examine this model analytically, asking about the degree distribution and clustering of the emergent network. Here we simply show their results graphically. Fig. 2 shows the degree distribution (top panel) and expected degree (bottom panel). (In each case there are 100 ﬁrms in the industry and 100 potentially relevant facts.) The left panel shows the link distribution for different values of p, with d ﬁxed at 15. What we see in general is that the distributions are skewed to the right (note that this is a log-linear plot), having long tails, but that typically these tails are not as heavy as a power law. For values of p near 0.36 the distribution is not monotonic, having a strictly positive mode. The bottom panel shows expected degree as a function of p, for four values of d. We observe that for

p = 0.2 p = 0.3 p = 0.4 p = 0.5 p = 0.25 p = 0.35 p = 0.45 p = 0.55

10-1

Probability of observation

10-2 10-3 10-4 10-5 10-6 10-7 10-8 10-9 10-10 10

20

30

40

Degree

δ = 10 δ = 15 δ = 20 δ = 25

12

Expected degree

10

8 6

4

2 0 0.2

0.4

0.6

0.8

p

Fig. 2. Top: Degree Distribution for Various Values of p; Bottom: Expected Degree as Function of p for Four Values of d (Note that in Both Cases w ¼ 100, n ¼ 100).

435

Innovation and Network Dynamics

Ratio conditional/unconditional probability

every value of d, the expected degree rises and then falls with p. When p is small, ﬁrms have trouble ﬁnding ﬁrms with the right overlap. For example if p is very small, almost every ﬁrm will have fewer than d pieces of knowledge, and thus will not be able to participate in any alliance. When p is slightly larger, ﬁrms will have close to d pieces of knowledge so (in the case easiest to see), if a ﬁrm has exactly d pieces, it can only partner with a ﬁrm that knows the same pieces, and possibly more. At the other extreme, when p is large, ﬁrms know too much, and typically any pair of ﬁrms matches in too many places to have a partnership. Even in the restricted model it is not possible to calculate clustering coefﬁcients directly. An indication of whether or not a network will be clustered though, can be obtained by examining the ratio of two probabilities: the probability that a link ij forms conditional on its being a triangle closing link (i.e., Pr{ij|ik and jk exist}) to the unconditional probability that ij forms. Fig. 3 shows this ratio as a function of p for four values of d (againp there ﬃﬃﬃﬃﬃﬃﬃﬃﬃ are 100 ﬁrms and 100 facts). What we observe is that except when p d=w this ratio is very large, and no matter the value of p, always greater than 1. This implies that the networks will always be

δ = 10 δ = 15 δ = 20 δ = 25

1020

1015

1010

105

100 0.2

0.4

0.6

0.8

Probability of knowing a fact, p

Fig. 3. Ratio of Conditional to Unconditional Probability as a Function of p (w ¼ 100, n ¼ 100).

436

ROBIN COWAN AND NICOLAS JONARD

clustered, and for large regions of the parameter space even extremely clustered. These results show, or suggest that in this special case, the alliance networks that emerge will have skewed link distributions, and are likely to display small-world properties. By generating numerically many networks, and covering the parameter space, Cowan and Jonard show that this is indeed the case. When d is extremely large or small, there ispalmost ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ no alliance activity (it is too hard to ﬁnd partners); when p ðd=wÞ the network is almost complete (it is too easy to ﬁnd partners), but for other values of p, the network is in fact a small world with a skewed link distribution, corresponding nicely to what is observed empirically. Finally, an interesting question has to do with ‘‘who is partnering’’; in particular, whether it is ﬁrms with a lot or a little knowledge that form partnerships. Again this is possible to answer analytically in the special case, and we show the result in Fig. 4. What we observe is that when the industry average knowledge levels are low relative to the optimal overlap, ﬁrms that are knowledge-rich form many partnerships; when industry

Covariance between knowledge and number of partners

20

δ = 10 δ = 15 δ = 20 δ = 25

10

0

-10

-20 0.2

0.4

0.6

0.8

p

Fig. 4.

Correlation between Knowledge Levels and Number of Partners as a Function of p.

Innovation and Network Dynamics

437

knowledge levels are high, ﬁrms that are knowledge-poor are the ones that partner. This discussion concerned a special case of the model. We return now to the full model as described. It is complex enough that it cannot be treated analytically, but numerical analysis of the model yields results that are both consistent with the analysis of the special case, and interesting empirically.

2.3. Computational Experiment A challenge with numerical simulation generally is to disentangle general effects from artefacts due to, for example, idiosyncracies in the initial conditions that have been assigned. In the model described above, however, this is not an issue, as the model represents a stationary Markov process. Standard results from Markov process theory imply that initial conditions are unimportant – in the long run the system has a unique steady state, regardless of where it starts. We must be careful with the expression ‘‘steady state’’, though, as this does not imply that the system stops moving, but rather that the proportion of time the system spends at each point of the state space settles down to an unchanging value. To expand slightly, each ﬁrm has a binary knowledge vector vi of ﬁxed length w. The state space of each ﬁrm is the set of all possible values of that vector, Vi . Since for every ﬁrm vi 2 Vi , the state space of our system of ﬁrms is the product of ﬁrms’ state spaces, V ¼ xj Vj . Any time a ﬁrm innovates, knowledge vectors change, and the system moves from one state ðv01 ; . . . ; v0n Þ to another state ðv01 ; . . . ; v0n Þ in the system state space V. (We move to a point where the innovating ﬁrm has one more piece of knowledge, and possibly where other ﬁrms have less knowledge, if the innovation has changed the frontier.) The transition from one point to another in this space is random, but with well-deﬁned transition probabilities. Further, these probabilities do not depend on which period we are in, nor do they depend on where the system was at any point in the past other than where it was yesteday. Thus the system is a (ﬁnite) ﬁrst-order stationary Markov chain. If we ignore the unreachable states (examples of which: no ﬁrm knows any fact; no ﬁrm knows the frontier fact) it is easy to show that if we choose any two states of the system ðv1 ; . . . ; vn Þ and ðv01 ; . . . ; v0n Þ, there is some (possibly several) ﬁnite sequence(s) of joint and autarchic innovations that will take us from one to the other.9 This is the deﬁnition of an irreducible Markov chain. As a consequence the system has a unique stationary

438

ROBIN COWAN AND NICOLAS JONARD

distribution. A convenient implication is that the expected value of any statistic of interest is arbitrarily well approximated by its time average. For instance, the expected number of ideas held by agents in the stationary distribution can be estimated by the time average of the number of ideas held. For this reason, numerical (Monte Carlo) simulation can be efﬁciently applied to study the complex dynamic system representing the industry.

2.4. Parameters Effects The characteristics of the industry network and the knowledge production process are affected by ﬁve parameters, which can essentially be discussed in terms either of their effect on the proﬁtability of allying (d and r), or in terms of their effect on the dynamics of innovation (D and l). The ﬁfth parameter, p0, will be seen to have only a transient inﬂuence. First, joint innovation is proﬁtable in expected terms (and so the alliance is worth creating for both partners) if and only if the partners’ overlap lies in (dr, dþr). This interval depends on d and r, which control the position and width of the inverted-U. d controls the optimal overlap when combining knowledge endowments. Suppose i and j hold ai and aj ideas, and suppose further these ideas are uniformly distributed over the possible w locations for the two ﬁrms. Then the number of hits between i and j is binomially distributed with parameters w and aiaj/w2. The expected size of i and j’s intersection is aiaj/w, and thus the likelihood of them allying peaks when d ¼ aiaj/w. At the outset, all ﬁrms hold each idea independently with probability p0. Thus partnering andﬃ pﬃﬃﬃﬃﬃﬃﬃﬃ structural similarity will peak when d ¼ wp20 ; or equivalently p0 ¼ d=w (see also the discussion in Section 2.2). As time passes, ﬁrms knowledge endowments change, and the ratio of what they know to the total relevant knowledge changes (which is roughly equivalent to changing p), which will affect their networking decisions and innovative success. r controls the width of the inverted-U, i.e., the extent to which deviating from the optimal distance is acceptable for joint innovation. Having r close to 0 implies very little networking, as the condition for establishing an alliance is very tight. Larger values of r sustain partnering, possibly to an artiﬁcially large extent. Second, innovation itself, whether joint or autarchic, depends on D, which scales the magnitude of innovative jumps, and l, which scales innovation rates, i.e. the pace of technical progress.

Innovation and Network Dynamics

439

D is the largest possible innovative step taken from the current frontier at any point in time. It exerts a central inﬂuence on the processes of knowledge accumulation and networking. Accumulated knowledge dictates how easily a ﬁrm can ﬁnd partners, in turn partnering affects the speed of knowledge accumulation, which again affects the ﬁrm’s standing in terms of future partnering. The effects of increasing D are easily understood. All else equal, a larger D increases the probability of the next innovation being beyond the current frontier, which increases the probability that every ﬁrm’s endowment falls. So at any point in time we would expect on average a smaller number of ideas held by the ﬁrms when D is larger. Consider a single ﬁrm i. Suppose it innovates. This can happen either within the frontier or beyond the frontier. If it holds a total of at ¼ a elements at time t, there are waþD possible locations at which it can innovate, of which wa are within the frontier. Thus the probability it innovates inside the frontier is (wa)/(waþD). If this happens then the variation in the ﬁrm’s knowledge stock is 1: atþ1 ¼ aþ1. With probability D/(waþD) the ﬁrm innovates outside the frontier. In this case, the location of the innovation is l, wþ1rlrwþD, and the ﬁrst lw locations are dropped and replaced by zeroes. Assuming that the a facts it knows are uniformly distributed over the w locations,10 in expected value the ﬁrst lw locations hold (lw)a/w pieces of knowledge. As the expected value of l is (2wþDþ1)/2, the expected change in the knowledge stock is simply 1(Dþ1)/2 a/w. Thus if a ﬁrm innovates, the expected variation of its knowledge stock is wa D að D þ 1Þ þ 1 waþD waþD 2w In a world of homogeneous ﬁrms, if there is an innovation the probability that ﬁrm i makes it is simply 1/n. If a different ﬁrm, j, makes it, i’s knowledge stock is unaffected unless j innovates beyond the frontier, which it does with probability D/(waþD).11 In this case, in expected value, i’s knowledge stock varies by a(Dþ1)/2w. Thus following any innovation, the expected change in i’s knowledge stock from period t to period tþ1 is 1 1 aðD þ 1Þ n1 D að D þ 1Þ waþD 1 nw a þ D 2w n w a þ D 2w Deﬁning a stationary number of ideas as a value of a such that the expected change is zero yields 2wðw þ DÞ an ¼ (2) 2w þ D2 n þ nD

440

ROBIN COWAN AND NICOLAS JONARD

It is immediately seen that a is falling with D, and decreasing with n. Both effects are intuitive: a larger upper bound to innovative jumps means a more rapid obsolescence of ideas, and thus on average an emptier idea set; similarly, more innovations made by more other ﬁrms in the industry means fewer ideas entering my vector relative to the number of ideas exiting my vector, and so on average an emptier idea set. We will see in Section 3.3.1 that the stationary value a is a fairly good approximation of the real behaviour of the system. From the above calculation we can also immediately derive a few straigthforward implications. The stationary number of ideas will interact with d the optimal overlap required for innovation. Suppose the variance of the distribution of the number of ideas across ﬁrms is small, so that every ﬁrm holds roughly pﬃﬃﬃﬃﬃﬃ the same number of ideas a. If a is much less or much more than wd it will be very unlikely that networking pﬃﬃﬃﬃﬃﬃ is active. By contrast, if the stationary number of ideas is close to wd networking will be more intense (see also the discussion in Section 2.2).12 A second aspect of innovation is encapsulated in l, the arrival rate of innovations. This parameter controls the probability (independent across ﬁrms and partnerships) of each innovative attempt yielding the discovery of a new idea. The larger l, the more rapid the pace of innovation and the obsolescence of previous ideas. However, the stationarity of industry behaviour (in the sense of it being a stationary Markov process) is unaffected by the speciﬁc value of l, and so the only real impact of l on the system is that changes in knowledge stocks and thus alliance networks from one period to the next are of greater magnitude. As a result, industry statistics would behave much less smoothly over time with larger l than with smaller. Finally, the initial amount of knowledge in the industry is controlled by p0, the probability of any ﬁrm holding any idea at the industry birth independently from other ideas and ﬁrms. A larger p0 implies on average a higher number of ideas held by the ﬁrms early in history. This parameter however only has a transient effect. Indeed, as argued in Section 2.3, the industry is a stationary Markov process and such processes are initial-conditions independent. However, how fast a particular ﬁrm and as a consequence the industry as a whole will approach (and then ﬂuctuate around) the stationary knowledge stock identiﬁed above will be affected by p0, through the effect of the latter on the intensity of networking and thus on the number of innovations per period.

441

Innovation and Network Dynamics

2.5. Settings We consider a population of n ¼ 100 ﬁrms and a relevant knowledge base of w ¼ 100 ideas. At the outset individual knowledge endowments are randomly drawn over {0,1} with initial probability p0 ¼ 1/2, independently for every element in each ﬁrm’s idea set. The autarchic innovation rate l is set to 5% (20 periods is the expected waiting time between two consecutive innovations for a given ﬁrm). For the inverted-U we use 1 f ð yÞ ¼ 2 y2 þ 2dy d2 þ 2r2 ; r40 (3) lr At the optimal overlap f(d) ¼ 2/l (twice the innovation rate of internal R&D) and c ¼ 1/l. Let f(dr) ¼ f(dþr) ¼ c, using r to control the width of the inverted-U. As alliances form if f(y)Zc, the base width of the inverted-U (2r) and c play equivalent roles: a larger c is equivalent to a smaller r, both reducing the number of partnerships that can form. The equilibrium network in each period is thus simply g ¼ {ij : f(yij)Zc} ¼ {ij : |yijd|rr}. Finally set r to 10, so that 10% of the knowledge base is the largest acceptable deviation from the optimal amount of overlap. Each period ﬁrms form pairs (or stay alone), attempt to innovate, possibly discover new knowledge which makes old ideas obsolete, and part company. In the ﬁrst stage of the analysis, our interest will be static: given the state of ﬁrms’ knowledge, what are the characteristics of the network that forms? Then we will turn to a longer term perspective. Preliminary experiments have suggested that the system reaches its stationary regime after roughly speaking 500 periods. So we will focus on the periods 500–800, collecting averages of the statistics each period. We will also look at time series, to understand the type of cyclical behaviour displayed by network structure. There remain two independent parameters, d and D, which we vary from 0 to 60 and 1 to 10, respectively. For each point in the parameter space, we generate one history of length 800 periods and retain ﬁrst period results and ﬂuctuations around the stationary values from the periods after 500.

3. RESULTS In this section we begin by examining the static, ﬁrst period industry in which knowledge is identically and independently distributed across ﬁrms.

442

ROBIN COWAN AND NICOLAS JONARD

Then we turn to the long-run behaviour of the industry, both examining cyclical patterns in time series and analyzing the stationary distribution of several statistics of the system.

3.1. First Period Outcome The ﬁrst period results are summarized in the two ﬁgures below. The intuitions gathered from the restricted, special case (Section 2.2) suggest that small worlds, as deﬁned by the conjunction of high clustering and low average path lengths, are likely to be present in some parts of the parameter space. To conﬁrm this conjecture numerically, we have created a large number of networks and computed several network statistics. Sticking to w ¼ 100 and n ¼ 100, we generate 1,000 equilibrium networks for different values of p0 and with d ¼ 20. To abstract from issues of isolated ﬁrms, when discussing clustering and characteristic path length we focus on the largest component, i.e. the largest subset of the industry such that any two ﬁrms in it are linked by a path. Clustering coefﬁcient and average distance have to be rescaled by a random ‘‘counterpart’’ in order to distinguish the presence of structure from mere randomness. The approximations used in the literature for path length and clustering of random graphs13 are only reasonable for very large populations. Our smaller networks demand a specially tailored normalization. For each network we wish to rescale, we record density and then generate 1,000 random networks with exactly that density. Taking the average clustering and path length over that sample, we use these as the random counterpart with which to normalize our statistics. The small-world ratio is then simply the ratio of rescaled clustering over rescaled distance. In addition we show the skewness of the link distribution and the correlation between knowledge levels and degree. pﬃﬃﬃﬃﬃﬃﬃﬃﬃ 5ﬃ illustrates the special role of p0 ¼ d=w; which is equal to pFig. ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 20=100 ¼ 0:447 in the case at hand. At this level of initial endowment, the likelihood of partnering is maximized, and so is the value where the largest component size and average degree are maximized. Moving away from pﬃﬃﬃﬃﬃﬃﬃﬃﬃ d=w in either direction implies a decline in these two measures. Note that skewness (asymmetry in the degree distribution) and the SW ratio behave in the opposite way: an almost complete graph is uninteresting in terms of skewness (everyone has approximately n-1 connections) and pﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ structure. Rather, it is when initial endowments move away from d=w that sparser

443

Innovation and Network Dynamics Largest component (left axis) Average degree (left axis) Small world ratio (left axis) Skewness (right axis) 100 600 80

500 400

60

300 40 200 20

100 0

0 0.2

0.4

0.6

0.8

1.0

p0

Fig. 5.

Network Characteristics in the First Period.

networks can form, with a potential for richer architectures. Speciﬁcally, around 0.2 and 0.65 the equilibrium graphs displays the largest amounts of asymmetry and clustering. These are the regions where our model produces results that resemble the networks observed empirically. Fig. 6 shows four curves: rescaled clustering, rescaled distance, joint discoveries as a proportion of total discoveries, and the correlation between knowledge and degree. The ﬁrst two simply decompose the ratio, shown in Fig. 5. Not surprisingly the proportion of innovations that are made in an alliance as opposed to being made by ﬁrms acting individually tracks degree faithfully. The more alliances there are, the more innovations will be made jointly. Finally we can see that the correlation between knowledge and pﬃﬃﬃﬃﬃﬃﬃﬃﬃ degree changes sign as p0 crosses the d=w threshold. Left of that value ﬁrms hold on average too few ideas, so a more successful ﬁrm is one with more ideas than average. Right of that value ﬁrms hold on average too much knowledge, thus ﬁrms that are successful in ﬁnding partners are those with fewer than average ideas.

444

ROBIN COWAN AND NICOLAS JONARD Correlation of degree and knowledge (left axis) Share of isolated innovations (left axis) Rescaled average distance (left axis) Rescaled clustering coefficient (right axis) 1.5 50 1.0 40 0.5 30 0.0

20

-0.5

10

0 0.2

0.4

0.6

0.8

1.0

p0

Fig. 6.

Network Characteristics in the First Period Continued.

3.2. Fluctuations in the Stationary Regime In this section we observe the ﬂuctuations of the system in its stationary regime. Again (as stated in Section 2.3), stationary refers to the fact that the system spends a ﬁxed proportion of time (encapsulated in the stationary distribution) in each possible state. It should not be understood as convergence to a single absorbing ﬁxed state. Thus there will always be ﬂuctuations for any statistic of interest. In addition, the time average of the ﬂuctuations of a given statistic is an unbiased estimator of the expectation of that statistic computed with the stationary distribution. Thus we can proceed in two ways. In this section we freeze d and D (at d ¼ 35 and D ¼ 2) and ask how the system changes over the longer time horizon, once it has reached its stationary distribution. In the next section we ask about how average behaviour (over time) responds to the two parameter d and D. This we do by running the model for a large range of the two parameters and looking at averages of the interesting statistics (degree, clustering, and so on) over the last 300 periods.

445 14

14 12

13

10 12 8 6

11

4 10 2

Average degree (left) Average number of ideas (right)

0 500

550

600

650

700

Time

Fig. 7.

750

Average number of ideas

Innovation and Network Dynamics

12.5 12.0 11.5 11.0 10.5 10.0

9 800

4

6

8

10

12

14

Average degree

Average Degree and Number of Ideas.

The left panel of Fig. 7 displays the behaviour of the average number of ideas held and average degree over time. Both time series display ample ﬂuctuations, with the average degree oscillating between 2 and 15, while the average number of ideas evolves between 10 and 12. The two statistics are negatively correlated, as illustrated by a scatter-plot of ideas versus degree provided, in the right panel of Fig. 7. So the industry displays a variety of patterns, marked with outbursts and collapses in network activity (for a discussion of a simple model generating comparable patterns, see Marsili, Vega-Redondo, & Slanina, 2004). A batch of innovations by several ﬁrms will impoverish the industry in general by making obsolete much pre-existing knowledge. pﬃﬃﬃﬃﬃﬃ This will trigger a decline in networking (as a has moved away from wd) and the slow accumulation of knowledge through internal R&D until the point where networking increases again, which creates the possibility for the next collapse through a lucky sequence of innovations. Additional aspects of network structure are shown in Fig. 8. In the left panel, the clustering coefﬁcient shows strong oscillations between 0.3 and 1, with larger values of clustering being more prevalent. In parallel, average distance ﬂuctuates between 1 and 2. The general situation is strongly small worldish, with the small-world ratio of rescaled clustering over rescaled path length being systematically large as seen in the right panel of Fig. 8. We conclude this section with some elements on the occurrence of repeated ties. Fig. 9 depicts the evolution of the proportion of ties existing at time t which were in place at t1. Thus this proportion is a measure of the extent of turmoil at the micro-level, which direct observation of aggregate variables cannot capture: typically a constant number of edges in a graph does not imply that edges are held by the same ﬁrms over time.

446

ROBIN COWAN AND NICOLAS JONARD

1.2

Clustering coefficient Average distance

2.0

1.0

6 5

Log of small world ratio

4 1.8 3

0.8

1.6

0.6

1.4

2

1.2 0.4 1.0 500

Fig. 8.

550

600

650 700 Time

750

100

800

9

500

550

600

650 700 Time

750

800

Average Clustering and Distance in the Network, and the Small World Ratio.

1.0

0.8

0.6

0.4

Proportion of today's ties which repeat yesterday's ties

500

Fig. 9.

550

600

650 Time

700

750

800

Stability at the Micro-Level Measured by the Proportion of Repeats.

The time series in Fig. 9 (again) display signiﬁcant ﬂuctuations: periods of stability in the network (in which all of today’s ties are repeats) can be followed by more disruptive periods where the network reorganizes. Again these changes are triggered by particular sequences of innovation, in which repeated ties tend to be associated with unlucky innovative attempts, while successful innovations in the knowledge space tend also to create innovations in network (re-)organization.

447

Innovation and Network Dynamics

3.3. Long-Run Average Behaviour In this section time averages are computed to obtain a more compact representation of the relationship between our two independent parameters d and D, and various industry statistics. We begin with a check of our intuitions and calculations regarding the existence of a stationary number of ideas with limited ﬂuctuations around it. 3.3.1. Knowledge Fig. 10 displays the characteristic features of knowledge accumulation: the average number of ideas and the stationary prediction derived in Section 2.4. To construct the box-plots below, we have pooled all the observations for the average number of ideas across time and all d values. The middle of the box-plot is the sample median, the box top (bottom) is the 75th (25th) percentile and the whisker top (bottom) is the 90th (10th) percentile. The stationary value a turns out to be a very reliable upper bound to the average number of ideas held by the ﬁrms for any D value, and displays the same sensitivity to changes in D as the computed average. The median number of ideas however is always signiﬁcantly lower than a. The reason is that the calculation of a does not take into account the effects of alliance, namely joint innovation. We have derived a under the (strong)

50 40 Median number of ideas Stationary prediction

30

20 10

0 2

Fig. 10.

4

Δ

6

8

10

Box-Plots for the Population Average Number of Ideas Pooled across all Values of d and all Final 300 Periods and the Stationary Value a.

448

ROBIN COWAN AND NICOLAS JONARD

assumption of exclusively autarchic and independent innovation. Adding joint innovation will result always, though to a varying extent, in more rapid technological change, that is, more rapid obsolescence and thus fewer relevant ideas held on average. Regarding ﬁrms’ heterogeneity, the coefﬁcient of variation in number of ideas across ﬁrms (not shown pﬃﬃﬃﬃﬃ here) is always small (below 1.5, for a maximum possible value of 99 ¼ 9:9 with n ¼ 100 ﬁrms) so that the average number of ideas held is representative of the behaviour of the population of ﬁrms. 3.3.2. Network Fig. 11 displays for each network statistic the time average of the ﬁnal 300 periods. As discussed in Section 2.3, this is a good representation of the general behaviour of the system. Average degree within the largest component indicates differentiated levels of network activity across the (d, D) space. There is a region of intense partnering: a very dense graph (almost complete) obtains when dr10 and DZ2. When DZ2 the average number of ideas held by ﬁrms is very small. Thus partnering is only possible if the demands of knowledge overlap are very weak. The width of the inverted-U is 20 (d710) so when dr10 even zero overlap results in a proﬁtable alliance. As d increases from 10, though, the probability of ﬁnding a partner falls very rapidly, especially for larger D values. For lower values of D, the average number of ideas becomes relatively large, and networking is sustainable at larger values of d.14 Consequently, the case D ¼ 1 is markedly different from all other D values, with an interior peak in networking (around d ¼ 20), and partnering existing for the whole range of d values. The logic of these patterns arise entirely from two properties of the model: As D increases, the average number of ideas held falls, relatively quickly; and for a given average knowledge level (provided it is not too low), the probability of ﬁnding a partner (all else equal) rises and falls with d. This all follows from Section 2.2. We can observe a frontier between a region of autarchy, in the upper right, and networking ﬁrms in the lower left. The system has interesting behaviour along the border between these two regions (the contour line 1, in the average degree panel). In this boundary region all indicators (rescaled path length and clustering, small-world ratio of rescaled clustering over rescaled path length, skewness) point to the presence of skewed small worlds. Finally, in this zone of small worlds, the correlation between knowledge and degree is positive, showing the value to a ﬁrm of holding more ideas.

449

Innovation and Network Dynamics 10

10

20 8

8 4000

40 20

2

60 0

10

20

30

δ

40

500 500

0

10

10

10

Δ

80

5

60

4

40

5

20

2

0

5 0

10

20

30

40

50

δ

30

40

50

8

1.2

0.5

1.0

6

0.8

Δ

6

20

1

0.6

0.5

4

0.4

0.5 2

1

1

0

0.2 0.0

1

10

20

δ

30

40

Rescaled average distance

100

Rescaled clustering coefficient

120

5

-4000

0.5

5 8

-2000

500 500

50

5

0

4

2

0

20

2000

Δ

Δ

1

4

6

Skewness

60

Average degree

80 6

50

δ

10

10

5

8

8 140 100 80

5

60

4

40 20

5 2

0

5 0

10

20

δ

30

Fig. 11.

40

50

0.6

0.5

6

0.4

Δ

Δ

6

0.8

Small world ratio

120

0.2

4

2

0.5 -0.2

0

0.0 -0.2

0.5 10

0.5 20

30

40

50

δ

Network Characteristics in the Steady State.

From the discussions in Section 2.2 (spike restriction) and Section 3.1 (period 0), the results reported here tell us that similarity-driven joint innovation still produces small world type of networks when embedded in a larger model of the creation (discovery) and destruction (obsolescence) of

Correlation of degree and knowledge

5

450

ROBIN COWAN AND NICOLAS JONARD

ideas. Over some parts of the parameter space an almost complete or an empty, and thus relatively uninteresting graph forms. However there are also regions where the emerging structure is sparser, leaving room for a richer organization of ties. There, clustered groups tend to form and persist, disbanding after a particular sequence of innovations, before such new groups form again.

4. CONCLUSION Our model of joint innovation lies at the heart of the alliance formation process. It has the feature of being ‘‘tunable’’ in terms of the degree to which the knowledge frontier can advance in response to a single innovation (our parameter D). One thing we observe here is that in an industry in which there is large innovative potential (large D) rapid progress in terms of the frontier implies that knowledge is made obsolete more quickly, and so at any point in time the average ﬁrm knows fewer of the relevant facts. (Recall that we have assumed that at the industry level there is a constant number of relevant facts, w.) This suggests greater differentiation among ﬁrms when the innovative potential is large. In addition, ﬁrms that are successful in pushing the frontier will have more relevant knowledge than other ﬁrms, and so will have two competitive advantages: they will possess the most recent discoveries, and they will be more knowledgeable generally. In industries with smaller innovative potential (as measured by size of potential jumps beyond the frontier) frontier ﬁrms will lose the second advantage, as the average ﬁrm knows a higher proportion of the relevant facts. Firms are more similar, and the differentiating feature generally is how recently a ﬁrm has pushed the frontier in a way that makes other knowledge obsolete. A second observation regarding innovative potential is that over the life cycle of an industry we might expect innovative potential to fall. Because our model is stationary, the results we have produced regarding how the model behaves in different parts of the states space can be used to make conjectures about how an industry network might change over time. If the industry transits through its life cycle relatively slowly, then as parameters change due to its aging, it will fairly quickly respond and move to a new steady state. Thus in essence, as an industry ages it will move through the parameter space of Fig. 11. Focussing on the declining innovative potential, as an industry ages D will fall – we move vertically down the panels in Fig. 11. As is well known, entry and exit form an important part of the patterns we observe over time, so the top left panel of the ﬁgure is likely to

Innovation and Network Dynamics

451

be misleading, due to its heavy dependence on the number of ﬁrms being constant. The others, however are less dependent on the number of ﬁrms. If it is generally the case that innovative potential falls over the life of an industry, then we can see, by tracing vertically down the panels of Fig. 11 that we can expect clustering to rise and then fall as an industry ages (middle left panel). By contrast, average distances do not respond strongly to D. These jointly suggest that a young industry will not look like a small world, a middle-aged one will, and an old one probably will, but in a less strong fashion (bottom left panel of Fig. 11). Finally, the skewness of the link distribution also rises and falls as time passes. Because many things apart from innovative potential change over the course of a life cycle this is certainly not the whole story, but it could be suggestive of patterns that might emerge empirically. In this chapter we have argued that the nature of joint innovation alone is enough to produce industry networks that share many properties of empirically observed strategic alliance networks. When choosing alliance partners, ﬁrms clearly take many things into consideration. The empirical literature, however, has focussed very heavily on issues of social capital, and has paid much less attention to the fact that partnering ﬁrms must also have a ‘‘cognitive ﬁt’’. In this way our model is complementary to the work of Reagans and McEvily in this volume (Reagans & McEvily, 2008). They use knowledge overlap as a (linear) control in examining formation of knowledge transfer and knowledge-seeking links within an organization. Their main interest is in network positioning effects, but this control is highly signiﬁcant in their results. One interesting extension to their work would be to ask whether expertise overlap can ‘‘do more work’’ in explaining link formation with a richer (non-linear) speciﬁcation. Our model does present a strong contrast to two other chapters in this volume by van Liere et al. and Amburgey et al. (van Liere, Koppius, & Verest, 2008; Amburgey, Al-Laham, & Tzabbar, 2008). In both of those chapters alliance formation is constrained by network distance – in effect ﬁrms cannot see very far along the network, and so partner choice is constrained to ﬁrms that are close by in network space (typically friends of friends). This is consistent with the idea that many alliances are formed through referrals by common third-party ties. In our model, every ﬁrm has good knowledge about every other ﬁrm in the industry, both that it exists, and precise details about its knowledge portfolio. This is clearly a strong assumption. But this assumption means that a ﬁrm can chose as a partner any other ﬁrm in the industry, not just those ﬁrms already close in network space. In a sense, this makes our results that much stronger: with no

452

ROBIN COWAN AND NICOLAS JONARD

considerations of network distance constraining ﬁrms to form links within a small neighbourhood, we still see the emergence of clustered networks. While not denying the importance of social capital considerations, we show here that cognitive ﬁt can be a very strong inﬂuence on the types of networks that form.

NOTES 1. See for instance the studies of Powell, White, Koput, and Owen-Smith (2005) on the biotech industry, Baum, Shipilov, and Rowley (2003) on bank syndicates, Riccaboni and Pammolli (2002) on the life sciences and ICT industry, and Ahuja (2000) on the international chemicals industry. 2. The value of that form of social capital is empirically observed by Dyer and Nobeoka (2000) in the automobile industry, Gulati and Gargiulo (1999) in several industries, Powell, Koput, and Smith-Doerr (1996) in the biotechnology sector, and Rowley, Behrens, and Krackhardt (2000) in the steel and semiconductor industries. See also the ﬁndings and detailed discussions on the value of embeddedness in Walker, Kogut, and Shan (1997). 3. The value of structural holes is examined in Ahuja (2000) in the context of the international chemical industry (structural holes have a negative impact on industry performance, whereas indirect and direct ties have a positive impact on ﬁrm innovative performance). Gargiulo and Benassi (2000), in a study of Italian IT ﬁrm, ﬁnd that a trade-off emerges, which is associated with the safety conferred by cohesive ties – social capital – and the ﬂexibility conferred by ties that connect different parts of a network; Baum et al. (2003) in the Canadian merchant banking industry. 4. See for example Ahuja and Katila (2001), Gulati and Gargiulo (1999), Mowery, Oxley, and Silverman (1998, 1996), or Schoenmakers and Duysters (2006). In this volume, Reagans and McEvily use ‘‘expertise overlap’’ as a control variable, and ﬁnd that it is strongly signiﬁcant for formation both of knowledge transfer and knowledge-seeking links. They include only a linear effect, though, so we cannot tell whether it is possible to have ‘‘too much’’ overlap from their results. 5. Distance is measured differently in different studies, though often based on patent data. The inverted-U relationship seems to hold nonetheless. 6. In fact ‘‘overlap’’ is a negative measure of distance, but given the symmetry in the inverted-U, the intuition and arguments are unaffected. Overlap also implies similarity, and we argued above that both similarity and complementarity were necessary for successful alliances. But complementarity is guaranteed to exist, except in the exceptional case where the two partners have exactly the same knowledge stocks (which happens with very small probability, approximately pw, where p is the probability that a ﬁrm knows any particular fact). 7. Implicitly we are assuming that a fact new to a ﬁrm and a fact new to the world have the same value. It could be argued that in a second stage in which ﬁrms use the knowledge they have discovered to create competitive advantage, facts ‘‘new to

Innovation and Network Dynamics

453

the world’’ are more valuable than facts ‘‘new to me’’, but here we only focus on the immediate production of knowledge. A second implicit assumption is that ﬁrms are indifferent about which facts they discover. To include the more realistic idea that ﬁrms get more value from one fact than another, would demand both a detailed speciﬁcation of how ﬁrms turn facts into proﬁts and how facts interact in that process, and a detailed speciﬁcation of how ﬁrms turn existing knowledge into new knowledge. These are of course possible in principle, and material for an extension of the model, but the complication they would add at this point would distract signiﬁcantly from the main message. 8. There is a second-order effect that we ignore. If i has a partnership with j, the partnership ij can innovate in any location in L ¼ W vi vj . If i now considers k as a potential partner, any innovations that ik might make in L would be duplicates, and therefore of less value than innovations that take place outside L. Thus the evaluation of k as a partner should involve this second-order effect, and in general, the evaluation of a portfolio of partners should include these interactions. We ignore this here and in what follows, since in the numerical experiment below success probabilities are small enough that these second-order effects will have very little effect on decisions. In the experiment later, a ﬁrm innovates on average once per period, thus the risk of duplicate innovations is very remote. Including this effect would on average lower the expected value of an alliance, and so decrease the degree of the network. 9. This can be seen by construction. Find a ﬁrm, i, in ðv01 ; . . . ; v0n Þ that knows the frontier fact. From ðv1 ; . . . ; vn Þ have ﬁrm i alone innovate at wþD until all other ﬁrms’ vectors are empty. Then have each ﬁrm innovate autarchically in the locations where it has knowledge under state ðv01 ; . . . ; v0n Þ. Each of these events has positive probability, so the transition from ðv1 ; . . . ; vn Þ to ðv01 ; . . . ; v0n Þ has positive probability. 10. At ﬁrst glance this assumption seems incorrect, since after an innovation that takes place at wþz, expanding the frontier, all of i’s knowledge is held in the ﬁrst wz places. However, the position of i’s knowledge in the long run is determined by when he innovates relative to when other ﬁrms have frontier expanding innovations. All ﬁrms are equally likely to innovate in each period, and so there is no pattern. i’s knowledge, on average, will be spread evenly over the w positions. 11. The assumption of homogeneous ﬁrms seems strong. In Section 3.3.1, we show that in terms of the number of facts held by a ﬁrm, there is in fact little variation, so this assumption is not as strong as it seems at ﬁrst glance. 12. The above calculation is an approximation of what happens in the more complex situation in which ﬁrms partner and jointly innovate, but the general intuitions are correct, as will be seen later. We have also made the implicit assumption that ﬁrms are homogeneous in terms of where innovation takes place in their idea sets, so a should be seen as the expected value of the stationary number of ideas rather than the stationary value itself. 13. If n is average degree, the standard approximations are n=ðn 1Þ for the clustering coefﬁcient and lnðnÞ= lnðnÞ for the characteristic path length. 14. From Section 2.2, when the inverted-U is a spike of zero width, networking pﬃﬃﬃﬃﬃﬃﬃﬃ peaks when p d=w. D determines the average steady-state knowledge quantity a, which can be seen as a ¼ pw. Thus we can calculate directly the value for d for which networking should peak for different values of D. A list of (D, d) ordered pairs

454

ROBIN COWAN AND NICOLAS JONARD

illustrates: (1, 25), (2, 7), (3, 2), (4, 1), and (5, 0). The reason we see any networking at all for values of D above 4 has to do with the width of the inverted-U, and only for D ¼ 1 will we observe the rise and fall of networking as d increases, that we would expect from the special case calculations.

ACKNOWLEDGMENTS We gratefully acknowledge comments from Terry Amburgey, Joel Baum, Ronald Burt, Tim Rowley, Gordon Walker and other participants to the Network Strategy Conference held in Toronto, May 2007. We acknowledge support from the European FP6-NEST-Adventure Programme, contract no. 028875, Network Models, Governance and R&D collaboration networks (NEMO).

REFERENCES Ahuja, G. (2000). Collaboration networks, structural holes and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Ahuja, G., & Katila, R. (2001). Technological acquisitions and the innovation performance of acquiring ﬁrms: A longitudinal study. Strategic Management Journal, 22, 197–220. Amburgey, T. L., Al-Laham, A., & Tzabbar, D. (2008). The structural evolution of multiplex organizational networks: Research and commerce in biotechnology. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 171–209). Bingley, UK: JAI/Emerald Group. Baum, J. A. C., Shipilov, A. V., & Rowley, T. J. (2003). Where do small worlds come from? Industrial and Corporate Change, 12, 697–725. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Coleman, J. A. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Cowan, R., & Jonard, N. (2008). Knowledge portfolios and the organization of innovation networks. Academy of Management Review, forthcoming. Dyer, J. H., & Nobeoka, K. (2000). Creating and managing a high performance knowledge sharing network: The Toyota case. Strategic Management Journal, 21, 345–367. Gargiulo, M., & Benassi, M. (2000). Trapped in your own net? Network cohesion, structural holes, and the adaptation of social capital. Organization Science, 11, 183–196. Gulati, R., & Gargiulo, M. (1999). Where do inter-organizational networks come from? American Journal of Sociology, 104, 1439–1493. Jackson, M. O., & Wolinsky, A. (1996). A strategic model of social and economic networks. Journal of Economic Theory, 71, 44–74. Marsili, M., Vega-Redondo, F., & Slanina, F. (2004). The rise and fall of a networked society: A formal model. Proceedings of the National Academy of Science of the USA, 101, 1439–1442.

Innovation and Network Dynamics

455

Mowery, D. C., Oxley, J. E., & Silverman, B. S. (1996). Strategic alliances and interﬁrm knowledge transfer. Strategic Management Journal, 17, 77–91. Mowery, D. C., Oxley, J. E., & Silverman, B. S. (1998). Technological overlap and interﬁrm cooperation: Implications for the resource-based view of the ﬁrm. Research Policy, 27, 507–523. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Inter-organizational c-ollaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41, 116–145. Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network dynamics and ﬁeld evolution: The growth of inter-organizational c-ollaboration in the life sciences. American Journal of Sociology, 110, 1132–1205. Reagans, R., & McEvily, B. (2008). Contradictory or compatible? Reconsidering the ‘tradeoff’ between brokerage and closure on knowledge sharing. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 275–313). Bingley, UK: JAI/Emerald Group. Riccaboni, M., & Pammolli, F. (2002). On ﬁrm growth in networks. Research Policy, 31, 1405–1416. Rowley, T., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21, 369–386. Schoenmakers, W., & Duysters, G. (2006). Learning in strategic technology alliances. Technology Analysis and Strategic Management, 18, 245–264. van Liere, D., Koppius, O., & Verest, P. (2008). Strengthening of bridging positions: Network horizon and network horizon heterogeneity. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 595–639). Bingley, UK: JAI/Emerald Group. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8, 108–125.

BRINGING THE FIRM BACK IN: NETWORKING AS ANTECEDENT TO NETWORK STRUCTURE Ravi Madhavan, Turanay Caner, John Prescott and Balaji Koka ABSTRACT In the network strategy view, relative competitive advantage stems not merely from opportunity structures embedded in networks but also from the distribution of ability and motivation among ﬁrms. Thus, there is a need to ‘‘bring the ﬁrm back in’’ to the network strategy narrative. We demonstrate that a mixed-methods design, blending large-sample data with micro-data on speciﬁc ﬁrms and their networks, can increase our understanding of the interplay of network structure and actor mechanisms, thus bridging the chasm between theory and practice in network strategy. We believe this is a critical step toward the ‘‘strategic design of networks.’’

BRINGING THE FIRM BACK IN Suppose a network study establishes that centrality predicts innovation (as some network studies have). How does one take that ﬁnding and apply it to Network Strategy Advances in Strategic Management, Volume 25, 457–501 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25013-8

457

458

RAVI MADHAVAN ET AL.

the speciﬁc situation of a manager who might ask, ‘‘Does it mean that I should increase my ﬁrm’s centrality? If yes, how should I go about doing it?’’ Clearly, what is involved in such a translation is that various ‘‘adjustments’’ may need to be made to the aggregate ﬁnding – for example, the ﬁrm’s current centrality level, its level of complementary assets, etc. (As an analogy, consider how ﬁnancial valuation proceeds from the identiﬁcation of comparable assets – whether homes or companies – and then through a successive series of adjustments to arrive at a usable number.) Inspired by this volume’s theme of adopting a dynamic view of networks including network entrepreneurship, we propose that understanding the action implications for a particular ﬁrm is somewhat different from (although complementary to) understanding the statistical relationships between variables at the aggregate level. Thus, bringing ﬁrm agency back into the network analysis narrative is a crucial ﬁrst step in understanding how to move toward a workable model for ‘‘translating’’ aggregate ﬁndings to meaningful managerial prescriptions and entrepreneurial opportunities at the level of the speciﬁc ﬁrm. Similarly, understanding individual ﬁrms’ networking activity will suggest new opportunities for multi-level research studies. Indeed, Hagedoorn and Frankfort (2008) make the theoretical argument that interaction among dyadic, interorganizational, and environmental levels of embeddedness affects ﬁrms’ networking behavior, or partnership formation. This reciprocal interaction between theory and practice creates a virtuous upward cycle of evidencebased learning applicable to scholars and managers alike. In this way, bringing the ﬁrm back into the network perspective contributes to bridging the chasm between theory and practice (Raelin, 2007).

STRUCTURALISM, NON-STRUCTURALISM AND MIXED-METHODS DESIGN Competitive advantage in a network context stems not merely from opportunity structures embedded in networks, but also from the distribution of ability and motivation between ﬁrms (Koka & Prescott, 2008). Thus, it is intuitively evident that ﬁrm attributes matter, not only in terms of speciﬁc ﬁrm characteristics inﬂuencing network structure, but also in terms of the networking behavior that they engage in. However, as Kilduff and Krackhardt (1994) pointed out, the history of network thinking has been marked by the structuralist refrain that ‘‘students of social structure need

Bringing the Firm Back In

459

not be concerned with individuals or individual-level variables’’ (p. 88). The structuralist view was captured most forcefully by Mayhew (1980, p. 339) when he argued that ‘‘Structuralists and individualists are asking different questions. They are attempting to explain different things y no shared language and no line of communication unites them in any common discourse.’’ (In our context, the ‘‘individualist’’ should be understood broadly as ‘‘non-structuralist,’’ i.e., emphasizing actor-level effects, thus encompassing ﬁrm-level constructs as well.) The structuralist dominance in the literature is highlighted by the fact that extant network studies have largely examined how network opportunity – operationalized as structure – inﬂuences ﬁrm performance, implicitly assuming that ﬁrm attributes, such as ability and motivation, were constant across ﬁrms or at least randomly distributed (Koka, Prescott, & Canella, 2007). Given such a background, it is not surprising that the nascent network evolution literature in strategy and organization is also largely structuralist in orientation. When exogenous inﬂuences on network evolution have been investigated, that has largely been at the industry level (Madhavan, Koka & Prescott, 1998; Owen-Smith & Powell, 2004). Yet, the structuralist–individualist schism is especially unfortunate in the inter-ﬁrm networks literature because of the explicit belief that ﬁrms behave strategically with respect to creating and managing networks. If such strategic behavior is indeed the case, then ﬁrm ability and motivation clearly matter (i.e., they are not randomly distributed). Supporting this view, Soda, Zaheer, and Carlone (2008) found that, if the motivation of the actor is to acquire diverse information from the network relationships, then network closure will reduce imitation among network members. Their ﬁnding goes against the structuralist argument that network closure is always associated with high levels of imitative behavior, which does not take into account the individualistic approach in network formation. In fact, we assert that ﬁrm attributes and network structure co-evolve – which is consistent with one of the themes of this volume, that understanding endogeneity effects in causal explanations is necessary to advance network theory. Interestingly, the challenge in moving toward an integration of structuralist and non-structuralist positions is less theoretical than methodological. After all, there is no dearth of qualitative and micro-level studies in the network literature. However, there would appear to be a natural division between studies that employ large data sets and quantitative analysis at the aggregate level (structuralist) and studies that focus on more intensive analysis at the actor-level (non-structural). Motivated by the above considerations, we seek to demonstrate that a mixed-methods design that

460

RAVI MADHAVAN ET AL.

blends large-sample data and analysis with a micro-level treatment zooming in on speciﬁc ﬁrms and their networks can help increase our understanding of network structure and process. Such an approach has two beneﬁts. One, mixed-methods designs help further our understanding of network evolution by focusing on both exogenous and endogenous drivers of network change. Two, it helps us develop ‘‘translational’’ techniques for taking aggregate ﬁndings and tailoring them to speciﬁc ﬁrms and visa-versa. We proceed by leveraging a ﬁnding from the dissertation work of one of the authors (Caner, 2007) that centrality positively impacts patent productivity, but that ﬁrm’s centrality is endogenous with respect to geographical location (in-cluster vs. out-cluster). When faced with such a ﬁnding, one lens to adopt is that of the endogeneity explanation – that is, that network structure may be reﬂective of ﬁrm attributes (such as location), which in turn calls for methodological countermeasures. Equally plausibly, however, we could adopt the stance of seeking greater theoretical clarity about the processes by which ﬁrm attributes lead to network structures. We began by inspecting the centrality- patent productivity curves (see Fig. 2) of four speciﬁc ﬁrms as they relate to the overall family of such curves, and what explains the particular features of those speciﬁc curves. We went from there to map the networks of those four ﬁrms, asking how ﬁrm attributes might explain the similarities and differences between them. After illustrating some of the ﬁrm-level processes and mechanisms evident from the micro-level analysis, we return to the topic of how mixed-methods designs can illuminate the central questions of network strategy.

CENTRALITY AND PATENT PRODUCTIVITY A central premise of studies involving geographical clusters is that cluster membership inﬂuences ﬁrms’ strategic choices such as alliance network structure. In particular, ﬁrms located within geographical clusters are purported to have advantages both as an initiator and as a target of alliance activities compared to ﬁrms that are located outside geographical clusters due to economic, cultural and social beneﬁts associated with cluster membership1 (Bagchi-Sen, 2004; Porter, 1998; Saxenian, 1994; Scott, 2005). Yet, approximately 53% of the US ﬁrms in the biopharmaceutical industry are headquartered outside a biopharmaceutical cluster and they do not appear to be signiﬁcantly disadvantaged. For example, on average, outcluster ﬁrms are more productive and innovative than in-cluster ﬁrms (approximate sales per employee for out-cluster ﬁrms is 0.3 MM vs.

Bringing the Firm Back In

461

0.2 MM; average number of patents granted for out-cluster ﬁrms is 94 vs. 33 for in-cluster ﬁrms, results are also statistically signiﬁcant, po0.01). We explore these conﬂicting conjectures as it relates to the relationship between cluster membership, R&D alliance network structure, and the moderating effect of absorptive capacity on innovation. Our framework has two main arguments. First, we argue that cluster membership, deﬁned as ﬁrms’ geographic location within a cluster area, affects alliance network structure by means of two alternate mechanisms: a complementary mechanism and a substitution mechanism. We theoretically establish the endogenous nature of alliance network structure with respect to ﬁrms’ cluster membership. Second, given the endogenous nature of alliance network structure with respect to ﬁrm’s cluster membership we examine the implications of alliance network structure for innovation contingent on ﬁrms’ absorptive capacity. Following prior literature, we suggest a contingent view on alliance network structure and innovation relationship. Similar to Tsai’s (2001) study, we suggest that absorptive capacity moderates the effect of alliance network structure on ﬁrms’ innovation because if ﬁrms have not developed the capabilities for acquisition, assimilation, transformation, and commercialization of new information they are unlikely to translate opportunities in their alliance network structure into innovation. Network structure in this chapter is focused on centrality to allow us to more fully explain, compare and contrast our large sample deductive conclusions with our inductive assessment of four biopharmaceutical ﬁrms. Firms are more central in their alliance networks when they are involved in a large number of alliances with other organizations (Wasserman & Faust, 1994).

Cluster Membership and Centrality There appears to be an intriguing relationship between network centrality and cluster location. Is geographic clustering an antecedent to network clustering (which leads to centrality) or is it a substitute? Key to this question is how the ﬁrm gains information within these two clustering contexts. Thus, for example, it could be that the diffusion mechanism consists of informal relationships in geographic clustering while it consists of formal relationships in network clustering. In such a view, informal ties between members of geographically proximate organizations may include friendship and other personal ties, through which useful technical knowledge may also ﬂow. To the extent that pure network clustering rules out

462

RAVI MADHAVAN ET AL.

geographic clustering, the former may depend more on formal ties based on the governance structure of the network. However, there are complications: although informal relationships, such as shared club membership, may indeed play a greater role in geographic clustering, it is plausible that such ties may also lead to formal inter-ﬁrm relationships between geographically proximate ﬁrms. Consequently, there are two possible theoretical perspectives on the relationship between cluster membership and centrality: a complementary view and a substitution view. According to the complementary view, ﬁrms located within a cluster area become more central in their alliance networks, while the substitution view suggests that in-cluster ﬁrms have less of a need to be central in their alliance networks. According to the complementary view, a ﬁrm’s in-cluster location affects its number of alliances, and thereby its centrality, by means of three mechanisms: (1) in-cluster ﬁrms form alliances with other in-cluster organizations due to reduced transaction and communication costs associated with proximate alliance partners (transaction costs); (2) in-cluster ﬁrms serve as attractive alliance partners for out-cluster ﬁrms who need access to cluster resources (resource complementarities); and (3) social networks within a cluster reduce the moral hazard problem associated with alliance partners (embeddedness). These three mechanisms are elaborated below. Firms enter into agreements in which the transaction costs are at a minimum (Williamson, 1975). Physical proximity facilitates inter-ﬁrm cooperation (Dyer, 1996; Enright, 1995). Naturally, for in-cluster ﬁrms it is less costly to form alliances with other in-cluster organizations because of the absence of long distance search, monitoring and formal communication costs. Moreover, ﬁrms utilize their local advantages in a manner to maximize the performance beneﬁts. Saxenian (1994) states that Hewlett Packard and other physically proximate ﬁrms have improved performance by setting up alliances with other ﬁrms in the Silicon Valley technology cluster. McKelvey, Alm, and Riccaboni (2003) found that ﬁrms are more likely to collaborate with co-located organizations than with international ones in the Swedish biopharmaceutical sector when they are involved in research and development. Anecdotal evidence also suggests that geographical proximity is important in alliance relationships. For example, an executive from a medical device manufacturing company states that ‘‘it is a lot easier to drive across town and visit a supplier than it is to pick up the phone and try to talk through some complicated issue’’ (Aeppel, 2006). In-cluster ﬁrms have more alliances than out-cluster ﬁrms because in-cluster ﬁrms are attractive alliance partners for out-cluster ﬁrms. First,

Bringing the Firm Back In

463

in-cluster ﬁrms have access to cluster resources. For example, factor endowments such as production inputs are localized (Feldman, 2000). Second, clusters provide various opportunities for collaboration among universities, research-intensive biotechnology ﬁrms, and large pharmaceutical corporations located within a cluster. Indeed, Arora and Gambardella (1990) state that there are systematic linkages among universities, biotechnology ﬁrms, and large pharmaceutical corporations. Kogut, Walker, Shan, and Kim (1994) suggest that ﬁrms within a region share both tradable and un-tradable resources such as knowledge. This provides the motive for out-cluster ﬁrms to enter into alliances with in-cluster ﬁrms in order to beneﬁt from these un-tradable resources within clusters. Aharonson, Baum, and Feldman (2004) provides evidence that R&D alliances are a complement to in-cluster location. Bagchi-Sen (2004) shows that in-cluster ﬁrms also initiate alliances with out-cluster ﬁrms.2 Third, ﬁrms might enter into alliances with other in-cluster ﬁrm in order to beneﬁt from others’ component knowledge. Knowledge transfer among in-cluster ﬁrms happen in the form of component knowledge transfer (Tallman, Jenkins, Henry, & Pinch, 2004). Clusters provide the necessary condition for the development of social networks among employees of ﬁrms located in one location. Firms entering into alliances face moral hazard problems due to uncertainty and the likely costs of opportunistic behavior by partners (Das & Teng, 1998). Social networks help in becoming aware of such moral hazard problems associated with partners (Gulati, 1999). This enables ﬁrms to establish alliances with organizations they already know from employees’ social networks. Similarly, Inkpen and Tsang (2005) argue that social ties among employees in an industrial district serve as a foundation for formal connections among ﬁrms in these areas. Giuliani (2007) argues that business interactions which include any business related interactions (e.g., participating in fairs, vertical, horizontal trade of goods, etc.) among in-cluster ﬁrms and knowledge ﬂows among these ﬁrms are not highly co-occurring phenomena. In other words, social interaction among individuals within a cluster area does not substitute for formal alliances. Therefore, ﬁrms’ in-cluster location affects the number of alliances they form leading to more central positions in their alliance network. In contrast, the substitution perspective would suggest that in-cluster location and having access to resources in a cluster are substitutes for entering into alliances. This reasoning paves the way for the spillover mechanism in explaining the relationship between in-cluster location and ﬁrm centrality. Spillover arguments suggest that ﬁrms beneﬁt from one another due to their proximity because resources such as knowledge

464

RAVI MADHAVAN ET AL.

spillover to other parties and there are geographic boundaries to spillovers (Audretsch & Feldman, 1996; Jaffe, 1989; Jaffe, Trajtenberg, & Henderson, 1993; Krugman, 1991). Firms located within a geographically close group of ﬁrms, institutions, suppliers, and service providers have different beneﬁts than ﬁrms located elsewhere. For example, clusters tend to have social networks due to both intentional and unintentional frequent interaction among employees working for different in-cluster ﬁrms. Information ﬂows from its source through social networks in a cluster and therefore spillovers are localized. This enables in-cluster ﬁrms to beneﬁt from spillover effects. Thus, if there are spillovers then ﬁrms might not form inter-ﬁrm alliances because the need to form alliances is already satisﬁed by the spillover effects. This leads to in-cluster ﬁrms forming fewer alliances, thereby being less central in their alliance networks. Owen-Smith and Powell (2004) show that membership in a cluster provides access to information and other resources through informal channels in the region. This suggests that in-cluster ﬁrms need not form alliances with other in-cluster ﬁrms because centrality in their local network will not bring any unique beneﬁts to in-cluster ﬁrms. In summary, spillover effects and social networks in a cluster substitute for formal alliances. The contradictory effects of complementary and substitution mechanisms thus prompt two competing predictions with respect to the relationship between cluster membership and centrality. Within the complementary view, cluster membership promotes or adds to the alliance network structure and hence, increases ﬁrms’ centrality. Conversely, cluster membership substitutes for the alliance network structure leading to less centrality in the alliance network structure. Thus, we suggest that Hypothesis 1a. In-cluster ﬁrms are more central in their alliance network than ﬁrms located elsewhere. Hypothesis 1b. In-cluster ﬁrms are less central in their alliance network than ﬁrms located elsewhere.

Centrality, Absorptive Capacity and Innovation Firms’ absorptive capacity positively moderates the endogenous alliance network structure with respect to cluster membership and innovation relationship. This is important because prior work has established the relationship between alliance network structure and innovation based on the

Bringing the Firm Back In

465

strict exogeneity of alliance network structure. However, the relationship between network structure and innovation can be subject to various ﬁrm speciﬁc factors such as cluster membership. Following the contingency view in the alliance network literature, we suggest that ﬁrms’ absorptive capacity is a contingency factor that inﬂuences the relationship among centrality and innovation (McEvily & Yao, 2005; Owen-Smith & Powell, 2004). While ﬁrms’ centrality inﬂuenced by ﬁrms cluster membership provides access to information volume, absorptive capacity determines how efﬁcient and effective this information will be utilized towards innovation. Recent research suggests that ﬁrms innovate to the extent that they have absorptive capacity (Cohen & Levinthal, 1990). Absorptive capacity is the ability to acquire, assimilate, transform, and exploit external information for ﬁrm advantage (Cohen & Levinthal, 1990; Lane, Koka, & Pathak, 2006; Zahra & George, 2002). We suggest that the effect of ﬁrms’ centrality inﬂuenced by cluster membership on their innovation depends upon their absorptive capacity. Thus, centrality is a necessary but not sufﬁcient condition for innovation given the effect of cluster membership on ﬁrms’ centrality. We draw on social capital arguments to explain the relationship between ﬁrms’ centrality and innovation. We build on the arguments that alliance network structure provides information resources that are deﬁned as social capital (Bourdieu & Wacquant, 1992). Performance implication of social capital has been previously explored in the literature. For example, Koka and Prescott (2002) theoretically argued and empirically demonstrated that information resources ﬁrms acquire due to their alliances are social capital and this is contingently related to ﬁrm performance in the steel industry. Therefore, we suggest that central ﬁrms have more social capital in terms of information volume. Yet, social capital by itself does not sufﬁce to explain centrality and innovation relationship in a comprehensive way. Thus, we complement this view with a contingency argument by looking at the moderating effect of absorptive capacity for the relationship between centrality and innovation, given the endogenous nature of ﬁrms’ centrality. Based on this reasoning we suggest that although ﬁrms’ centrality in their alliances provides information volume the use of this information is likely to be contingent on ﬁrms’ absorptive capacity. That is, ﬁrms that have the ability to acquire, assimilate, transform, and commercialize information are likely to have greater innovation beneﬁts from their central position in their alliance networks. The process through which absorptive capacity is developed provides support for this line of reasoning. For example, centrality provides access to and in turn acquisition of information but

466

RAVI MADHAVAN ET AL.

this does not necessarily mean that information will be productive in terms of innovation. Firms need to have capabilities to comprehend, interpret (Lane & Lubatkin, 1998); to internalize and convert information into usable form and ﬁnally to commercialize it (Cohen & Levinthal, 1990). If ﬁrms have access to information but do not have the sufﬁcient level of capabilities to convert this information into a usable form then it is likely that they will not receive positive beneﬁts. Tsai’s (2001) study supports this reasoning; he found that the interaction between centrality and absorptive capacity has a positive signiﬁcant effect on ﬁrms’ innovation in the context of intra-ﬁrm networks. Extending his arguments to alliance networks (or inter-ﬁrm networks) we suggest that absorptive capacity will positively inﬂuence the relationship between ﬁrms’ centrality in their alliance networks and their innovation. Hypothesis 2. Given the endogeneity of ﬁrms’ centrality in their alliance network structure with respect to their cluster membership, ﬁrms’ absorptive capacity positively moderates the relationship between ﬁrms’ centrality in their alliance networks and innovation. Fig. 1 represents the research model graphically.

METHODS Sample and Data Collection The context for our research is the US biopharmaceutical sector. Both clustering and alliance networks have been shown to inﬂuence ﬁrms’ innovation in this sector (Arora & Gambardella, 1990; Audretsch & Feldman, 1996; McKelvey et al., 2003; Owen-Smith & Powell, 2004). Data Cluster Membership • In-Cluster • Out-Cluster

Centrality

Innovation Performance

Absorptive Capacity

Fig. 1.

Large Sample Study Research Model.

467

Bringing the Firm Back In

used to test our hypotheses were obtained from several sources. First, using the Mergent Online, Compustat, and Thomson databases we identiﬁed 147 public companies that designate their primary business as pharmaceutical preparations (SIC 2834). Our sample includes 847 observations for 147 ﬁrms for seven years. Since the number of observations for each ﬁrm is not the same our dataset is therefore unbalanced. Second, we relied on Mergent online, Compustat, Edgar database, Worldscope, US Census Bureau, and company websites in order to gather ﬁnancial, metropolitan statistical areas (MSA), and geographic location data including both headquarter and R&D facilities location to determine cluster membership. Third, data on R&D alliances were obtained from Recombinant Capital (Recap), a private database that tracks and analyzes alliances including biopharmaceutical ﬁrms. The Recap database has been extensively used in prior studies (Lane & Lubatkin, 1998; McEvily & Yao, 2005). Although we collected cumulative R&D alliance data pertaining to discovery and development of a pharmaceutical end product for the period between 1995 and 2004, we conducted the analysis for the 1998–2004 period in order to address time censoring issues. Our alliance database includes 5,367 alliances (including 3,086 organizations) since 1995. Finally, we gathered patent data to measure innovation from the US Patents-Granted Collection on Delphion. Delphion’s database includes information on all patents granted by the US Patent and Trademark Ofﬁce (USPTO) since 1971. Delphion has a comprehensive patents database and has also been used in prior studies (Chacar & Lieberman, 2003; Furman, Kyle, Cockburn, & Henderson, 2005).

Measures Cluster Identiﬁcation We identify the US clusters based on the Milken Institute’s America’s Biotech and Life sciences Clusters (2004) study. The Milken Institute Research Reports (2004) has identiﬁed 12 metropolitan statistical areas (MSA) in the US as biopharmaceutical clusters3. Other scholars have also used MSA to identify geographic units or clusters (Jaffe et al., 1993). In order to validate our use of Milken study we also compared clusters we employed in this study to the biopharmaceutical clusters identiﬁed by the Cluster Mapping Project at the Institute for Strategy and Competitiveness, Harvard Business School. The results of the comparison show that clusters

468

RAVI MADHAVAN ET AL.

that are identiﬁed by the Milken Study were also identiﬁed as biopharmaceutical clusters by the Cluster Mapping Project. Independent Variable Cluster Membership. Previous studies identiﬁed cluster membership based on ﬁrms’ headquarter (HQ) location in a cluster area (Aharonson et. al., 2004; Bell, 2005; DeCarolis & Deeds, 1999; Owen-Smith & Powell, 2004). Birkinshaw, Braunerhjelm, Holm, and Terjesen (2006) consider three relevant indicators of HQ location: the legal domicile, the location of top management team, and the location of the various HQ functions. Following their study, ﬁrst, we identify the location of corporate HQ in a cluster area by ﬁrms’ registered address.4 Second, based on the location of HQ functions approach we also look at the location of R&D functions because biopharmaceutical sector is highly research intensive, and research is conducted in formal R&D laboratories. In general, ﬁrms have R&D labs in their corporate HQ locations (more than two thirds of ﬁrms in our sample have R&D labs in HQ locations). Further, most of the biopharmaceutical companies that have corporate HQ location outside clusters, particularly big ﬁrms, have R&D labs located within cluster areas. By the same token, Chacar and Lieberman (2003) found that geographic organization of ﬁrms’ R&D labs have signiﬁcant effect on research productivity of pharmaceutical ﬁrms. This indicates that by having R&D labs in cluster areas ﬁrms might beneﬁt from clusters to some extent if not in full. Thus, we measure cluster membership in two different ways. First, we use HQ location to determine in-cluster and out-cluster ﬁrms. We assign a value of 1 to ﬁrms that have HQ within a cluster, 0 otherwise. Second, we also identify in-cluster and out-cluster ﬁrms based on R&D facilities’ location data. Again, we use a dummy variable which takes the value of 1 if a ﬁrm has at least one R&D lab located within a cluster area, 0 otherwise. Endogenous (Network Structure) Variable We calculate network measures based on the entire cumulative alliance network matrices for R&D alliances for the period of 1998 and 2004.5 We used UCINET-6 network analysis software in order to calculate the network measures (Borgatti, Everett, & Freeman, 2005). Centrality In hypotheses 1a and 1b we argue that ﬁrms’ in-cluster location is positively associated with its number of alliances. This actually indicates how active a

Bringing the Firm Back In

469

ﬁrm is in its alliance network. Thus, to measure ﬁrms’ centrality in their alliance network we use Freeman’s degree centrality measure. A ﬁrm’s degree centrality is the sum of its alliances with its partners. The degree centrality measure is a well accepted measure and it is widely used in studies where the focus is on ﬁrms’ alliance activity (Ahuja, 2000; Madhavan, 1996). Moderating Variable Absorptive Capacity. We measure ﬁrm’s absorptive capacity by its R&D intensity. Since some of the ﬁrms in our sample have zero sales for some years we take R&D intensity as a ratio of R&D expenditures over total assets. R&D intensity as a measure of absorptive capacity is appropriate measure for our context, because using absolute number of R&D expenditure might bias our results as our sample consists of large and small biopharmaceutical ﬁrms. It is also a common practice to measure ﬁrms’ absorptive capacity with R&D expenditure or R&D intensity (Cohen & Levinthal, 1990; Dushnitsky & Lenox, 2005). Interaction Variables We create a multiplicative interaction variable to measure the moderating effect of absorptive capacity for the centrality and innovation relationship. To represent the interaction between centrality and absorptive capacity we ﬁrst centered the centrality and absorptive capacity variables and then multiplied the centered values. Dependent Variable Innovation. We adopt the deﬁnition of innovation as the creation of new processes and products that are developed by building on ﬁrms’ existing technological trajectory or by a shift to a completely new technological trajectory (Benner & Tushman, 2002). We measure innovation performance of ﬁrms using a ﬁrm’s annual count of patents granted. Patents are effectively used to protect innovations in the biopharmaceutical industry. Patent counts, patent citations, new product announcements, and R&D inputs are used as indicators of ﬁrms’ innovation performance in the empirical literature (Griliches, 1990). Hagedoorn and Cloodt (2003) found that the correlations among these variables are so strong that any of these indicators can be used as a measure of innovation performance in the high tech industries. This supports our use of the number of patents granted as an appropriate measure in the biopharmaceutical industry context.

470

RAVI MADHAVAN ET AL.

Control Variables Firm Size. We measure ﬁrm size by number of employees since previous studies have indicated that ﬁrm size can inﬂuence innovation performance in the high tech industries including biopharmaceutical industry (Shan, Walker, & Kogut, 1994; Stuart, 2000). Biopharmaceutical ﬁrms with large number of employees are perceived to have more scientists devoted to R&D and therefore a greater amount of patent output. We use log of number of employees. Firm Age. Since ﬁrm age affects the rate at which ﬁrms patent (Sorenson & Stuart, 2000) we control for the number of years passed since founding of ﬁrm i to the year of the observation of the dependent variable. Indirect Ties. A focal ﬁrm’s partners can bring information from their alliances with other partners to the focal ﬁrm. Firms’ alliance networks may act as an information gathering or as an information processing device (Ahuja, 2000). Therefore, both the amount and diversity of information that is acquired by a focal ﬁrm is affected by its indirect ties within its alliance network. We calculate reach centrality by a built in function in UCINET 6 to control for the effect of indirect ties on innovation. This measure provides the proportion of ﬁrms that a focal ﬁrm can reach in j or fewer steps in its alliance network (Borgatti et al., 2005; UCINET 6). Time (Year Effects). Biopharmaceutical industry is research intensive industry. Scientiﬁc breakthroughs during certain years might affect ﬁrms’ innovation in the following years. This is controlled by including a dummy variable for all but one year (Wooldridge, 2002, p. 427). We also estimate our models without the time effects. The results do not change. Analysis To assess the relationships between cluster membership and centrality (hypothesis 1a, 1b) we use a random effects negative binomial model due to nonnegative integer values of the centrality variable. This speciﬁcation accounts for issues associated with heteroskedastic, non-normal residuals of nonnegative integer data (Hausman, Hall, & Griliches, 1984). Since our independent cluster membership variable (HQ or R&D-lab location) is time invariant, ﬁxed effects model is not appropriate. To analyze the moderating effect of absorptive capacity for the centrality– innovation relationship (hypothesis 2) we use a ﬁxed effects instrumental variable (IV) regression with General Method of Moments (GMM) estimation.6 As we do not have time invariant regressors in this model we use ﬁxed effects IV regression to control for the inﬂuence of omitted variables that differ from one ﬁrm to another but are constant over time.

471

Bringing the Firm Back In

Since, our dependent variable consists of count patent data we estimate the ﬁxed effects IV regression in an exponential form.7 Instrumental variables include HQ-location, R&D-lab location, and lagged patent variable. Recent empirical research in economics and ﬁnance has largely used GMM estimation (Greene, 2003). While negative binomial and Poisson models are commonly used in models with a count dependent variable such as number of patents they are not appropriate for our model because they do not account for endogeneity of regressors (Iwasa & Odagiri, 2004).8 In models with endogenous explanatory variables, ﬁxed effects IV regression in exponential form with GMM estimation has several advantages over Poisson and negative binomial models with maximum likelihood estimation (MLE). First, GMM estimation does not assume equality of variance and mean of count data (in Poisson Model) or over-dispersion (variance is larger than mean) of data in negative binomial model. Second, GMM estimation accounts for heteroscedasticity and autocorrelation of error terms. Finally and most importantly, it relaxes the strict exogeneity assumption of explanatory variables (Cincera, 1997; Crepon & Duguet, 1997). We also include negative binomial analysis for hypothesis 2 for comparison purposes.

RESULTS Descriptive statistics and correlations for the 847 observations in the sample are available from the authors. The descriptive statistics indicate that the ﬁrms are characterized by signiﬁcant diversity on key variables. The correlation matrix shows low to moderate correlations among variables. Table 1 presents the results of negative binomial regression analysis for hypothesis 1. Model 1 includes effect of HQ-location (cluster membership deﬁned as having HQ inside cluster) on centrality. Model 2 includes effect of R&D-lab location (cluster membership deﬁned as having at least one R&D lab inside cluster) on centrality. In hypothesis 1 we propose competing predictions for the effect of ﬁrms’ in-cluster location on their central position in their alliance network structure. The data indicate, in support of hypothesis 1a that ﬁrms’ in-cluster location when measured by HQ location is positively associated with being more central in their R&D alliance network structure compared to ﬁrms located elsewhere (po0.05). That is, in-cluster location complements ﬁrms’ central position in their alliance networks structure. This result also implies that the competing hypothesis 1b that in-cluster ﬁrms are less central than out-cluster ﬁrms due to the

472

RAVI MADHAVAN ET AL.

Table 1. Hypothesis 1a, 1b: Cluster Membership and Centrality Relationship – Dependent Variable (DV): Centrality (Random Effect Negative Binomial Model). Variables Independent Variables HQ-location

Model 1: HQ 0.309 (0.151)

R&D-Lab location Control Variables Absorptive capacity Firm size Firm age Indirect ties Time effects Log likelihood Wald w2

0.0000894 (0.00033) 0.375 (0.0732) 0.016 (0.005) 4.197 (0.489) Yes –1610.72 495.77

Model 2: R&D Lab

0.346 (0.151) 0.00007 (0.0003) 0.354 (0.074) 0.015 (0.005) 4.195 (0.488) Yes –1610.21 496.05

Coefﬁcient (Standard Error) wpo0.10; po0.05; po0.01; po0.001

substitution mechanism is not supported. Therefore, the competing hypothesis 1b is rejected. Results of hypothesis 1a are also signiﬁcant when we estimate the model with R&D-lab location variable as a proxy for cluster membership (po0.05). Both Models 1 and 2 include control variables of absorptive capacity, ﬁrm size, company age, indirect ties, and time effects. Except the absorptive capacity variable all control variables are signiﬁcant in both HQ-location and R&D-lab location models. Next, we provide the results of the exponential ﬁxed effects IV regression with GMM estimation. Ideally, one would estimate the coefﬁcients of hypothesis 2 in the same regression model as the dependent variable is the same. However, given the obstacles in ﬁnding relevant and strictly exogenous instruments in addition to our existing instruments we chose to run the models for hypothesis 2 separately.9 Table 2 provides the results of analysis for hypothesis 2 predicting that ﬁrms’ absorptive capacity positively moderates the centrality and innovation relationship. Coefﬁcient of the interaction variable (0.001), centrality X absorptive capacity in Model 1, Table 2 provides support for hypothesis 2, that the interaction between

473

Bringing the Firm Back In

Table 2. Hypothesis 2: Centrality and Patent Relationship – Dependent Variable (DV): Patent (Fixed Effects Model with Endogenous Regressors with GMM Estimation for Count DV). Variables

Fixed Effects Model with Random Effects Negative Binomial Endogenous Regressors with Model: Models do not Account for GMM Estimation for Count DV: Endogeneity Models Account for Endogeneity Model 1: HQ

Endogenous Variables Centrality Centrality X absorptive capacity Independent Variables Absorptive capacity Firm size Firm age Indirect ties GMM objective functiona HQ

Model 2: R&DLab

Model 3: HQ

Model 4: R&DLab

0.3 (0.138) 0.001w (0.0005)

0.04 (0.017) 0.0006 (0.006)

0.007 (0.003) –0.00003 (0.00005)

0.008 (0.003) –0.00002 (0.00005)

–0.012 (0.004) 0.607 (2.89) –0.103 (0.083) 0.036 (18) 0.07

–0.009 (0.002) 1.44 (0.573) –0.04w (0.024) 0.03 (3) 0.03

0.0003 (0.0005) 0.138w (0.083) –0.004 (0.0043) 0.310 (0.147) NA

0.0004 (0.0005) 0.104 (0.084) –0.007 (0.004) 0.371 (0.149) NA

0.574 (0.153)

R&D-Lab location Log likelihood

–2858.66

0.087 (0.147) –2865.4217

Coefﬁcient (Standard Errors) po0.10; po0.05; po0.01; po0.001 w

a

GMM objective function is calculated by the optimization program in Gauss Software. The closer the function to zero the better the model ﬁt is.

ﬁrms’ absorptive capacity and centrality is positive and signiﬁcant. This indicates that based on ﬁrms’ absorptive capacity central ﬁrms innovate more than other ﬁrms when endogeneity of centrality is controlled with HQ location. However, when we do not control for endogeneity the moderating effect is not signiﬁcant (Model 3, Table 2). This ﬁnding is important because it shows

474

RAVI MADHAVAN ET AL.

that the relationship between centrality and ﬁrm innovation is more complex than it is originally studied. Although it is not one of our hypothesis, the positive and signiﬁcant coefﬁcient of centrality variable (0.3) supports the prior literature that the higher the centrality of ﬁrm in its alliance network the higher the innovation (po0.05) (Ahuja, 2000). This further, provides a strong support for earlier literature, where centrality and innovation has been positively linked (Ahuja, 2000; Owen-Smith & Powel, 2004). We should note that there might still be differences in the magnitude of coefﬁcients in models where endogeneity is controlled and the magnitude of coefﬁcients in models where endogeneity is not controlled. We also estimate the models by controlling for endogeneity with R&D-lab location variable. The results are similar. Control variables are not signiﬁcant except absorptive capacity and time effects. Thus, at the aggregate level, we have examined the endogenous impact of cluster membership deﬁned as the geographical location of a ﬁrm within a cluster area on ﬁrms’ alliance network structure and how this alliance network structure inﬂuences innovation contingent on ﬁrms’ absorptive capacity. We ﬁnd that ﬁrms’ cluster membership affects ﬁrms’ position in their overall alliance network structure. In-cluster ﬁrms are more central than other ﬁrms. The results provide support for the basic premise that network position can be enhanced by locating ﬁrms’ HQ (or at least one R&D lab) in a cluster area. Further, our ﬁndings suggest that network structure is necessary but not a sufﬁcient condition for innovation. Speciﬁcally, ﬁrms need to develop capabilities to acquire, assimilate, transform, and commercialize external information that is acquired through their network structure.

EXPLANATORY MICRO-ANALYSIS The above section has summarized the base study that we seek now to extend, employing one type of mixed methods design: explanatory analysis. Explanatory designs involve the collection and analysis of aggregate quantitative data followed by ﬁner-grained, often qualitative, data (Creswell & Plano Clark, 2007). The purpose of the ﬁner-grained data is typically to help further explain signiﬁcant ﬁndings, anomalies, non-signiﬁcant ﬁndings and surprises. Thus, among the 147 biopharmaceutical ﬁrms (SIC 2834) in the large sample, we selected four ﬁrms to analyze more intensively. Our selection criteria included: (1) data availability for a comparable period of multiple years, (2) minimum differences in ﬁrm age and size, and

475

Bringing the Firm Back In

(3) maximum range on the main ﬁrm attribute of geographic location within a biopharmaceutical cluster area. Based on the above criteria we identiﬁed four ﬁrms: Immunogen, Alexion, Endo, and Neurogen.

Analysis In the large sample study, we ﬁnd that in-cluster ﬁrms (ﬁrms that have HQ or at least one R&D facility within a geographic cluster area) are more central in their alliance network structure and are more innovative than ﬁrms that are located outside cluster areas. According to the above ﬁnding, one suggestion to managers would be to locate their companies in cluster areas. However, when we look at speciﬁc examples in our data set, it is clear that just locating in the cluster might not be enough to increase innovation More importantly, locating outside the cluster might not be an obstacle to innovation in all cases, either. Such deviations from aggregate quantitative ﬁndings are neither new nor rare in empirical research. The general response is to recognize outliers and remove them from the analysis (Andriani & McKelvey, 2007). However, our objective is to explain such departures from the norm since it is critical to the practical question of how to strategically design networks. We thus conduct a more ﬁne-grained analysis by focusing on ﬁrm-speciﬁc factors that complement the ﬁndings from our large-scale quantitative study. As a ﬁrst cut, we plot the centrality – patent productivity relationship for the sample as well as our four ﬁrms (Fig. 2) to provide a visual representation. The plots can be used to identify anomalies as well as gain a better appreciation for the underlying structure of the centrality – patent productivity relationship. We classiﬁed our sample ﬁrms into four cells based on where their HQs and/or at least one R&D facility were in or out of a biopharmaceutical cluster. Table 3 indicates that there are anomalies in the general ﬁndings as we move from the quantitative large sample study to qualitative analysis at the ﬁrm level. For example, according to the quantitative large sample study, one would expect that Neurogen’s innovation performance would lag behind Immunogen’s. Neurogen is not located in a cluster nor does it have more ties. However, as the ﬁrm level analysis shows, Neurogen is the most innovative company among these four biopharmaceutical companies. Immunogen, on the other hand, is located in a cluster. But it is not very central nor is it very innovative. In fact, Immunogen’s innovativeness is on par with those of Endo and Alexion, even though these other ﬁrms do not have as many ties as Immunogen. Clearly, we need to examine these

476

RAVI MADHAVAN ET AL.

Fig. 2.

Centrality – Patent Productivity Relationship.

anomalies more closely if we are to make managerial recommendations that are both relevant and useful. We focus on two ﬁrm speciﬁc characteristics for the purpose of this analysis: ﬁrm business strategy and ﬁrm alliance network strategy. Examining these two together is important because the ﬁrm’s business strategy is at the root of a ﬁrm’s actions and choices. More importantly, business strategy also drives the alliance strategy of the ﬁrm. Alliance-related decisions such as alliance formation timing, the choice of partners, the purpose of the alliance and the type of alliance are driven by

477

Bringing the Firm Back In

Table 3.

Comparison of Sample Firms Based on HQ and R&D Facility Location. HQ

R&D

IN Cluster

OUT Cluster

IN-Cluster

OUT-Cluster

Immunogen Centrality: 11 Innovation: 19 R&D intensity: 86% of sales Endo Centrality: 8 R&D intensity: 8% of sales Innovation: 18

Alexion Centrality: 6 Innovation: 15 R&D intensity: 1,051% of sales Neurogen Centrality: 5 Innovation: 103 R&D intensity: 169% of sales

the ﬁrm’s needs and overall objective, which in turn are a function of the strategy of the ﬁrm. In other words, the alliance strategy of the ﬁrm has to be consistent with the overall purpose and strategy of the ﬁrm. We thus need to examine these two characteristics together in order to explore and explain alliance activity and ﬁrm level difference in performance outcomes for each of these ﬁrms. Our approach is consistent with the explanatory mixed methods design in which the sequence of analysis is quantitative followed by qualitative – as a way to provide follow-up explanations. In this design, the quantitative analysis dominates the qualitative analysis. In the next section, we provide a brief overview of each ﬁrm followed by an analysis of their business and alliance strategy. The alliance network structures of all four ﬁrms are depicted in Fig. 3. We also provide information on their strategy and their partner characteristics in Tables 4–10.

Firm-Level Analysis Immunogen focuses on developing new, targeted therapeutics for the treatment of various types of cancer. In the cancer treatment process, Immunogen uses the cancer biology, monoclonal antibodies, and small molecule cell-killing agents. The company focuses on its proprietary TumorActivated Prodrug (TAP) technology and its antibody expertise to develop cancer treatments. Immunogen’s TAP technology uses antibodies, to deliver a potent cytotoxic agent speciﬁcally to cancer cells. The TAP technology is developed to manufacture effective, well-tolerated anticancer products.

Fig. 3.

Evolution of Firm Ego Networks.

1998–2004

1998–2004

1998–2004

Neurogen

Endo Pharmaceuticals

Alexion Pharmaceuticals Inc.

Biotechnology company

Specialty pharmaceuticals

Biotech

Biotech

Business Focus

Employees, total sales, assets, patents refer to 2004.

1998–2004

Immunogen

a

Observation Period

Focal Firms

Table 4.

204

549

170

146

No of Employeesa

48.46

50.55

32.50

22.22

4.61

615.1

19.18

25.96

320

947.49

183.82

122.63

R&D Total Total Exp. ($ Sales ($ Assets ($ Million)a Million)a Million)a

Sample Firms: Summary.

15

18

103

19

Innovation (No of Patents)a

Eleven R&D alliances with four big pharma, ﬁve biotech, one specialty pharma, and one university Five R&D Alliances with big pharma companies Eight R&D alliances with pharmaceutical companies Six R&D alliances with biotech, pharmaceutical companies, and universities

R&D Alliance Activity During Observation Period

Bringing the Firm Back In 479

2000

Formed two new alliances SmithKline

1999

Exclusive right to Genentech to use TAP

None

Mid1990s

Abandoned the use of plant protein

Formed ﬁve new alliances Vernalis Morphosys

None

1990s

Developed TAP technology: Innovation in cancer treatment

NA

1981 1989 1980s

Founded Went public Developed plant protein to be used in cancer treatment

Network Events

Time

Announcement of SVP $41 million, private placement

NA/None

NA/None

NA/None

NA/None

Founded

Strategic Events

2000

1999

1998

1997

1996

1987

Time

No change in alliances Pﬁzer Schering plough Wyeth-Ayerst

Formed one new alliance Cubist Phar. No change in alliances Pﬁzer Schering plough Wyeth-Ayerst Cubist Phar.

No change in alliances Pﬁzer Schering plough Wyeth-Ayerst

Formed three alliances Pﬁzer Schering plough Wyeth-Ayerst

Network Events

Business Focus: Develop new drugs for pain, insomnia, inﬂammation, depression, and obesity

Business Focus: Develop targeted therapeutics for cancer treatment

Strategic Events

Neurogen

Immunogen

Table 5. Immunogen and Neurogen: Analysis of Firm Strategic Events and Network Events Over Time.

480 RAVI MADHAVAN ET AL.

Change in CFO position regains development & commercialization rights for a new therapy (mertansine)

Favorable clinical results for the treatment of smallcell lung cancer Joins Russell 3000 index Favorable clinical results for the treatment of colorectal, pancreatic, and certain non-small-cell lung cancers Appoints new CFO Announce share repurchase program ImmunoGen has developed taxane derivatives that can be linked to tumortargeting monoclonal antibodies and are more potent than competitor’s Taxol(R) (paclitaxel)

2003

2002

2001

Formed one new alliance Aventis

Formed two new alliances Raven Biotech Boehringer Ingelheim

GTC Biotech Abgenix SUNY

One member elected to board of directors

One member elected to board of directors Appointments to strategic management Announces staff reduction

Start of a novel antiinﬂammatory drug 2001

Formed one new alliance Merck

No change in alliances Pﬁzer Schering plough Wyeth-ayerst Cubist phar. Aventis

Formed one new alliance Aventis

Cubist Phar

Bringing the Firm Back In 481

Time

2004

Announces addition to board of directors

Formed two new alliances Biogen Idec Centocor

Network Events

Time 2004

Strategic Events VP ﬁnance, appointed $100 million, private placement

No change in alliances Pﬁzer Schering plough Wyeth-Ayerst Cubist Phar. Aventis Merck

Network Events

Business Focus: Develop new drugs for pain, insomnia, inﬂammation, depression, and obesity

Business Focus: Develop targeted therapeutics for cancer treatment

Strategic Events

Neurogen

Immunogen

Table 5. (Continued )

482 RAVI MADHAVAN ET AL.

Formed two new alliances with Algos and Lavipharma No Change in alliance activity

1999

2000

Acquired Algos Pharmaceuticals

Formed a new alliance with Penwest Pharmaceutical

1997 1998

Founded

Network Events

Time

Acquired Proliferon Inc. Major organizational changes and top management promotions and replacements takes place

Founded

Strategic Events

2000

1999

1992 1998

Time

No Change in alliance activity

Formed new alliances with GTC Biotherapetics Proctor & Gamble

Formed new alliances with Tufts University Genetic Therapy

Network Events

Business Focus: Develop innovative therapeutic agents to treat immune mediated diseases, infection diseases, inﬂammation, and cancer

Business Focus: Develop specialty pharmaceuticals to treat and manage pain

Strategic Events

Alexion Pharmaceuticals Inc.

Endo and Alexion Pharmaceutical: Analysis of Firm Strategic Events and Network Events Over Time.

Endo Pharmaceuticals Inc.

Table 6.

Bringing the Firm Back In 483

Formed a new alliance with Novartis

Formed new alliances with Skye Pharma and Durect Formed a new alliance with Collegium Pharma

Formed a new alliance with Mak Scientiﬁc

2001

2002

2004

Sold interest in DuPont Pharmaceutical to Bristol Myers Squibb Co.

Acquired BML Pharmaceuticals Inc.

2003

Network Events

Time

Co proposed public offering in order to fund working capital, R&D expenses, and clinical trial expenses Change in company’ board of directors

A new vice president was appointed

Strategic Events

2004

2003

2002

2001

Time

No change in alliance activity

Formed a new alliance with Xoma

Formed a new alliance with University Medical Center of Nijmegen in Netherlands No change in alliance activity

Network Events

Business Focus: Develop innovative therapeutic agents to treat immune mediated diseases, infection diseases, inﬂammation, and cancer

Business Focus: Develop specialty pharmaceuticals to treat and manage pain

Strategic Events

Alexion Pharmaceuticals Inc.

Endo Pharmaceuticals Inc.

Table 6. (Continued )

484 RAVI MADHAVAN ET AL.

None Smithkline Abgenix GTC biotherapeutics Morphosys Smithkline SUNY Vernalis

Boehringer-Ingelheim Raven biotech None Aventis Centocor

2001 2001 2002 2003 2004

New partners

1998 1999 2000 2000 2000 2000 2000 2000

Year

Partners

Big Pharma Biotech

Big Pharma Biotech Biotech Biotech Big Pharma Research U. Specialty Pharma Big Pharma Biotech-small

Partner type

1,494

12 213 21

Partner R&D productivity (%)

9 7 20 19

9 10 11

2 0 5

No of patents

9

0 2 7

No of direct ties

152.16 118.03 122.63

159.16

5.88 7.17 19.34

Total asset (million)

5.88 7.63 25.96

4.48

0.31 3.4 11.18

Total sales (million)

Immunogen

17.69 23.43 22.22

15.21

6.62 6.1 8.88

R&D expend. (million)

Table 7. Partner Information for Focal Firm Immunogen.

105 117 146

76

55 57 60

No of employees

22 23 24

21

18 19 20

Company age

Bringing the Firm Back In 485

New partners

Cubist Pharma Pﬁzer Schering Plough Wyeth-Ayerst None None Aventis None Merck None

Year

1998 1998 1998 1998 1999 2000 2001 2002 2003 2004

Big Pharma

Big Pharma

Big Pharma Big Pharma Big Pharma

Partner type

Partners

Table 8.

15

642 17

Partner R&D productivity (%)

4 4 5 5 6 5

4

No of direct ties

79 100 102 138 126 103

66

No of patents

92.13 142.59 145.96 115.78 95.37 183.82

101.81

Total asset (million)

10.21 20.41 11.51 15.73 6.79 19.18

11.08

Total sales (million)

Neurogen

Partner Information for Focal Firm Neurogen.

24.04 32.69 35.4 34.14 33.16 32.5

20.91

R&D expend. (million)

168 187 200 163 155 170

148

13 14 15 16 17 18

12

No of Company employees age

486 RAVI MADHAVAN ET AL.

Penwest Pharmaceutical Lavipharma

Algos

None Novartis Skye Pharma

Durect Collegium Pharma

MakScientiﬁc

1998

1999

2000 2001 2002

2002 2003

2004

1999

New partners

Pharmaceutical Specialty pharmaceutical Pharma Specialty pharmaceutical co., Private Co. Privately held biotech co.

Pharma Private Company Biotech, Private Company

Pharma

Partner type

NA

176 NA

14 37

NA

NA

395

Partner R&D productivity (%)

8

7

3 4 6

3

1

No of direct ties

18

20

7 6 5

9

2

No of patents

947.49

753.88

467.84 471.0 51.20

329.44

287.61

Total asset (million)

615.10

595.61

194.43 251.98 398.97

138.50

108.37

Total sales (million)

50.55

44.06

159.21 38.44 77.12

9.37

5.89

R&D expend. (million)

Endo Pharmaceuticals

Partners

Year

Partner Information for Focal Firm Endo Pharmaceuticals.

Table 9.

549

492

140 167 277

98

NA

8

7

4 5 6

3

2

No of Company employees age

Bringing the Firm Back In 487

2002 2003 2004

2000 2001

1999

1998 1999

Genetic Therapy

1998

Partner type

1,380

NA

303

5 6 6

4 5

4

2

NA

NA 32

No of direct ties

Partner R&D productivity (%)

15 26 15

4 10

9

2

No of patents

354.0 270.0 320.0

192.0 40.0

44.0

42.0

6.54 0.88 4.61

21.44 11.81

18.75

5.04

60.01 71.04 59.84

40.19 59.87

23.71

12.32

Total Total R&D asset sales expend. (million) (million) (million)

Alexion Pharmaceuticals

Partner Information for Focal Firm Alexion Pharmaceuticals.

Privately held Pharmaceutical Co. Tufts University Research University Proctor & Gamble Consumer Products Co. GTC Biotherapeutics None Medical Center of Research University Nijmegen, Netherlands None Xoma None

New partners

Year

Partners

Table 10.

172 191 204

113 140

90

69

No of employees

11 12 13

9 10

8

7

Company age

488 RAVI MADHAVAN ET AL.

Bringing the Firm Back In

489

Immunogen’s TAP technology uses antibodies targeting only tumor cells to minimize harm to healthy tissue. The company also licenses its TAP technology to other companies. Both its HQs and its R&D are within the biotech cluster in Cambridge, MA. Immunogen seems to have pursued a very focused strategy not just in terms of its research area but also in terms of the technology it is willing to pursue within that research stream. As we noted above, Immunogen has focused on cancer therapeutic research. More speciﬁcally, it is interested only in leveraging those compounds that utilizes its TAP technology. Such restrictions both narrow the set of potential partners as well as the opportunity to innovate. Second, its alliance strategy seems to have multiple dimensions. On the one hand, it is involved in partnering with ﬁrms to discover and develop new compounds. In that sense, innovation is a critical element of its alliance strategy. On the other hand, its alliance strategy also involves the licensing of its TAP technologies as well as commercialization of the more promising drugs, both of which have relatively little to do with patenting innovation. Given its multiple objectives, locating within the cluster seems to be an appropriate choice. First, its willingness to license its TAP technology suggests that leakage due to spillover may not be a big concern. Second, given its focus on licensing and innovation, collaboration assumes considerable importance in its strategy, consistent with the decision to co-locate. Finally, its multiple ties resulting in moderate level of centrality seems to be also a function of its multiple objectives. In other words, the lower patent rate is consistent with the possibility that only some of its alliances are designed for patenting innovation. In contrast, Neurogen is involved in the discovery and development of new drugs for a range of disorders including metabolic, neurological diseases, pain and inﬂammation. Neurogen’s drug development business strategy involves focusing on diverse drug development programs rather than focusing on one blockbuster product. Neurogen conducts its R&D independently and also collaborates with pharmaceutical companies during the pharmaceutical R&D process to access complementary expertise and to obtain additional ﬁnancial resources. Both the HQs and R&D of Neurogen are outside the Boston cluster. Neurogen strategy as reported on its website stresses multiple themes that are useful in understanding its actions and outcomes. Our strategy is to advance a mix of proprietary drugs independently and, when advantageous, collaborate with world-class pharmaceutical companies during the drug development process to obtain additional resources and to access complementary expertisey (Neurogen website, 2007)

490

RAVI MADHAVAN ET AL.

First, it positions itself as a drug discovery and development organization. Second, its patented Accelerated Integrated Drug Discovery (AIDD) system enables it to be more efﬁcient in the trial and error process, resulting in the discovery of multiple drug candidates. Third, its AIDD process enables it to operate in multiple research streams. As these themes indicate, Neurogen is pursuing its drug discovery agenda independently. Alliance activity is restricted to big pharmaceutical ﬁrms with a view to leveraging and learning their manufacturing and clinical trial capabilities. In other words, Neurogen’s alliance activity has little to do with discovery and patenting of new compounds. Instead, its alliance strategy is geared towards the commercialization of the discovered compounds using the capabilities of the large pharmaceutical ﬁrms. Examining the ﬁrm’s strategy thus enables us to understand that Neurogen decision to locate itself outside the biotech cluster may be a deliberate choice. Such a decision seems to, on the one hand, downplay the beneﬁts of spillovers from co-location. More importantly, locating outside the cluster may be a way of preventing accidental leakage of its AIDD process because of such spillovers. In terms of alliance strategy, alliance formation takes place after the discovery of the drugs rather than as a precursor to discovery. Finally, its large innovation rate may also be attributed to the diversity of its targets. As noted on its website, Neurogen believes that the value of its portfolio depends on diversity rather than a focus on a single blockbuster drug. In sum, Neurogen’s innovation is independently built, and seemingly has little to do with its alliance strategy. Moving on to the third ﬁrm, Endo Pharmaceuticals Inc. is a specialty pharmaceutical company. The company is engaged in R&D, sales and marketing of prescription pharmaceuticals to treat and manage pain. Company HQs are located in Chads Ford, PA. Endo Pharmaceutical serves both patients and healthcare professionals. The company has also several products in the development stage. Of the development stage products four are at the phase II and III of clinical trials. Endo was founded in 1997 as a spin-off from Dupont Merck. Endo’s business strategy, as reported on its website, is different from the other two companies in that it has a well established sales and marketing infrastructure, a focus on marketing generic analgesics and an R&D focus centered around the commercialization of analgesic products that are reﬁnements and extensions of current products. Much of their focus is on leveraging their existing brand names and/or acquiring and licensing complementary products. In other words, the company seems to be narrowly focused on three dimensions. First, their research is limited to

Bringing the Firm Back In

491

pain-related compounds and technologies. Even within this area, their R&D focus is limited to extensions of related and complementary products rather than developing completely new compounds and/or technologies. Finally, there seems to be a focus on products/areas that can be easily and quickly commercialized which arises from their need to leverage already existing sales and marketing infrastructure. Given this focus, it is not surprising that their patent activity is quite limited which also seems to be a reﬂection of their R&D spending relative to sales (8%). Endo’s alliance strategy also seems to reﬂect this focus. First, their portfolio comprises a mixture of biotech, specialty pharmaceutical and pharmaceutical ﬁrms, indicating the focus of their collaborative efforts that range from discovery, testing, and commercialization. Second, the number of partners is also limited, which again seems to reﬂect the focused nature of their business strategy. Third, the limited nature of their patent productivity also seems to reﬂect the relative unimportance of cluster location. Finally, Alexion Pharmaceuticals Inc. is engaged in the discovery and development of therapeutic products to treat hematologic diseases, cancer, and autoimmune disorders. The company allocates its resources to drug discovery, research, and product and clinical development. It is located in Cheshire, CT and has R&D facility in San Diego, CA. Alexion’s experimental efforts have focused on treating disease by blocking formation of the terminal components of ‘‘complement.’’ Complement is a natural part of the human immune system that is comprised of a cascade of proteins, one protein leading to formation of the next. The existing components of the ‘‘complement cascade’’ are generally beneﬁcial, but the later-formed components can have signiﬁcant harmful effects. Similar to Endo and Immunogen, Alexion also seems to be a focused company. While its focus is on developing antibody therapeutics, it is unlike Endo in that it is focused on the discovery of novel therapies. For instance, its ﬁrst product, Soliris is the ﬁrst ever therapy approved for the reduction of hemolysis in patients. To the extent that much of its recent strategy has been focused on getting Soliris approved through the FDA as well as launching it in the USA and Europe market, it is not surprising that both its patent productivity and alliance ties are limited. Second, the focus on novel therapies also seems to be a function of the research university background of its founding members, as is its partnership with other research universities. As these brief reviews of the ﬁrms and their strategies reveal, there appear to be several ﬁrm-speciﬁc factors that may explain why these ﬁrms do not completely ﬁt the curve found in our quantitative analysis. For instance,

492

RAVI MADHAVAN ET AL.

Immunogen’s location of its R&D and HQs in a biotech cluster does not seem to have contributed greatly to its possessing many ties. More importantly, Endo, Alexion, and Immunogen seem to be on the lower end of the centrality-patent curve. In contrast, Neurogen, in spite of lacking cluster afﬁliation as well as possessing very few ties, seems to have been very productive. Examining these ﬁrms together thus suggest that there could be several possible explanations for the anomalies in the performance of these four ﬁrms. First, deviations from the aggregate quantitative ﬁndings may be due to differences in the motivations of the ﬁrm with respect to its alliance network. As noted above, an implicit assumption of our quantitative study is that all ﬁrms share the same propensity and/or motivation to innovate. Neurogen’s description of itself as a drug discovery and development organization suggests that its propensity to innovate is tied to its identity, which in turn is reﬂected in its higher performance. Closely related to this is another assumption that the alliances are speciﬁcally formed for the purpose of innovation and patents. Here again, Neurogen’s alliance activity is related to commercialization rather than to innovation, explaining its poorer ﬁt to the centrality-patent curve found in this study. A similar explanation for Immunogen can be seen in the fact that its alliances seem to serve both innovation and commercialization. A second explanation may be related to the ability of the ﬁrm. Not only does its high patenting rate suggest that Neurogen has both the ability and motivation to innovate, it also forces us to recognize that such a rate is a function of its own in-house capabilities rather than a function of learning/ knowledge acquisition from its partners. Part of the ability argument is also related to the breadth of the innovation strategy. For instance, ﬁrms pursuing a focused strategy may have fewer opportunities to innovate compared to a ﬁrm that is pursuing a broad diversiﬁed strategy. More importantly, a focused strategy may suggest that, over time, opportunities for the ﬁrm may lie in pursuing exploratory and/or radical innovations as it exhausts the potential for exploitative and incremental innovations. Alexion’s lower patenting may be a function of its focus on exploratory innovations while, Neurogen’s performance may be a function of its broad therapeutic portfolio. Third, some of the anomalies may be related to path dependence and history. For instance, Endo’s focus on generics and exploitative innovation may be a function of its founding as a management spin-off with relatively well established capabilities, market and strategy. Alexion’s focus, on the other hand, on exploratory innovation may be a function of the fact that

Bringing the Firm Back In

493

many of its founders belonged to research universities prior to founding the ﬁrm. A ﬁnal explanation may lie in recognizing that the ﬁrm may not have the ability and/or the motivation to take advantage of its opportunities in the network. Thus, a low patenting rate may suggest that ﬁrm might not have the ability to learn from its partners in order to innovate. In the absence of either a base to compare its innovating rate to, or other information to the contrary, we need to keep the null as a possible explanation for Immunogen’s poor patenting performance in spite of its cluster location and centrality. Our arguments on ability and motivation as critical complementary factors in explaining network structure and ﬁrm outcome are reﬂected in other studies in this volume as well (Dagnino, Levanti, & Li Destri, 2008; Fund, Pollock, Baker, & Wowak, 2008). The Fund et al. (2008) argument on the cognitive ability of Benchmark Capital funds highlights the importance of motivation as well as the ability of the Benchmark Capital partners in determining their network strategy. Similarly, ST Microelectronics’ design of its particular network arose from choices it made in forming ties with a diverse set of partners in order to access and acquire critical knowledge worldwide (Dagnino et al., 2008). While these authors have usefully incorporated such issues in their analysis, extant network studies in general tend to ignore these ﬁrm-speciﬁc factors of ability and motivation in favor of purely structural explanations, leading to incomplete explanations of the antecedents and consequences of network structure.

MIXED-METHODS DESIGNS TO BRING THE FIRM BACK IN One of our primary objectives in this chapter is to demonstrate that, by bringing the ﬁrm back into structural analysis, we can provide a richer explanation of network strategy than either approach could individually. A limitation of any large sample quantitative approach is that while it provides very good generalizations regarding the centrality – innovation relationship, it does not make speciﬁc predictions for an individual ﬁrm or explain the underlying alliance strategies and mechanisms that lead to the generalized conclusions. Dubin (1978) has called this the precision paradox. In contrast, focusing on the alliance strategies of individual ﬁrms clariﬁes how the development of an alliance portfolio inﬂuences a ﬁrm’s centrality and innovation portfolio but creates signiﬁcant generalizability problems, or what Dubin (1978) refers to as the power paradox. We have tentatively

494

RAVI MADHAVAN ET AL.

addressed these paradoxes by adopting an explanatory mixed methods approach to studying network strategy. As our brief overview suggests, using a mixed methods design enables us to provide for deeper understanding of network structure and processes. As we noted above, employing mixed methods enables us to both explain anomalies and identify key constructs that we need to consider in order to develop actionable implications from our research. First, it highlights that network structure is an important factor affecting ﬁrm outcomes. But it is not the only determinant or sometimes even the most important determinant of ﬁrm actions and outcomes. Centrality and co-location do matter but they matter only to the extent that they provide opportunities for action. Second, a ﬁrm’s ability and motivation may be as important in determining its actions and outcomes. Assessing the relative impact of opportunity, ability, and motivation as well as their functional relationships (additive, multiplicative, curvilinear, etc.) are worthy research initiatives. Speciﬁcally, our brief analysis highlights the role of ﬁrm’s strategy in affecting its actions and outcomes. More importantly, it suggests that network structure is both cause and consequence of ﬁrm’s ability and motivation and illustrates the network dynamics theme of this volume. For example, Neurogen’s network activity may be subsequent to its innovating activities rather than an antecedent. Finally, some of our measures of absorptive capacity such as R&D activity may not be an accurate reﬂection of the ability of the ﬁrm. For instance, both Neurogen and Immunogen had roughly the same R&D intensity. But Neurogen’s ability to innovate seems to be a function of its AIDD process. In other words, it might not be the quantity but the quality of R&D expenditure that determines the ﬁrm’s ability. Clearly, one response to this analysis would be to argue that we have surfaced additional constructs for inclusion in the model (i.e., internal innovation capability, innovation range, and alliance intent). Incorporating these constructs and their operationalizations would lead to better theories of how network structure leads to performance. While much of network theory uses process arguments to hypothesize for the effect of structure on performance, empirical analysis has typically been limited to and focused on network structure. In doing so, extant studies have merely tested whether access to information and knowledge in the network leads to performance. However, we need to examine whether ﬁrms are able to leverage such access. Identifying constructs such as motivation and ability allow us to more fully specify our empirical models (Koka et al., 2007). However, there is a limit to how many such constructs we can include in our models. Increasing the grain of analysis will always lead to the need for

Bringing the Firm Back In

495

successively more elaborated but eventually intractable regression models. We suggest that the conventional approach is well suited for explaining observed network patterns (as apparent from our quantitative analysis) but not so for explaining differences in results. Thus, by analogy in medicine, epidemiology studies help to understand patterns of incidence of a disease but offer very little guidance as to whether a given patient will respond to a given treatment. The former is the province of science (generalizing from particulars) and the latter is the province of practice (particularizing from generalities). Mixed methods designs in network studies thus offer one way to develop guidelines for practice (going back to our motivating question). Micro-analyses of the sort attempted here may lead to exhaustive checklists rather than parsimonious models, and to trial-and-error learning rather than deﬁnitive prescriptions, but that is precisely how practice is different from academic science. Not surprisingly, the mixed methods approach of combining quantitative and qualitative data has been gaining currency among scholars (Creswell & Plano Clark, 2007). There are essentially four mixed methods designs: triangulation, embedded, explanatory, and exploratory. The differences among the four designs reside in the timing of the data collection for each method, the relative emphasis given to each type of data in the analysis and how the two types of data will be mixed (Creswell & Plano Clark, 2007). Applying these criteria to our study, we adopted the explanatory design approach because we collected the quantitative data prior to the stratiﬁed random selection of our four ﬁrms, the quantitative data is weighted more heavily and the purpose of the mixing is to connect and better understand the quantitative results by exploring the alliance strategies of our four ﬁrms. While we adopted the explanatory design approach, network research can beneﬁt from each of the other mixed methods research designs. Below, we brieﬂy describe each design to stimulate thinking along these lines. Triangulation designs collect complimentary quantitative and qualitative data on the same topic area. There is an equal weighting of the data and it is mixed during the analysis and interpretation phase to allow comparisons and contrast across the two methods. This approach is very applicable to multilevel analysis. For example, by collecting motivation, opportunity- and ability-related data for each alliance, one could then explore how the performance and outcomes of individual alliances compares to a ﬁrm’s alliance constellation. One research question might be framed in the following way: if both our HQs and R&D facilities are located out of cluster, how many and of what types of alliances do we need to form with ﬁrms located in-cluster to gain cluster beneﬁts?

496

RAVI MADHAVAN ET AL.

Embedded designs are used when one data set is used in a supportive role for the primary data collection effort. Embedded designs are used with multiple research questions that are best addressed by different types of data but one type is primary and the other supportive. Experimental designs often embed qualitative data to determine a treatment or to explore an intervention process. In our sample, a qualitative dominated embedded design could explore in-depth motivations for alliance partner selection and how it impacted the development of a ﬁrm’s alliance constellation. A series of in-depth cases would be complemented with quantitative indicators of partner availability, partner capabilities, industry alliance networks, strategy, structure, and processes of alliance partners, etc. In this way, a historical assessment of a ﬁrm’s alliance constellation would provide a thick description supported by quantitative evidence. An exemplar of this type of analysis is proved by Robert Gross’s (2001) analysis of the Battle of Concord. Explanatory designs involve the sequential collection of quantitative and qualitative data. The purpose of the qualitative data is to help further explain signiﬁcant ﬁndings, anomalies, non-signiﬁcant ﬁndings, and surprises. Establishing analytical and interpretive connections between the two phases of the data collection occurs during the mixing of the data sets. This is the approach adopted in this chapter. One way to extend our approach is to conduct ﬁeld interviews with each ﬁrm to gather participant insights. Exploratory designs involve the collection and analysis of qualitative data as an input to the subsequent quantitative phase of the research. This design is particularly useful when research instruments have not been developed for measures of interest, there is an absence of theory, or the desire to develop taxonomies. An example of the explicit connection of the results from the ﬁrst phase as an input to the second phase is the development of a questionnaire instrument from the results of a qualitative investigation. The development of diagnostic tools related to the selection of alliance partners or geographical location is an example of how one might apply this approach in our context. Given our set of partners, ‘‘What types of partners should we form new alliances with to maximize our patent productivity?’’ is a type of question that can be addressed with diagnostic tools.

CONCLUSION We began by proposing that understanding the action implications for a particular ﬁrm is somewhat different from understanding the statistical

497

Bringing the Firm Back In

relationships between variables at the aggregate level. Further, we proposed that bringing the ﬁrm back into network studies is a crucial ﬁrst step in understanding how to move toward a workable model for ‘‘translating’’ aggregate ﬁndings to meaningful managerial prescriptions at the level of the speciﬁc ﬁrm. We are particularly interested in questions related to the strategic design of alliances and how alliance constellations evolve over time to achieve strategic objectives, a central theme of this volume. Adopting a mixed-methods explanatory design, we complemented our aggregate ﬁndings from a large sample study with ﬁner-grained analysis of the network strategies of four selected ﬁrms. While our efforts captures some emerging insights from this process, our ultimate goal in this chapter is to demonstrate how mixed-methods designs may be the methodological portal through which networks research may ‘‘bring the ﬁrm back in’’ and thereby provide an important step in crossing the bridge that links the scholarly and practitioner worlds.

NOTES 1. Throughout the chapter, the terms cluster membership, in-cluster, and location within a cluster are used interchangeably. 2. In-cluster ﬁrms might enter into alliances from two directions: as initiator of alliance activity and as a target of alliance activity. 3. An MSA is ‘‘an area containing a recognized population nucleus and adjacent communities that have a high degree of integration with that nucleus’’ (Executive Ofﬁce of the President of the United States, Ofﬁce of Management and Budget, 2000). According to Milken Study 12 clusters are San Diego, Boston, Raleigh-Durham-Chapel Hill, San Jose, Seattle-Bellevue-Everett, Washington – DC, Philadelphia, San Francisco, Oakland, Los Angeles – Long Beach, Orange County, Austin-San Marcos. 4. One drawback of choosing corporate HQ address is that it may not show the physical location where the HQ functions are performed (Birkinshaw et al., 2006). For example, some companies may form a shell holding company in an offshore location to beneﬁt from tax advantages. Fortunately, with our sample this is not an issue because we focused on US biopharmaceutical companies and an examination of our dataset indicates that none of our sample ﬁrms have HQ locations outside of US. 5. Our network structure variables are calculated based on the complete network that includes the entire biopharmaceutical ﬁrms as well as other organizations with which they have alliances. 6. GMM only involves structural equation estimation with the help of instruments, instruments are orthogonal to structural errors and by beneﬁting from this property no reduced form equation is estimated in the GMM estimation. In contrast, in 2SLS a reduced form equation is estimated ﬁrst.

498

RAVI MADHAVAN ET AL.

7. The exponential model takes the general form of Pit ¼ exp (Xitbþei)þuit. Where Pit represents dependent variable, number of patents granted; Xit indicates vector of independent and control variables, and ei represents the unobserved effects not captured by the variables in the model. 8. In the ﬁrst stage of our model, our analysis shows that both HQ location inside cluster or at least one R&D lab inside cluster has signiﬁcant positive effect on both the centrality and structural holes (HQ and R&D-Lab location, po0.05) these signiﬁcant results establish the endogenous nature of the centrality variable in the model. We also tested for endogeneity by using the Durbin-Wu-Hausman test in STATA. Results of this test also conﬁrm the endogeneity of centrality variable. 9. For instrumental regression to work, there must be at least as many instrumental variables as endogenous regressors. It is not easy to ﬁnd perfect instruments as instrument relevance and instrument exogeneity are crucial for obtaining efﬁcient coefﬁcient estimates (Stock & Watson, 2003). For this reason we chose to run two separate models instead of having to use instruments that are not relevant and strictly exogenous, which may not provide efﬁcient coefﬁcient estimates.

REFERENCES Aeppel, T. (2006). Still made in the USA: Sticks and stones may break bones, but Warsaw, Ind., makes replacements (Eastern ed., p. B.1). New York, NY: Wall Street Journal. Aharonson, B., Baum, J. A. C., & Feldman, M. A. (2004). Industrial clustering and returns to inventive activity: Canadian biotechnology ﬁrms, 1991–2000. Working Paper, University of Toronto. Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Andriani, P., & McKelvey, B. (2007). Beyond Gaussian averages: Redirecting international business and management research toward extreme events and power laws. Journal of International Business Studies, 38, 1212–1230. Arora, A., & Gambardella, A. (1990). Complementarity and external linkages: The strategies of the large ﬁrms in biotechnology. The Journal of Industrial Economics, 38, 361–379. Audretsch, D. B., & Feldman, M. (1996). Innovative clusters and industry life cycle. Review of Industrial Organization, 11, 253–273. Bagchi-Sen, S. (2004). Firm-speciﬁc characteristics of R&D collaborators and non-collaborators in US biotechnology clusters and elsewhere. International Journal of Technology & Globalization, 1, 92–118. Bell, G. G. (2005). Clusters, networks, and ﬁrm innovativeness. Strategic Management Journal, 26, 287–295. Benner, M. J., & Tushman, M. (2002). Process management and technological innovation: A longitudinal study of the photography and pain industries. Administrative Science Quarterly, 47, 676–706. Birkinshaw, J., Braunerhjelm, P., Holm, U., & Terjesen, S. (2006). Why do some multinational corporations relocate their headquarters overseas? trategic Management Journal, 27, 681–700. Borgatti, S. P., Everett, M. G., & Freeman, L.C. (2005). Ucinet 6 for Windows: Software for social network analysis. Analytic Technologies.

Bringing the Firm Back In

499

Bourdieu, P., & Wacquant, L. J. D. (1992). An invitation to reﬂexive sociology. Chicago, IL: University of Chicago Press. Caner, T. (2007). Geographical clusters, alliance network structure, and innovation in the U.S. biopharmaceutical industry. Unpublished doctoral dissertation, University of Pittsburgh. Chacar, A., & Lieberman, M. (2003). Organizing for technological innovation the U.S. pharmaceutical industry. In: J. A. C. Baum & O. Sorenson (Eds), Geography and strategy: Advances in strategic management (Vol. 20, pp. 317–340). Kidlington UK: JAI Press. Cincera, M. (1997). Patents, R&D, and technological spillovers at the ﬁrm level: Some evidence from econometric count models for panel data. Journal of Applied Econometrics, 12, 265–280. Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35, 128–152. Crepon, B., & Duguet, E. (1997). Estimating the innovation function from patent numbers: GMM on count panel data. Journal of Applied Econometrics, 12, 243–263. Creswell, J., & Plano Clark, V. (2007). Designing and conducting mixed methods research. Thousand Oaks, CA: Sage Publications. Dagnino, G. B., Levanti, G., & Li Destri, A. M. (2008). Evolutionary dynamics of interﬁrm networks: A complex systems perspective. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 67–129). Bingley, UK: JAI/Emerald Group. Das, T. K., & Teng, B. S. (1998). Between trust and control: Developing conﬁdence in partner cooperation in alliances. Academy of Management Review, 23, 491–512. DeCarolis, D. M., & Deeds, D. L. (1999). The impact of stocks and ﬂows of organizational knowledge on ﬁrm performance: An empirical investigation of the biotechnology industry. Strategic Management Journal, 20, 953–968. Dubin, R. (1978). Theory building (revised ed.). New York, NY: Free Press. Dushnitsky, G., & Lenox, M. J. (2005). When do incumbents learn from entrepreneurial ventures? Corporate venture capital and investing ﬁrm innovation rates. Research Policy, 34, 615–639. Dyer, J. H. (1996). Does governance matter? Keiretsu alliances and asset speciﬁcity as sources of Japanese competitive advantage. Organization Science, 7, 649–666. Enright, M. (1995). Organization and coordination in geographically concentrated industries, in coordination and information. In: D. Raff & N. Lamoreux (Eds), Historical perspectives on the organization of enterprise (pp. 103–142). Chicago, IL: University of Chicago Press. Executive Ofﬁce of the President of the United States, Ofﬁce of Management and Budget (2000). Standards for deﬁning metropolitan and micropolitan statistical areas; notice. In: Federal Register (Vol. 65, No. 249, pp. 82227–82238). Washington, DC (December 27). Feldman, M. (2000). Location and innovation: The new economic geography and innovation, spillovers, and agglomeration. The Oxford handbook of economic geography. Oxford, UK: Oxford University Press. Fund, B. R., Pollock, T. G., Baker, T., & Wowak, A. J. (2008). Who’s the new kid? The process of developing centrality in venture capitalist deal networks. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 563–593). Bingley, UK: JAI/Emerald Group.

500

RAVI MADHAVAN ET AL.

Furman, J. L., Kyle, M., Cockburn, W. E., & Henderson, R. (2005). Knowledge spillovers, geographic centralization and the productivity of pharmaceutical research. Available at http://people.bu.edu/furman/, last accessed March 3, 2008. Giuliani, E. (2007). The selective nature of knowledge networks in clusters: Evidence from the wine industry. Journal of Economic Geography, 7, 139–168. Greene, W. H. (2003). Econometric analysis (5th ed.). Prentice Hall. Griliches, Z. (1990). Patent statistics as economic indicators: A survey. Journal of Economic Literature, 28, 1661–1707. Gross, R. A. (2001). The minutemen and their world (25th anniversary ed.). NY: Hill and Wang. Gulati, R. (1999). Network location and learning: The inﬂuence of network resources and ﬁrm capabilities on alliance formation. Strategic Management Journal, 20, 397–420. Hagedoorn, J., & Cloodt, M. (2003). Measuring innovative performance: Is there an advantage in using multiple indicators? Research Policy, 32, 1365–1379. Hagedoorn, J., & Frankfort, H. (2008). The gloomy side of embeddedness: The effect of overembeddedness on partnership formation. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 503–530). Bingley, UK: JAI/Emerald Group. Hausman, J., Hall, B., & Griliches, Z. (1984). Econometric models for count data with application to the patents – R&D relationship. Econometrica, 52, 909–938. Inkpen, A. C., & Tsang, E. W. K. (2005). Social capital, networks, and knowledge transfer. Academy of Management Review, 30, 146–165. Iwasa, T., & Odagiri, H. (2004). Overseas R&D, knowledge sourcing, and patenting: An empirical study of Japanese R&D investment in the US. Research Policy, 33, 807–828. Jaffe, A. (1989). Real effects of academic research. American Economic Review, 79, 957–970. Jaffe, A. B., Trajtenberg, M., & Henderson, R. (1993). Geographic localization of knowledge spillovers as evidenced by patent citations. Quarterly Journal of Economics, 434, 578–598. Kilduff, M., & Krackhardt, D. (1994). Bringing the individual back in: A structural analysis of the internal market for reputation in organizations. Academy of Management Journal, 37, 87–108. Kogut, B., Walker, G., Shan, W., & Kim, D. J. (1994). Platform technologies and national industrial networks. In: J. Hagedoorn (Ed.), Technical change and the world economy (pp. 58–82). London: Edward Elgar. Koka, B. R., & Prescott, J. E. (2002). Strategic alliances as social capital: A multidimensional view. Strategic Management Journal, 23, 795–811. Koka, B. R., & Prescott, J. E. (2008). Designing alliance networks: The inﬂuence of network position, environmental change and strategy on ﬁrm performance. Strategic Management Journal, 29(6), 639–661. Koka, B. R., Prescott, J. E., & Canella, B. (2007). Does network position really matter?: A meta-analysis of the network position-ﬁrm performance relationship in interﬁrm networks. Working paper, Arizona State University. Krugman, P. (1991). Geography and trade. Cambridge, MA: MIT press. Lane, P., Koka, B., & Pathak, S. (2006). The reiﬁcation of absorptive capacity: A critical review and rejuvenation of the construct. Academy of Management Review, 31, 833–863. Lane, P. J., & Lubatkin, M. (1998). Relative absorptive capacity and interorganizational learning. Strategic Management Journal, 19, 461–477.

Bringing the Firm Back In

501

Madhavan, R. (1996). Strategic ﬂexibility in the steel industry: The role of interﬁrm linkages. Unpublished doctoral dissertation, University of Pittsburgh. Madhavan, R., Koka, B. R., & Prescott, J. E. (1998). Networks in transition: How industry events (re)shape interﬁrm relationships. Strategic Management Journal, 19, 439–459. Mayhew, B. (1980). Structuralism versus individualism: I – Shadowboxing in the dark. Social Forces, 59, 335–375. McEvily, S., & Yao, B. E. (2005). Getting the right mix: The roles of absorptive capacity and network connections in technology innovation. Working paper, University of Pittsburgh. McKelvey, M., Alm, H., & Riccaboni, M. (2003). Does co-location matter for formal knowledge collaboration in the Swedish biotechnology–pharmaceutical sector?. Research Policy, 32, 483–502. Milken Institute Research Reports. (2004). America’s biotech and life sciences clusters. Prepared by DeVol, R., Wong, P., Ki, J., Bedroussian, A., & Koepp, R. Available at http:// www.milkeninstitute.org/publications/publications.taf?function ¼ list&cat ¼ ResRep& start ¼ 21, last accessed on March 5, 2008. Owen-Smith, J., & Powell, W. W. (2004). Knowledge networks as channels and conduits: The effects of spillovers in the Boston biotechnology community. Organization Science, 15, 5–21. Porter, M. E. (1998). On competition. Boston, MA: HBS Press Book. Raelin, J. (2007). Toward an epistemology of practice. Academy of Management Learning and Education, 6, 495–519. Saxenian, A. (1994). Regional advantage: Culture and competition in Silicon Valley and Route 128. Cambridge, MA: Harvard University Press. Scott, A. (2005). On Hollywood: The place, the industry. Princeton: Princeton University Press. Shan, W., Walker, G., & Kogut, B. (1994). Interﬁrm cooperation and startup innovation in the biotechnology industry. Strategic Management Journal, 15, 387–395. Soda, G., Zaheer, A., & Carlone, A. (2008). Imitative behavior: Network antecedents and performance consequences. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 531–560). Bingley, UK: JAI/Emerald Group. Sorenson, J. B., & Stuart, T. E. (2000). Aging, obsolescence, and organizational innovation. Administrative Science Quarterly, 45, 81–112. Stock, J. H., & Watson, M. W. (2003). Introduction to Econometrics. Boston: Addison Wesley Longman. Stuart, T. E. (2000). Interorganizational alliances and the performance of ﬁrms: A study of growth and innovation rates in a high technology industry. Strategic Management Journal, 21, 791–811. Tallman, S., Jenkins, M., Henry, N., & Pinch, S. (2004). Knowledge, clusters, and competitive advantage. Academy of Management Review, 29, 258–271. Tsai, W. (2001). Knowledge transfer in intraorganizational networks: Effects of network position and absorptive capacity on business unit innovation. Academy of Management Journal, 44, 996–1004. Wasserman, S., & Faust, K. (1994). Social network analysis. Cambridge, UK: Cambridge University Press. Williamson, O. (1975). Markets and hierarchies. New York, NY: The Free Press. Wooldridge, J. (2002). Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press. Zahra, S., & George, G. (2002). Absorptive capacity: A review, reconceptualization, and extension. Academy of Management Review, 27, 185–203.

THE GLOOMY SIDE OF EMBEDDEDNESS: THE EFFECTS OF OVEREMBEDDEDNESS ON INTER-FIRM PARTNERSHIP FORMATION John Hagedoorn and Hans T. W. Frankort ABSTRACT We discuss the ‘gloomy’ side of ﬁrms’ embeddedness in networks of interﬁrm partnerships. We propose a nested understanding of the effects of three levels of overembeddedness – environmental, inter-organizational and dyadic overembeddedness – on subsequent inter-ﬁrm partnership formation and argue for a joint examination of these three levels and their interactions over time. As a whole, increases in ﬁrms’ embeddedness will generate decreasing returns to the ﬁrms involved, prompting (i) the search for and attachment to novel partners and (ii) the dissolution of extant partnerships. On the ﬂipside, overembeddedness thus sparks network evolution – by cueing ﬁrms to look beyond their embedded partnerships.

Network Strategy Advances in Strategic Management, Volume 25, 503–530 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25014-X

503

504

JOHN HAGEDOORN AND HANS T. W. FRANKORT

INTRODUCTION The past few decades have witnessed an unprecedented growth in the formation of inter-ﬁrm partnerships in a wide variety of ﬁelds (Hagedoorn, 2002; Powell & Grodal, 2005). Increasingly, scholars have come to view the formation of such inter-ﬁrm partnerships as socially embedded events (e.g., Gulati, 1999; Hagedoorn, 2006; Walker, Kogut, & Shan, 1997). That is, the literature increasingly shows appreciation of existing social structures that surround potential partner ﬁrms, and the history of prior ties between them, which has signiﬁcantly furthered our understanding of inter-ﬁrm partnership formation. For example, Gulati (1995a) demonstrates that past partnerships between two ﬁrms cue the formation of subsequent ones, Garcia-Pont and Nohria (2002) ﬁnd that the density of ties in the group of ﬁrms surrounding two potential partners affects the likelihood of tie formation between them, and Hagedoorn (1993) shows that ﬁrms’ sectoral embeddedness signiﬁcantly inﬂuences their propensity to engage in new partnerships. The key message of contributions like these is that inter-ﬁrm partnership formation does not ﬁnd place in isolation, but rather does so in an embedded manner (Granovetter, 1985; Gulati & Gargiulo, 1999). Many contributions demonstrate that social embeddedness positively affects inter-ﬁrm partnership formation because it provides ﬁrms with, e.g., information on available partners, their capabilities, and credibility. However, a small number of contributions also suggest that the effect of embeddedness on new partnership formation is not necessarily positive. Under conditions of increasing social embeddedness, ﬁrms could face decreasing opportunities for the formation of valuable new partnerships with others embedded in the same partnership network (e.g., Burt, 1992; Duysters, Hagedoorn, & Lemmens, 2003; Hagedoorn, Letterie, & Palm, 2007; Uzzi, 1997). In other words, there may be a ‘gloomy’ side to ﬁrms’ embeddedness in their partnership network due to the over-entrenched nature of well-embedded inter-ﬁrm ties. In this chapter, we propose that this over-entrenchment cues ﬁrms to establish partnerships with un-embedded others, and gradually dissolve those with extant ones. Together, these spark a network’s evolution. We explore the gloomy side of embeddedness by distinguishing several distinct yet interrelated levels of overembeddedness and their separate and joint effects on inter-ﬁrm partnership formation. In so doing, we follow up on extant work by, e.g., Dacin, Ventresca, and Beal (1999), Gnyawali and Madhavan (2001), Hagedoorn (2006), and Simsek, Lubatkin, and Floyd (2003), who each propose interactive, multi-level conceptions of embeddedness that might provide us with a more in-depth understanding of the

Gloomy Side of Embeddedness

505

relationship between social embeddedness and inter-ﬁrm partnership formation. We illustrate that we can further our understanding of inter-ﬁrm partnership formation and, more speciﬁcally, ﬁrms’ choice of partners, through a multilevel, longitudinal analysis of the increasing embeddedness and eventual overembeddedness of ﬁrms in their networks of external partnerships. Our contribution is broadly twofold. First, we contribute a number of explanations of changes in the patterns of inter-ﬁrm network ties that stress their highly endogenous nature (cf. Gulati & Gargiulo, 1999; Madhavan, Caner, Prescott, & Koka, 2008). In a large number of settings, network patterns are path-dependent. We show how an understanding of the mechanisms that underpin such path-dependencies is important for explaining the dissolution of extant ties and the formation of novel ones. Second, we argue that the effects of embeddedness on a network’s evolution are complex because of their multi-level nature. One cannot study the effects at one level without properly accounting for variance at, and interactions with and between, other levels. Thinking about the evolution of networks in a multi-level fashion brings to the fore the complex dynamics at and between the individual embeddedness levels (cf. Hagedoorn, 2006). In this chapter, we proceed as follows. First, we present an outline of our understanding of several levels of social embeddedness, the interactions between these different levels, and their individual and combined effects on inter-ﬁrm partnership formation. Second, we subsequently discuss individual levels of overembeddedness, their possible effects on future inter-ﬁrm partnership formation, and the consequences of the interaction effects between different levels of overembeddedness. Lastly, we formulate some propositions that serve to guide theoretical and empirical development. Although this chapter is conceptual and theoretical in nature, we provide illustrative evidence to exemplify our main line of reasoning. In particular, we present illustrations of the effects of overembeddedness in the context of R&D partnership networks in the information technology industry during the 1990s. Our speciﬁc focus is on IBM, one of the major players in the industry. We took the information for these examples from the MERITCATI database on cooperative R&D agreements (see Hagedoorn, 2002).

EMBEDDEDNESS AND INTER-FIRM PARTNERSHIP FORMATION A vast body of previous work has introduced a differentiation of several levels of social embeddedness that affect the formation of relatively new

506

JOHN HAGEDOORN AND HANS T. W. FRANKORT

forms of economic organization such as inter-ﬁrm partnerships (Dacin et al., 1999; Granovetter, 1992; Gulati & Gargiulo, 1999; Hagedoorn, 2006; Hite, 2003; Lam, 1997; Simsek et al., 2003; Uzzi, 1997). Most of the extant work has, in broad terms, distinguished between structural embeddedness and relational embeddedness (Gulati, 1998). Structural embeddedness refers to the broader environmental setting of the social relationships in which economic organizations ﬁnd themselves. Relational embeddedness refers to the speciﬁc dyadic relationships in pairs of related economic organizations. In line with Hagedoorn (2006), we take this differentiation one step further by making a distinction into three separate levels of embeddedness of economic organizations, i.e. their environmental embeddedness, their inter-organizational embeddedness, and their dyadic embeddedness. One of the main advantages of such a differentiated understanding of the concept of embeddedness is that it allows for empirical tests that peal apart the micro-, meso-, and macro-level dimensions of embeddedness. Our particular differentiation of embeddedness resonates the recommendations by, among others, Dacin et al. (1999), Gnyawali and Madhavan (2001), Gulati and Gargiulo (1999), and Smelser and Swedberg (1994) to further specify the concept of embeddedness in such a way that it can be applied in a speciﬁc and empirically relevant context.

Environmental Embeddedness and Inter-Firm Partnership Formation At the most wide-ranging level of social embeddedness that affects inter-ﬁrm partnership formation, i.e. environmental embeddedness, we think of the sectoral, industry-speciﬁc propensity to build inter-ﬁrm partnerships.1 The larger the environmental embeddedness, the more ﬁrms are tied together beyond their immediate circle of partnerships into an overarching industry network. A considerable body of work has established that sectors of industry differ widely with respect to the degree to which ﬁrms engage in partnerships (Contractor & Lorange, 2002; Hagedoorn, 2002; Oster, 1999; Yu & Tang, 1992). In general, ﬁrms in high-tech industries engage in partnerships more frequently than those in medium- and low-tech industries. This has led to a lop-sided distribution of inter-ﬁrm partnerships across industries. DiMaggio and Powell (1983) explain that it is through a process of informed imitation, or ‘mimetic isomorphism’, that ﬁrms cope with uncertainty and ambiguity. By modeling their actions after successful others, ﬁrms avoid

Gloomy Side of Embeddedness

507

unnecessary investments in searching for and weighing the alternative avenues to economic success. Abrahamson and Rosenkopf (1993) also argue that such modeling, or the existence of so-called bandwagons, can occur in a vast array of situations in which ambiguity with respect to economic returns is the common denominator. Moreover, in such ambiguous and uncertain environments, inter-ﬁrm partnerships may serve to obtain legitimacy in the market (Dacin, Oliver, & Roy, 2007). This may be one of the possible explanations for the omnipresence of inter-ﬁrm partnerships in high-tech industries, which are typically characterized by higher levels of uncertainty, ambiguity, and change than medium- and low-tech industries. As an example, in the beginning of the eighties the call for compatibility between different parts of information systems prompted many incumbents to join their efforts in co-owned ventures (Harrigan, 1985). The consecutive examples set by ﬁrms like IBM and Hitachi in terms of engaging in these inter-ﬁrm partnerships were soon ampliﬁed at the industry level (see e.g., Hagedoorn, 2002). Gulati (1995a) found results that are consistent with this line of reasoning as he determined that, in three different industries, aggregate alliance formation at the industry level signiﬁcantly inﬂuenced dyadic alliance formation between 1980 and 1989. The existing partnership distribution for industries does not necessarily imply that the sectoral opportunity to engage in partnerships is given and stable over time. The research mentioned in the above merely indicates that in many high-tech industries and dynamic sectors inter-ﬁrm partnerships currently are a more familiar phenomenon than in other industries. Such familiarity is relevant as it indicates the degree to which ﬁrms ﬁnd themselves in larger industrial settings where many other ﬁrms are also engaged in inter-ﬁrm partnerships.

Inter-Organizational Embeddedness and Inter-Firm Partnership Formation The inter-organizational embeddedness of inter-ﬁrm partnerships is the next level of embeddedness where inter-ﬁrm networks are created by groups of ﬁrms or strategic blocks. These groups or strategic blocks refer to ‘y a set of ﬁrms that are connected more densely to each other than to other ﬁrms in the industry y’ (Nohria & Garcia-Pont, 1991, p. 106). Early work by Harrigan (1985) already indicated the relevance of understanding the role of these groups, which she described as constellations of interacting ﬁrms (see also Gomes-Casseres, 1996; Granovetter, 1994). In such groups, ﬁrms are tied together by a network of relatively strong ties where ﬁrms maintain and

508

JOHN HAGEDOORN AND HANS T. W. FRANKORT

replicate multiple ties within their group (Gimeno, 2004; Nohria & GarciaPont, 1991; Vanhaverbeke & Noorderhaven, 2001). This understanding of embeddedness in terms of groups relates to the concept of small worlds where inter-ﬁrm networks are clique-like settings of ﬁrms. The density and tie strength within these groups is typically high, whereas the strength of ties connecting different groups remains low (Schilling & Phelps, 2007; Watts, 1999). Most studies on groups of ﬁrms largely echo the seminal contribution by Coleman (1988), which argues that being part of a dense group of network actors is advantageous since it fosters trust development and cooperation among group members. The dense structure gives rise to obligations and sanctioning behavior conducive to the functioning of the group. In addition, it contributes to increased exchange efﬁciencies among group members (Soda, Zaheer, & Carlone, 2008). The inter-organizational embeddedness of inter-ﬁrm partnerships leads to a form of generalized exchange, which we understand to represent an intricate web of dependencies and informational spillovers that reaches beyond mere dyadic reciprocity (Levine & Kurzban, 2006; Takahashi, 2000). Research by, among others, Anand and Khanna (2000), Dyer and Singh (1998), and Gulati (1998) indeed indicates that both the familiarity of ﬁrms with partnering and their interactions within groups of relatively densely connected ﬁrms increase the likelihood that ﬁrms will engage in new partnerships. Both ﬁrms’ familiarity and their interactions establish embeddedness mechanisms that prompt the endogenous evolution of inter-organizational network ties (cf. Gulati & Gargiulo, 1999). More generally, based on insights from social network theory, we note that ﬁrms that become well-embedded in these networks accumulate informational advantages that increase their propensity to engage in new partnerships (Freeman, 1979; Gulati, 1999; Wasserman & Faust, 1994).

Dyadic Embeddedness and Inter-Firm Partnership Formation At the third level of embeddedness we ﬁnd dyadic embeddedness, which can be understood in the context of repeated ties within pairs of ﬁrms (Gulati, 1995b; Gulati & Gargiulo, 1999). Dyer and Singh (1998) and Gulati (1995a) explain that ﬁrms will most probably enter into partnerships with ﬁrms with which they have collaborated before. This repeated tie effect likely creates strong cohesive ties between ﬁrms through frequent interactions. These strong ties are solid and reciprocal relationships that create a basis for trust

Gloomy Side of Embeddedness

509

and closeness between partners. Trust and closeness are believed to lead to dyadic reciprocation over time, which entails the ‘y extension of beneﬁts to transacting partners y’ and vice versa ‘y when a new situation arises y’ (Uzzi, 1996b, p. 678). Unlike the abovementioned generalized exchange inherent in inter-organizational embeddedness, the notion of dyadic embeddedness thus entails a more restricted form of exchange in which the two actors in a tie reciprocate to each other only (Takahashi, 2000; Uzzi, 1996b). Apart from the repeated nature of partnerships, dyadic embeddedness also refers to the simultaneous operation of multiple partnerships between two parties, or a combination of elements comprising, what Gimeno and Woo (1996) label ‘economic multiplexity’ (for a discussion of the evolution of dyadic ties, see Hite, 2008). In a partnership that is multiplex, several partnerships in a dyad may exist that serve to regulate symbiotic, competitive, and commensalistic interdependencies (Pfeffer & Nowak, 1976). Evolving routines in, e.g., the commensalistic ‘dimension’ of a multiplex partnership may aid in straightening out possible complications in, e.g., the competitive dimension. As such, multiplexity adds to the dyadic reciprocation over time by enabling the contemporaneous conservation and continuation of simultaneous ties. Possible reasons for a sustained preference for repeated or simultaneous dyadic partnerships are, among others, the costly and time-consuming nature of both the search for trustworthy and valuable partners and the ﬁnal selection process of those partners. In addition, changing partners increases the risk that other relationships with existing partners will be dissolved. As relational trust becomes embedded in repeated ties between ﬁrms, it positively affects the prolongation and stability of their relationship (Gulati, 1995b; Nooteboom, Berger, & Noorderhaven, 1997). Zollo, Reuer, and Singh (2002) argue that the development of inter-organizational routines is both an antecedent to and consequence of the occurrence of dyadic embeddedness: routines develop through repeated interactions with the same partner and serve to smoothen the interactions in subsequent partnerships. Hence, such routines serve as an endogenous partnership development mechanism. Consequently, ﬁrms prefer to engage in local search and replicate their existing ties rather than search for novel ones.

Interactions among Levels of Embeddedness Recent contributions (Dacin et al., 1999; Dansereau, Yammarino, & Kohles, 1999; Hagedoorn, 2006; Hite, 2003) stress that individual levels of

510

JOHN HAGEDOORN AND HANS T. W. FRANKORT

embeddedness are indeed important for understanding the effect of social embeddedness on inter-ﬁrm partnering. However, we can gain a more intricate understanding of the complex nature of this relationship in the analysis of the nested interaction of multiple levels of embeddedness (Dacin et al., 1999; Gnyawali & Madhavan, 2001). The core argument for such a nested perspective is that the combined environmental, inter-organizational and dyadic embeddedness of partnering ﬁrms exercise a multiplicative, interacting effect on future joint partnering (Hagedoorn, 2006). This implies that patterns of sectoral inter-ﬁrm partnership formation, as well as the speciﬁcs of groups of partnering ﬁrms and the history of pairs of ﬁrms, jointly affect future partnership formation. Inter-ﬁrm partnership formation is rooted in the dyadic embeddedness between partnering ﬁrms, which itself is affected by inter-organizational embeddedness in terms of the broader experience of ﬁrms with partnering and their surrounding networks. This combination of different levels of embeddedness is overarched and reinforced further by an environmental embeddedness that is characterized by a set of industry-based forces that additionally shape the nature of ﬁrms’ partnering activities (Hagedoorn, 2006). However, these effects should be seen in a dialectic context, where it is not only the effect of the higher levels of embeddedness on embeddedness at lower levels, the process also works in the opposite direction. The more ﬁrms engage in repeated ties, increasing dyadic embeddedness, the more this affects inter-organizational embeddedness as the density of in-group ties between ﬁrms increases (Hite, 2008). This, in turn, has an effect on the environmental embeddedness of inter-ﬁrm partnerships, as partnership formation in an industry increases.

OVEREMBEDDEDNESS AND INTER-FIRM PARTNERSHIP FORMATION Many contributions – whether considering individual-level embeddedness effects or the interaction of multiple levels – stress the positive effects of social embeddedness on partnership formation. Note that this is exactly what we have done to this point. At some point in time, however, increased partnership formation might create overembeddedness (Uzzi, 1997) in which case ﬁrms face fewer opportunities for entering into valuable new partnerships. As explained by, e.g., Burt (1992) and Gargiulo and Benassi (2000), a concrete effect of overembeddedness would eventually be the declining

Gloomy Side of Embeddedness

511

propensity of network players to form, what could be considered as, redundant partnerships. An over-dependence on a particular group of partners and diminishing information gains through additional partnerships within the same group of ﬁrms are known to play a role in this overembeddedness effect (Chung, Singh, & Lee, 2000; Duysters et al., 2003; Gulati, 1995b; Saxton, 1997; Uzzi, 1996b). This effect is most apparent at the level of pairs of ﬁrms, i.e. the level of dyadic embeddedness (Chung et al., 2000; Gulati, 1995b), but depending on the number of (potential) partners in networks and industries, the effect of overembeddedness can take place at each level of embeddedness. Our understanding of overembeddedness is that, up to a certain threshold, the embeddedness of inter-ﬁrm partnerships parallels a growth of new partnerships that generates useful new information (see Fig. 1). Beyond a certain point, where social embeddedness reaches its maximum in terms of valuable new partnerships – the gray area in Fig. 1 – there is an increasing decline of new information gains (Hagedoorn, et al. 2007). Additional partnerships then lead to decreasing marginal returns to the ﬁrms involved. Consequently, the potential for useful new partnerships with existing partners, within existing groups of interconnected ﬁrms, and within the industry declines (Duysters et al., 2003; Kenis & Knoke, 2002; Uzzi, 1996b, 1997). In short, the three embeddedness mechanisms introduced in the above gradually alter the opportunity structure faced by the ﬁrms in the partnership network. Here, we propose that a direct consequence of this

Fig. 1.

The Relationship between (Over)embeddedness and Valuable New Partnership Formation.

512

JOHN HAGEDOORN AND HANS T. W. FRANKORT

process is the shift in a ﬁrm’s partner choice (cf. Hagedoorn et al., 2007), which will coincide with the dissolution of overembedded partnerships. As continued local cooperation with existing partners, within existing groups of interconnected ﬁrms, and within the industry’s web of partnerships, becomes less fruitful, ﬁrms will increasingly detach from such existing, ‘local’ partners and select ‘nonlocal’ ones, i.e. those in different groups and/ or industries. The process of embedding thus eventually leads to overembeddness, which sparks the dissolution of extant ties and the concurrent formation of novel ones. This cues a new cycle of embedding, overembeddedness, dissolution of extant ties, and the simultaneous search for and formation of novel ones. That the search for novel partners most likely coincides with a decline in the number of extant, embedded partnerships has several reasons, three of which we mention here. First, a ﬁrm’s capacity to maintain external partnerships is limited. Therefore, investing time and effort in certain partnerships necessarily bounds investments in others.2 Assuming that ﬁrms ultimately prefer to enter the most productive partnerships, they thus face the trade-off of maintaining less productive partnerships versus engaging in more productive others. Second, closely related to the previous reason, unlike many interpersonal relationships, inter-ﬁrm partnerships ultimately serve economic purposes to the ﬁrms involved. Although these economic purposes need not be reﬂected in immediate or direct returns – e.g., status consequences of afﬁliation with certain (groups of) ﬁrms may take time to surface – it is fair to assume that trust and shared norms are of little use in the absence of (longer-term) economic beneﬁts. Third, the endogenous mechanisms driving ﬁrms’ embeddedness in their partnership networks cause local informational, cognitive and normative convergence within such networks (cf. Baum & Ingram, 2002). The ﬂow of novel information and the concurrent emergence of new mental images of the cooperative landscape that result from a ﬁrm’s involvement with nonlocal ﬁrms are, more likely than not, in dispute with extant local representations of this landscape.

Dyadic Overembeddedness and Inter-Firm Partnership Formation At the level of possible dyadic overembeddedness, empirical work by, e.g., Gulati (1995a) and Rosenkopf, Metiu, and George (2001) indicates an inverted U-shaped, curvilinear, relationship between the number of previous partnerships between two ﬁrms and the likelihood of valuable new partnership formation between them. Fear of over-dependence on speciﬁc

Gloomy Side of Embeddedness

513

partners, declining useful information exchanges and decreasing opportunities for learning from these partners appear to be main indicators of possible overembeddedness of partnerships and their sponsoring ﬁrms. Continuous partnership formation and the concomitant information exchanges between two ﬁrms might eventually create certain similarities between these partners (cf. Gomes-Casseres, Hagedoorn, & Jaffe, 2006; Mowery, Oxley, & Silverman, 1996). At some point in time, this will have a negative effect on the potential value of an ever-increasing partnership formation process between the ﬁrms (Mowery et al., 1996; Saxton, 1997; Uzzi, 1996b). Consequently, in terms of the aforementioned possibly multiplex nature of dyadic embeddedness, ﬁrms may at a certain point in time decide to ‘de-multiplexify’ their simultaneous partnerships in a dyad as to maintain a certain degree of ﬂexibility for future tie formation (Uzzi, 1996a). Although multiplexity adds to appropriate governance of interdependencies between ﬁrms, it may also saturate the dyad and lead to the loss of its momentum. Rosenkopf et al. (2001) provide a detailed understanding of the dyadic embeddedness of partnership formation by relating both joint technical committee activity and previous dyadic partnerships to new partnership formation between two ﬁrms. Their data show that both joint technical committee activity and previous dyadic alliances individually have an inverted U-shaped relationship to new partnership formation. Moreover, the interaction of these two phenomena also negatively inﬂuences dyad formation – suggesting that, beyond a certain threshold, the interplay between various elements of dyadic embeddedness apparently leads to overentrenchment of the dyad, decreasing informational returns, and to consecutive decreases in the formation of valuable new partnerships. A direct consequence, we believe, will be that the ﬁrms involved start searching for novel partners and gradually dissolve extant patnerships. The formation of partnerships by IBM in the information technology industry provides an interesting illustration of this potential dyadic overembeddedness effect. Figs. 2 and 3 present IBM’s ego network based on newly formed R&D partnerships for the periods 1990–1994 and 1995– 1999, respectively.3 Dotted lines represent 1–3 R&D partnerships between ﬁrms, whereas solid lines indicate 4–9 partnerships and thick lines represent 10 or more R&D partnerships. One of the main observations in these network plots is that during the ﬁrst half of the 1990s, IBM appeared to be well embedded in close-knit R&D cooperation through a series of multiple dyadic alliances with four ﬁrms: Apple, Siemens, Toshiba and HewlettPackard (see Fig. 2).

514

Fig. 2.

JOHN HAGEDOORN AND HANS T. W. FRANKORT

IBM’s Ego Network Based on Newly Established R&D Partnerships in 1990–1994.

Most illustrative is the case of the IBM-Apple collaboration. The two created eight R&D partnerships in 1991, followed by ﬁve partnerships in 1993. In the following three years, IBM and Apple annually introduced a small number of two or three partnerships but the opportunities for further collaboration at such an extensive scale appear to have diminished during the second half of the 1990s, when most partnerships were terminated and no new R&D partnerships were established (see also Hagedoorn, Carayannis, & Alexander, 2001). The R&D partnerships formed by IBM and Siemens, IBM and Toshiba, and IBM and Hewlett–Packard portray a somewhat similar pattern. In a short period during the ﬁrst half of the 1990s,

Gloomy Side of Embeddedness

Fig. 3.

515

IBM’s Ego Network Based on Newly Established R&D Partnerships in 1995–1999.

IBM created 17 R&D partnerships with Apple, 10 with Siemens, 9 with Toshiba, and 7 with Hewlett-Packard. During the second half of the 1990s, most of the existing R&D partnerships were only continued for some time but R&D collaboration was not extended at the same level and with the same degree of intensity (see Fig. 3). IBM started extensive collaborative efforts on joint R&D with a different set of ﬁrms with which it had no or only few prior R&D partnerships. During that period, IBM established multiple R&D partnerships with AT&T, Intel, Motorola, Netscape, Novell, Oracle, and Sun-Microsystems. Apparently, opportunities for further R&D cooperation with individual ﬁrms from the ﬁrst local group of partners – in which IBM was well-embedded through multiple dyadic ties – had dried up in a relatively short period of time and other ﬁrms became attractive partners for R&D collaboration.

516

JOHN HAGEDOORN AND HANS T. W. FRANKORT

Inter-Organizational Overembeddedness and Inter-Firm Partnership Formation At the level of inter-organizational embeddedness, the phenomenon of overembeddedness occurs in networks of partnering ﬁrms when they become locked-in within densely connected sub-networks (groups, cliques, or blocks). In such case, groups of well-connected ﬁrms become isolated from others in the broader network of (potential) partners. Some contributions show the nonlinear, or inverted U-shaped, effects of interorganizational embeddedness on various performance outcomes at the ﬁrm level. For example, Uzzi (1996b) found that high levels of interorganizational embeddedness through ‘embedded ties’ result in signiﬁcantly higher organizational failure rates. As suggested by, e.g., Duysters et al. (2003), Gargiulo and Benassi (2000), and Gomes-Casseres (1996), inter-organizational overembeddedness leads to excessively cohesive networks that concentrate on inter-ﬁrm partnerships within existing groups of partners. Information about potential partners outside the existing sub-networks does not reach the participants, and the group of partners becomes inﬂexible and inert, while the number of valuable new partnerships declines over time. For example, Garcia-Pont and Nohria (2002) show, in the global automobile industry, that initially inter-organizational embeddedness positively inﬂuences new partnership formation but that it is beyond a certain intra-group density threshold that the probability for new tie formation decreases signiﬁcantly. Firms may even implicitly experience social pressures from their partners to replicate their ties within their own network, e.g., to prevent knowledge spillover effects outside their existing network. This is somewhat akin to Portes and Sensenbrenner (1993, p. 1340) who mention ‘y the constraints that community norms put on individual action and receptivity to outside culture y’. Thus, an implicit expectation of loyalty to other network members can prevent ﬁrms from allying with ﬁrms from competing groups (Gulati, Nohria, & Zaheer, 2000) as this might result in conﬂicting interests among partners (Nohria & Garcia-Pont, 1991). Hence, certain potential outside partners are not part of a ﬁrm’s choice set when they have ties to competing groups. In this way, competing partnership groups can foreclose further partnering opportunities with non-group members (Gimeno, 2004; Gomes-Casseres, 1996). As a consequence, potentially interesting outside partners are simply excluded from partner selection and, based on their initial choices, ﬁrms can become locked-in within their own group of partners (Kim, Oh, & Swaminathan, 2006).

Gloomy Side of Embeddedness

517

Gargiulo and Benassi (2000) and Uzzi (1997) point at the danger of being cognitively locked-in when the rigidity among group members will increase the likelihood that they are isolated from ﬁrms outside of their own network. The group of partners functions as a ﬁlter that selects the information and new perspectives for its members. In the end, an interorganizationally over-embedded group of partners consists mainly of exclusively connected, strongly embedded inter-ﬁrm ties (Uzzi, 1997) where ﬁrms face a strategic gridlock (Gomes-Casseres, 1996) as they only engage in local search for partners within their own network of partnerships. The understanding of the sub-optimal cohesiveness in strategic groups within particular industries (where strategic groups are deﬁned as collectives of interacting ﬁrms) also reﬂects the notion of inter-organizational overembeddedness. As Thomas and Carroll (1994) explain, stable and dense networks of ﬁrms can be seen as robust building blocks of strategic groups. However, once these dense networks within strategic groups become insensitive to external signals of potentially valuable change, their robustness and stability become sub-optimal. Also, the level of inertia frustrates further economic growth within these strategic groups. Thus, after a certain threshold level of inter-organizational embeddedness has been reached, the likelihood that social structural mechanisms supersede rationality with respect to external initiatives – such as inter-ﬁrm partnership formation – will steadily amplify and, consequently, hinder effective economic action (cf. Gulati & Westphal, 1999; Uzzi, 1997, p. 59). Firms’ cognitive lock-in and the decreasing marginal informational and substantial returns they experience will inﬂuence their performance and partnering behaviors. We expect that ﬁrms who become cognitively lockedin within a group of partners will only endure the negative informational consequences of such lock-in up to a certain threshold. Although research documents that ﬁrms allow such overembedded alliances to persist (Inkpen & Ross, 2001), their negative performance impact will at some point in time cue the search for novel, nonlocal partners (see e.g., Baum, Rowley, Shipilov, & Chuang, 2005), even despite group-level pressures to replicate local ties. Even absent such dramatic negative performance effects, ﬁrms may start to look for nonlocal contacts as to avoid the over-dependence on key local players in case such players malfunction themselves (cf. Uzzi, 1997). The network in the information technology industry from the 1990s, in which IBM was well placed, represents an interesting example of interorganizational overembeddedness. During the ﬁrst half of the 1990s, the core of the wider inter-ﬁrm network in which IBM participated consisted of

518

JOHN HAGEDOORN AND HANS T. W. FRANKORT

multiple partnerships between Apple, Hewlett-Packard, IBM itself, Siemens, and Toshiba. This created a dense multi-dyadic network of computer hardware manufacturers with a variety of interests in other sub-sectors of information technology. IBM became well embedded in a network of R&D partnerships of ﬁrms with similar research objectives that were all well connected to each other. However, this group of ﬁrms, with IBM as a major player, appeared to have become over-embedded. During the early years of the 1990s, the ﬁrms mentioned in the above created a densely populated clique of cooperating ﬁrms that quickly dismantled during the second half of the 1990s. During that period, a new inter-organizational network gradually emerged through a variety of new R&D partnerships within another group of ﬁrms. IBM also became part of this new network and so did many other computer hardware ﬁrms. Because of these changes in the mid-1990s, IBM became embedded in a new network of multiple R&D partnerships with other players such as AT&T, Intel, Motorola, Netscape, Novell, Oracle, and Sun-Microsystems. As a further illustration, Fig. 4 shows the number of newly established R&D partnerships with ﬁrms from both the original as well as the new group of partners. Whereas during the ﬁrst half of the 1990s IBM created 43 partnerships with Apple, Hewlett-Packard, Siemens, and Toshiba, it only formed 24 with these ﬁrms between 1995 and 1999. In contrast, the partners that obtained prevalence in IBM’s partnership portfolio during the second half of the 1990s – through the formation of 40 new partnerships – participated in only 12 partnerships with IBM between 1990 and 1994. IBM’s group of most important allies (in terms of numbers of partnerships) thus changed drastically, especially considering the fact that our data indicate that the biggest changes actually occurred only after 1996.

Environmental Overembeddedness and Inter-Firm Partnership Formation Obviously, given the relatively large number of potential partnerships at the level of environmental embeddedness, we expect that the potential degree of overembeddedness at the level of industries is limited. Nevertheless, some research indicates that the ﬁnite possibility of increasing partnership formation at this level is not just a theoretical notion. For instance, a study of partnership formation in the electronics industry by Park and Ungson (1997) demonstrates that, given the degree of partnership formation in that industry, inter-sectoral partnerships with ﬁrms

519

Gloomy Side of Embeddedness 50

Newly Established R&D Partnerships (#)

45 40 35 30 25 20 15 10 5 0 1990-94

1995-99 Time Period

Fig. 4. Numbers of IBM’s Newly Established R&D Partnerships with Firms in Two Main Groups, 1990–1994 and 1995–1999. Source: The gray columns represent IBM’s R&D partnerships with Apple, Hewlett–Packard, Siemens, and Toshiba. The black columns represent its R&D partnerships with AT&T, Intel, Motorola, Netscape, Novell, Oracle, and Sun-Microsystems.

from other industries appear to have a higher likelihood of continuation than intra-sectoral partnerships that focus on ﬁrms from the electronics sector per se. The tendency to ‘cavalierly’ use inter-ﬁrm partnerships in the belief that they are the key to success in particular industries (Inkpen & Ross, 2001) may lead to saturation and decreasing numbers of newly established interﬁrm partnerships. Partnerships that, often unconsciously, result from herd behavior in some way lack an inherent ‘raison d’eˆtre’ in terms of their substantial and relational aspects. This may lead ﬁrms to be increasingly dissatisﬁed with given partnerships as ﬁrms are unconscious of the discrepancy between, on the one hand, the exact environmental forces that

520

JOHN HAGEDOORN AND HANS T. W. FRANKORT

drove them to engage in certain partnerships in the ﬁrst place and, on the other hand, their (misplaced) expectations given these forces.4 Ceteris paribus, we expect that the dissatisfaction with intra-industry partnerships will lead to shifts in the ﬁrm-level pattern of intra- versus inter-industry partnership formation. Relatedly, increasing numbers of ties across groups of densely connected ﬁrms will lead to decreases in the requisite differentiation among (groups of) ﬁrms that is foundational to the achievement of competitive advantages (Baum & Ingram, 2002; Schilling & Phelps, 2007). Driven by increased numbers of industry-wide connections among ﬁrms, the increased convergence of the knowledge base underlying an industry’s main activities likely drives out the variety necessary for ﬁrms to progress and differentiate themselves from others. The inter-ﬁrm R&D partnerships in the information technology industry illustrate the possible effect of environmental overembeddedness on changes in the network. During the early 1990s the emphasis in the inter-ﬁrm network in information technology in general, and for IBM in particular, was on R&D partnerships in computer hardware and related activities such as computer-based telecommunication systems and supporting software. Given the limited number of ﬁrms that were active in these particular activities and given their focus on somewhat similar interests, many of them started to establish R&D partnerships in other information technology ﬁelds and in sectors outside information technology. These new intersectoral R&D partnerships concentrated on related ﬁelds such as microelectronics, software, various internet-related products and services, and a host of multimedia technologies. Some additional data on the R&D partnerships of IBM illustrate the ramiﬁcations of environmental overembeddedness for IBM’s ego network. Although IBM’s ego network does not present the industry’s whole partnership network, it reﬂects very clearly the trend that can be observed in the industry at large (see also Cloodt, Hagedoorn, & Roijakkers, 2006, 2007). Between 1990 and 1999, IBM alone established new R&D ties with 163 different ﬁrms. During the period 1990–1994, it tied to 95 ﬁrms. Of the 93 ﬁrms it tied up with between 1995 and 1999, it had only set up R&D partnerships with 25 during the previous period 1990–1994. What this shows is that of IBM’s entire R&D partnership portfolio during 1990–1999, only 15% of the ﬁrms served as R&D partner in both 1990–1994 as well as in 1995–1999. Most of IBM’s newly established ties in the latter half of the 1990s were thus of an un-embedded nature (68 out of 93), see also Figs. 2 and 3.

Gloomy Side of Embeddedness

521

Interactions among Levels of Overembeddedness As indicated by, e.g., Hagedoorn (2006) and Uzzi (1997), the nested interaction of different levels of social embeddedness can jointly affect the negative impact of overembeddedness even further than in the case of singlelevel effects of overembeddedness. Fig. 5 summarizes our understanding of the effects of the growth in embeddedness and its effect on the ﬁrm-level choice of local, embedded partners, versus nonlocal, novel partners. Similar to Hagedoorn’s (2006) theoretical understanding of the strengthening, positive, and multiplicative effects across levels of embeddedness in determining rates of inter-ﬁrm partnership formation, we also expect such effects at increasing levels of embeddedness. Essentially, such multiplicative

Fig. 5. The Interaction of the Growth in Environmental, Inter-organizational, and Dyadic Embeddedness and their Effects on Overembeddedness and Partnership Formation.

522

JOHN HAGEDOORN AND HANS T. W. FRANKORT

effects progress along a continuum, the effects of which we understand to reﬂect an aggregation of parabolic, inverted U-shaped effects at the individual embeddedness levels as depicted in Fig. 1. In other words, at each of the three levels of embeddedness we described, we expect a parabolic effect to ﬁnd place. The difference among the progression of these three parabolic effects is the time it takes for the individual curves to evolve, which should aid in the empirical identiﬁcation of such effects in a longitudinal research design.5 As presented in Fig. 5, we expect a positive effect of the growth of embeddedness at each level on the degree of overembeddedness, which will ultimately cause a shift in the nature of newly established partnerships. As indicated before, this shift has two faces, the order and intertwinement of which is to be identiﬁed empirically. First, ﬁrms will quit forming local partnerships and gradually dissolve extant local partnerships. Second, ﬁrms will start engaging in nonlocal partnerships. The expected curvilinear effect of increasing levels of embeddedness on the growth of valuable new local partnerships (see Fig. 1) is more pronounced for various combinations, i.e. interactions, of different levels of embeddedness than for individual levels of social embeddedness. Given the expected effects of overembeddedness at different levels, based on theory development and the empirical evidence from previous research, we postulate that at a given point in time the effects of overembeddedness on partnership formation will differ for each of these different levels of embeddedness. Ceteris paribus, dyadic and inter-organizational embeddedness will see the effects of overembeddedness at an earlier point in time than environmental embeddedness. We also expect differences between dyadic and inter-organizational levels of embeddedness, where the stage of dyadic overembeddedness will be reached earlier than the stage of inter-organizational overembeddedness. We brieﬂy return to the illustration of the possible effects of overembeddedness for IBM. It is obvious that, after a number years of intense R&D collaboration, the number of options for continued R&D partnering between IBM and Apple had grown limited compared to the potential number of other interesting options in IBM’s network that were still open to the ﬁrm. In addition, given the somewhat limited scope of the core of the network of R&D partnerships of computer ﬁrms in which IBM operated during the ﬁrst half of the 1990s, there were still multiple other opportunities for R&D partnership formation outside its existing network. Hence, Proposition 1. Growth in the dyadic embeddedness of inter-ﬁrm partnerships will have an earlier impact on overembeddedness than the growth in

Gloomy Side of Embeddedness

523

environmental embeddedness and inter-organizational embeddedness will. As such, dyadic embeddedness will lead to (i) decreasing opportunities for valuable new local partnership formation, and (ii) increasing opportunities for valuable nonlocal partnership formation sooner than environmental embeddedness and inter-organizational embeddedness will. Along similar lines, the development of groups of connected ﬁrms will have an earlier impact on patterns of partnership formation than industrybased forces that develop rather sluggishly over time. Therefore, Proposition 2. Growth in inter-organizational embeddedness will have an earlier impact on overembeddedness than the growth in environmental embeddedness will. As such, inter-organizational embeddedness will lead to (i) decreasing opportunities for valuable new local partnership formation, and (ii) increasing opportunities for valuable nonlocal partnership formation sooner than environmental embeddedness will. In addition to the expected greater impact of the interaction of different levels of embeddedness compared to the effect of individual levels of embeddedness, we also anticipate that the interaction of various levels of social embeddedness will have alternative effects. Following the various expected increasing effects at different levels of embeddedness and the empirical evidence from other studies, we postulate that at a given point in time the effects of the interaction for different levels of embeddedness will have an increasing effect on the overembeddedness of new partnership formation. However, again, there are a larger number of potential partnerships at the level of environmental embeddedness, where the risk of overembeddedness is smaller than at the level of inter-organizational embeddedness and certainly at the level of dyadic embeddedness. This implies that various combinations of interacting levels of social embeddedness of inter-ﬁrm partnerships generate differential outcomes as to their aggregate effect on new partnership formation. In the context of the exemplary setting of IBM and the information technology industry, this implies that the level of overembeddedness of IBM’s R&D partnerships with Apple, in combination with IBM’s wellembedded network with other computer hardware manufacturers during the ﬁrst half of the 1990s, was very high and with increasingly limited opportunity for useful future partnership formation. Consequently, Proposition 3. The interaction of dyadic embeddedness with interorganizational embeddedness will have (i) a greater negative impact on

524

JOHN HAGEDOORN AND HANS T. W. FRANKORT

new local partnership formation and (ii) a greater positive impact on new nonlocal partnership formation than the interaction of environmental embeddedness with the other levels of embeddedness will.

CONCLUSION To date, the literature on the relationship between social embeddedness and inter-ﬁrm partnerships has typically stressed the positive effects of embeddedness on inter-ﬁrm partnership formation. However, increasing numbers of inter-ﬁrm partnerships at different levels of social embeddedness can generate negative effects that we characterized as the gloomy side of embeddedness through overembeddedness. We point at some of the speciﬁc multi-level effects of the gloomy side of embeddedness on new partnership formation and the temporal nature of developments of these effects. In particular, we expect that the interaction between dyadic overembeddedness – the redundant relationship between two ﬁrms through long-term repeated ties – and inter-organizational overembeddedness – through crowded groups or congested cliques of exclusively cooperating ﬁrms – is a major cause of overembeddedness. In the end, the overembeddedness will become evident through the actual redundancy of newly created local inter-ﬁrm partnerships, the increased dissolution of such partnerships, and the subsequent formation of – what are to the ﬁrm – novel, nonlocal partnerships. Necessarily, our conception of ﬁrms’ embeddedness in networks of external partnerships has limitations. Two are conspicuous. First, our model does not speciﬁcally identify exogenous drivers of partnership formation. Our focus on the gloomy side of embeddedness has led us to focus predominantly on endogenous mechanisms that determine the choice of local versus nonlocal partners. We note, however, that ﬁrms’ speciﬁc choice of nonlocal partners will without doubt reﬂect more exogenous factors, such as, e.g., the distribution of technological and ﬁnancial resources among ﬁrms (cf. Ahuja, 2000; Baum & Ingram, 2002; Gulati & Gargiulo, 1999). Second, we cannot pinpoint the exact sequencing of the dissolution of local partnerships on the one hand, and the formation of nonlocal partnerships on the other. At any rate, the empirical identiﬁcation of how ﬁrms sequence these actions, and the mechanisms that underlie this sequencing, poses a formidable research challenge. After all, ﬁrms seldom – if at all – make public the dissolution of partnerships with the same aplomb that

525

Gloomy Side of Embeddedness

characterizes the announcement of new partnerships. This thus requires careful quantitative and qualitative analyses of networking processes (such as the mixed-methods explanatory design in Madhavan et al., 2008). Overall, we hope to have illustrated that building an understanding of the nested, multi-level implications of overembeddedness ultimately necessitates the consideration of complex interactions between those levels over time. Although temporal differences in the impact of levels of embeddedness on inter-ﬁrm partnership formation require longitudinal research designs spanning decades, the upshot is that they allow for pealing apart the macro-, meso-, and micro-level drivers of interest. We hope to have encouraged efforts in this direction.

NOTES 1. We deﬁne inter-ﬁrm partnerships as all forms of cooperation between ﬁrms that share knowledge and resources through formal agreements that can be governed through either equity-based joint ventures or a range of non-equity, contractual agreements. 2. This does not necessarily mean that a ﬁrm’s capacity to manage partnerships is constant over time. It may increase as cooperative experience feeds into its ability to evaluate and absorb external information, see e.g., Gulati (1999), Powell, Koput, and Smith-Doerr (1996), and Powell and Grodal (2005). 3. The network plots result from a two-step procedure. First, all ﬁrms’ MDS coordinates are generated in a two-dimensional space, based on R&D partnering data from the MERIT-CATI database. Second, we use the network visualization software tool Najojo (see the appendix) to add ﬁrm labels to the nodes and connect the nodes based on the number of partnerships among the ﬁrms in IBM’s ego network. 4. We note, however, that the performance consequences of imitation are likely contingent on the speciﬁc nature of the cooperative environment, see e.g., Soda et al. (2008). 5. Such a design ideally spans several decades; see e.g., Gulati and Gargiulo (1999, p. 1478) who indicate that their design that spans 10 years was effectively too short to observe parabolic sector-level phenomena of interest (see also Hagedoorn, 2002).

ACKNOWLEDGMENTS We would like to acknowledge editors Joel Baum and Tim Rowley for their insightful suggestions, and Sarianna Lundan and participants in the Advances in Strategic Management Developmental Conference at the

526

JOHN HAGEDOORN AND HANS T. W. FRANKORT

Rotman School of Management, University of Toronto, May 2007, for commenting on the ideas developed in this chapter. We also thank Marc van Ekert for research assistance. This chapter was written in part while the second author was a visiting doctoral student at the Institute of Management, Innovation and Organization (IMIO) at the Walter A. Haas School of Business, University of California at Berkeley, on grants generously provided by the Maastricht Research School of Economics of Technology and Organizations (METEOR) and Dr. Hendrik Muller’s Vaderlandsch Fonds.

REFERENCES Abrahamson, E., & Rosenkopf, L. (1993). Institutional and competitive bandwagons: Using mathematical modeling as a tool to explore innovation diffusion. Academy of Management Review, 18, 487–517. Ahuja, G. (2000). The duality of collaboration: Inducements and opportunities in the formation of interﬁrm linkages. Strategic Management Journal, 21, 317–343. Anand, B., & Khanna, T. (2000). Do ﬁrms learn to create value? The case of alliances. Strategic Management Journal, 21, 295–315. Baum, J. A. C., & Ingram, P. (2002). Interorganizational learning and network organizations: Toward a behavioral theory of the interﬁrm. In: M. Augier & J. G. March (Eds), The economics of change, choice, and structure: Essays in the memory of Richard M. Cyert (pp. 191–218). Cheltenham, UK: Edward Elgar. Baum, J. A. C., Rowley, T. J., Shipilov, A. V., & Chuang, Y.-T. (2005). Dancing with strangers: Aspiration performance and the search for underwriting syndicate partners. Administrative Science Quarterly, 50, 536–575. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Chung, S., Singh, H., & Lee, K. (2000). Complementarity, status similarity and social capital as drivers of alliance formation. Strategic Management Journal, 21, 1–22. Cloodt, M., Hagedoorn, J., & Roijakkers, N. (2006). Trends and patterns in interﬁrm R&D networks in the global computer industry: An analysis of major developments, 19701999. Business History Review, 80, 725–746. Cloodt, M., Hagedoorn, J., & Roijakkers, N. (2007). Inter-ﬁrm R&D networks in the global software industry: An overview of major trends and patterns. Working paper, Maastricht University, The Netherlands. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Contractor, F. J., & Lorange, P. (Eds). (2002). Cooperative strategies and alliances. Oxford, UK: Elsevier. Dacin, M. T., Oliver, C., & Roy, J.-P. (2007). The legitimacy of strategic alliances: An institutional perspective. Strategic Management Journal, 28, 169–187. Dacin, M. T., Ventresca, M. J., & Beal, B. D. (1999). The embeddedness of organizations: Dialogue & directions. Journal of Management, 25, 317–356.

Gloomy Side of Embeddedness

527

Dansereau, M., Yammarino, F. J., & Kohles, J. C. (1999). Multiple levels of analysis from a longitudinal perspective: Some implications for theory building. Academy of Management Review, 24, 346–357. DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational ﬁelds. American Sociological Review, 48, 147–160. Duysters, G., Hagedoorn, J., & Lemmens, C. (2003). The effect of alliance block membership on innovative performance. Revue d’Economie Industrielle, 103, 59–70. Dyer, J. H., & Singh, H. (1998). The relational view: Cooperative strategy and sources of interorganizational competitive advantage. Academy of Management Review, 23, 660–679. Freeman, L. C. (1979). Centrality in social networks: Conceptual clariﬁcation. Social Networks, 1, 215–239. Garcia-Pont, C., & Nohria, N. (2002). Local versus global mimetism: Dynamics of alliance formation in the automobile industry. Strategic Management Journal, 23, 307–321. Gargiulo, M., & Benassi, M. (2000). Trapped in your own net? Network cohesion, structural holes, and the adaptation of social capital. Organization Science, 11, 183–196. Gimeno, J. (2004). Competition within and between networks: The contingent effect of competitive embeddedness on alliance formation. Academy of Management Journal, 47, 820–842. Gimeno, J., & Woo, C. Y. (1996). Economic multiplexity: The structural embeddedness of cooperation in multiple relations of interdependence. In: J. A. C. Baum & J. E. Dutton (Eds), The embeddedness of strategy: Advances in strategic management (Vol. 13, pp. 323–361). Greenwich, CT: JAI Press. Gnyawali, D. R., & Madhavan, R. (2001). Cooperative networks and competitive dynamics: A structural embeddedness perspective. Academy of Management Review, 26, 431–445. Gomes-Casseres, B. (1996). The alliance revolution: The new shape of business rivalry. Cambridge, MA: Harvard University Press. Gomes-Casseres, B., Hagedoorn, J., & Jaffe, A. B. (2006). Do alliances promote knowledge ﬂows? Journal of Financial Economics, 80, 5–33. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Granovetter, M. (1992). Problems of explanation in economic sociology. In: N. Nohria & R. G. Eccles (Eds), Networks and organizations (pp. 25–56). Boston, MA: Harvard Business School Press. Granovetter, M. (1994). Business Groups. In: N. J. Smelser & R. Swedberg (Eds), The handbook of economic sociology (pp. 453–475). Princeton, NJ: Princeton University Press. Gulati, R. (1995a). Social structure and alliance formation: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R. (1995b). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38, 85–112. Gulati, R. (1998). Alliances and networks. Strategic Management Journal, 19, 293–317. Gulati, R. (1999). Network location and learning: The inﬂuence of network resources and ﬁrm capabilities on alliance formation. Strategic Management Journal, 20, 397–420. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Gulati, R., Nohria, N., & Zaheer, A. (2000). Strategic networks. Strategic Management Journal, 21, 203–215.

528

JOHN HAGEDOORN AND HANS T. W. FRANKORT

Gulati, R., & Westphal, J. D. (1999). Cooperative or controlling? The effects of CEO-board relations and the content of interlocks on the formation of joint ventures. Administrative Science Quarterly, 44, 473–506. Hagedoorn, J. (1993). Understanding the rationale of strategic technology partnering: Interorganizational modes of cooperation and sectoral differences. Strategic Management Journal, 14, 371–385. Hagedoorn, J. (2002). Inter-ﬁrm R&D partnerships: An overview of major trends and patterns since 1960. Research Policy, 31, 477–492. Hagedoorn, J. (2006). Understanding the cross-level embeddedness of interﬁrm partnership formation. Academy of Management Review, 31, 670–680. Hagedoorn, J., Carayannis, E., & Alexander, J. (2001). Strange bedfellows in the personal computer industry: Technology alliances between IBM and Apple. Research Policy, 30, 837–849. Hagedoorn, J., Letterie, W., & Palm, F. (2007). On the information value of (un)embedded network ties. METEOR research memorandum 2007-004, Maastricht University, The Netherlands. Harrigan, K. R. (1985). Strategies for joint ventures. Lexington, MA: Lexington Books. Hite, J. M. (2003). Patterns of multidimensionality among embedded network ties: A typology of relational embeddedness in emerging entrepreneurial ﬁrms. Strategic Organization, 1, 9–49. Hite, J. M. (2008). The evolution of strategic dyadic network ties: Strategically navigating bounded agency within multi-dimensional and dynamic dyadic relationships. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 133–170). Bingley, UK: JAI/Emerald Group. Inkpen, A. C., & Ross, J. (2001). Why do some strategic alliances persist beyond their useful life? California Management Review, 44, 132–148. Kenis, P., & Knoke, D. (2002). How organizational ﬁeld networks shape interorganizational tie-formation rates. Academy of Management Review, 27, 275–293. Kim, T.-Y., Oh, H., & Swaminathan, A. (2006). Framing interorganizational network change: A network inertia perspective. Academy of Management Review, 31, 704–720. Lam, A. (1997). Embedded ﬁrms, embedded knowledge: Problems of collaboration and knowledge transfer in global cooperative ventures. Organization Studies, 18, 973–996. Levine, S. S., & Kurzban, R. (2006). Explaining clustering in social networks: Towards an evolutionary theory of cascading beneﬁts. Managerial and Decision Economics, 27, 173–187. Madhavan, R., Caner, T., Prescott, J., & Koka, B. (2008). Bringing the ﬁrm back in: Networking as antecedent to network structure. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 457–501). Bingley, UK: JAI/Emerald Group. Mowery, D. C., Oxley, J. E., & Silverman, B. S. (1996). Strategic alliances and interﬁrm knowledge transfer. Strategic Management Journal, 17, 77–91. Nohria, N., & Garcia-Pont, C. (1991). Global strategic linkages and industry structure. Strategic Management Journal, 12, 105–124. Nooteboom, B., Berger, H., & Noorderhaven, N. G. (1997). Effects of trust and governance n relational risk. Academy of Management Journal, 40, 308–338. Oster, S. M. (1999). Modern competitive analysis (3rd ed). New York: Oxford University Press. Park, S. H., & Ungson, G. R. (1997). The effect of national culture, organizational complementarity and economic motivation on joint venture dissolution. Academy of Management Journal, 40, 279–307.

Gloomy Side of Embeddedness

529

Pfeffer, J., & Nowak, P. (1976). Joint ventures and interorganizational interdependence. Administrative Science Quarterly, 21, 398–418. Portes, A., & Sensenbrenner, J. (1993). Embeddedness and immigration: Notes on the social determinants of economic action. American Journal of Sociology, 98, 1320–1350. Powell, W. W., & Grodal, S. (2005). Networks of innovators. In: J. Fagerberg, D. C. Mowery & R. R. Nelson (Eds), The Oxford handbook of innovation (pp. 56–85). Oxford, UK: Oxford University Press. Powell, W. W., Koput, K., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41, 116–145. Rosenkopf, L., Metiu, A., & George, V. P. (2001). From the bottom up? Technical committee activity and alliance formation. Administrative Science Quarterly, 46, 748–772. Saxton, T. (1997). The effects of partner and relationship characteristics on alliance outcomes. Academy of Management Journal, 40, 443–461. Schilling, M. A., & Phelps, C. C. (2007). Interﬁrm collaboration networks: The impact of largescale network structure on ﬁrm innovation. Management Science, 53, 1113–1126. Simsek, Z., Lubatkin, M. H., & Floyd, S. W. (2003). Interﬁrm networks and entrepreneurial behaviour: A structural embeddedness perspective. Journal of Management, 29, 427–442. Smelser, N. J., & Swedberg, R. (Eds). (1994). The handbook of economic sociology. Princeton, NJ: Princeton University Press. Soda, G., Zaheer, A., & Carlone, A. (2008). Imitative behavior: Network antecedents and performance consequences. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 531–560). Bingley, UK: JAI/Emerald Group. Takahashi, N. (2000). The emergence of generalized exchange. American Journal of Sociology, 105, 1105–1134. Thomas, H., & Carroll, C. (1994). Theoretical and empirical links between strategic groups, cognitive communities, and networks of interacting ﬁrms. In: H. Daems & H. Thomas (Eds), Strategic groups, strategic moves and performance (pp. 7–29). Oxford, UK: Pergamon. Uzzi, B. (1996a). Coase encounters of the sociological kind: Organizational ﬁelds as markets. In: J. A. C. Baum & J. E. Dutton (Eds), The embeddedness of strategy: Advances in strategic management (Vol. 13, pp. 419–430). Greenwich, CT: JAI Press. Uzzi, B. (1996b). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Vanhaverbeke, W., & Noorderhaven, N. G. (2001). Competition between alliance blocks: The case of the RISC microprocessor technology. Organization Studies, 22, 1–30. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8, 109–125. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge, UK: Cambridge University Press. Watts, D. J. (1999). Networks, dynamics, and the small-world phenomenon. American Journal of Sociology, 105, 493–527. Yu, C.-M., & Tang, M.-J. (1992). International joint ventures: Theoretical considerations. Managerial and Decision Economics, 13, 331–342. Zollo, M., Reuer, J. J., & Singh, H. (2002). Interorganizational routines and performance in strategic alliances. Organization Science, 13, 701–713.

530

JOHN HAGEDOORN AND HANS T. W. FRANKORT

APPENDIX. NAJOJO To visualize the ego networks of IBM, we make use of our own network visualization software tool Najojo. This tool is capable of visualizing large, dense networks involving more than 500 ﬁrms. There are two separate input (text) ﬁles underlying the generation of networks in Najojo: one holding the MDS coordinates for each of the individual ﬁrms participating in the network and a different one holding all unique ﬁrm pairs and their numbers of partnerships. Based on the ﬁrst input ﬁle, Najojo determines whether it visualizes the particular network in landscape or portrait orientation. As a second step in the visualization process, Najojo divides the landscape in X points and it then maps the ﬁrms’ coordinates held by the ﬁrst input ﬁle onto those points, visualizing them as dots. While creating this ‘scatter’ plot, the program makes sure that the relations among dots remain constant and that dots belonging to different ﬁrms do not overlap. Next, the program places ﬁrm labels with the dots in such a way that they do not overlap with other labels or dots. Najojo variably determines the font size of ﬁrm labels depending on network density and the number of ﬁrms participating in the network. Based on the second input ﬁle, Najojo then visualizes the total number of partnerships entered into by all unique ﬁrm pairs making up the network. The tool ﬁrst identiﬁes both research partners, i.e. the beginning and ending dots, and subsequently draws polybezier lines between these dots, making sure that these lines do not cross dots belonging to ﬁrms that are not part of the partnership. The type and thickness of lines indicate the number of partnerships between ﬁrms.

IMITATIVE BEHAVIOR: NETWORK ANTECEDENTS AND PERFORMANCE CONSEQUENCES Giuseppe Soda, Akbar Zaheer and Alessandra Carlone ABSTRACT Organizational networks are generally considered major antecedents of mutual inﬂuence in adopting similar practices, typically via a structure of dense ties, or closure. We propose that under conditions of competitive interdependence, closure may be associated with links established to access resources and knowledge and become a possible source of differentiation rather than imitation. We test these and other antecedents of imitative behavior and performance in the Italian TV industry with 12 years of data on 501 productions. We ﬁnd that network closure is associated with lower imitation, centrality, but not status, leads to imitation, and that imitation lowers performance.

Similarities among ﬁrms or organizational actors, including individuals and teams, and the antecedents of such imitative behaviors, have received a great deal of attention in the organization and strategic management literature (Haunschild, 1993; Rumelt, Schendel, & Teece, 1994), in innovation Network Strategy Advances in Strategic Management, Volume 25, 531–560 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25015-1

531

532

GIUSEPPE SODA ET AL.

research (e.g. Rogers, 1965), and in sociology (DiMaggio & Powell, 1983). Several studies on imitation and innovation examine the mechanisms by which choices are inﬂuenced in the process of the generation and implementation of new ideas in a number of contexts such as banks, telephone equipment, airlines and hospitals (see Lieberman & Asaba, 2006, for a review). More recently, scholars have emphasized the processes through which individuals and organizations may be inﬂuenced by others in the adoption of common practices, as well as the conditions under which imitative behavior occurs and its consequences for organizations (e.g. Haunschild, 1993; Haunschild & Miner, 1997). Speciﬁcally, scholars have examined the processes and concomitants of imitative behavior and its consequences in connection with ﬁrm strategies (Fligstein, 1991; Haveman, 1993; Haunschild & Miner, 1997), organizational structure (Fligstein, 1985; Burns & Wholey, 1993) and other organizational processes (Sutton & Dobbin, 1996). Research has also shown that one of the most powerful sources of inﬂuence for imitative behavior and mimetic processes is an organization’s network of ties (e.g. Haunschild & Miner, 1997). For example, Galaskiewics and Wasserman (1989) argue that organizational actors are more likely to mimic organizations to which they are linked through interpersonal ties via boundary-spanning personnel. At a micro-level of analysis, research on behavioral conformity shows that individuals are heavily inﬂuenced by the actions and beliefs of others (Asch, 1956; Cialdini, Reno, & Kallgren, 1990; Moscovici, 1985; Sherif, 1936). More generally, how the economic actions of organizational and individual actors are inﬂuenced by social relations is one of the most classic questions in the social sciences (Granovetter, 1985). While at a general level there is wide acceptance of the idea that networks of relations inﬂuence imitative behavior, research also suggests that a speciﬁc kind of social structure – closure, deﬁned as a structure of dense interconnections between and among the actors in the network – is associated with exerting strong pressures on actor behavior (Portes & Sensenbrenner, 1993; Soda & Usai, 1999). As Coleman (1988) points out, dense, overlapping ties curb opportunistic behavior among actors through reputation effects and sanctions. In this vein, Baum and his colleagues (Baum, Rowley, & Shipilov, 2003; Baum, Rowley, Shipilov, & Chuang, 2005) note that dense ties ‘‘tend to stabilize inter-organizational networks’’ and such stabilization process may amplify imitation among actors. These arguments suggest that a network structure with high levels of closure generates social and institutional mechanisms – such as shared

Imitative Behavior

533

norms, trust, obligations and reciprocity – which constrain actors’ ability to differentiate themselves and thereby induce conformity. From an institutional perspective, DiMaggio and Powell (1983) adopt a similar argument, observing that ‘‘y highly structured organizational ﬁelds provide a context in which individual efforts to deal rationally with uncertainty and constraint often lead y to homogeneity in y culture and output’’ (p. 147). Following this argument, imitation should be seen as the outcome of conformist behaviors caused by the norms, rules and sanctions associated with high network closure (Donaldson, 1997). At the same time, research in a number of ﬁelds suggests that the quest for resources is one of the most important considerations in the formation of network links (Dyer & Singh, 1998; Gulati, 1998). Resources held by potential partners, such as human and ﬁnancial capital and technology, are critical determinants of link formation and partner selection (Soda, Perrone, & Usai, 2001). When organizations operate in a competitive interdependence setting, connections may be established for seeking synergistic complementarities among the diverse resource content available from the network. The different resources and knowledge that a focal organization accesses and combines by forming ties with other organizations then become a valuable source of the creation of a differentiated product. Put differently, by connecting with multiple other actors, who are also competitors, focal organizations are made aware of their respective product decisions, but rather than imitate them, are inﬂuenced to consciously distinguish themselves from them to heighten their chances of success. Thus, access to multiple other alters’ resources and knowledge helps focal organizations explore a wider set of opportunities and thereby generate differentiated products. Relatedly, using an argument from population ecology, if we consider the diffusion of product imitation as an increase in the number of players in the same competitive niche, differentiation will help these players reduce the density of the niche, enhancing their likelihood of success and survival (Baum & Singh, 1994). In the present chapter we develop this idea, which runs counter to received wisdom, that competitive forces, combined with a pattern of dense ties that provide network access to resources held by competing organizations, will make focal organizations differentiate themselves from, rather than imitate, one another. Moreover, in light of the importance of understanding the antecedents of imitative behavior, we also search for other network structural explanations for the imitation phenomenon, stemming from favored actor position due to centrality, as well as deriving from a network positional measure of status (Podolny, 1993).

534

GIUSEPPE SODA ET AL.

We also tackle two further issues in the organizational network literature; the ﬁrst is a general neglect of network dynamics and network evolution; and the second is that since structure, content and outcomes co-evolve, without an explicit consideration of the endogeneity inherent in such coevolution, results from prior studies are questionable (Mouw, 2006). In this chapter, we use lags to account for network dynamics. Further, by using a methodology that explicitly splits apart causes and outcomes by means of a Two-Stage Least Square Analysis, we are able to make much more robust assertions about the relationships between network structure and network outcomes (Shaver, 2005). Put differently, we tease out the endogeneity inherent in network structures and outcomes and thereby are able to predict network outcomes, speciﬁcally performance, with greater conﬁdence. A further critical question in the research on imitation concerns performance outcomes arising from imitative behaviors. While on the one hand, research in institutional theory argues that mimetic or imitative behaviors enhance the performance of ﬁrms by imparting them legitimacy (DiMaggio & Powell, 1983), research in strategic management speciﬁcally cautions against the perils of imitation in competitive situations, exhorting ﬁrms to carve out differentiated and distinct positions for themselves (Porter, 1996). While this latter question has been broadly addressed by Deephouse (1999), placing it in the context of imitative behaviors arising from network structures, which we do in this chapter, represents a signiﬁcant advance over the extant literature. Thus, our contribution in this chapter is twofold; ﬁrst, we use social network analysis to model theoretically and assess empirically imitative processes; more precisely, we address the issue of how network structures, in which competitive teams or ﬁrms are embedded, inﬂuence imitative behavior; and we propose a new approach to the investigation of imitative behavior based on product imitation rather than the imitation of practices at the macro or organization strategy level. As well, by presenting a theory linking network closure to imitation that runs counter to conventional wisdom, and testing competing theories about the imitative outcomes of social structure, we advance understanding on the social structural antecedents of economic action. In addition, we investigate two other major structural explanations for imitative behavior; centrality and status. Second, by including performance, we contribute to a better understanding of the relationship between imitative behavior and performance outcomes. We tease apart the potential endogeneity among network structures, imitative behaviors and performance through the adoption of an appropriate estimation technique, a Two-Stage Least Squares model

535

Imitative Behavior

(Shaver, 2005). By mapping the causal chain of the relationships between network structures, imitation and performance, we advance research and understanding with regard to the mechanisms underlying the role and value of network closure and imitative behavior. We investigate these issues in the longitudinal context of the performance of TV production teams in Italy. In contrast to work that presumes a link exists between types of network structures and alter content, such as that linking content diversity to structural holes (Burt, 1992), we are able to use the content produced by TV production teams to empirically test the link between content and structure.We obtained data on all the 501 TV projects that were produced over a 12-year period (1988–1999) for Italian TV and constructed networks of the 4,793 individuals involved in the projects. In the following sections we ﬁrst discuss our research context, and thereafter elaborate on our theory, develop hypotheses, present our methodology and results, and discuss the implications of our work for future research.

RESEARCH CONTEXT Our research context is the Italian TV production industry, which includes TV movies and serials, sitcoms, soaps and made-for-TV specials. The industry is composed by of a large group of specialists including musicians, actors, producers, screenwriters and ﬁnanciers. These specialists combine their expertise and manage all the steps in the value chain, from idea generation to screen-writing, to pre-production, shooting, editing and postproduction. Specialists work in temporary teams (TV production teams), for several weeks or months, to create the TV product. The TV production industry can be viewed as a network of interconnected nodes linked through shared memberships in the production teams. By employing a specialist from another team, a focal team acquires and shares the critical resources of this industry, which are knowledge and experience. Thus, as in other types of networks, like the well-investigated setting of interlocking directorates, the network deriving from co-membership across teams allows them to access and exchange key resources. The resources described above are critical due to the ‘‘cultural-prototype’’ nature of TV productions. Rather than merely ‘‘applying their skills,’’ individual specialists that form a team bring to bear on the project their experiences, memories, ideas and in general their creative inputs, to cocreate an original expression of joint teamwork in a non-additive, nonsequential but reciprocal process. Despite sharing specialists, no issue of

536

GIUSEPPE SODA ET AL.

‘‘coordination’’ across teams arises since teams actually compete with one other for the TV audience. In fact, the relationships across teams may be described as competitive interdependence rather than symbiotic interdependence. Resources and ideas brought to a team by the concurrent membership of the individual specialist in other teams helps in the creative process. As a consequence, TV production content, particularly in the Italian TV industry, is the result of a team process wherein a number of specialists, including but not limited to just the screenwriters, are involved in creating the product, including actors, the director and even the ﬁnanciers. In this way, the Italian TV context is quite different from Hollywood movies where screenwriters play a more signiﬁcant role. Broadly, a TV production is characterized by high levels of uncertainty and complexity. As well, a need for coordination within teams exists in directing the combination and recombination of both team-generated resources as well as those acquired from other teams (Bielby & Bielby, 1999). Starkey, Barnatt, and Tempest (2000) posit that ‘‘the long term survival of ﬁrms in cultural industries depends heavily on replenishing their creative resource.’’ The role of group creativity is key. The TV production product is distributed by another industry, the TV broadcasting industry, which in Italy is dominated by two major players: RAI and Mediaset, each owning three non-cable, non-satellite channels. The two major players account for roughly 93% of the entire TV audience, including cable, satellite and local channels. The oligopolistic market structure of the TV broadcasting industry in Italy is similar to that of the industry in other countries.

THEORY AND HYPOTHESES Network Closure and Imitative Behavior Since, as we mentioned earlier, the relationship between closure and imitation among interconnected actors can be viewed as presenting two opposing arguments, in this section we develop theory for the two competing hypotheses that derive from this reasoning. We begin by making the case for the positive link between closure and imitation. Coleman (1988) argues that dense social structures, or closure, is a form of social capital because it generates group norms and obligations that facilitate the extension of trust as a form of social credit, thereby expanding a system’s action capacity (Marsden, 2005). However, strong mutual

Imitative Behavior

537

connections between members of a network may create effects of groupthink, with detrimental effects on creative ideas (Janis, 1972). Furthermore, higher social pressures from high closure produce social norms through conformity, constraining and restricting individual creativity and expression, as Portes and Sensenbrenner (1993) suggest. Some empirical results support Coleman’s thesis on network closure. Ahuja (2000a) ﬁnds that in a context of innovation, inter-organizational networks with high closure likely provide beneﬁts of increased trust, lowered opportunism and superior routines that overcome the disadvantages of lower information variety and heterogeneity to enhance innovation output. Moreover, an increased degree of adoption of alters’ ideas and content may reﬂect social learning efﬁciencies rather that just obeying social norms (Westphal, Gulati, & Shortell, 1997). Thus, conformity is not just the outcome of potential social sanctions, but can be seen as a strategy through which connected actors increase efﬁciency by exploiting common languages and shared routines. Several recent studies suggest a link between diversity (or the lack of conformity) and density that supports this reasoning. For example, Reagans and Zuckerman (2001) ﬁnd a correlation between diversity and density; higher diversity is associated with lower density, suggesting that a pattern of dense network ties, characteristic of high closure, is related to greater conformity. Reagans, Zuckerman, and McEvily (2004) ﬁnd a link between demographic diversity and ‘‘range’’ (a variant of structural holes, associated with low closure). Similarly, Balkundi et al. (2007) ﬁnd a positive relationship between diversity and structural holes in the network. All these studies ﬁnd results consistent with the argument that higher density is associated with lower diversity, implying greater conformity or imitation. The theoretical reasoning supporting such relationships stems from the Coleman (1988) thesis of closure as imposing norms of conformity. In either case, whether through social pressures for conformity or via efﬁciency through shared routines, the expected outcome among densely connected actors is higher imitation and conformity. In both explanations, we consider conformity behavior as the underlying mechanism that links network closure to imitative outcomes. Accordingly, we hypothesize that: H1a. The higher the closure in the focal TV production team’s ego network, the higher the degree of imitation among the focal team and its alters. We believe that the previous arguments are built on an assumption implicit in the literature that needs to be both theoretically challenged and

538

GIUSEPPE SODA ET AL.

empirically addressed. The idea that a network structure reﬂects necessarily certain actor characteristics is a presumption which should be better investigated (Burt, Horgarth, & Michaud, 2000). Put differently, we do not believe that a dense network must necessarily reﬂect high homogeneity and similarity among the linked actors. As some authors point out (e.g. Suitor, Wellman, & Morgan, 1997), a network may increase member homogeneity over time when its evolution ampliﬁes the stability and eventually the multiplicity of relations. However, with some notable exceptions (Reagans et al., 2004; Soda, Usai, & Zaheer, 2004), the bulk of research on networks does not address this issue longitudinally. Thus, in a cross-sectional research setting the presumption that network closure should be associated with high similarity has to be empirically tested than being simply assumed. Even theoretically, actors may establish multiple and interconnected links to search for diversity and novelty. Thus, density could be considered as the infrastructure surrounding and enabling the successful integration of resources across production teams. Resources held by potential alters, such as knowledge and ideas, are therefore critical as determinants of link formation. An extensive ﬁeld of research suggests that the quest for resources is one of the most important dimensions in the inter-organizational relationships or alliance formation process (Pfeffer & Salancik, 1978; Nohria & GarciaPont, 1991; Shan & Hamilton, 1991. Thus, network links are conduits through which organizations access critical resources. The need for resources is consequently an inducement to establish linkages (Ahuja, 2000). Since resources may span organization boundaries and are often embedded in inter-organizational relationships (Dyer & Singh, 1998), multiple complementary connections to access the resources may contribute to creating dense networks. While the resource access argument has been frequently employed in the literature on alliance and network tie formation (Gulati & Gargiulo, 1999; Soda, Usai, & Perrone, 2001), rarely have its consequences on behavioral conformity been explored. Particularly in competitive contexts like our research setting, where it is important to capture new ideas from alters, connections may be established for creating resource recombinations and complementarities through access to diverse content rather than imitating alter behavior. It is important to note that complementarity in this setting is more properly seen as access to the resources held by network alters, rather than the traditional deﬁnition of complementarity as a synergistic combination of resources (Milgrom & Roberts, 1995). The combination of different resources and knowledge that focal teams access by forming shared membership links with other teams is a

Imitative Behavior

539

source of the creation of a differentiated product, which the strategy literature suggests, is associated with ﬁrm success (Porter, 1980). In particular, while a group of production team members working together on multiple productions may recognize the need to differentiate their products (or productions), we suggest that rival teams also use network connections to consciously ‘‘match’’ their productions to avoid niche overlaps (e.g. one produces a police drama, another a medical drama and a third a sitcom). In this vein, Gnyawali and Madhavan (2001) view the network as a ‘‘search and monitoring mechanism for each other’s strategies and actions, increasing, in the process, the cognitive salience of some competitors relative to others’’ (p. 432). In our research context the prototypical nature of the product increases the risks and the costs associated with failure because teams only get one shot at creating the product. Since, as we mentioned earlier, competition across teams is strong, focal teams will differentiate themselves from alters in order increase their chances of success while the dense interconnections with alters will enable them to gain a good understanding of what other teams are doing. The motive for differentiation here is akin to the early and well-established idea from strategy not to attack competition head on (von Clausewitz, 1997, [1832]). In this vein, research shows that niche overlap density, tantamount to many teams producing similar products, is positively associated with mortality rates (Baum & Singh, 1994). Moreover, in our industry large, generalist organizations are not dominant, suggesting that with differentiation, the viability of small, specialized, creative organizations increases. In sum, if focal teams imitate, the downside from competition and niche density may outweigh the potential beneﬁts of imitation, such as efﬁciency. Thus, access to resource and knowledge through network relations will reduce teams’ propensity to conform and increase the creative combination of teams’ own knowledge with that accessed from network alters. This creative combination of knowledge and experience, combined with an awareness of alters’ products and the negative consequences of imitation, will reduce the overall similarity among highly connected teams. In sum, the search for novelty and new ideas will create connections among content diverse teams and the diversity and novelty will be embodied in products which will result in low imitation among alters. Accordingly, we hypothesize that: H1b. The higher the closure in the focal TV production team’s ego network, the lower the degree of imitation among the focal team and its alters.

540

GIUSEPPE SODA ET AL.

Actor Centrality and Imitative Behavior Different network positions correspond to different opportunities for an actor to access knowledge and resources (Brass, 1992; Tsai, 2001). In fact, like formal authority, network centrality implies a high position in status position (Ibarra, 1993). Actors’ positions within a social network inﬂuence their degree of access to strategic resources and resulting potential beneﬁts. A crucial construct indicating how an actor is positioned relatively to others in a network is centrality (Scott, 2000). By occupying a central position in a social network, an actor is likely to more easily and directly access desired strategic resources that fuel its innovation. In this vein, Madhavan, Caner, Prescott, and Koka (2008) suggest that central ﬁrms have more social capital in terms of information volume. Research ﬁndings at individual, group, organizational and interorganizational levels of analysis strongly indicate that network centrality is a signiﬁcant source of power and inﬂuence (Brass, 1992). Organizations that have links with the greatest number of others tend to be larger and more prestigious (Lieberman & Asaba, 2006). As Gulati and Gargiulo (1999) point out, ‘‘The more central an organization’s network position, the more likely it is to have better information’’ (p. 1,448). Haunschild (1993) has shown that centrality in terms of a higher number of board interlocks is associated with the focal ﬁrm carrying out a larger number of acquisitions, although it is unclear the extent to which such behavior was due to imitation or access to private information. Following this argument, Hansen (1999) demonstrates that a central organizational actor has more possibilities to access unique knowledge, including a better understanding of where such knowledge is located and how to obtain it. Thus, actors in central network positions have greater access to, and potential control over, relevant resources, such as information in a communication network (Brass, Galaskiewicz, Greve, & Tsai, 2004). Moreover, as Balkundi and Harrison (2006) ﬁnd, central actors in a social network tend to be better performers because of their more efﬁcient access to resources (Tsai, 2001). In network terms, central actors are more prominent and, as a consequence, are perceived by alters as owners of critical competences and resources. This strategic positioning will provide central actors with a strong inﬂuence on network alters since, under conditions of uncertainty, as in our industry context, central and more prominent actors will be perceived as being more legitimate. From an institutional theory perspective, such actors will induce greater mimetic isomorphism, or imitation by their alters

541

Imitative Behavior

in the ego network (DiMaggio & Powell, 1983). Accordingly we hypothesize that: H2. The higher the network centrality of a focal TV production team the higher the degree of imitation among the focal team and its alters.

Status and Imitative Behavior The underlying presumption for the centrality argument is that a central actor seen as more prominent and legitimate will be able to inﬂuence alters’ behavior toward imitation (Ibarra, 1993; DiMaggio & Powell, 1983). Nevertheless, another explanation may be that when actors are perceived as being of higher status than others they also inﬂuence alters’ imitation (Powell, Koput, & Smith-Doerr, 1996). While centrality refers to an actor’s position in a network, status is a concept related to the past performance of its alters. Speciﬁcally, status is deﬁned as a position within a social structure that confers rights, prestige, or honor upon an individual according to various criteria (Parsons, 1970). From a sociological perspective, status relates to the position of an actor and is derived through network links with successful teams from the past (Benjamin & Podolny, 1999). Thus, connections with successful actors may be an important template of action (Marquis, 2003). Under conditions of uncertainty and risk, distinct evaluations by external audiences – like network ties with successful actors from the past in our research context – sustain status hierarchies and lead to the emergence of a status-based homophily structure (Podolny, 1993). In our research context, we believe that TV production teams connected with high status alters will imitate them, and consequently the degree of homogeneity among the focal team and its alters will increase. Accordingly, we hypothesize that: H3. The higher the status of a focal TV production team the higher the degree of imitation among the focal team and its alters.

Performance Consequences of Imitation Research on conformity behaviors at the individual level (Cialdini & Goldstein, 2004), shows that the tendency to imitate is sometimes so swift and mindless that it is almost automatic (Griskevicius, Goldstein,

542

GIUSEPPE SODA ET AL.

Mortensen, Cialdini, & Kenrich, 2006). Different explanations have been suggested in explaining conformity arising from imitative behavior. First, conformity is assumed as an efﬁcient and adaptive strategy for actions under conditions of uncertainty. Following successful others may lead to better and generally more accurate decisions when actors face uncertainty (Cialdini, 2001; Crutchﬁeld, 1955). This conformity is known as ‘‘accuracy-based conformity’’ (Deutsch & Gerard, 1955), and it persists because in many cases it is considered an efﬁcient form of behaving (Gigerenzer & Todd, 1999). A second line of thought points to the role of social pressures and normative inﬂuence (Deutsch & Gerard, 1955) that serves and facilitate the goal of beneﬁcial afﬁliation (Baumeister & Leary, 1995; Insko et al., 1985; Martin & Hewstone, 2003). Although conformity can confer numerous beneﬁts, nonconformity may also be advantageous (Argyle, 1957; Hollander, 1958). Nonconformity includes two types of behavior: independence, or resisting inﬂuence; and anticonformity, or rebelling against inﬂuence (Nail, MacDonald, & Levy, 2000; Willis, 1963). Both types of nonconformity tend to be effective in differentiating people from others, which can satisfy a need for individuation or uniqueness (Maslach, Stapp, & Santee, 1985; Snyder & Fromkin, 1980). For example, when a person’s uniqueness is threatened by an encounter with a highly similar individual, such a situation increases the tendency to nonconform (Duval, 1972; Snyder & Fromkin, 1980). If at the individual level of analysis both conformity and nonconformity can be beneﬁcial, this duality raises the important question of the impact on performance at macro level of analysis and in contexts of inter-organizational relationships. Chen and Hambrick (1995) speculate on the contradictory prescriptions related to conformity of strategic management and institutional theory. For the latter, greater strategic and organizational similarity increases performance because the organizational ﬁeld institutionalizes and legitimates a range of normal strategies through an iterative isomorphic process (Scott, 1995). Conversely, strategic management theory proposes that strategic conformity increases both risks and opportunities of competitive advantage (Abrahamson & Hegeman, 1994; Baum et al., 2005). Deephouse (1999) describes the tensions between the need for an organization ‘‘to be the same or to be different’’ (p. 147), by investigating the consequences on performance of ﬁrm-level strategic similarity. He proposes a strategic balance between the two conﬂicting perspectives by hypothesizing an inverted U-shaped relationship and arguing that: ‘‘y moderately differentiated ﬁrms have higher performance that either highly conforming or highly differentiated ﬁrms’’ (p. 148).

543

Imitative Behavior

In a complex, dynamic and creativity-based environment, decisionmakers have to deal with large amounts of ambiguous information, and managers use cognitive simpliﬁcation processes to overcome uncertainty and conquer the information processing nightmare (Huff, 1982; Reger & Huff, 1993; Schwenk, 1984). They simplify by stereotyping – creating cognitive categories or groups of ﬁrms with relevant similarities, and identifying their actions with other belonging to the same group category or group. We believe that legitimacy-driven forces may lead ﬁrms to inappropriate decisions (Westphal et al., 1997). This argument implies the existence of a ‘‘normative rationality,’’ based on social justiﬁcation, in contrast to ‘‘economic rationality,’’ based on proﬁtability (Oliver, 1997). Further, as we have argued earlier, greater imitation is analogous to greater niche density, which as the population ecology literature suggests, is associated with negative outcomes (Baum & Singh, 1994). Overall, in a context of competition and high uncertainty we argue that imitation among alters will reduce performance by inhibiting innovation and creativity. Accordingly we hypothesize that: H4. The higher the degree of imitation within the ego network of a focal TV production team, the lower its performance.

METHODOLOGY We test the hypotheses on data from TV productions in Italy during 1995–1999. We included in our dataset all types of TV productions, such as TV movies, serials and so on, broadcast by the six national TV channels (covering 95% of the Italian TV audience) during this time. Speciﬁcally, we collected three kinds of data; ﬁrst, from the annual report of TV movies and serials in Italy, published by the state-owned broadcaster RAI, we obtained data over time on production teams and their networks of ties; second, detailed synopses of each production were assembled from the publication’s appendix; third, from Auditel, an independent agency that tracks viewership, we retrieved audience data for the TV productions in our dataset. In the rare event that a production was broadcast over multiple years, we only included the ﬁrst broadcast year. Overall, our dataset encompasses 249 television productions created and broadcast over the period. Moreover, we also examined production content by means of a content analysis of the script. Because of the need of controlling for path dependency effects on our endogenous variable of

544

GIUSEPPE SODA ET AL.

imitative behavior, we also measured this variable for past relations by projecting current teams over the previous 252 TV production teams that operated between 1988 and 1994.1 The TV production or project team is our level of analysis. We explain how we analyzed TV production content in the following section.

Analysis of TV Production Content Content analysis has been widely employed as a replicable and systematic means of collapsing a large amount of text into content categories based on a set of explicit rules of coding (Berelson, 1952; Krippendorff, 1980). For our purposes, identity can be effectively assessed through content analysis since authors express their artistic identity and the messages they wish to communicate through their creative output. The TV production script thus becomes the key document for understanding a team’s meaning and identity in cultural contexts such as ours. Based on industry experts’ recommendations, we used a procedure whereby we established coding categories prior to the analysis. The synopsis of the script (on average three pages) created by the team for the national TV archive was used as a basis for the analysis. We did the analysis in the following stages. One, we created a set of 19 categories and variables after reviewing the script synopses ourselves. Two, we identiﬁed and invited six industry experts to validate our list of 19 categories and variables, which we reduced to 12 based on their input. Three, we used two members of our research team to independently code content using a 1/0 schema (agreement equals 1, disagreement equals 0). Four, we used Cohen’s kappa (Cohen, 1960) to test for the interrater reliability across our variables. The kappa coefﬁcient measures the percent of agreement between the raters and is computed as: k¼

PA Pc 1 Pc

where PA is the proportion of items on which the raters agree and Pc proportion of items for which agreement is expected by chance. In case of disagreement, a third researcher was called in to repeat the steps and the majority value was chosen. Overall, the kappa coefﬁcient was (0.80) (Table 1) which is favorable given that the literature indicates that a value of 0.61 is generally acceptable as representing good agreement (Kvalseth, 1989).

545

Imitative Behavior

Table 1. Variable

Coding of Contents.

Type of Variable

Theme

Categorical variable

Relations

Dummy

Values

Dummy

Pain

Dummy variable

Power and success Professions of characters Positive or negative character Ending

Dummy variable Categorical variable

Setting

Categorical variable

Time period Schema

Categorical variable Categorical variable

Ordinal variable Categorical variable

Contents Detective story (whodunit); dramatic; life story; friendship; love; family; Bible theme; religion; sport; fantasy; power/money/career; others 1 for love, friendship, kindship, afﬁliation, afﬁnity, consanguinity, and liaison; 0 otherwise 1 for Human justice (i.e. story of a crime prosecutors), Religion (i.e. stories about the life of Saints and Blessed), freedom and independency, social ﬁghts against evil (i.e. citizen or consumer against powerful organizations for environmental protection) or against social prejudice and discrimination; 0 otherwise 1 for dramatic, telling stories of diseases, suffering, conﬂicts; 0 otherwise 1 for power, money, career, elites; 0 otherwise Dominant professions in which the production in set Weighted number of protagonists, antagonists and secondary protagonists The nature of the epilogue: happy, ambiguous, and unhappy Context in which the story has been located: Itlay, abroad Time period the story is set The conﬂicts schema in the sentimental relationship

Two-Stage Least Squares Analysis Our theory indicates a mediation effect, joining ego network imitation antecedents with performance mediated by ego network imitation. Moreover, there is the issue of possible endogeneity between performance and the imitation variable because a third variable may be inﬂuencing both imitation and performance. As a consequence, we ﬁrst estimate the imitation variable as an endogenous variable and thereafter use it to predict performance, using a Two-Stage Least Squares (2SLS) Analysis to control for the effects of correlation of errors across equations. Shaver (2005) suggests that the 2SLS methodology is appropriate in mediated models

546

GIUSEPPE SODA ET AL.

because of its power to deal with the potential correlation among errors, and produces consistent and unbiased coefﬁcients. We begin by estimating 2SLS with random effects using the Hausman (1978) test to check whether unobserved heterogeneity is an issue. Next, we use the robust variance estimator providing accurate assessments of parameter estimates even with misspeciﬁed models and heteroskedasticity which in Stata is called the Huber/White/sandwich estimator of variance (Huber, 1967; White, 1980). Further, we checked the VIF (Variance Inﬂation Factor) estimates to ascertain the presence of multicollinearity but found no evidence of it.

Two-Stage Least Square Analysis (2SLS): First Stage Variables Endogenous Variable: Ego Network Imitation We computed ego imitation for TV production teams by measuring the content similarity among all ego network members, including the ego. As a result, we obtained a measure of content similarity among all members of the ego network. For example, in Fig. 1 for team A, we extracted its alters B, C and D (teams linked to the ego A by sharing some specialists), and then measured the degree of homogeneity among all members of the ego network (B, C, D and A) by applying the procedure described below. To measure imitation, we used the 12 content variables mentioned above. We assessed the content of the 249 TV productions by transforming the two-mode matrix of Production by Content (with dimensionality 249 * 12) into a onemode Production by Production matrix, where xij is the degree of imitation between productions i and j. Thereafter, we used the similarity procedure of

B

C

D

Ego network

A

Fig. 1.

Alters extracted from ego A

Representation of the Ego Network.

Imitative Behavior

547

UCINET VI for this transformation and adopted the measure of similarity as ‘‘matches’’ that computes the proportion of cases in which xi ¼ yi for all i. (Borgatti, Everett, & Freeman, 2002). We measured the extent of homogeneity in the ego network by calculating the average content similarity among all ego network members. Instrumental Variables Team network centrality. Network centrality measures the degree centrality (Freeman, 1979) of each TV production team in the global network of the relations among teams based on sharing specialists. The degree centrality measure is a well-accepted measure and it is widely used (Madhavan et al., 2008; Ahuja, 2000). Degree centrality assesses the prominence of a team and indicates the extent to which its specialists are also involved in other TV production teams. Ego network closure. Ego network closure is the density of current ties among the members of the ego network for a TV production team. We computed this measure by applying the density procedure to all TV production teams. Square matrices represent the networks of ties for each year in which xij is the number of specialists which teams i and j share. Team status. Focal TV production team status is the accumulation of past performance of past alter teams for seven years. Benjamin & Podolny’s (1999) conception of status as a connection with high-performing alters is the model for our measure. Since our teams are composed of several specialists, including those performing technical tasks, we decided to use only those specialists in critical roles. The specialists we considered were the directors, screenplay writers, original authors, producers and actors in major starring roles. We added up the performance of past alter teams to calculate the status of focal teams. More recent successes were weighted by a decay function based on the age of ties, since the effect of more recent successes is stronger than that of older ones.

Two-Stage Least Square Analysis (2SLS): Second Stage Variables Dependent Variable: Team Performance As mentioned earlier, we used audience data from Auditel to assess the performance of TV productions. In light of the highly skewed nature of the audience variable, we used its natural log as the measure of performance and our dependent variable.

548

GIUSEPPE SODA ET AL.

Controls In stage one of the 2SLS, we control for time effects by using a series of dummy variables corresponding to 1995, 1996, 1997 and 1998 (1999 is the omitted category). We also control for path dependency effects of past ego network imitation because scholars have pointed out how the past continues to inﬂuence the present perpetuating locally imprinted patterns over time (Marquis, 2003). The network of past relations might therefore be supposed to exert a strong effect on outcomes (Walsh & Ungson, 1991). We measure past ego network imitation applying the same procedure explained above for the computation of the ego network imitation but referred to the past. Using measures of past ties we also control for the age of the ties by weighting every tie by the inverse of its age. Speciﬁcally, for each team we computed a weighted average of their past relations. We also control in the ﬁrst stage for the size of the TV production team to take into account the effect of team size on network measures. In the second stage of our analysis we employ controls of several types. First, as several research ﬁndings highlight (Granovetter, 1973; Burt, 1992; Soda et al., 2004; Perry-Smith, 2006) open social structures create brokerage advantages and new opportunities for creativity. Hence, we also consider the structural holes spanned by each focal team. The measure of current structural holes has been computed using the network of current ties to identify the alters of focal TV production teams. We use Burt’s (1992) measure of constraint to compute structural holes, deﬁned as the degree to which a particular node’s network is ‘‘directly or indirectly concentrated in a single’’ node. We transformed the value of constraint into structural holes by multiplying its value by –1 (which might be considered the ‘‘opposite’’ of constraint). We also used some industry-speciﬁc controls. All the six TV channels we considered (which cover 95 % of the total audience) do not have the same audience potential. Speciﬁcally, there are two major channels that have far higher viewership than the others. Accordingly, we use a dummy variable for major channel, coded 1 when the production was broadcast on either major channel and 0 otherwise. Moreover, we control for prime time with a dummy variable. Our ﬁnal set of variables controls for the effect on performance of different formats (such as TV movies, soaps and sitcoms) and the number of episodes. We control for such task characteristics by computing two additional variables: a TV movie dummy and the number of episodes. The TV movie dummy equals 1 when the production is a TV movie and 0 if it is a serial-like production (sitcoms, soaps and so on). The number of episodes control variable is based on the number of episodes actually broadcast.

549

Imitative Behavior

RESULTS Table 2 reports the descriptive statistics (means and standard deviations) and correlations among our variables. In the ﬁrst stage of 2SLS analysis (Table 3), two of the explanatory variables show consistent results with hypothesized effects for ego network imitation antecedents. Model 1 accounts for controls (dummies for years, age, team size and past ego network imitation). In Model 2 we introduced the measure of team network centrality (H2) and we found that it is robustly and positively related with the ego network imitation (b ¼ .004, p ¼ .000). Thus, due to their central position in the network, central actors are considered to be teams with more prominence and power and as a result they more strongly inﬂuence alters to be imitative. In Model 3 we introduced the measure of ego network closure (H1a, H1b). Results show that closure is negatively and signiﬁcantly associated to ego network imitation (b ¼ .167 p ¼ .000), supporting our counter hypothesis H1b rather than H1a. In Model 4 we add the measure of status (H3) and while centrality exerts a strong and positive inﬂuence on ego content similarity, status is not signiﬁcant (b ¼ 002, n.s.). Thus the past success of team members does not incent imitation by alters. In the second stage of 2SLS (Table 4), we ﬁnd that ego network imitation is detrimental for performance (b ¼ .116, p ¼ .077), thus supporting H4. This result suggests that when homogeneity among alters is also associated with conformity to their ideas and identity, the efﬁciency effect turns into a detrimental effect from mere imitation. In a context of high uncertainty and dense inter-organizational relationships, imitative behavior among alters reduces performance by inhibiting innovation and creativity. Moreover, the control for structural holes supports the extant literature as regards a positive effect on team performance (b ¼ .191, p ¼ .001). Fig. 2 reports the ﬁndings of both stages of our 2SLS analysis.

DISCUSSION In this chapter, we make a contribution to the literature by theoretically challenging and empirically testing the well-known idea that network closure is reﬂected in imitation and subsequent homogeneity among actors linked in a network. In addition, we also examined other network antecedents of imitative behavior – centrality and status. To disentangle the causal chain linking network structures with its performance

Mean

Standard Deviation

13. 14. 15. 16. 17. 18.

Status Past ego network imitation Ego network imitation Structural holes Team network centrality Team performance

Variables

6.47 8.349786 .4832571 .1172289 30.16867 8.373815

Mean

1

1 5.299695 .0609587 .041996 14.73187 .598522

0.0944 0.0032 0.0250 0.0620 0.2254 0.1393

12

0.2499 0.2248 0.2778 0.0475 0.2669 0.0798 0.0048 0.1032 0.1002 0.0417 0.0961 0.2470 0.4008 0.3666 0.1832 0.0307

2

4

0.3137 0.0656 0.0792 0.0521 0.0084 0.0617 0.2481 0.1016 0.0662 0.4421 0.1211 0.0528 0.1674 0.0155

0.0527 0.0055 0.3700 0.2772 0.0431

13

0.2822 0.3488 0.0251 0.0209 0.0032 0.0457 0.0694 0.0166 0.0311 0.0923 0.0725 0.1915 0.0897 0.2491 0.0577

3

0.1274 0.2788 0.2985 0.0180

14

0.1169 0.1958 0.0436 0.0180 0.0234 0.2162 0.1052 0.1504 0.1187 0.2697 0.2235 0.3318 0.0143

5

7

15

0.1549 0.1651 0.1978 0.0702 0.1324 0.2971 0.1585 0.0658 0.7249 0.2619 0.1027

0.0690 0.5723 0.1012

0.0759 0.2118 0.3631 0.0758 0.1159 0.0355 0.0465 0.0439 0.1224 0.0251 0.0202 0.2037

6

Descriptive Statistics and Correlations.

0.1595 0.2003 0.1801 0.2226 0.0604 0.0362 0.0383 0.0198 0.0173 0.5252 0.1110 0.0598 0.3409 0.1928 0.0737 0.3775 0.1142

Standard Deviation

1. Year 95 .1124498 .3165553 2. Year 96 .1686747 .3752186 3. Year 97 .2369478 .4260664 4. Year 98 .2048193 .4043823 5. Year 99 .2771084 .4484719 6. Number of episodes 8.935743 25.12252 7. Team size 25.36546 7.697859 8. TV movie .3052209 .4614284 9. Prime time .8674699 .3397491 10. Major channel .6144578 .4877034 11. Age 2.684136 .6167447 12. Ego network closure .0490598 .0838544 13. Status 6.47 1 14. Past ego network imitation 8.349786 5.299695 15. Ego network imitation .4832571 .0609587 16. Structural holes .1172289 .041996 17. Team network centrality 30.16867 14.73187 18. Team performance 8.373815 .598522

Variable

Table 2. 9

10

11

0.5441 0.231

16

0.1131

17

18

0.0672 0.0917 0.0051 0.0826 0.1060 0.0819 0.1459 0.0439 0.2086 0.0560 0.1801 0.0629 0.0090 0.0011 0.1451 0.0933 0.0642 0.4864 0.0753 0.1736 0.1561 0.0627 0.1914 0.2560 0.1302 0.0375 0.1080 0.1511 0.0787 0.2505 0.1427 0.5328 0.4668 0.2450

8

550 GIUSEPPE SODA ET AL.

551

Imitative Behavior

Table 3.

First Stage.

Results of Two Stages Least Square: First Stage Endogenous Variable – Ego Network Imitation Instrumental variables

Model 1

Model 2

Model 3

Model 4

Constant Year 95

.408 .020 (.015) .101** (.012) .056** (.010) Dropped .001 (.009) .001 (.007) .000 (.000) .004** (001)

.390 .031** (.010) .079** (.007) .002 (.007) Dropped 0.05** (.006) .002 .004 .001** (.000) .001* (.000) .003** (.000)

.39 .027** (.008) .079** (.007) .001 (.006) Dropped .048** (.006) .001 (.003) .001** (.000) .001* (.000) .004** .000 .153** (.022)

.26 .025** (.007) .069** (.006) .002 .006 Dropped .052** .005 .003 .003 .000 .000 .001** .000 .004** .000 .167** (.021) .002 (.002)

.3387 .3194 17.49

.7591 .7510 93.74

.7983 .7906 104.20

.8395 .8291 80.58

Year 96 Year 97 Year 98 Year 99 Age Team size Past ego network imitation Team network centrality Ego network closure Status R2 Adjusted R2 f-value

Note: Standard errors are in parentheses. w po.10; * po.05; ** po.01.

consequences, we further investigated how imitative behavior is related to performance. To capture and resolve the endogeneity embodied in this research question, we address the problem of the endogenous relationships between social structure, subsequent behaviors and performance using an appropriate methodology (2SLS) (Shaver, 2005). We aimed to resolve the puzzle of two competing arguments for the relationship between network structure – in this case, network closure – and the degree of imitative behavior as assessed by the similarity between the content produced by a focal team and its alters. While on the one hand,

552

GIUSEPPE SODA ET AL.

Table 4.

Second Stage.

Results of Two Stages Least Square: Second Stage Dependent Variable – Project Team Performance (Audience) (H4) Explanatory Variables Year 95 Year 96 Year 97 Year 98 Year 99 Prime time TV movie Number of episodes Major channel Structural holes Ego network imitation R2 Adjusted R2 f-value

Coefﬁcients Standardized Dropped .064 (.116) .107 (.099) .093 (.097) .195* (.092) .505 (.088) .096* (.059) .005 (.001) .495 (.056) .191** (.754) .116* (.613) .555 .536 29.61

Note: Standard errors are in parentheses. w po.10; ** po.05; *** po.01.

research has argued for the homogenizing effects of closure, though conformity to norms and social pressures, on the other hand, drawing on a vigorous stream of inter-organizational network and alliance research, we present and test an alternative thesis for the creation of dense network links which serve to access resources and capabilities as well as information which organizational actors use to differentiate themselves under conditions of competitive interdependence. In fact, our results show that, consistent with this idea, network closure is associated with less rather than greater homogeneity, suggesting that, contrary to received wisdom, imitation is reduced rather than increased through network density. As Hagedoorn and

553

Imitative Behavior

Team network centrality

+

Ego Network closure

-

Ego network imitation

Team Performance

Team Status

Fig. 2.

2SLS Findings.

Frankort (2008) argue, although the literature on social embeddedness has typically stressed its positive effects, overembeddedness creates cognitive lock-in and reduced marginal information that organizational actors can acquire from their partners, which have negative consequences. From a population ecology perspective, if we conceive of product imitation as a mechanism that increases the number of players in a speciﬁc product niche (niche density), our ﬁndings suggest that a dense network infrastructure linking organizations reduces niche density. By connecting with multiple other competitive teams, focal teams are made aware of others’ product decisions, and consciously generate differentiated products because access to multiple other alters’ resources and knowledge helps focal organizations explore more resource combination opportunities (see also Hite, 2008). From a theoretical standpoint, our ﬁnding supports a contingency view of the outcomes of closure. This contingency perspective suggests that it is important to ascertain the context and the content of network ties in order to make inferences about the effect of structures on action such as product imitation. Speciﬁcally, if ties are formed to access different resources and ideas in order to create a differentiated and unique product, and if network actors are operating under conditions of competitive interdependence, we should not expect that greater imitation and homogeneity will result from network closure. Conversely, when ties are formed for homophily or social similarity reasons and organizations are operating under conditions of symbiotic interdependence, actors may exhibit highly imitative behavior because the tight and dense ties in this case are conduits of social meaning and values rather than of complementary resources. Our ﬁnding is also

554

GIUSEPPE SODA ET AL.

interesting because it contradicts the standard structural holes argument that diversity is the result of accessing novelty through non-redundant ties since we actually ﬁnd the opposite link, one between redundancy (closure) and diversity (low homogeneity). We also found that the network centrality was a strong predictor of imitative behavior among the networked teams. On the contrary, we did not ﬁnd the expected relationship between status and imitative behavior. It appears that teams imitate the most prominent team amongst them, rather than one that is composed of successful members of past teams, suggesting that a central structural position is a more powerful inﬂuence on alters’ imitation than status. The negative relationship between imitation and performance supports the argument that in a competitive context in which creativity and innovation are beneﬁcial, imitative behavior and isomorphic actions might be detrimental for ﬁnal performance. Again, this ﬁnding argues for a nuanced understanding of the context. In particular, legitimacy is often conferred by engaging in mimetic or imitative behavior (DiMaggio & Powell, 1983). However, Deephouse (1999) suggested an inverted U-shaped relationship between imitation and performance. Our results suggest that when creativity is a key goal and actors compete with one another, imitation results in lowered performance. Overall, by assessing the effects of imitation on performance we contribute to a better understanding of the relationship between structure and outcomes via imitative actions. We show that it is not status that creates imitative homogeneity but rather centrality and the imitation of powerful and prominent actors. Contrary to conventional wisdom, network closure is not associated with imitation but with differentiation.

Limitations and Directions for Future Research Our research ﬁndings may be context-speciﬁc because of the creative and temporary nature of the teams that we study. However, since the social structure and content in our research context is recreated in every period, it may represent a strength for the study. A further possible limitation lies in our measurement of content because we only capture outcomes in terms of the creative output – i.e. the ﬁlm or serial – rather than the process teams use or the knowledge and skills that they bring to the production. Of course, in such a cultural industry, the product is the most important element of imitative behavior. Further, we consider only the current ties of each TV

555

Imitative Behavior

production team without considering, except for status, for effects deriving from the past. Future research should build on our results to investigate other characteristics of network structure or content affecting the source of homogeneity, and the possible boundary conditions of the negative link between closure and conformity.

NOTE 1. Consistent with the standard procedure in longitudinal analysis, past measures are computed by using the same time window for each variable (7 years in our data set) (see for examples Ahuja, 2000; or Soda et al., 2004).

ACKNOWLEDGMENTS The authors would like to thank the editors Joel Baum and Tim Rowley. Versions of this chapter were presented at the New York University Stern School (NYU) Mini-conference on Cultural Production in a Global Context: The Worldwide Film Industries Conference and at the Academy of Management Annual Meetings in Philadelphia in 2007. Comments from participants are gratefully acknowledged. All errors are ours.

REFERENCES Abrahamson, E., & Hegeman R. (1994). Strategic conformity: An institutional theory explanation. Paper presented at the annual meeting of the Academy of Management, Dallas, TX. Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Argyle, M. (1957). Social pressure in public and private situations. Journal of Abnormal and Social Psychology, 54, 172–175. Asch, S. E. (1956). Studies of independence and conformity: A minority of one against a unanimous majority. Psychological Monographs, 70 (9, Whole No. 416). Balkundi, P., & Harrison, D. A. (2006). Ties, leaders, and time in teams: Strong inference about network structure’s effects on team viability and performance. Academy of Management Journal, 49(1), 49–68. Balkundi, P., Kilduff, M., Barsness, Z. I., & Michael, J. H. (2007). Demographic antecedents and performance consequences of structural holes in work teams. Journal of Organizational Behavior, 28, 241–260. Baum, J. A. C., Rowley, T. J., & Shipilov, A. (2003). Where do small worlds come from? Industrial and Corporate Change, 12, 697–725.

556

GIUSEPPE SODA ET AL.

Baum, J. A. C., Rowley, T. J., Shipilov, A., & Chuang, Y-T. (2005). Dancing with strangers: Aspiration performance and the search for underwriting syndacate partners. Administrative Science Quarterly, 50, 536–575. Baum, J. A. C., & Singh, J. V. (1994). Evolutionary dynamics of organizations. New York: Oxford University Press. Baumeister, R. F., & Leary, M. R. (1995). The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin, 122, 38–44. Benjamin, B. A., & Podolny, J. M. (1999). Status, quality, and social order in the California wine industry, 1981–1991. Administrative Science Quarterly, 44(3), 563–589. Berelson, B. (1952). Content analysis in communication research. Glencoe, III: Free Press. Bielby, W. T., & Bielby, D. D. (1999). Organizational mediation of project-based labor markets: Talent agencies and the careers of screenwriters. American Sociological Review, 64, 64–85. Borgatti, S., Everett, M. G., & Freeman, L. C. (2002). ‘‘Ucinet VI’’, Harvard: Analytic technologies. Cambridge,: MA. Brass, D. J. (1992). Power in organizations: A social network perspective. In: G. Moore & J. A. Whitt (Eds), Research in politics and society (Vol. 4, pp. 295–323). Greenwich, CT: JAI Press. Brass, D. J., Galaskiewicz, J., Greve, H. R., & Tsai, W. (2004). Taking stock of networks and organizations: A multilevel perspective. Academy of Management Journal, 47(6), 795–817. Burns, L. R., & Wholey, D. R. (1993). Adoption and abandonment of matrix management programs: Effects of organizational characteristics and inter-organizational networks. Academy of Management Journal, 36(1), 106–138. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge: Harvard University Press. Burt, R. S., Horgarth, R., & Michaud, C. (2000). The social capital of French and American managers. Organization Science, 11, 123–147. Chen, M., & Hambrick, D. C. (1995). Speed, stealth, and selective attack: How small ﬁrms differ from large ﬁrms in competitive behavior. Academy of Management Journal, 38, 453–482. Cialdini, R. B. (2001). Inﬂuence: Science a practice (4th ed). New York: Allyn and Bacon. Cialdini, R. B., & Goldstein, N. J. (2004). Social inﬂuence: Conformity and compliance. Annual Review of Psychology, 55, 591–621. Cialdini, R. B., Reno, R. R., & Kallgren, C. A. (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58, 1015–1026. Cohen, J. (1960). A coefﬁcient of agreement for nominal scales. Educational and Psychological Measurement, 20, 37–46. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94(Suppl.), 95–120. Crutchﬁeld, R. S. (1955). Conformity and character. American Psychologist, 10, 191–198. Deephouse, D. L. (1999). To be different, or to be the same? It’s a question (and theory) of strategic balance. Strategic Management Journal, 20, 147–166. Deutsch, M., & Gerard, H. B. (1955). A study of normative and informational social inﬂuences upon individual judgment. Journal of Abnormal and Social Psychology, 51, 629–636.

Imitative Behavior

557

DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational ﬁelds. American Sociological Review, 48(2), 147–160. Donaldson, L. (1997). A positivist alternative to the structure–action approach. Organization Studies, 18(1), 77–92. Duval, S. (1972). Conformity of a visual task as a function of personal novelty on attitudinal dimensions and being reminded of the object status of the self. Unpublished doctoral dissertation, University of Texas. Dyer, J. H., & Singh, H. (1998). The relational view: Cooperative strategy and sources of interorganizational competitive advantage. Academy of Management Review, 23(4), 660–679. Fligstein, N. (1985). The spread of the multidivisional form among large ﬁrms, 1919–1979. American Sociological Review, 50, 377–391. Fligstein, N. (1991). The structural transformation of American Industry: An institutional account of the causes of diversiﬁcation in the largest ﬁrms, 1919–1979. In: W. W. Powell & P. J. DiMaggio (Eds), The institutionalism in organizational analysis (pp. 294–310). Chicago: University of Chicago Press. Freeman, L. C. (1979). Centrality in social networks: Conceptual clariﬁcation. Social Networks, 1, 215–239. Galaskiewics, J., & Wasserman, S. (1989). Mimetic processes within an inter-organizational ﬁeld: An Empirical test. Administrative Science Quarterly, 34, 454–479. Gigerenzer, G., & Todd, P. M. (1999). Simple heuristics that make us smart. London: Oxford University Press. Gnyawali, D. R., & Madhavan, R. (2001). Cooperative network and competitive dynamics: A structural and embeddedness perspective. Academy of Management Review, 26(3), 431–445. Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Griskevicius, V., Goldstein, N. J., Mortensen, C. R., Cialdini, R. B., & Kenrich, D. T. (2006). Going along versus going alone: When fundamental motives facilitate strategic (non)conformity. Journal of Personality and Social Psychology, 91(2), 281–294. Gulati, R. (1998). Alliances and networks. Strategic Management Journal, 19(4), 293–317. Gulati, R., & Gargiulo, M. (1999). Where do inter-organizational networks come from? American Journal of Sociology, 104(5), 1439–1493. Hagedoorn, J., & Frankort, H. T. W. (2008). The gloomy side of embeddedness: The effects of overembeddedness on inter-ﬁrm partnership formation. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 503– 530). Bingley, UK: JAI/Emerald Group. Hansen, M. T. (1999). The search-transfer problem: The role of weak ties in sharing knowledge across organizational subunits. Administrative Science Quarterly, 44, 82–111. Haunschild, P. R. (1993). Interorganizational imitation: The impact of interlocks on corporate acquisition activity. Administrative Science Quarterly, 38, 564–592. Haunschild, P. R., & Miner, A. S. (1997). Modes of inter-organizational imitation: The effects of outcome salience and uncertainty. Administrative Science Quarterly, 42, 472–500. Hausman, J. A. (1978). Speciﬁcation tests in econometrics. Econometrica, 46(6), 1251–1271.

558

GIUSEPPE SODA ET AL.

Haveman, H. A. (1993). Follow the leader: Mimetic isomorphism and entry into new markets. Administrative Science Quarterly, 38, 593–627. Hite, J. (2008). The evolution of strategic dyadic network ties: Strategically navigating bounded agency within multidimensional and dynamic dyadic relationships. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 133–170). Bingley, UK: JAI/Emerald Group. Hollander, E. P. (1958). Conformity, status, and idiosyncrasy credits. Psychological Review, 65, 117–127. Huber, P. J. (1967). The behavior of maximum likelihood estimates under non-standard conditions. Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability. University of California Press, Berkeley, CA (Vol. 1, pp. 221–223). Huff, A. S. (1982). Industry inﬂuences on strategy reformulation. Strategic Management Journal, 3(2), 119–131. Ibarra, H. (1993). Network centrality, power, and innovation involvement: Determinants of technical and administrative roles. Academy of Management Journal, 36(3), 471–501. Insko, C. A., Smith, R. H., Alicke, M. D., Wade, J., & Taylor, S. (1985). Conformity and group size: The concern with being right and with being liked. Personality and Social Psychology Bulletin, 11, 41–50. Janis, I. L. (1972). Victims of groupthink. Boston: Houghton Mifﬂin. Krippendorff, K. (1980). Content analysis: An introduction to its methodology. Newbury Park, CA: Sage. Kvalseth, T. O. (1989). Note on Cohen’s kappa. Psychological Reports, 65, 223–226. Lieberman, M. B., & Asaba, S. (2006). Why do ﬁrms imitate each other? Academy of Management Review, 31(2), 366–385. Madhavan, R., Caner, T., Prescott, J., & Koka, B. (2008). Bringing the ﬁrm back in: Networking as antecedent to network structure. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 457–501). Bingley, UK: JAI/Emerald Group. Marquis, C. (2003). The pressure of the past: Network imprinting in inter-corporate communities. Administrative Science Quarterly, 48, 655–689. Marsden, P. V. (2005). The sociology of James S. Coleman. Annual Review of Sociology, 31, 1–24. Martin, R., & Hewstone, M. (2003). Social-inﬂuence process of control and change: Conformity, obedience to authority and innovation. In: M. A. Hogg & J. Cooper (Eds), The Sage handbook of social psychology. Thousand Oaks, CA: Sage. Maslach, C., Stapp, J., & Santee, R. T. (1985). Individuation: Conceptual analysis and assessment. Journal of Personality and Social Psychology, 49, 729–738. Milgrom, P., & Roberts, J. (1995). Complementarities and ﬁt strategy, structure, and organizational change in manufacturing. Journal of Accounting and Economics, 19(2–3), 179–208. Moscovici, S. (1985). Social inﬂuence and conformity. In: G. Lindzey & E. Aronson (Eds), The handbook of social psychology (3rd ed., Vol. 2, pp. 347–412). New York: Random House. Mouw, T. (2006). Estimating the causal effect of social capital: A review of recent research. Annual Review of Sociology, 32, 79–102. Nail, P. R., Macdonald, G., & Levy, D. A. (2000). Proposal of a four dimensional model of social response. Psychological Bulletin, 126, 106–116.

Imitative Behavior

559

Nohria, N., & Garcia-Pont, C. (1991). Global strategic linkage and industry structure. Strategic Management Journal, 12, 105–124. Oliver, C. (1997). Sustainable competitive advantage: Combining institutional and resourcebased views. Strategic Management Journal, 18(9), 697–713. Parsons, T. (1970). Equality and inequality in modern society, or social stratiﬁcation revisited. In: E. O. Laumann (Ed.), Social stratiﬁcation: Research and theory for the 1970s (pp. 22–40). Indianapolis: Bobbs-Merrill. Perry-Smith, J. E. (2006). Social yet creative: The role of social relationships in facilitating individual creativity. Academy of Management Journal, 49(1), 85–101. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations. New York: Harper and Row. Podolny, J. M. (1993). A status based model of market competition. American Journal of Sociology, 107(1), 33–60. Porter, M. E. (1980). Competitive strategy. New York: The Free Press. Porter, M. E. (1996). What is strategy? Harvard Business Review, 74(6). Portes, A., & Sensenbrenner, J. (1993). Embeddedness and immigration: Notes on the social determinants of economic social action. American Journal of Sociology, 98(6), 1320–1350. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Inter-organizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41(1), 116–145. Reagans, R., & Zuckerman, E. W. (2001). Networks, diversity, and productivity: The social capital of RandD teams. Organization Science, 12(4), 502–517. Reagans, R., Zuckerman, E. W., & McEvily, B. (2004). How to make the team: Social networks vs. demography as criteria for designing effective teams. Administrative Science Quarterly, 49, 101–133. Reger, R. K., & Huff, A. S. (1993). Strategic groups: A perspective. Strategic Management Journal, 14(2), 103–124. Rogers, E. M. (1965). Diffusion of innovations. New York: Free Press. Rumelt, R. P., Schendel, D. E., & Teece, D. J. (1994). Fundamental issues in strategy. In: R. P. Rumelt, D. E. Schendel & D. J. Teece (Eds), Fundamental issues in strategy: A research agenda (pp. 9–53). Boston, MA: Harvard Business School Press. Schwenk, C. (1984). Cognitive simpliﬁcation processes in strategic decision-making. Strategic Management Journal, 5(2), 111–128. Scott, J. (2000). Social network analysis: A handbook (2nd ed.). Thousand Oaks, CA: Sage. Scott, W. R. (1995). Institutions and organizations. Thousand Oaks, CA: Sage. Shan, W., & Hamilton, W. (1991). Country-speciﬁc advantage and international cooperation. Strategic Management Journal, 12(6), 419–432. Shaver, J. M. (2005). Testing for mediating variables in management research: Concerns, implications, and alternative strategies. Journal of Management, 31(3), 330–353. Sherif, M. (1936). The psychology of social norms. New York: Harper. Snyder, C. R., & Fromkin, H. L. (1980). Uniqueness: The human pursuit of difference. New York: Plenum Press. Soda, G., & Usai, A. (1999). The dark side of dense networks. In: A. Grandori (Ed.), Interﬁrm networks: Organization and industrial competitiveness (pp. 276–303). London and New York: Routledge.

560

GIUSEPPE SODA ET AL.

Soda, G., Perrone, V., Usai, A. (2001). Bringing the task back in: A network investigation on the contingent value of resource complementarity in partner selection and alliance formation. Paper presented to SUNBELT XXI International Sunbelt Social Network Conference in Budapest, Hungary. Soda, G., Usai, A., & Zaheer, A. (2004). Network memory: The inﬂuence of past and current networks on performance. Academy of Management Journal, 47(6), 893–906. Starkey, K., Barnatt, C., & Tempest, S. (2000). Beyond networks and hierarchies: Latent organizations in the UK television industry. Organization Science, 11(3), 299–305. Suitor, J. J., Wellman, B., & Morgan, D. L. (1997). It’s about time: How, why, and when networks change. Social Networks, 19, 1–7. Sutton, J. R., & Dobbin, F. (1996). The two faces of governance: Responses to legal uncertainty in U.S. ﬁrms, 1955 to 1985. American Sociological Review, 61(5), 794–811. Tsai, W. (2001). Knowledge transfer in transfer in intra-organizational networks: Effects of network position and absorptive capacity on business unit innovation and performance. Academy of Management Journal, 44(5), 996–1004. Von Clausewitz, C. (1832, republished 1997). On war, Ware: Wordsworth Editions Limited. Walsh, J. P., & Ungson, G. R. (1991). Organizational memory. Academy of Management Review, 16(1), 57–91. Westphal, J. D., Gulati, R., & Shortell, S. M. (1997). Customization or conformity? An institutional and network perspective on the content and consequences of TQM adoption. Administrative Science Quarterly, 42(2), 366–394. White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817–830. Willis, R. H. (1963). Two dimensions of conformity–nonconformity. Sociometry, 26, 499–513.

WHO’S THE NEW KID? THE PROCESS OF DEVELOPING CENTRALITY IN VENTURE CAPITALIST DEAL NETWORKS Bret R. Fund, Timothy G. Pollock, Ted Baker and Adam J. Wowak ABSTRACT In this chapter we examine the process by which new ﬁrms become central actors within their industry networks. We focus, in particular, on how relatively new venture capital (VC) ﬁrms become more central within investment syndication networks. We present a model that captures the relationships among (1) the social capital and status of the new VC ﬁrm’s founders, (2) the VC ﬁrm’s resource endowments, (3) the VC ﬁrm’s ability to forge relationships with other prestigious and central venture capital ﬁrms, (4) the visibility-enhancing performance of portfolio ﬁrms, and (5) the urgency and effort exhibited by the new VC as it pursues these opportunities. These factors combine to shape a new VC’s journey from the periphery to the center of its industry network. To illustrate these processes, we develop in-depth case studies of Benchmark Capital and August Capital, two VC ﬁrms founded in 1995. We then elaborate upon the enacted nature of resource and opportunity constraints and Network Strategy Advances in Strategic Management, Volume 25, 563–593 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25016-3

563

564

BRET R. FUND ET AL.

conclude with a discussion of how new ﬁrms create their own self-fulﬁlling prophecies.

Conventional wisdom holds that new entrants to existing industries often struggle to gain access to important resources while established players typically enjoy much easier access. Merton (1968a) labeled this phenomenon the ‘‘Matthew Effect,’’ and described how prominent actors are able to acquire resources, attention, and rewards more readily than less well-known actors – in other words, how the rich get richer while the poor get poorer. Social networks scholars have devoted considerable attention to this phenomenon, arguing that ﬁrms occupying more central positions in networks have informational advantages (Freeman, 1979; Gulati, 1998) and more accurate views of existing networks (Krackhardt, 1990) that provide them with more detailed knowledge of and access to larger pools of potential network partners (Gulati & Gargiulo, 1999). These informational advantages grant central ﬁrms better access to critical resources while simultaneously depriving less central ﬁrms of similar opportunities. However, both the Matthew Effect and network perspectives on the role of centrality in partner selection and network formation tend to assume fairly static and stable network structures. As a consequence, their focus is on how central actors entrench their positions by gaining early access to new and valuable resources while relegating less central organizations to either surviving on the periphery of industry networks or exiting these networks completely (Kim, Oh, & Swaminathan, 2006). These perspectives argue that established players behave in ways that maintain stability – thereby ensuring central actors’ continued positional advantages – while the resulting network inertia (Kim et al., 2006) leads to underperformance or failure for new ﬁrms seeking to enter and establish their own positions within industry networks. Yet casual observation of all but the most stagnant industries suggests that new ﬁrms routinely navigate their way from the periphery to the center of industry networks in spite of their presumed disadvantages. The original biblical context that inspired Merton’s ‘‘Mathew Effect’’ (King James Bible, Matthew, XXV: 14–34) can be instructive in this regard, although not necessarily for the reasons Merton articulated. In Matthew’s parable, a man leaving the country gave his goods to his servants to care for while he was away. He gave one servant ﬁve talents (a measure of currency equal to several years’ wages), another two talents and a third one

Process of Developing Centrality in Venture Capitalist Deal Networks

565

talent. The two servants who received multiple talents invested them, and were subsequently able to return double the talents to their master upon his return. In contrast, the servant who received only one talent was fearful and buried the talent in the ground rather than investing it. Upon returning, their master rewarded the ﬁrst two servants for their efforts but castigated the servant who hid rather than invested his money, taking the one talent away from this servant and giving it to the servant who had achieved the highest return on his investment. The lesson Merton took from this story and reiﬁed in the public consciousness is that the rich get richer while the poor get poorer. However, the second lesson from this parable, which generally goes unacknowledged, is that sometimes the poor become poorer when they fearfully bury and hide what little they own rather than using their existing endowments, however meager, to trade to their advantage. This suggests the poor may be more likely to escape poverty if they invest and leverage, rather than horde, whatever endowments they may possess. It is this second lesson on the effects of individual agency that we will use to understand how new ﬁrms can become central players in industry networks. In recent years scholars have begun to pay more attention to the role of individual agency in network building activities (e.g., Emirbayer & Goodwin, 1994; Lin, 1999; Madhavan, Caner, Prescott, & Koka, 2008; Pollock, Porac, & Wade, 2004); however, we are unaware of any network research that has explored or developed theory to explain how new organizations can leverage available resources to overcome network inertia and gain more central or prominent network positions. Accordingly, this issue is the focus of our chapter. We explore this question by examining differences in how two new venture capital ﬁrms founded in the mid-1990s navigated their way to central positions within industry deal networks. In doing so, our chapter joins the work of others in this volume in enhancing our understanding of the processes through which interorganizational networks are developed and navigated (Dagnino, Levanti, & Li Destri, 2008; Madhavan et al., 2008; Rowley & Baum, 2008; van Liere, Koppius, & Vervest, 2008). The venture capital industry is an ideal context in which to study the process of developing network centrality primarily because of the importance of network structures to ﬁrm performance. Network structures strongly inﬂuence which ﬁrms become members of the prestigious investment syndicates that form around the most promising deals. The ability to access and invest in the most promising new ventures is a key to venture fund performance and also inﬂuences the ability to raise future funds.

566

BRET R. FUND ET AL.

This chapter is organized as follows. We initially provide a short background on the venture capital industry, highlighting its usefulness as a context in which to investigate the phenomenon under study. We then introduce our process model of centrality achievement and summarize the history and evolution of two venture capital ﬁrms – Benchmark Capital and August Capital – to illustrate the elements and relationships in our model. We conclude with a brief discussion of the model’s implications and potential future research directions.

THE VENTURE CAPITAL INDUSTRY: A VERY BRIEF HISTORY1 Although some form of venture capital ﬁnancing has existed for millennia (Gompers & Lerner, 2006), the modern venture capital industry is a relatively new organizational form and ﬁeld. Today, thousands of professionals manage hundreds of funds that are geographically concentrated on the east and west coasts of the United States (Gompers & Lerner, 2006). In comparison with the assets of even a single large bank or mutual fund company, the approximately $150 billion in capital managed by the entire VC industry is relatively small. Nonetheless, because of its focused investments in highly risky but also highly lucrative young ventures, the venture capital industry is considered an important contributor to the overall health and growth of the U.S. economy. The contemporary venture capital industry is in many ways a modern extension of the role played by wealthy families for hundreds of years. Families with vast accumulations of personal wealth have long maintained the practice of investing part of their fortunes in risky ventures while leveraging their extensive networks and strong ties among similarly positioned elites to both identify investment opportunities and shape investment outcomes to their families’ advantage (Padgett & Ansell, 1993). Modern venture capital ﬁrms began to appear in the United States by the end of World War II, but these VC ﬁrms still relied primarily on wealthy individuals and families for much of their capital (Gompers & Lerner, 2006). At the same time, many large ﬁnancial institutions had moved beyond family control and were managing investment portfolios that overshadowed even dynastic family wealth (although legal restrictions and tradition severely limited institutional investments in risky VC funds).

Process of Developing Centrality in Venture Capitalist Deal Networks

567

The VC industry received a signiﬁcant boost in 1979 when an amendment to the U.S. Employee Retirement Income Security Act (ERISA) modiﬁed the ‘‘prudent man’’ rule governing pension fund investments. Before that time, pension funds were prohibited from investing signiﬁcant sums of capital in high-risk assets such as venture capital funds. Responding in part to changes in academics’ understanding of the role of proper portfolio diversiﬁcation (Markowitz, Elton, & Gruber, 1979), the 1979 amendment relaxed this prohibition. Increased amounts of capital began to ﬂow into venture funds from institutional investors eager for higher returns. Indeed, while new investments in venture funds in 1978 totaled only $495 million (in 2002 dollars), by 1984, this amount had increased to $5.6 billion, and at its peak in 2000 a total of $108 billion was invested in new VC funds (Gompers & Lerner, 2006). Not surprisingly, industry growth has historically been unsteady and subject to booms and shakeouts triggered by overall economic and equity market downturns (Kindleberger & Aliber, 2005). Fresh capital inﬂows in the early 1980s spurred the founding of new VC ﬁrms and investments in riskier industries such as personal computing, software, and biotechnology. A strong IPO market further increased inﬂows and resulted in even riskier investments by VCs. The market crash of 1987 and subsequent recession reduced capital inﬂows, however, and ultimately drove many inexperienced VC ﬁrms from the industry. Beginning in 1992, the IPO market began once again to gain steam, stimulating a fresh ﬂow of capital into the industry. New venture capital ﬁrms proliferated: in 1995 alone 112 new venture capital ﬁrms were founded.2 By the mid-1990s, the rules of the VC game were fairly well established and familiar to participants. The limited partnership structure, which allows venture capitalists to raise multiple investment funds with deﬁned life spans (typically 10 years), while also protecting investors from management responsibilities and legal liabilities, remains the dominant organizational form (Gompers & Lerner, 2006). The industry is characterized by a large body of taken-for-granted legal, accounting and managerial routines that quickly became legitimated as the industry developed (Suchman, 1995) and which helped to make the process of starting a new VC ﬁrm relatively straightforward. Nevertheless, at least two general characteristics of the early, pre-ERISA form of the VC industry remained consistent throughout this period and helped maintain advantages for established VC ﬁrms. First, reminiscent of the times when venture capital was provided and managed by wealthy families and their trusted ﬁnancial advisors, the VC industry continued to exhibit a strong sense of status hierarchy among member ﬁrms

568

BRET R. FUND ET AL.

(Gompers, 1995). Second, the formation of investment syndicates, or deal networks (Pollock et al., 2004), remained common, allowing multiple VC ﬁrms to share the risks and rewards of funding entrepreneurial start-ups. Complementing the well-developed legal structure of the industry, embedded ties among elite participants remained important to the functioning of the industry while simultaneously creating status- and network-based boundaries that limited access to the industry’s core deal networks. In addition to these established norms, new VC ﬁrms in the mid-1990s faced another challenge. Growth and an increasing focus on capitalintensive technology industries rapidly increased the size of the ‘‘entry level’’ fund a VC ﬁrm needed to raise in order to become a serious industry participant and pursue what the industry now perceived as the most promising deals. VC money moved away from the highest risk, ‘‘seed stage’’ deals and toward meeting the larger capital needs of later-stage ﬁrms. Moreover, the public and private pension funds that had so eagerly invested in the 1980s became more selective and reluctant to commit capital to new and untried VC ﬁrms. Despite the fact that the number of existing VCs increased almost fourfold between 1980 and 1990 (from 87 in 1980 to 325 in 1990), there were approximately one third fewer ﬁrst-time VC funds raised in 1990 than in 1980 (23 ﬁrst-time funds in 1980 vs. 14 ﬁrst-time funds in 1990). This situation afforded a number of beneﬁts to established VC ﬁrms. These key players were able to use their status and position to attract institutional investors while also leveraging their centrality to increase the compensation they earned via management fees and to claim a larger proportion of the funds’ proﬁts for themselves.3 New entrants, on the other hand, were often expected to accept less attractive compensation terms in order to secure sufﬁcient capital (Stross, 2000). Central, high status VC ﬁrms also enjoyed advantages through superior access to the most lucrative investments. Finding blockbuster deals is not easy; venture capitalists may examine hundreds of business plans and management teams for each investment they make. However, centrally positioned VC ﬁrms gain privileged access to the most promising deals because of their presumed experience in developing new ﬁrms and their networks of relationships with key resource providers (Sahlman, 1990), potential alliance partners (Pollock & Gulati, 2007), and talented and experienced executives (Jain & Kini, 2000). As the venture capital industry has developed, many ventures seeking ﬁnancing have come to prefer partnering with central and high-status VC ﬁrms, often to the point of

Process of Developing Centrality in Venture Capitalist Deal Networks

569

accepting lower valuations in exchange for the perceived added value of these investors (Hsu, 2004; Sapienza, Manigart, & Vermeir, 1996). Taken together, these factors suggest that individuals attempting to establish a new VC ﬁrm in the period between the early 1990s recession and the late 1990s boom faced a number of challenges in ﬁnding institutional investors and gaining access to investment syndicates for the best deals. Both of the ﬁrms we will examine were founded and raised their ﬁrst funds in 1995. The individual founders of both ﬁrms were experienced venture capitalists, but the question still remained: Could they do it on their own? In the following section, we present our model of centrality development and then delve into the histories of each ﬁrm to explore how the components of the model shaped the speed and success with which each ﬁrm enhanced its centrality in VC deal networks.

THE PROCESS OF DEVELOPING NETWORK CENTRALITY Early network studies in the management literature, such as ‘‘old’’ institutional theory (Selznick, 1948), contingency theory (Lawrence & Lorsch, 1967; Thompson, 1967) and resource dependence theory (Pfeffer & Salancik, 1978), all argued that ﬁrms create ties to obtain desired resources and manage uncertain environments. Network formation was therefore thought to be driven by ﬁrms’ needs to connect with other organizations possessing the resources and capabilities required to help these ﬁrms cope with exogenous constraints. Organizations were assumed to be only mildly constrained by the characteristics of networks themselves in seeking and building their own ties. This perspective promoted an understanding of the factors that inﬂuenced the likelihood organizations will attempt to create new ties, but did little to explain how networks or prior ties determined the possibilities open to new organizations (Gulati & Gargiulo, 1999). More recently, scholars have turned their focus to factors endogenous to networks that inﬂuence and constrain how organizations are able to embed themselves within existing network structures (Gulati & Gargiulo, 1999; Nohria, 1992). This literature often focuses on an actor’s centrality within the network as an important endogenous factor, typically deﬁning centrality as ‘‘degree centrality’’ (Wasserman & Faust, 1999), or the number of direct ties an actor has with other members of the network. From this perspective, the network itself is the key social constraint, providing the context that

570

BRET R. FUND ET AL.

shapes how organizations’ actions recreate or alter the characteristics of the network within which they are embedded (Gulati & Gargiulo, 1999; Koka, Madhavan, & Prescott, 2006). Managers of the ﬁrms in a network are limited in the actions available to them as a result of the network structure they have helped to create (Nohria, 1992). Research from this perspective suggests that a ﬁrm is likely to form relationships with other organizations in the network with which it has had prior dealings, or with its partners’ partners (Gulati, 1995). Further, this literature argues that central players are more likely to seek relationships with similar others – or with more central players – rather than with peripheral organizations (Chung, Singh, & Lee, 2000; Gulati & Gargiulo, 1999). But what happens to a ﬁrm that does not occupy such a privileged position? Are these organizations doomed to a life on the periphery of industry networks? Whereas earlier perspectives focused on the general agency of organizational actors in ﬁnding ways to obtain resources and capabilities by building network ties, more recent work emphasizes that central ﬁrms maintain their privileged positions through securing additional network resources via the formation of valuable new relationships while dissolving old relationships that have become less valuable (Gulati, 1998; Gulati & Gargiulo, 1999; Inkpen & Beamish, 1997; Mizruchi & Galaskiewicz, 1994). Observation of real network dynamics suggests that new organizations are neither free to write their own tickets in becoming central within a network nor are they forever banished to the sidelines. New organizations can and do become central players within existing networks. The question, then, is how does this occur? To address this question, we differentiate between two types of centrality: structural centrality and cognitive centrality. Structural centrality is based on the traditional notion of degree centrality in network analysis. Simply put, a structurally central actor is one who ‘‘is the most active in the sense that they have the most ties to other actors in the network’’ (Wasserman & Faust, 1999, p. 178). Cognitive centrality, on the other hand, is a social psychological concept originally used to describe how actors in work groups (whose knowledge and expertise is shared with larger numbers of group members) obtain greater inﬂuence within the network (Kameda, Ohtsubo, & Takezawa, 1997). More recent work has extended the concept to describe globally shared cognitive network representations (Batchelder, 2002) and has explored the various cues used by network participants to identify other members who are more expert within a network of actors (Bunderson, 2003). In reviewing the literature on cognitive centrality, Bunderson (2003, p. 559) notes that ‘‘groups seem to perform better and make better decisions

Process of Developing Centrality in Venture Capitalist Deal Networks

571

when members share an accurate understanding of one another’s expertise, and y because expertise cannot be directly observed, group members are frequently either unaware of each others’ expertise or inaccurate in their assessments of that expertise.’’ He goes on to say that ‘‘yit seems clear that subjects in these studies attended carefully to any potentially available and potentially relevant expertise signals’’ and notes that these signals can come from a variety of factors, including expert role assignment, past experience, and assertive behaviors. Bunderson argues that those actors in the group for whom others hold high performance expectations will become more cognitively central within the work group, that these expectations are based upon both speciﬁc and diffuse status characteristics (Humphreys & Berger, 1981), and that more speciﬁc characteristics will be relied upon for longertenured group members while more diffuse characteristics will be used when group tenure is short. This work is also consistent with research on cognitive availability (Hoffman & Ocasio, 2001; Kuran & Sunstein, 1999; Pollock & Gulati, 2007; Pollock, Rindova, & Maggitti, 2008; Tversky & Kahneman, 1973) that argues the ability to attract attention and become cognitively available, or salient and easy to recall, can enhance an actor’s access to resources and perceptions of their capabilities and expectations regarding future performance. In this study we argue cognitive centrality initially is likely to precede structural centrality for new actors in industry networks, but that cognitive and structural centrality will subsequently enhance each other in a recursive and mutually beneﬁcial fashion. As an actor becomes more cognitively central and available its opportunities to participate in deals will increase, thereby enhancing its structural centrality. This increase in structural centrality will further enhance a ﬁrm’s cognitive availability by providing an important status characteristic signal that others will use in evaluating its expertise and likely future performance (Bunderson, 2003). Because other network members rely on more diffuse sets of signals for shorter-tenured members, new ﬁrms are likely to have an easier time molding their own cognitive centrality as a lever to gain structural centrality than they would have trying to push their way into a structurally central position more directly. This sets the stage for the important role played by urgency and effort that we describe below. Fig. 1 summarizes our process model of network centrality development. We suggest that this recursive process involves four key components that shape the extent to which a new ﬁrm becomes central in its industry network: (1) the personal social capital and prestige the founders bring to the ﬁrm; (2) the resource endowments the ﬁrm is able to acquire; (3) the

572

BRET R. FUND ET AL.

Fig. 1.

Process Model of Centrality Development.

organization’s ability to gain certiﬁcation and afﬁliation beneﬁts from relationships with other prestigious industry members; and (4) the extent to which the ﬁrm is able to achieve visibility-enhancing performance outcomes. Each component of the model affects the others. The social capital and status a founder brings to the ﬁrm can inﬂuence both the prestigious afﬁliations the ﬁrm develops as well as its ability to raise capital. Similarly, the more resources a ﬁrm has, the more opportunities it will be able to pursue and thus the greater its ability to develop relationships with prestigious industry members and achieve attention-grabbing performance. The better the ﬁrm’s performance, the easier it becomes to raise capital for future investments and the greater its chances of investing alongside prestigious VC ﬁrms in deals. We also argue that the ‘‘hub’’ around which these four factors revolve – and that determines the rate at which centrality is achieved – is the time urgency the founders experience and the effort they put into acquiring and leveraging the four components of the ‘‘wheel.’’ The greater the founders’ cognitive, emotional, and behavioral

Process of Developing Centrality in Venture Capitalist Deal Networks

573

urgency, which we label their urgency and effort, the more torque generated by the wheel, and the more quickly centrality is likely to develop. In the following sections, we introduce Benchmark Capital and August Capital and use their histories to explore each of these mechanisms and the relationships among them. As will become apparent, the two ﬁrms shared much in common and were in many ways quite similar to each other along the four exterior dimensions of our wheel, although they differed to some extent in the initial status characteristics they possessed. This allowed us to focus on how the effects of the four elements of the wheel depicted in Fig. 1 were shaped by differences in urgency and effort (Yin, 2003) while constructing a more nuanced picture of how the elements of the wheel interact to determine a VC’s journey toward the center of the industry network.

BENCHMARK AND AUGUST: A TALE OF TWO VENTURE CAPITAL FIRMS By 1995, the venture capital industry was becoming increasingly attractive to talented, enterprising individuals and potential investors alike. Excitement stemmed from what appeared to be a robust IPO market in conjunction with the enticing commercial possibilities of the Internet and its promise to change everyday life and business in fundamental – though extremely unpredictable – ways. This uncertainty and unpredictability created space for many true believers in the future success of the Internet tech sector – and a similarly large group of skeptics. Around this time, two prestigious VC ﬁrms disbanded when their founding partners decided to retire. One of these organizations, Merrill Pickard Anderson & Eyre (MPAE), was a pioneering venture ﬁrm that had grown out of Bank of America in the 1970s. MPAE had a long history of investment success and its founders were well respected within the VC industry. The second ﬁrm, Technology Venture Investors (TVI), was also founded in the 1970s. TVI had been the sole VC ﬁrm to invest in Microsoft and had invested in other notable high-tech companies. TVI had a strong reputation among entrepreneurs and investors alike. The breakup of these two established players displaced many of their partners and associates and spawned a number of new VC ﬁrms, including Benchmark Capital and August Capital.

574

BRET R. FUND ET AL.

Founders’ Personal Social Capital and Status Founders bring their personal prestige and accumulated social capital to their new ﬁrms. Founders’ social capital includes both relationships among the founders, and also their ties to individuals and organizations outside the new ﬁrm (Adler & Kwon, 2002; Bolino, Turnley, & Bloodgood, 2002; Koka & Prescott, 2002). This social capital can provide considerable value to a new ﬁrm from the very beginning and can have lasting effects on the ﬁrm’s performance and prospects of survival (Fischer & Pollock, 2004; Neergaard & Madsen, 2004). These social resources can also be instrumental in new ﬁrms’ attempts to overcome the challenges posed by network inertia. In this particular case, the founding partners of August and Benchmark shared employment histories at venture ﬁrms of similar status (some even overlapped at TVI), but nevertheless varied regarding the quality and types of social capital they brought to their respective ﬁrms. August Capital (August) was founded by partners David Marquardt and John Johnston, two former partners of TVI whose early stage investment experience prior to founding August included investments in Microsoft, Adaptec, Compaq, Sun Microsystems, Seagate, Intuit, Sybase, Visio, Actel, and ViewLogic. David Marquardt is a prominent and high-status member of the VC community; he was a co-founder of TVI and the lead VC for the Microsoft deal. To this day he continues to serve on Microsoft’s board. Benchmark Capital (Benchmark), in contrast, was founded by four individuals with respectable – but not extraordinary – records and reputations: Robert Kagle from TVI; Bruce Dunlevie and Andrew Rachleff, both from MPAE; and Kevin Harvey, a software company entrepreneur with no previous venture capital experience. While not as well-known as their August counterparts, the Benchmark founders were active and visible in VC industry organizations. Bob Kagle, for example, was elected to the board of the National Venture Capital Association and served as the program chair for its 1998 annual meeting. Benchmark also possessed some unusual sources of social capital. Kevin Harvey was well-known among software executives due to his previous experience as an entrepreneur. And in 1997 Benchmark welcomed a ﬁfth partner, David Bierne, who possessed extensive contacts with technology executives worldwide as a consequence of his experience founding executive recruiting ﬁrm Ramsey Beirne, which specialized in recruiting executives for technology ﬁrms (Stross, 2000). From their earliest days, the two ﬁrms took signiﬁcantly different approaches to building their networks and leveraging their initial social resource endowments. In the simplest terms, August seemed to take its time,

Process of Developing Centrality in Venture Capitalist Deal Networks

575

moving at a very deliberate pace. In the several months following the close of its inaugural fund, August made only one small investment for about $1 million (representing approximately 1% of its total fund). Our reading of a variety of contemporary descriptions of August’s behavior and our examination of their investment behavior suggests the self-conﬁdent manner of a ‘‘master of the universe’’ that felt little urgency or compulsion to hurry in making investments and putting the new ﬁrm on the map. In contrast, Benchmark’s behavior was more assertive and exhibited urgency and effort from the beginning. The founders set for themselves the audacious goal of becoming the world’s number one venture capital ﬁrm within 10 years (hence the ‘‘Benchmark’’ name). They subsequently entered the market in an aggressive manner. Within six months they had funded six startups, investing about a quarter of their total fund. These ventures ranged from Silicon Gaming, a casino software and solutions company, to Xantel, a manufacturing resource planning company focused on the medical device industry. The broad range of these investments – and the fact that none of these early investments turned out to be a big win – reinforces the sense that these investments were made in a hurry. Benchmark’s behavior also suggests the founding partners understood they would be unlikely to win by acting like prima donnas; rather it would require they work together to effectively leverage the resources at their disposal. To this end, when establishing their ownership structure they broke with the industry tradition of a star system and decided instead that proﬁts would be distributed equally among the partners. This practice, they believed, would eliminate the inﬁghting among partners common at other ﬁrms. Even as they added partners over the years, the four original founders never claimed titles such as ‘‘senior’’ or ‘‘founding’’ partners for themselves and they continued the policy of sharing proﬁts equally (although their joint share of the ﬁrm’s proﬁts were proportionately reduced with each new partner). They also eschewed the common practice of hiring a second tier of non-partner associates to handle much of the grunt work and sifting of deals required in the venture capital business. While limiting their ability to leverage the partners’ skills and experience, this policy increased their urgency and effort in making the best use of their limited time and resources and heightened their ability to identify promising ‘‘diamonds in the rough’’ that might be overlooked by less experienced associates. Further, recognizing that their new ﬁrm did not have the status or cachet that typically accompanies a strong performance history or the presence of a star VC, Benchmark sought to overcome these deﬁcits when going after the most promising new ventures by offering the highest levels of personal

576

BRET R. FUND ET AL.

dedication, advice, resources and mentoring. Although this promise carried a lot of weight with some, it was nevertheless no match against the reputations of established, central players in the industry. Indeed, Benchmark was shut out of a number of early deals they actively pursued. Nor was Benchmark able to depend heavily on its own network to gain access to central players’ promising deals; instead, the partners mostly hustled up their own deals. Of the six investments Benchmark made in 1995, four were ﬁrst round investments. Benchmark was the sole investor in three of these deals (Broadbase Software, Compact Devices, and CW Gourmet). The other three investments included PointCast Inc, Silicon Gaming, and Xantel. Both Silicon Gaming and PointCast were seeking a second round of funding and had been seeded by established VCs, while Xantel was seeking seed round funding. Kevin Harvey led the investments in Broadbase Software and PointCast, leveraging the social capital accrued from his experience founding his own successful software ﬁrm. Fig. 2 below contrasts each VC ﬁrm’s deal networks in 1995.

Firm’s Resource Endowments In addition to its founders’ initial social capital, a ﬁrm’s other resource endowments signiﬁcantly inﬂuence its ability to become central in industry networks. In the venture capital industry the amount of capital a ﬁrm can raise for its investment funds will shape the number and magnitude of the deals in which it can participate, as well as the extent to which it is able, and seen by entrepreneurs as able, to keep at hand ‘‘dry powder’’ – the capital necessary for participating in or leading later rounds of ﬁnancing for ﬁrms in its portfolio. The size of a VC ﬁrm’s fund is also often an indicator of its status and success in selling its capabilities to institutional investors, and one common standard for determining VC ﬁrm success is a pattern of raising larger investment funds over time (Gompers, 1995; Lee & Wahal, 2004). Both Benchmark and August were able to raise moderately large initial funds by the standards of 1995 (later, many funds became much larger). The initial Benchmark fund closed at $85 million while August’s ﬁrst fund was $98 million. It is interesting to note that some investors expected new ﬁrms trying to raise their ﬁrst funds to provide investors more generous terms than established funds in return for their participation, even if they have had prior relationships with the new VC ﬁrm’s partners and had invested with them at their previous ﬁrms. In 1995, the standard proportion of a fund’s proﬁts retained by the general partners in a venture fund was 20%, with

KPCB

Compact Devices

Broad Base

Fig. 2.

Xantel

One Liberty

TVI

Legend

-Portfolio Investment Company

-Prestigious Venture Capital Partner

-Venture Capital Partner

Print Paks

August

Hummer Winblad

-Focal Firm (Benchmark or August)

CW Gourmet

Mohr Davidow

1995 Deal Networks for Benchmark and August Capital.

Silicon Gaming

Benchmark

Point Cast

MPAE

Process of Developing Centrality in Venture Capitalist Deal Networks 577

578

BRET R. FUND ET AL.

only one superstar ﬁrm – Kleiner Perkins Cauﬁeld and Beyers – able to command a 30% ‘‘carry’’ from its investors (Stross, 2000). In an audacious move, Benchmark decided that it would also demand a 30% carry from its investors; they did, however, give this demand a creative twist. They agreed to accept a 20% carry until investors had fully earned back their initial investments and would also accept a declining management fee (the service fee VC ﬁrms take not from proﬁts but directly from investable capital) as the investments started to yield additional returns and the carry grew. They also put 3% of their own capital into the fund, which was triple the industry standard 1% (Stross, 2000). Despite these ‘‘performance-based’’ incentives for the VCs, some investors still balked. As a result, Benchmark’s brash behavior and failure to ‘‘know its place’’ as a new venture capital ﬁrm alienated some institutional investors. For example, the endowment manager at Stanford (the alma mater of several of the partners) refused to invest in their fund and successfully lobbied two other institutional investors to withdraw their commitments. Benchmark nonetheless successfully raised its ﬁrst fund on terms equivalent to those extracted by VC ﬁrms at the very pinnacle of the industry. As we discuss in greater detail below, Benchmark used this initial resource endowment effectively to begin building both its ties to other VC ﬁrms and a portfolio of high potential ﬁrms while August continued to engage in its safer and more measured pursuit of investment opportunities.

Firm’s Afﬁliations with Prestigious Industry Members Gaining legitimacy via the institutional support of powerful external players can be an important determinant of future network formation opportunities for new ﬁrms (Baum & Oliver, 1992; Miner, Amburgey, & Stearns, 1990). Within the VC industry, venture capitalists often collaborate with other VC ﬁrms when seeking ﬁnancing for potential portfolio investments. Over time, networks of VC ﬁrms tend to develop in a manner similar to that described by Gulati and Gargiulo for alliance formations (1999). These networks are often referred to as ‘‘VC syndicates’’ and provide access to novel information and lucrative opportunities (Podolny, 2001; Sorenson & Stuart, 2001). Not surprisingly, gaining entrance to a prominent syndication network can be quite difﬁcult for new VC ﬁrms. One way to break into these networks is to be invited by established or central players. These established players may be willing to welcome a non-central partner if the invitee offers unique resources not available from other available partners (Ahuja, 2000;

Process of Developing Centrality in Venture Capitalist Deal Networks

579

Mitchell & Singh, 1992). Invitations to partner with prominent VCs can not only directly affect a new VC’s future network formation opportunities, they can also change the existing network structure for the ﬁrm, thereby moderating the original relationships between existing network structure and future network formation opportunities. As the two ﬁrms entered their second year, August continued its more conservative approach and made no additional investments in the ﬁrst three months of 1996. It appeared, rather, that the August partners continued to work with ventures they knew from their TVI days but in which August had not yet made investments. Finally, in April of 1996, August invested along with six other VC ﬁrms in Be, Inc., a company that TVI originally funded in 1992. In contrast, Benchmark wasted no time in putting together deals and creating a wide set of investment network relationships for their ﬁrm. Benchmark made its ﬁrst investment of 1996 on January 1st – a second round investment in Xantel, which was completed a scant month and a half after the seed funding round. This second round also provided an opportunity to seek out other potential investment partners and Benchmark ended up co-investing with two additional VC ﬁrms (St. Paul Ventures and Lightspeed Ventures). Barely pausing between investments, by the end of the ﬁrst quarter of 1996 Benchmark had invested a total of $41 million in seven companies (three of which they initially funded in 1995) and had partnered with more than ten additional VC ﬁrms. Thus, while Benchmark was hustling riskier but more affordable investments in earlier stage companies with big upside potential, August was carefully seeking later-stage deals that required more capital but bore lower risks per dollar of investment. The remaining three quarters of 1996 would see this pattern continue, as Benchmark invested aggressively (still mostly in early stage ventures) while August maintained its more measured approach. Benchmark would ultimately invest in 18 companies in 1996 for a total outlay of $92 million. Benchmark’s ﬁrst $85 million fund had been completely invested by September of 1996, providing a brief pause in investing activity as the partners began to raise their second fund. The second fund would not be completed until early 1997, but that did not stop Benchmark from making a few small seed round investments in December 1996. August ultimately invested in ﬁve companies during 1996 for a total outlay of $30 million, less than a third of the fund it had raised in 1995. Fig. 3 contrasts Benchmark’s and August’s 1996 deal networks. Benchmark’s urgency and effort resulted in a series of rapid-ﬁre investments and an extraordinary increase in the number of relationships

STAR

AVI

Alta CA Partners

Be, Inc.

Fig. 3.

CV Soffinova

Draper Fischer

Jamba Juice Company

Aperture Associates

Oak Investments

TVI

August

Newtek

Walden

Artisan Components, Inc

US Ventures

Crosspoint Ventures

New Enterprise Assc

Technology Crossover

-Prestigious Venture Capital Partner

-Venture Capital Partner

-Portfolio Investment Company

-Focal Firm (Benchmark or August)

Legend

St. Paul Ventures

Lightspeed Ventures

1996 Deal Networks for Benchmark and August Capital.

Pacven Walden

Crosspoint

PointCast PictureVision, Inc. EnCompass, Inc. Mohr Davidow Institutional Venture Partners Wink Communications, Inc. Snap Track, Inc. Broad Base Mayfield Wilson, Sonsini, Goodrich & Rosati Epigram, Inc. Benchmark Allegis Corporation KPCB Juniper Networks, Inc. Advanced Tech Ventures Silicon Gaming Ariba, Inc Genesys Telecom Labs, Inc. Blue Coat Systems, Inc Integral Capital Xantel

Knight Ridder

Rosewood Capital Trinity Ventures

Mobius Venture Capital Rustic Canyon Ventures

Adobe Ventures

ULSI Systems, Inc. Tumbleweed Communications Apache Systems

Robertson Stevens

TVM

Burr, Egan, Deleage

Associated Ventures

InterWest

580 BRET R. FUND ET AL.

581

Process of Developing Centrality in Venture Capitalist Deal Networks

with prominent players. As Figs. 1 and 2 show, Benchmark’s founders were able to leverage their individual-level social capital and the large set of deals they put together to attract high-proﬁle co-investors. For example, Silicon Gaming had received its ﬁrst round of funding ($1 million) from TVI, where Bob Kagle had managed the investment. After co-founding Benchmark, Kagle continued to work with Silicon Gaming and used funds from Benchmark’s ﬁrst fund to provide a second round of ﬁnancing in 1995. Silicon Gaming carried some status of its own; in part because of TVI’s prior investment, Kagle was able to co-invest with Kleiner, Perkins on the deal, creating the ﬁrst linkage between Benchmark and the 800-pound gorilla of the industry. Beginning in 1996, Benchmark was starting to co-invest with a number of central and high-status venture capital ﬁrms. For example, they partnered several more times with Kleiner, Perkins (Accept.com, Handspring, Impresso ,and others), Sequoia (Scient, Mahi Networks, WebVan), and Greylock (NorthPoint, RedHat, Send.com), among others. Although not as active as Benchmark, August also made investments alongside several high proﬁle venture capital ﬁrms during the period of this study, including New Enterprise Associates (Atheros Communications, Be Inc., and Seagate Technology Holdings), Sequoia (ebates, Shopping.com), and Kleiner, Perkins (Escalade Corporation and VLSI Technology). In absolute numbers of relations formed Benchmark was more promiscuous; however, on a percentage basis Benchmark and August partnered with high status VC ﬁrms at approximately equal rates. Of the 100 ﬁrms that Benchmark invested in by the end of 2000, 46% also received ﬁnancing from prominent and central players in the industry, while the comparable ﬁgure for August was 18 of 44, or 41% of its deals. Further examination of this data yields some interesting insights. Table 1 presents information (compiled from VentureXpert) on total partner and prestigious partner4 formations for each VC ﬁrm on an annual basis. The Table 1.

Number of VC Partners by Year. 1995

August (Prestigious Partners) 1 (0) Benchmark (Prestigious Partners) 5 (3)

1996

1997

1998

16 (2) 35 (9)

35 (8) 40 (11)

17 (2) 53 (14)

Prestigious Partners as Percentage of Total Partners August (%) 0.0 12.5 Benchmark (%) 60.0 25.7

22.9 27.5

11.8 26.4

1999 93 (13) 163 (17) 14.0 10.4

2000 75 (6) 155 (8) 8.0 5.2

582

BRET R. FUND ET AL.

data in this table reveal that during the ﬁrst four years of each ﬁrm’s existence, Benchmark generally partnered with twice as many prestigious venture capital ﬁrms as August on a percentage basis and had more prestigious partners in every year on an absolute basis. In Table 2, we aggregated the total number of deals that each ﬁrm did with prestigious VC ﬁrms. Combined, Benchmark and August partnered with 27 of the 47 prestigious VC ﬁrms on Pollock and colleagues’ list of prestigious VCs

Table 2.

Building Relationships with Prestigious VC Firms.

Prestigious VC Partners Accel Ventures Advent Alta Asset Management Austin Ventures Brentwood Associates Canaan Capital Centennial Charles River Frontenac Greylock Highland Capital Institutional Ventures InterWest Partners KPCB Matrix Partners Mayﬁeld MPAE Mohr, Davidow New Enterprise Norwest Oak Investment Sequoia Sigma Partners Sierra Ventures JH Whitney WPG Ventures

Benchmark August 5

1 4 2

1 1 4 1 1 1

2 6 1 5 1 9 2 1 1 5 2 1 3 8

1 5 1 1 3 1 1 2 1 3 1 1

1 1

Total Deals with Prestigious VCs

61

30

Number of Unique Prestigious VCs Partners

20

18

Percentage of Prestigious VCs with whom Firm did Multiple Deals (%)

60

28

Process of Developing Centrality in Venture Capitalist Deal Networks

Fig. 4.

583

Prestigious and Non-Prestigious Partners by Year.

(Pollock, Chen, Jackson, & Hambrick, 2007). However, Benchmark appears to have been more focused on developing stronger ties. Benchmark formed multiple partnerships with 60% of the 20 prestigious VC ﬁrms it partnered with during its ﬁrst ﬁve years. In contrast, August, which began with more founder social capital, focused less on building strong ties with prestigious actors; it only formed multiple deal networks with 28% of its 18 prestigious partners. Thus, Benchmark was more proactive than August in developing embedded ties (Uzzi, 1996) with prestigious ﬁrms, especially during its early years. However, once its network position had become more established, Benchmark focused less on overcoming its initial status deﬁcits and began to partner with a wider variety of less prestigious ﬁrms. Fig. 4 graphically presents the pattern of prestigious and non-prestigious tie formations by August and Benchmark annually between 1995 and 2000.

Firm’s Visibility-Enhancing Performance Firms must often rely on the resources generated by their founders’ social capital and effort until they have an actual performance record to which they can point (Starr & MacMillan, 1990). As a venture is able to achieve visibility via successful performance it should also gain access to more future

584

BRET R. FUND ET AL.

opportunities, thereby increasing future performance prospects. Pollock and Gulati (2007) note that the ability to achieve attention and generate hype, or ‘‘buzz,’’ was essential for technology start-ups during this period to become successful. They quoted one tech company CEO as stating that ‘‘ﬁrms had to create their own self-fulﬁlling prophecies’’ (Pollock & Gulati, 2007, p. 346), and that an obvious way to accomplish this is through outsized market performance. It stands to reason that such visibility-enhancing performance would also be important for new venture capital ﬁrms (Gompers, 1995; Lee & Wahal, 2004). With all this talk of effort, status, endowments, and positioning, it is easy to forget that hitting a home run and making a lot of money counts for a great deal in the world of VC investments! The reputational advantages that accrue to VCs involved with IPOs attract both co-investors and entrepreneurs alike, and VCs – especially during hot markets, such as existed in the late 1990s – have typically viewed IPOs as the ultimate, and most proﬁtable, endgame for a portfolio company (Guler, 2007). Indeed, in a recent study conducted using IPO data from the late 1980s and early 1990s, Guler (2007) found that the average return to VCs on investments in companies that eventually went public was 1,608%, compared to a return of 446% on acquisitions. Other studies indicate that just 10–30% of VC investments result in IPOs, and that the top 10% of a VC’s investments account for 62% of the venture fund’s total return (Fenn, Liang, & Prowse, 1997; Guler, 2007; Scherer, Harnoff, & Kukies, 2000). Thus, to the extent a VC ﬁrm’s efforts allow it to unearth and/or develop a ﬁrm that becomes a home run investment, it enhances its ability to become more visible and successful, and hence cognitively central and available, to other industry actors. August’s ﬁrst two funds (with a combined total of $300 million) were fully invested in 34 companies by 1999. Overall, August invested in 44 companies from 1995 to 2000 with an average investment of $6.8 million. Among these companies were big names such as Epinions.com, Cobalt Networks, and Be, Inc. As Fig. 5 shows, during our period of study seven of August’s investments underwent initial public offerings (IPOs). The median return for the seven ﬁrms that August took public was 585%. Their two most successful IPOs during this period were Cobalt Networks and Silicon Image. August’s investment of $10 million in Cobalt Networks was worth $336 million at the end of the day Cobalt went public – a 3,360% return. Silicon Image was similarly successful; August’s $8.3 million investment in this ﬁrm was worth $119 million after the ﬁrst day of trading, generating a 1,444% return.

Process of Developing Centrality in Venture Capitalist Deal Networks

585

24 21 20

16

12

8

7

4

0 August

Fig. 5.

Benchmark

IPO Activity, 1995–2000.

Benchmark likewise experienced a number of successes among its portfolio companies. Of the 100 ﬁrms Benchmark invested in during the 1995–2000 period, 21 underwent IPOs. The median return from the IPOs of these 21 ﬁrms was 3,148%. Among Benchmark’s most notable successes was the online auction service eBay, widely regarded as one of the best-performing venture investments of all time. In 1997, Benchmark invested $2.6 million from its $125 million Benchmark II fund into eBay, obtaining a 22% share in the company. On the day of its IPO in 1998 this stake was worth $414.4 million, representing a 15,707% return on investment. By early 1999, Benchmark’s stake had increased in value approximately six times since the IPO and was worth $2.5 billion. Although eBay was its most famous investment during this period, it was not its biggest winner. That honor goes to the $1.5 million it used to purchase a 9.9% stake in NorthPoint Communications, which was worth $413.4 million at the close of the day it went public – a stunning 32,055% return on investment. These and other early successes allowed Benchmark to raise its $175 million Benchmark III fund in 1998 and its $1 billion Benchmark IV fund in 1999. Fig. 5 compares the IPO activity of each VC ﬁrm during the period under study.

586

BRET R. FUND ET AL.

Although we do not want to focus excessively on Benchmark’s big eBay hit, its urgency and effort clearly played a role in this ﬁrm’s success. eBay chose Benchmark over other VC ﬁrms in part because of its ‘‘high touch’’ approach and involvement with its portfolio ﬁrms, and also because of its ability to recruit top ﬂight executive talent. It is clear that eBay would not have attracted Meg Whitman as their CEO without the efforts and contacts of Benchmark partner David Bierne. Benchmark cannot claim total credit for eBay’s success, however; other factors, including some measure of luck also played a role. For example, no one could have grasped the role the eBay community would play in helping it fend off competition from more established ﬁrms like Amazon and Yahoo (Stross, 2000). Although eBay is the most prominent example, overall Benchmark was able to successfully leverage its initial resource endowments by engaging in sustained efforts that allowed it to overcome its initial status disadvantages and position itself as at least August’s equal. By the end of the period of our study, both ﬁrms had established themselves as legitimate players in the venture capital industry. Today, Benchmark and August hold $3 billion and $1.3 billion, respectively, under management. In light of the preceding discussion, it is useful to gauge how successful each VC ﬁrm ultimately was in achieving structural centrality within the overall VC deal network. Using VentureXpert, we collected data on all 112 VC ﬁrms founded in 1995 and traced the syndication activity of each ﬁrm over the entire period under study (1995–2000). We recorded symmetric ties as existing when a focal ﬁrm partnered with another VC ﬁrm in an investment syndicate. The aggregate networks for each year thus consisted of the 1995 ‘‘class’’ of VCs and all other VCs with whom they invested. Degree centrality scores were then calculated for each ﬁrm. Fig. 6 traces the evolution of August’s and Benchmark’s structural centrality. For sake of comparison, we note the structural centrality scores each year for the 75th percentile of the class of 1995 (August’s and Benchmark’s peers), which represents a reasonable benchmark for success. The graph indicates that both August and Benchmark were clearly at the top of their class in terms of achieving prominent and central positions in their industry network.

DISCUSSION AND CONCLUSION Social structure, including network structure, has been described as the locus of both opportunity and constraint (Bourdieu & Passeron, 1990; Dagnino et al., 2008; Giddens, 1984; Rowley & Baum, 2008). But the

587

Process of Developing Centrality in Venture Capitalist Deal Networks Degree Centrality 200 180 160 140 120 100 80 60 40 20 0 1995 August

Fig. 6.

1996

1997

Benchmark

1998

1999

2000

75th Percentile of 1995 Peer Group

Evolution of Structural Centrality, 1995–2000.

relationship between opportunity and constraint is not purely structural; it is also cognitive, social psychological, and behavioral (Madhavan et al., 2008; van Liere et al., 2008). In this chapter, we have suggested that the founders’ sense of time urgency and their willingness to be assertive in taking risks and exerting substantial effort play a strong role in shaping the meaning and inﬂuence of four important factors – founders’ social capital, initial resource endowments, prestigious afﬁliations, and visibilityenhancing performance – on the overall success and future viability of a new venture. These effects are cognitive in the sense that they depend on how the founders think about what they are doing and how they assess their current situation. It appears that the elite social position of August’s founders prompted them to take a careful and deliberate approach to managing their ﬁrm. Not only were they more circumspect in the investments they made, they also exhibited less concern with actively developing and strengthening ties with prestigious VC ﬁrms, perhaps because they felt access to these prestigious ﬁrms could largely be taken for granted and would be available when needed given the speciﬁc status signals offered by their past accomplishments at TVI (Bunderson, 2003). In contrast, we suspect that

588

BRET R. FUND ET AL.

Benchmark’s founders perceived their somewhat less-prestigious status as an opportunity to engage in audacious goal-setting and frame-breaking behaviors that increased their cognitive centrality within the industry. Rather than enacting the resource limitations suggested by their ‘‘objective’’ endowments (Baker & Nelson, 2005; Weick, 1979), Benchmark’s founding partners consciously structured their ﬁrm and initial investment fund in ways that belied the ﬁrm’s newness to the industry. Thus Benchmark became cognitively central, it seems to us, largely because it took assertive actions (Bunderson, 2003; Dagnino et al., 2008) and acted as if it was already a dominant, central player in the industry, or at least was destined to claim such a position. Perhaps the most prominent example of this was Benchmark’s boldness in claiming a 30% carry, which was previously charged only by Kleiner Perkins, the titan of the industry. Benchmark’s aggressiveness in repeatedly pursuing investment partnerships with prominent, established VC ﬁrms further demonstrated their own belief in their ability to play at the top of their industry. The relationship between opportunity and constraint in this narrative is also social-psychological because of the way in which the contrasting approaches of the two new VC ﬁrms apparently shaped the perceptions of others. Consistent with the Matthew Effect, August’s success is truly ‘‘the rich getting richer.’’ Forgive us for ignoring the already substantial wealth of the Benchmark founders – it’s a matter of context here. Others appear to have interpreted the speciﬁc status characteristics of August’s founders in ways that provided them access to deals with other high status actors at later stages of investment. In contrast, Benchmark’s success illustrates another Robert Merton conceptualization, the self-fulﬁlling prophecy (Merton, 1968b). In Benchmark’s case, it appears that investors, entrepreneurs, and other VCs responded in a manner that supports an alternative claim to the Matthew Effect: those who act like they have, get. Benchmark’s refusal to play the role of an underdog or ‘‘rookie,’’ including its aggressive pursuit of access to the best deal syndicates and its demands for a 30% carry illustrate how its self-perception became its enacted reality. This point also serves to reinforce the notion that cognitive centrality may precede, rather than simply mediate, structural centrality. For example, Benchmark’s demand for a 30% carry may have served as an indirect status signal to others who used it as an indicator of Benchmark’s performance capabilities. In doing so, Benchmark became a more cognitively available and central player, despite its lack of structural network centrality. Further, Benchmark’s efforts to identify and invest in lucrative deals and then invite

Process of Developing Centrality in Venture Capitalist Deal Networks

589

other prestigious VCs to invest in later rounds was another way of displaying expertise in deal ﬁnding and making, which are critical skills in the venture capital industry. This too may have served to increase its cognitive availability and centrality. This growing visibility and reputation for deal-making expertise then generated subsequent opportunities to collaborate with other high status VCs, thereby improving Benchmark’s structural centrality within the network. Finally, the relationship between opportunity and constraint is behavioral in the sense that although Benchmark could easily have ﬁgured it was constrained by its resource limitations and thus moved slowly and carefully to discover whatever narrow advantages its idiosyncratic experiences, talents, resources and position might have provided, it instead acted in ways that demonstrated little regard for objective constraints and social standing. In this sense, our chapter joins a growing chorus of voices arguing that resources and resource constraints are often not conducive to objective valuation, but are – to an extent that is often difﬁcult to predict – dependent upon the internal vision and subjective valuation of ﬁrms themselves (Baker & Nelson, 2005; Mishina, Pollock, & Porac, 2004; Penrose, 1959). Indeed, as Brass and Burkhardt (1993, p. 466) note, ‘‘strategic action can be used to compensate for relatively weak resources. Skillful political activity is one tool for overcoming a lack of resources or making less valuable resources more potent. Actors in powerful positions, who control ample resources, are less dependent on their capabilities to use resources strategically than are actors who lack ample resources.’’ Compared to established and prominent players, Benchmark did not have a wealth of ﬁnancial or social resources. It was, however, able to overcome its structural constraints through careful and deliberate actions and choices (and some luck). The venture capital industry may be a particularly good context for future study of these issues, as it continues to be home to a rare combination of entrenched, privileged elites and aggressive newcomers of all stripes, all seeking central positions within the industry network and the privileges such positions bring. In this chapter we set out to explore how new ﬁrms can become central players in industry networks. We developed and illustrated a process model of centrality development and illustrated its applicability using a case study of two venture capital ﬁrms. Future research using other data and methodological approaches should continue to explore the relationship between cognitive and structural centrality, as well as the forces and dynamics that drive network evolution and change over time.

590

BRET R. FUND ET AL.

NOTES 1. Much of the material for this section is drawn from the historical review of the industry provided by Gompers and Lerner (2006). 2. Based on data obtained from Thomson Financial’s VentureXpert database. 3. As the general partners of the venture funds, VCs typically receive an annual fee equal to 1.5–3% of the capital under investment in the fund, plus a speciﬁed percentage of the total proﬁts, or ‘‘carried interest’’ (referred to within the industry as ‘‘the carry’’) generated by the fund. The rest of the proﬁts are distributed to the fund’s limited partners. 4. Prestigious VC ﬁrms were identiﬁed based on the list compiled by Pollock et al. (2007), who identiﬁed the VC ﬁrms that raised the 10 largest funds annually between 1990 and 1995. The ﬁnal list included 47 VC ﬁrms.

ACKNOWLEDGMENTS This chapter has beneﬁted enormously from the helpful comments of Joel Baum, Tim Rowley, participants in the Penn State ORG working group, and our co-contributors who attended the Rotman School of Business Advances in Strategic Management Paper Development Workshop.

REFERENCES Adler, P. S., & Kwon, S.-W. (2002). Social capital: Prospects for a new concept. Academy of Management Review, 27(1), 17–40. Ahuja, G. (2000). The duality of collaboration: Inducements and opportunities in the formation f interﬁrm linkages. Strategic Management Journal, 21, 317–343. Baker, T., & Nelson, R. E. (2005). Creating something from nothing: Resource construction through entrepreneurial bricolage. Administrative Science Quarterly, 50(3), 329–366. Batchelder, E. (2002). Comparing three simultaneous measurements of a sociocognitive network. Social Networks, 24(3), 261–277. Baum, J. A. C., & Oliver, C. (1992). Institutional embeddedness and the dynamics of organizational populations. American Sociological Review, 57, 540–559. Bolino, M. C., Turnley, W. H., & Bloodgood, J. M. (2002). Citzenship behavior and the creation of social capital in organizations. Academy of Management Review, 27, 505–522. Bourdieu, P., & Passeron, J. C. (1990). Reproduction in education, society, and culture. London: Sage. Brass, D. J., & Burkhardt, M. E. (1993). Potential power and power use: An investigation of structure and behavior. Academy of Management Journal, 36(3), 441–470. Bunderson, J. S. (2003). Recognizing and utilizing expertise in work groups: A status characteristics perspective. Administrative Science Quarterly, 48, 557–591.

Process of Developing Centrality in Venture Capitalist Deal Networks

591

Chung, S., Singh, H., & Lee, K. (2000). Complementarity, status similarity, and social capital as drivers of alliance formation. Strategic Management Journal, 21(1), 1–22. Dagnino, G. B., Levanti, G., & Li Destri, A. M. (2008). Evolutionary dynamics of interﬁrm networks: A complex systems perspective. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 67–129). Bingley, UK: JAI/Emerald Group. Emirbayer, M., & Goodwin, J. (1994). Network analysis, culture, and the problem of agency. American Journal of Sociology, 99(6), 1411–1454. Fenn, G. W., Liang, N., & Prowse, S. (1997). The private equity market: An overview. Financial Markets, Institutions & Instruments, 6(4), 1–106. Fischer, H. M., & Pollock, T. G. (2004). Effects of social capital and power on surviving transformational change: The case of initial public offerings. Academy of Management Journal, 47(4), 463–481. Freeman, L. C. (1979). Centrality in social networks: Conceptual clariﬁcation. Social Networks, 1(3), 215–239. Giddens, A. (1984). The constitution of society: Outline of the theory of structuration. Cambridge: Polity Press. Gompers, P. A. (1995). Optimal investment, monitoring, and the staging of venture capital. Journal of Finance, 50(5), 1461–1489. Gompers, P. A., & Lerner, J. (2006). The venture capital cycle (2nd ed.). Cambridge, MA: MIT Press. Gulati, R. (1995). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R. (1998). Alliances and networks. Strategic Management Journal, 19(4), 293–317. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1439–1493. Guler, I. (2007). Throwing good money after bad? Political and institutional inﬂuences on sequential decision making in the venture capital industry. Administrative Science Quarterly, 52(2), 248–285. Hoffman, A., & Ocasio, W. (2001). Not all events are attended equally: Toward a middle range theory of industry attention to external events. Organization Science, 12(4), 414–434. Hsu, D. H. (2004). What do entrepreneurs pay for venture capital afﬁliation? Journal of Finance, 59(4), 1805–1844. Humphreys, P., & Berger, J. (1981). Theoretical consequences of the status characteristics formulation. American Journal of Sociology, 86, 953–983. Inkpen, P., & Beamish, P. (1997). Knowledge, bargaining power, and the instability of international joint ventures. Academy of Management Review, 22, 177–202. Jain, B. A., & Kini, O. (2000). Does the presence of venture capitalists improve the survival proﬁle of IPO ﬁrms? Journal of Business Finance and Accounting, 27, 1139–1183. Kameda, T., Ohtsubo, Y., & Takezawa, M. (1997). Centrality in sociocognitive networks and social inﬂuence: An illustration in a group decision-making context. Journal of Personality and Social Psychology, 73(2), 296–309. Kim, T., Oh, H., & Swaminathan, A. (2006). Framing interorganizational network change: A network inertia perspective. Academy of Management Review, 31(3), 704–720. Kindleberger, C. P., & Aliber, R. Z. (2005). Manias, panics, and crashes: A history of ﬁnancial crises. New York: Wiley.

592

BRET R. FUND ET AL.

Koka, B. R., Madhavan, R., & Prescott, J. E. (2006). The evolution of interﬁrm networks: Environmental effects on patterns of network change. The Academy of Management Review, 31(3), 721–737. Koka, B. R., & Prescott, J. (2002). Strategic alliances as social capital: A multidimensional view. Strategic Management Journal, 23, 795–816. Krackhardt, D. (1990). Assessing the political landscape: Structure, cognition, and power in organizations. Administrative Science Quarterly, 35(2), 342–369. Kuran, T., & Sunstein, C. (1999). Availability cascades and risk regulation. Stanford Law Review, 51, 683–768. Lawrence, P. R., & Lorsch, J. W. (1967). Differentiation and integration in complex organizations. Administrative Science Quarterly, 12(1), 1–47. Lee, P. M., & Wahal, S. (2004). Grandstanding, certiﬁcation and the underpricing of venture capital backed IPOs. Journal of Financial Economics, 73(2), 375–407. van Liere, D., Koppius, O., & Vervest, P. (2008). Strengthening of bridging positions: Network horizon and network horizon heterogeneity. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 595–639). Bingley, UK: JAI/Emerald Group. Lin, N. (1999). Building a network theory of social capital. Connections, 22(1), 28–51. Madhavan, R., Caner, T., Prescott, J., & Koka, B. (2008). Networking as antecedent to network structure. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 457–501). Bingley, UK: JAI/Emerald Group. Markowitz, H., Elton, E. J., & Gruber, M. J. (1979). Portfolio theory, 25 years after: Essays in honor of Harry Markowitz. Amsterdam: Elsevier. Merton, R. K. (1968a). The Matthew Effect in Science. Science, 159, 56–63. Merton, R. K. (1968b). Social theory and social structure. New York: Free Press. Miner, A. S., Amburgey, T. L., & Stearns, T. M. (1990). Interorganizational linkages and population dynamics: Buffering and transformational shields. Administrative Science Quarterly, 35, 689–713. Mishina, Y., Pollock, T. G., & Porac, J. F. (2004). Are more resources always better for growth? Resource stickiness in market and product expansion. Strategic Management Journal, 25, 1179–1197. Mitchell, W., & Singh, K. (1992). Incumbents’ use of pre-entry alliances before expansion into new technical subﬁelds of an industry. Journal of Economic Behavior and Organization, 18(3), 347–372. Mizruchi, M., & Galaskiewicz, J. (1994). Networks of organizations. In: S. Wasserman & J. Galaskiewicz (Eds), Advances in social network analysis (pp. 230–254). Thousand Oaks, CA: Sage. Neergaard, H., & Madsen, H. (2004). Knowledge intensive entrepreneurship in a social capital perspective. Journal of Enterprising Culture, 12(2), 105–125. Nohria, N. (1992). Is a network perspective a useful way of studying organizations? In: N. Nohria & R. C. Eccles (Eds), Networks and organizations: Structure, form, and action (pp. 1–22). Boston, MA: Harvard Business School Press. Padgett, J. F., & Ansell, C. F. (1993). Robust action and the rise of the Medici: 1400–1434. American Journal of Sociology, 98, 1259–1319. Penrose, E. G. (1959). The theory of the growth of the ﬁrm. New York: Wiley. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations. New York: Harper & Row.

Process of Developing Centrality in Venture Capitalist Deal Networks

593

Podolny, J. M. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107(1), 33–60. Pollock, T. G., Chen, G., Jackson, E. M., & Hambrick, D. C. (2007). Just how much prestige is enough? Assessing the value of multiple types of prestigious afﬁliates for young ﬁrms. Working Paper, Pennsylvania State University. Pollock, T. G., & Gulati, R. (2007). Standing out from the crowd: The visibility-enhancing effects of IPO-related signals on alliance formation by entrepreneurial ﬁrms. Strategic Organization, 5(4), 339. Pollock, T. G., Porac, J. F., & Wade, J. B. (2004). Constructing deal networks: Brokers as network ‘‘architects’’ in the US IPO market and other examples. Academy of Management Review, 29(1), 50–72. Pollock, T. G., Rindova, V. P., & Maggitti, P. G. (2008). Market watch: Information and availability cascades among the media and investors in the U.S. IPO market. Academy of Management Journal, 51(2): 335–358. Rowley, T. J., & Baum, J. A. C. (2008). The dynamics of network strategies and positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 641–671). Bingley, UK: JAI/Emerald Group. Sahlman, W. A. (1990). The structure and governance of venture capital organizations. Journal of Financial Economics, 27(2), 473–521. Sapienza, H. J., Manigart, S., & Vermeir, W. (1996). Venture capitalist governance and value added in four countries. Journal of Business Venturing, 11, 439–469. Scherer, F. M., Harnoff, D., & Kukies, J. (2000). Uncertainty and the size of distribution of rewards from innovation. Journal of Evolutionary Economics, 10, 175–200. Selznick, P. (1948). Foundations of the theory of organization. American Sociological Review, 13(1), 25–35. Sorenson, O., & Stuart, T. E. (2001). Syndication networks and the spatial distribution of venture capital investments. American Journal of Sociology, 106(6), 1546–1588. Starr, J. A., & MacMillan, I. C. (1990). Resource co-optation via social contracting: Resource acquisition strategies for new ventures. Strategic Management Journal, 11, 79–92. Stross, R. E. (2000). eBoys. New York: Crown. Suchman, M. C. (1995). Managing legitimacy: Strategic and institutional approaches. Academy of Management Journal, 20(3), 571–610. Thompson, J. D. (1967). Organizations in action. New York: McGraw-Hill. Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207–232. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61(4), 674–698. Wasserman, S., & Faust, K. (1999). Social network analysis: Methods and applications. New York: Cambridge University Press. Weick, K. E. (1979). The social psychology of organizing. Reading, MA: Addison. Yin, R. K. (2003). Applications of case study research (2nd ed.). Thousand Oaks, CA: Sage.

NETWORK HORIZON: AN INFORMATION-BASED VIEW ON THE DYNAMICS OF BRIDGING POSITIONS Diederik W. van Liere, Otto R. Koppius and Peter H. M. Vervest ABSTRACT We propose an information-based view of the dynamics of network positions and use it to explain why bridging positions become stronger. We depart from previous network dynamics studies that implicitly assume that ﬁrms have homogenous information about the network structure. Using network experiments with both students and managers, we vary a ﬁrm’s network horizon (i.e., how much information a ﬁrm has about the network structure) and the network horizon heterogeneity (i.e., how this information is distributed among the ﬁrms within the network). Our results indicate that ﬁrms with a higher network horizon occupy a stronger bridging position, especially under conditions of high network horizon heterogeneity. At a more general level, these results provide an indirect validation of what so far has been an untested assumption in interﬁrm network research, namely that ﬁrms are aware of their position in the overall network and consciously attempt to improve their position. Network Strategy Advances in Strategic Management, Volume 25, 595–639 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25017-5

595

596

DIEDERIK W. VAN LIERE ET AL.

The burgeoning literature on the effects of network position on performance suggests that certain network positions can confer at least a temporary competitive advantage (Burt, 1992; Powell, Koput, Owen-Smith, & SmithDoerr, 1999), especially due to its complex, causally ambiguous nature (Wilcox King, 2007). However, empirical observation shows that networks change (Baum, Shipilov, & Rowley, 2003; Powell, White, Koput, & OwenSmith, 2005), not just through exogenous events (Madhavan, Koka, & Prescott, 1998), but also endogenously as a result of purposeful ﬁrm action to improve their position and achieve a competitive advantage (Fund, Pollock, Baker, & Wowak, 2008; Rowley & Baum, 2008). For instance, studies in the airline and telecommunication industries show that ﬁrms respond to alliances of competitors by announcing alliances of their own aimed at countervailing the effects of the competitors’ alliances (Gimeno, 2004; Gulati, 1998). Given these and other network dynamics, how can we then understand why some ﬁrms are able to maintain and strengthen advantageous network positions, whereas others do not have such positions or are not able to hold on to them? In this chapter, we offer one possible explanation for the case when competitive advantage results from occupying a position rich in structural holes, that is a bridging position (Burt, 1992). The departure point of our explanation is that in order to achieve competitive advantage, ﬁrms need to have accurate expectations of future resource value (Barney, 1986), because such information is crucial for discerning entrepreneurial opportunities (Burt, 1992; Kirzner, 1997; Ozgen & Baron, 2007). In a networked setting, entrepreneurial opportunities for competitive advantage are presented by occupying certain speciﬁc network positions in the overall network structure (Burt, 1992, 2000). Thus, having information about the network structure enables a ﬁrm to locate and obtain valuable network positions (Gulati & Gargiulo, 1999) before others do and potentially achieving a competitive advantage. We therefore introduce the construct of network horizon that describes the information that a ﬁrm has about the network structure it is embedded in. We expect ﬁrms with an extensive network horizon to be able to maintain their favorable positions longer. Furthermore, since ﬁrms in general can be expected to differ in their expectations of future resource value (Makadok & Barney, 2001), speciﬁcally in a network setting this implies that different ﬁrms will have different network horizons. These information differences will translate into different entrepreneurial opportunities being perceived (Denrell, Fang, & Winter, 2003), and hence ﬁrms with

Network Horizon

597

different network horizons will strive for different network positions. The resulting dynamics are likely to affect the dynamics of ﬁrms’ network positions, and we therefore introduce the construct of network horizon heterogeneity that describes how the network horizon differs across ﬁrms. Previous work on the dynamics of network positions suggests that ﬁrms may shift their position for instance due to resource dependence motives (Pfeffer & Salancik, 1978), competitive pressure (Bae & Gargiulo, 2004) or competition within (Gimeno, 2004) or between networks (Gomes-Casseres, 1996). These changes in the network structure reﬂect competitive dynamics between ﬁrms that are pursuing a competitive advantage. All of these explanations have the assumptions in common that ﬁrms are aware of the changes in the network, have information about their potential partner ﬁrms and detect opportunities in the network to shift their network position. Investigating what happens when we allow for interﬁrm differences in these information assumptions is the main focus of this chapter. This is also what differentiates us from recent research that looked at the dynamics and sustainability of network positions – structural holes in particular (Bala & Goyal, 2000; Buskens & van de Rijt, 2007; Sorenson & Ryall, 2007), who investigate different inﬂuences on network dynamics, but assume that all actors have complete information of the network. Our focus on information in the form of network horizon and network horizon heterogeneity can add substantial empirical and theoretical insight to this emerging literature for several reasons. First, we have a limited understanding to what extent organizational-decision makers are actually aware of the networks their ﬁrms are embedded in (Rowley & Baum, 2004), so studying this information aspect is an important step to understand the causal mechanisms behind the dynamics of network positions. Second, awareness of and information about the network structure is important because it is a ﬁrst prerequisite to respond to or anticipate changes that will affect ﬁrm strategy (Chen, 1996; Daft & Weick, 1984). Without such information it is not possible for a ﬁrm to respond in a purposeful manner to changes in its interﬁrm network. Network horizon enables the identiﬁcation of a consideration set of potentially valuable network opportunities, from which a ﬁrm can subsequently derive its network strategy. Third, the dynamics through which interﬁrm networks change over time are not yet very well understood (Baum, Rowley, Shipilov, & Chuang, 2005; Gnyawali & Madhavan, 2001), and the information based view that we propose can

598

DIEDERIK W. VAN LIERE ET AL.

shed light on how ﬁrms come to obtain or maintain a particular network position. Finally, while the assumption that ﬁrms are aware of their network and strategize their network position accordingly would seem eminently plausible to an interﬁrm network researcher (and indeed this very volume is based on the assumption that such a thing as a network strategy exists), it may not be as obvious as that. Not unlike Honda’s entry into the U.S. motorcycle market (Pascale, 1984), retrospectively looking at ﬁrms from a predetermined network strategy lens potentially risks seeking a strategy where it was not (Inkpen & Choudhury, 1995). Do ﬁrms really map their network, choose a speciﬁc network position and aim their strategy towards it? Our chapter provides an indirect test of this assumption by examining its consequences. If ﬁrms exhibit different strategic behavior, yet they differ only in information that would be relevant for a network strategy but not for other strategy aspects, then this implies that ﬁrms use that purely network-speciﬁc information (such as the overall structure of the network) in formulating their strategy, which would be consistent with the assumption of the existence of a network strategy. Questioning the network awareness assumptions that organizational decision-makers have of their interﬁrm networks is one of the goals of this chapter. We aim to contribute to this topic by explicitly studying the effect of different levels of information completeness on the ability of organizational decision makers to shift their network position. We employ network experiments to study the impact of information about the network structure on the dynamics of network positions. Results from a series of controlled experiments indicate that participants with an extended network horizon are better able to strengthen their bridging position compared with participants with a limited network horizon, especially under conditions of high network horizon heterogeneity. When network horizon heterogeneity is low, the majority of ﬁrms are much less able to strengthen their bridging positions and have weaker positions overall, particularly when the average network horizon of their competitors is high. These ﬁndings suggest that an explanation for the strength of a bridging position can be found in the ﬁrms’ network horizon. The remainder of this chapter is organized as follows. First, we argue that ﬁrms differ in how much information they possess about the network structure. Next, we introduce the network horizon and network horizon heterogeneity constructs and accompanying hypotheses. Subsequently, we introduce the network experiments and results, and ﬁnally we conclude with a discussion and future research.

599

Network Horizon

FIRMS HAVE DIFFERENT MAPS OF THE INTERFIRM NETWORK Network scholars recognize the importance of having information about potential partner ﬁrms and the overall network structure. Even though the notion of having information about the network structure is often acknowledged, usually either complete or no information is assumed and this assumption has received little empirical attention. Tables 1 and 2 give an overview of the prevalence and importance of this assumption in the current interﬁrm network literature. Rowley and Baum (2004, p. 120) phrase it as follows: ‘‘The idea that managers are aware of their ﬁrms’ networks and the types of positions that provide social capital advantages – core assumptions underpinning the network strategy perspective – remains largely unexplored.’’ Both versions of the information assumption (complete and none) would seem to be too Table 1. Quotes about the Information Actors have about the Network Structure. Author(s)

Cook & Emerson (1978, p. 726) Cook, Emerson, & Gillmore (1983, p. 280) Hite & Hesterly (2001, p. 279)

Skvoretz & Willer (1993, p. 814)

Gould (2002, p. 1152)

Quotes about the Information actors have about the Network Structure ‘‘In the experiments in this series the subjects have no knowledge about structural arrangements remote from their own location.’’ ‘‘An important feature of our laboratory research is that the actors located in the structure have no knowledge of the network beyond their own opportunity set.’’ ‘‘Emerging ﬁrms [y] are less likely to know of the full range of potential market ties. Perhaps more importantly, however, newer ﬁrms are less likely to be seen as potential ties by other ﬁrms [y] because they lack visibility.’’ ‘‘Finally, future research should use weak power networks to systematically explore the relationship between information and the development of power [y]. Yet theorists have long suspected that information available to actors can inﬂuence power differentials [y]. Our experiments were conducted in an open information context in which actors knew how their positions were connected in the larger network. With this information, subjects could make a cognitive assessment of their chances of exclusion and calibrate their behavior accordingly.’’ ‘‘yassume a closed and ﬁnite population of individuals, each of whom can direct attachments in any way he or she chooses across others in the population.’’

600

DIEDERIK W. VAN LIERE ET AL.

Table 2. Quotes about the Importance of Information about the Network Structure. Author(s)

Quotes about the Importance of Availability of Information about Potential Partner Firms

Van de Ven (1976, p. 31)

‘‘Firms must be aware of possible sources in other agencies where their needed resources can be obtained; otherwise organizational directors are likely to conclude that the goal or need which motivates the search for resources cannot be attained. [y] Awareness is therefore a predictor of the formation of interagency relations. [y] This level of awareness identiﬁes the number of potential alternatives for obtaining needed resources.’’ ‘‘Discovering new alliance opportunities and ﬁnding an appropriate partner that desires an alliance requires very good access to market information. Firms need to know about the reliability of potential partners as well. Information thus has a twofold purpose: it makes ﬁrms aware of viable partners, and it serves as a basis for trust between partners. Firms can learn about potential alliance opportunities from many sources, and one important source is their network of prior alliances.’’ ‘‘Embedded ties primarily develop out of third-party referral networks and previous personal relations. In these cases, one actor with an embedded tie to two unconnected actors acts as their ‘go-between’.’’ ‘‘Because alliances are volitional relationships, a lack of access to a good set of willing exchange partners is a limitation on many ﬁrms’ ability to put into place a productive cooperative strategy.’’ ‘‘For ﬁrms to build alliancesy they must ﬁrst be aware of the existence of potential partners.’’ ‘‘While interdependence may help a ﬁrm to orient the search for an adequate alliance partner, it cannot offer sufﬁcient cues to determine with whom it should build such an alliance.’’ ‘‘Yet, this approach masks the considerable heterogeneity of available information on prospective partners across ﬁrms, which may inﬂuence the formation of ties between speciﬁc ﬁrmsy’’ ‘‘The idea that managers are aware of their ﬁrms’ networks and the types of positions that provide social capital advantages – core assumptions underpinning the network strategy perspective – remains largely unexplored.’’

Gulati (1995, p. 622)

Uzzi (1997, p. 48)

Stuart (1998, p. 671)

Gulati (1999, pp. 399–400) Gulati & Gargiulo (1999, p. 1444)

Rowley & Baum (2004, p. 120)

Network Horizon

601

strong assumptions in reality. Several theories underpin our starting point that ﬁrms will have varying degrees of completeness of their information about the interﬁrm network. The information processing view states that the acquisition and distribution of information within an organization is a key organizational activity (Daft & Weick, 1984). The organizational boundary functions as a ﬁlter of the information that comes to the attention of the ﬁrm. Even if information circulated freely, it would not always penetrate through the organizational boundaries because it is too costly to analyze all the available information. Pfeffer and Salancik (1978, p. 74) observed: ‘‘There is a great deal of information, but only some of this comes to the attention of the organization and is, therefore, relevant for understanding its behavior.’’ The information that does come to the attention of the ﬁrm is usually the result environmental scanning and competitor analysis although this will be rarely complete or accurate (Sutcliffe, 1994). Relationships are often invisible and information about relationships may be unavailable or incomplete. Most relationships are not announced publicly, large alliance and joint venture relationships are a few of the exceptions. Instead, relationships are developed over time with no need to broadcast the existence of the relationship, or the existence of a relationship is kept low proﬁle to minimize the risk of competitive signaling (Moore, 1992). A ﬁrm can detect relationships that are not publicly announced through observation and environmental scanning but this requires resources that are limited and costly. Or as Gulati, Nohria and Zaheer (2000, p. 208) phrase it ‘‘the private and invisible nature of the ties renders the network inimitable, and thus too the information that it provides.’’ The resource-based view stresses ﬁrm heterogeneity that contradicts both assumptions (complete information and no information) that ﬁrms are homogenous in how knowledgeable they are about the network structure. An important characteristic of ﬁrms is that they differ in their resources available to them; this difference in available resources is likely to translate into different levels of completeness of information about the network structure. For example, a ﬁrm may spend more or less resources on monitoring the environment compared with other ﬁrms in the network. Second, the current network position a ﬁrm occupies is a network resource (Gulati, 1999) that acts as a source of information about possible future partners (Uzzi, 1997). The idiosyncratic process that has lead to a particular network position is most likely to lead to differences in information about the network structure. Research on cognitive decision-making processes has shown that individuals have a bounded perception of their environment and it seems

602

DIEDERIK W. VAN LIERE ET AL.

too strong an assumption that ﬁrms are completely knowledgeable or uninformed about the network structure. Krackhardt (1990) demonstrated that managers occupying central positions in an organization have more accurate maps of the network structure than peripheral members. This stream of literature focuses on situational and personal factors that inﬂuence the ability of an individual to perceive the network accurately. For example, Casciaro (1998) ﬁnds that personality characteristics such as need for afﬁliation and need for achievement positively inﬂuences the perception of the network structure while a situational factor such as being part-time employed negatively inﬂuences the perception of the network structure. Summarizing, based upon information processing theory, the resource based view, and cognitive decision-making we conclude that ﬁrms differ in their information about the network structure and this is likely to have consequences for their network strategy.

NETWORK HORIZON Purposeful shifting of a network position requires that a ﬁrm has information about the network structure. A ﬁrm that only knows (randomly) of the existence of a potential partner but does not know how this ﬁrm is embedded in the overall network cannot accurately assess how partnering with that ﬁrm will inﬂuence its network position. One of the sources ﬁrms use to gather information about the network is to use their network position to gather information about the network structure. Information about the network is a network resource (Gulati, 1999), because the information is scattered among partner ﬁrms and the partners’ partner ﬁrms. We introduce the network horizon construct and deﬁne it as ‘‘the number of ﬁrms and their relationships that the focal ﬁrm knows to exist in an interﬁrm network.’’ The better the ﬁrm is in gathering information about the network structure, the more aware it will be of valuable network positions. Network horizon is distinct from previous work on network perception (Freeman, 1992). Network horizon is a construct that captures the completeness of information about the network structure while network perception focuses on the extent to which an individual is able to accurately map the existing network structure. Put differently, the accuracy aspect deals with the (in)correct perception of links between nodes, whereas completeness also takes into account the extent to which all the nodes in the

Network Horizon

603

network are perceived to exist. Although it is possible (and perhaps quite likely) for a perceived network in practice to be both incomplete and partially inaccurate, in the remainder of this chapter we assume that accuracy is 100% and focus on the effect of completeness of information about the network structure. While we acknowledge the importance of information accuracy, there are two reasons for shifting the focus to completeness. From a theoretical standpoint, bracketing the accuracy aspect allows for more analytical clarity regarding the completeness aspect. From a practical standpoint, inaccuracies are easier to detect than incompleteness: as a partnering decision is based on information about the network structure around the newly formed tie, that is the partner’s local network. When this information is inaccurate these inaccuracies are likely to surface after the relationship has been established and the ﬁrm’s information will subsequently be updated to increase its accuracy. Incompleteness of information additionally deals with the broader network structure that is outside the direct scope of the newly formed tie and hence its correctness is much harder to detect. We therefore leave information accuracy outside the scope of this study and focus on the (in)completeness aspect. The network horizon construct resembles Friedkin’s (1983) notion of horizon of observability. The horizon of observability is reached when observability approaches zero, e.g. an individual does not have any information beyond the horizon of observability. Friedkin (1983) observed in the context of informal control of role performance that the horizon of observability is generally two steps. Consider a chain of four nodes (A-B-CD) then A can monitor B and C but cannot monitor D. In other work employing a notion of network horizon, Anderson, Hakansson, and Johanson (1994) used it to delineate the business network from the environment, ﬁrms that fall within the network horizon are deemed to be relevant for managers’ decision-making while ﬁrms that fall outside the network horizon belong to the environment. This is consistent with the deﬁnition of network horizon, the ﬁrms that are part of the network horizon are being monitored or have gained the attention of the ﬁrm, and hence this information is part of the decision-making process.

NETWORK HORIZON HETEROGENEITY The network horizon of a ﬁrm is an important determinant of the opportunity set of ﬁrms to choose from when the focal ﬁrm is going to initiate a new interﬁrm relationship. The information a ﬁrm has about the

604

DIEDERIK W. VAN LIERE ET AL.

entire network structure constrains the possible choices regarding establishing new ties, less information means less ﬁrms to choose from and therefore limits the possibilities of the focal ﬁrm to shift its network position. A ﬁrm with a more extended network horizon will be aware of more valuable network positions and will therefore have an advantage compared with a ﬁrm with a smaller network horizon in occupying this network position. A ﬁrm that is not part of the network horizon can still be considered as a potential partner by the focal ﬁrm. However, since information about the network position of the potential partner ﬁrm is lacking it is hard for the focal ﬁrm to estimate the effect of linking with that potential ﬁrm on the focal ﬁrms’ bridging position. Firms outside the network horizon that are still known to the focal ﬁrm, so-called randomly known ﬁrms, can strengthen the bridging position of the focal ﬁrm. However, because these ﬁrms do not fall within the network horizon of the focal ﬁrm, it is less likely that these ﬁrms represent opportunities to strengthen the focal ﬁrms’ bridging position because the technological or product distance is too large (Ahuja, 2000) or other incompatibilities prevent either ﬁrm to beneﬁt from establishing such an interﬁrm relationship. Hence, we hypothesize: Hypothesis 1. Firms with a more extended network horizon at time t will occupy a stronger bridging position at time tþ1. The second advantage of a more extensive network horizon is that a ﬁrm is able to span structural holes at a faster rate. The reason for this is that the network position is an important source of new and novel information (Burt, 1992; Reagans & McEvily, 2003). Thus, a ﬁrm occupying a strong bridging position is more likely to detect opportunities to span new structural holes not only because it is able to recognize similar opportunities in the network (‘‘better eyes’’ in Burt’s (2008) analogy), but also the experience of reaping beneﬁts from previous brokerage opportunities may have given it skills to utilize new opportunities that lesser-experienced may not be able to utilize (‘‘better glasses’’ in Burt’s (2008) analogy). Burt (2002, p. 334) noted in the context of individuals that ‘‘people [without structural holes] ﬁnd it more difﬁcult [compared with people who have structural holes] to see the structural holes in a network, and those whose own networks contain bridge relationships across structural holes more quickly learn new networks that contain structural holes.’’ Network horizon is hypothesized to lead to a stronger bridging position (hypothesis 1) but this starts a virtuous cycle in which the ﬁrm is able to span new structural holes at a faster rate. These structural holes give information beneﬁts in the form of awareness of new brokerage opportunities and skills at utilizing them that

Network Horizon

605

allows the focal ﬁrm to strengthen its bridging position even more. As Burt (2002) notes, social capital accrues to ﬁrms which already possess social capital and this becomes a virtuous cycle. Therefore, we hypothesize: Hypothesis 2. Firms with a more extended network horizon at time t will strengthen their bridging position at a faster rate at time tþ1. While each ﬁrm has a network horizon, this network horizon is not equal for each ﬁrm. Private or sensitive information does not ﬂow freely through the network; rather it might only be shared between close partners or partner’s partners and thus ﬁrms will differ in the information they have available to them. Two aspects of the availability of information across ﬁrms are important, ﬁrst the information distribution (Moldoveanu, Baum, & Rowley, 2003): What information about the network structure is available to which ﬁrm? Second, the information heterogeneity (Moldoveanu et al., 2003): to what extent is particular information about the network structure shared (common) between ﬁrms. The more common information about the network structure is, the larger the number of ﬁrms that will be aware of that particular network position and competition for that opportunity will be more intense and this intensity of competition for structural holes in turn determines how many structural holes a single ﬁrm will be able to span. Network horizon heterogeneity captures the degree of competitiveness in the network for network positions. The more homogeneous (i.e., common) the information about a given opportunity to span a structural hole is distributed within the network, the more likely it is that more ﬁrms will pursue that particular opportunity. Because a ﬁrm is inherently uncertain about the partnering actions its rivals will undertake, it will try to move ﬁrst to occupy a bridging position. However, when multiple ﬁrms adhere to this logic then multiple ﬁrms will ‘‘jump in’’ the structural hole to bridge it and effectively turn the hole into a closed position. Fig. 1 illustrates this situation. Assume that ﬁrms A, B, C, and D have complete information of the network structure of nodes A–H, while ﬁrms E, F, G, and H only aware are of their own network component, but not of ﬁrms A–D. Firms A, B, C, and D realize the potential value that can be unlocked by bridging with ﬁrm E and ﬁrms F, G, and H have a limited network horizon that they are not aware of the existence of ﬁrms A, B, C, and D. Thus, this network is characterized by a high level of network horizon heterogeneity. Because ﬁrms A, B, C, and D are uncertain about the partnering actions each one is going to take, each ﬁrm individually decides to pursue the bridging position between ﬁrm E and ﬁrms A, B, C, and D. Furthermore, it is not realistic to expect that ﬁrms A, B, C, and D are going to coordinate

606

DIEDERIK W. VAN LIERE ET AL.

Time t

Time t+1

D

A

E

A

G

C

C

B

F

D

F

H

E

B

G

H

Effective size

Constraint

Effective size

Constraint

Firm A

1.667

0.840

Firm A

1.500

0.677

Firm B

1.000

0.889

Firm B

1.000

0.681

Firm C

1.667

0.840

Firm C

1.500

0.677

Firm D

1.000

0.889

Firm D

1.000

0.681

Firm E

3.000

0.333

Firm E

5.571

0.303

Firm F

1.000

1.000

Firm F

1.000

1.000

Firm G

1.000

1.000

Firm G

1.000

1.000

Firm H

1.000

1.000

Firm H

1.000

1.000

Fig. 1.

Illustration of the Effect of Network Horizon Heterogeneity.

who is going to bridge the position because being the tertius gaudens is a more favorable role (in terms of autonomy) than being exploited in a structural hole. The right panel illustrates the result. All four ﬁrms have established a tie with ﬁrm E (assuming E accepts all new ties in order to minimize its dependence on any single ﬁrm) and none of the four ﬁrms will enjoy the control beneﬁts of the structural hole because that is foregone in the new situation. Furthermore, because this network is highly redundant the information beneﬁts are minimal as well. Consequently, we hypothesize: Hypothesis 3. A high level of network horizon heterogeneity at time t will result in the weakening of the focal ﬁrms’ bridging position at time tþ1. Network horizon heterogeneity also provides an important moderator for the effect of network horizon itself, as the beneﬁts of information are not absolute, but rather depend on whether or not other ﬁrms have that same

Network Horizon

607

information. A focal ﬁrm with an extended network horizon, in a network with a heterogeneous distribution of network horizons is able to sustain its network position longer, because the limited network horizon of non-focal ﬁrms implies a lack of information regarding the potential location of structural holes and these ﬁrms are not able to compete for structural holes. This implies that the focal ﬁrm’s position is more likely to be stronger. Hypothesis 4. Firms with a more extended network horizon at time t will have a stronger bridging position at time tþ1 in a network with high network horizon heterogeneity than with low heterogeneity. The ﬁnal hypothesis concerns the diminishing effect of network horizon. As the network horizon increases, the number of potential network positions that can be pursued increases as well. This requires the ability to distinguish ‘‘very valuable’’ structural holes from merely ‘‘valuable’’ structural holes. Pollock, Porac, and Wade (2004) observed that not every structural hole is equally valuable, it takes training to be able to distinguish the very valuable from the merely valuable structural holes. Rowley and Baum (2004) found in the Canadian investment bank industry that managers have difﬁculty distinguishing unconstrained from constrained structural holes. Therefore, we hypothesize: Hypothesis 5. There is an inverse U-shaped (-) relationship between network horizon and the strength of the bridging position.

METHODOLOGY: NETWORK EXPERIMENTS To test the effects of network horizon and network horizon heterogeneity, we used laboratory experiments. Although perhaps not a very common methodology used in strategic management per se, experimental research has a long and rich history in two distinct streams within the literature on social networks. The ﬁrst stream on communication structures in small groups originated from the work by Bavelas and Leavitt at MIT (e.g., Bavelas, 1950; Leavitt, 1951), who showed that an experimentally manipulated communication structure substantially affected the group’s task performance (for other important early contributions see Shaw (1964) and Guetzkow & Simon, 1955). The second stream is more sociological in its orientation, focusing on exchange networks instead of communication networks. Starting from Emerson’s original formulation of exchange theory (Emerson, 1962, 1972a, 1972b), subsequent researchers have used

608

DIEDERIK W. VAN LIERE ET AL.

experimental approaches to study how in particular the centrality, power and status of different positions in the network structure affect exchange and lead to a bargaining advantage (e.g., Cook, 1977; Cook et al., 1983; Lucas, Younts, Lovaglia, & Markovsky, 2001; Markovsky, Willer, & Patton, 1988). However, in this stream of literature the network structure itself is ﬁxed and hence the dynamics of how actors attain and alter their position remain unexplored (Leik, 1992; Willer & Willer, 2000). We developed a network experiment environment (Hoogeweegen, van Liere, Vervest, Hagdorn van der Meijden, & de Lepper, 2006) to study how participants manipulate their network position based on information available to them. Participants play the role of an organizational decision-maker of a simulated ﬁrm and have the assignment that he/she should try to be the most proﬁtable ﬁrm in the network in terms of margin and net income. The network experiments are built in such a way that bridging positions are beneﬁcial for ﬁrm performance; however the participants are unaware of this. Hence, participants expanding their bridging positions are more likely to win an experiment. We have simpliﬁed the ﬁrms in our network experiments. Each ﬁrm possesses two or three capabilities depending on its role in the business network. A customer market generates demand for a product or service: this demand is deﬁned as a speciﬁc set of capabilities. An order is awarded to a participant (i.e. one of the ﬁrms in the business network) if this participant ﬁrm can produce the required set of capabilities either by producing the set of capabilities itself, or by having access to such capabilities through its relationships in the business network (or a combination of ﬁrm capabilities and partner capabilities). None of the ﬁrms possess the required capabilities to produce the product or services independently: they need to produce these policies jointly by way of establishing a relationship. Therefore, each ﬁrm needs to invest in a portfolio of interﬁrm relationships to access the required capabilities in the network. Once a ﬁrm has access to the capabilities that constitute a speciﬁc product or service then it will start receiving orders for that particular product or service and its proﬁt will increase. Each ﬁrm can also decide to invest in new capabilities or to specialize in existing capabilities. So three investment (and divestment) decisions impact if a participant ﬁrm will do better than others in the game: invest in/or divest interﬁrm relationships, invest in new capabilities, or invest in specializing existing capabilities, or divest existing capabilities. In essence, the two main levers to increase the ﬁnancial performance of a ﬁrm are either to invest/ divest in new interﬁrm relationships or to invest/divest in capabilities.

609

Network Horizon

Order Allocation Process During the experiment, a simulated market generates customer demand for a potential order; this demand is deﬁned as a collection of capabilities at a certain price. This collection of capabilities becomes a ﬁnalized order when a ﬁrm has a relationship with the consumer market and is able to deliver the required capabilities. Either a ﬁrm can deliver these capabilities itself or it can establish relationships with other ﬁrms in the network that are able to deliver the required capabilities, or there can be a combination of these two possibilities. If multiple ﬁrms can deliver the required order then the ﬁrm who offers the lowest total price receives the order. If multiple ﬁrms offer the same price then a ﬁrm with the oldest relationship with the consumer market is chosen. If multiple ﬁrms have the same age for their relationship with the consumer market then a ﬁrm is randomly chosen. Thus, a ﬁrm receives an order if it can deliver the required capabilities and if it can do so at the lowest possible price. The consumer market does not respond to changes in the network structure or request new, previously unavailable capabilities, so the demand side remains stable. In principle, this notion of a product or service as a set of capabilities can be applied in any industry, but the network experiment environment can be customized to different scenarios tailoring the experiment to that particular industry’s capabilities and roles. A role is based upon the capabilities a ﬁrm possesses at the start of a network experiment (often representing the existing division of labor in that industry) which ﬁrms can change by investing in new capabilities or in new interﬁrm relationships. The starting roles for the insurance scenario we use in our experiments are insurance advice, sales, customer acceptation, service center, and customer service. There are different capabilities within each role. Table 3 summarizes which role each ﬁrm has at the start of an experiment and with which capabilities it starts. Firms at the start of an experiment do not offer capabilities in italics but participants can decide to invest in these capabilities. The number in parentheses behind the capability indicates the id number. At the start of an experiment, the network contains 15 ﬁrms and this number remains ﬁxed throughout the experiment. However, we also have a scenario of 14 ﬁrms (due to space limitations in the lab, it was not always possible to use 15 computers). In this scenario, the ﬁrms Pluto and Hermes have been merged to one ﬁrm Hermes that possesses both the generic advice and product bronze capability. An order contains one capability from each role. This means that these 13 unique capabilities can be combined into 36 different orders (2 3 3 2 ¼ 36). The reason that there are only 36 unique orders and not 108 is

610

DIEDERIK W. VAN LIERE ET AL.

Table 3.

Relationship between Firms: Roles and Capabilities.

Firms Apollo Jupiter Pluto Miracle Diamond Hermes Archimedes Blazer Saturn Phoenix Rainbow Delphi Crystal Cosimo Star

Roles

Capabilities

Insurance advice

Generic advice (1) Speciﬁc advice (2)

Sales

Product line bronze (3) Product line silver (4) Product line gold (5) Accept car policy (6) Accept home policy (7) Accept travel policy (8) Claims settlement car (9) Claims settlement home (10) Claims settlement travel (11) Customer care online (12) Customer care call center (13)

Customer acceptation

Service center

Customer service

because once an order has a customer acceptation capability then the accompanying service center capability is determined as well. Customer acceptation and service center capabilities are dependent on each other. For example, it is not possible to have an insurance policy with a car acceptation capability and a travel settlement capability. Fig. 2 illustrates the initial distribution of capabilities and initial network structure. We did not model capacity constraints for the production of insurance policies in our experiments. Although ﬁrms clearly do have a capacity constraint, we think that in this particular context it is less relevant. The production of an insurance policy is fully automated, due to the embedded coordination of the quick connect capability (van Liere, Hagdorn, Hoogeweegen, & Vervest, 2004) and production times are measured in minutes or hours rather than weeks or months, and thus the daily production capacity of insurance policies will always surpass the daily demand for insurance policies. Thus, at the start of the experiment, ‘‘network 1’’ in Fig. 2 can only fulﬁll insurance policies consisting of capabilities 1, 3, 6, 9, and 12. Likewise, ‘‘network 2’’ can only fulﬁll insurance policies consisting of capabilities 1, 3, 7, 10, and 12 and ‘‘network 3’’ can only fulﬁll insurance policies consisting of capabilities 1, 3, 8, 11, and 12. This means that at the start of each experiment each network can only satisfy 1/36 of the total number of orders. The networks will be able to

611

Network Horizon

Archimedes Capability 6 Apollo Capability 1

The

Miracle Capability 3

Network 1 Crystal Capability 12

Phoenix Capability 9

consumer market generates

Blazer Capability 7

orders that are sent to a firm if

Jupiter Capability 1

Diamond Capability 3

Network 2 Cosimo Capability 12

Rainbow Capability 10

the firm has access to the

Saturn Capability 8

needed capabilities

Pluto Capability 1

Hermes Capability 3

Network 3 Star Capability 12

Delphi Capability 11

Fig. 2.

Initial Network Structure and Distribution of the Capabilities.

satisfy a wider range of orders as the number of offered capabilities in the network increases. When a ﬁrm is involved more often in the production of a particular insurance policy then it will increase its ﬁnancial performance. The ﬁnancial performance is measured with two metrics: net income (deﬁned as turnover – ﬁxed costs – variable costs) and gross margin (deﬁned as ððfixed costs þ variable costsÞ=turnoverÞ 100%. Participants should score high on both metrics in order to become the winner of an experiment. How often a ﬁrm is part of the production depends on two factors: the capabilities it can access through a ﬁrm’s network position and the capabilities a ﬁrm possesses itself. The consumer market generates customer orders for car, home, and travel insurance policies at a constant rate. An insurance policy is offered at three different levels: product line bronze, product line silver, and product line gold. The insurance network consists of 15 organizational units grouped into 5 separate roles and these roles are represented in the experiment as separate organizations. The three brokers (‘‘Apollo,’’ ‘‘Jupiter,’’ and ‘‘Pluto’’) have direct access to the consumer market. These ﬁrms are the insurance brokers. Apollo is the insurance

612

DIEDERIK W. VAN LIERE ET AL.

broker for the car insurance policies; Jupiter is the insurance broker for the home insurance policies and Pluto is the insurance broker for the travel policies. The second ﬁrm is called ‘‘Miracle’’ for the car insurance policies and ‘‘Diamond’’ for the home insurance policies and ‘‘Hermes’’ for the travel insurance policies and all three of them offer product line bronze. These three ﬁrms are responsible for the branding and marketing of the policies. The third ﬁrm is called ‘‘Phoenix’’ for the car insurance policies and ‘‘Rainbow’’ for the home insurance policies and ‘‘Saturn’’ for the travel insurance policies. They are responsible for the customer acceptation process, that is, screening a customer, assessing the risk of the customer and deciding whether the customer is accepted or not (in our experiments each customer is always accepted). The fourth ﬁrm is called ‘‘Archimedes’’ for the car insurance policies and ‘‘Blazer’’ for the home insurance policies and ‘‘Delphi’’ for the travel insurance policies. They are responsible for the settlement of all damage claims of insured customers. The ﬁfth ﬁrm is called ‘‘Cosimo’’ for car insurance policies and ‘‘Crystal’’ for the home insurance policies and ‘‘Star’’ for the travel insurance policies, and they take care of all administrative processes of handling policies and offer customer service either by using a call center or a website.

Time Sequencing Each run of the experiment lasts 43 time units, since several pilot tests revealed a relatively stable network structure by that time, indicating convergence (these 43 time units equaled to approximately 30 real-time minutes, which allowed us to do multiple runs in a single experimental session). The 43 time units are divided in two different periods. An experiment starts with an information-gathering period that lasts seven time units. A participant can evaluate its current ﬁnancial performance, evaluate past decisions, and formulate new decisions for the coming time units during an information-gathering period. An information-gathering period lasts for 3.5 real-time minutes. An information-gathering period is continued with a decision point. The decision point is used to execute actual decisions: an investment in a new interﬁrm tie can be made or a ﬁrm can divest capability or a new capability can be bought. A decision point lasts for two real-time minutes. An experiment consists of ﬁve information-gathering periods (IGP), ﬁve decision points (DP) and three time units to end the experiment. Fig. 3 visualizes the structure of an experiment.

613

Network Horizon

IGP1

IGP2

DP1 DP2 IGP = Information Gathering Period DP = Decision Point

Fig. 3.

IGP3

IGP4 DP3

IGP5 DP4

T = 30 DP5

Sequence of Information Gathering Periods and Decision Points.

A participant can only make one investment during each decision point. This compels participants to envision the possible consequences of the different decisions they can take and to make the decision they think will improve their performance the most.

Managing a Firm During the Experiments Each ﬁrm starts with an initial cash position to cover the initial ﬁxed costs of the ﬁrst time units, to pay for outsourced capabilities and to let participants invest in capabilities and relationships. For the purpose of the experiments, we set the initial cash position at h50.000. This money can be used to invest in a new interﬁrm relationship to increase the access to capabilities or to invest in a new capability or specializing an existing capability. There are costs associated with maintaining capabilities, these are ﬁxed costs per capability and are deducted from the cash position at the end of each time unit. Specializing an existing capability means that a ﬁrms’ ﬁxed costs increase, but that the variable production costs are reduced. Hence, this investment makes sense when a ﬁrm has a signiﬁcant market share for a particular capability it has specialized. We use three strategies as deﬁned by Treacy and Wiersema (1993) to assist participants in deciding how to run their simulated ﬁrm. Treacy and Wiersema (1993, p. 84) propose that organizations that have taken leadership positions within their industries or business networks have focused on ‘‘y delivering superior customer value in line with one of three value disciplines – operational excellence, customer intimacy, or product leadership.’’ Operational excellence refers to providing reliable products and services to customers against competitive prices and convenience. Customer intimacy refers to the ability to meet customized demand by tailoring production to the exact requirements of individual customers or market

614

DIEDERIK W. VAN LIERE ET AL.

niches. Product leadership refers to the offering of innovative, leading-edge products and services to customers that enhance the use or application of the product or service; this should offset the value of the competitors’ products and services. We chose these three strategies because they can be easily operationalized in capabilities and network positions and we found these three strategies to be well understood by both business administration students and industry professionals. These strategies are randomly distributed among the participants at the start of an experiment but a participant keeps the same strategy card throughout the whole session. All three strategies are potentially interesting in order to increase the ﬁnancial ﬁrm performance. For instance, by specializing production, a ﬁrm could become a main node within the network and be part of virtually every temporary alignment formed within the network (operational excellence). Another focus could be to approach the end-customer and act as a ‘‘network coordinator’’ to fulﬁll its customized demand by forming the right temporary alignment (customer intimacy). Yet another option could be to innovate and develop new capabilities to meet the customers’ demand for state-of-art products and services (product leadership). Participants of the Experiments We used three groups of participants to conduct our experiments. The ﬁrst group of participants is from a large Dutch insurance company. The second group of participants consists of two groups of business administration students. First, there are students who participated in a minor about business networks as part of their masters in business administrations and second there were students that participated at an in-house business course at the same insurance ﬁrm. The third group of participants was a team of 15 students who participated on a regular basis. This group was used to study whether there are learning effects in shifting a ﬁrm’s network position. Multiple replications with the same participants increases the internal validity of the research design (Shadish, Cook, & Campbell, 2002).

Reliability and Validation Experiments For the experiments reported here, we tailored our scenario to the insurance industry. Through standardization of the data ﬂows between ﬁrms it has become easier for ﬁrms to establish and disband network ties. This increases

615

Network Horizon

Table 4.

Results Validation Network Environment.

Statement asked on a Likert scale from strongly disagree (1) to strongly agree (5) (N ¼ 98) The modeling of the insurance products in the Business Network Engine is realistic The modeling of the initial business network is realistic The Business Network Engine simulates convincingly a business network The modeling of the trade-off between internal production and outsourcing is realistic The Business Network Engine provides a valid learning experience about managing a business network I have conﬁdence in the Business Network Engine results

Mean

4.4 4.4 4.3 4.2 4.6 4.4

the rate of change in a network and makes it a suitable context to study the dynamics of network positions. The insurance scenario has been developed in close cooperation with a strategy consultancy ﬁrm, industry experts, and board members of an insurance ﬁrm. We asked 98 participants from 7 different pilot tests to rate, using a Likert scale, the experimental environment on various aspects of its modeling of the insurance industry and the network aspects. The participants of the pilot tests were senior and middle-level managers of a large Dutch insurance ﬁrm and its insurance brokers. The results in Table 4 indicate that the participants generally agree that the experimental environment is realistic in both simulating the insurance industry (products and capabilities) and the insurance network structure and different roles that are needed to sell an insurance policy. We conducted numerous tests to validate our network experiment setting. First, we calibrated the insurance scenario by consulting extensively with industry experts about the relative proﬁtability of each type of customer order (compared to other types of customer orders) to avoid biases of ﬁrm performance. Second, we conducted numerous computer interface tests to investigate whether participants understood the user interface and whether presented information was unambiguous. The results of these tests led to improvements of the software that were subsequently tested again.

Participants and Research Design We employed a randomly assigned within-subject design (Shadish et al., 2002). Three types of participants participated in the experiments. In total

616

DIEDERIK W. VAN LIERE ET AL.

Table 5. Participants

Managers and insurance brokers Students Student team Total

Descriptives of the Participants of the Experiments. Number of Sessions

Number of Number of Years of Industry Scenario Experiments Participants Experience Used (Average/s.d.)

7

12

210

15.49/9.19

15 ﬁrms

9 8 24

19 22 53

126 15 351

0.0/0.0 0.0/0.0

14 ﬁrms 15 ﬁrms

we had 210 managers from a large Dutch insurance ﬁrm and their insurance brokers participate as well as 126 students from different business administration graduate courses participate. Finally, we used a ﬁxed student team (15 participants) that participated in multiple experiments. For the managers, insurance brokers, and the student team we used a scenario consisting of 15 ﬁrms, for the graduate students we used a scenario of 14 ﬁrms. The students participated individually in the experiments, the managers teamed up with an insurance broker. We will control for these differences by including dummy variables in the analysis. We randomly assigned each participant to the condition of either a limited network horizon or a complete network horizon. We replicated the experiment 53 times in 24 sessions. In order to replicate experiments, we conducted multiple experiments in a single session. Table 5 summarizes the descriptions of the participants. Experimental Procedure The participants were seated behind a computer after arrival in the computer lab where the experiments were conducted. Before the experiments started, an extensive one-and-half hour explanation of the network experiment was given. During this explanation, the participants played one round of the network experiment to teach them where to ﬁnd relevant information, how to make decisions, and how to assess the impact of their decisions. The data obtained from this practice round was discarded. Participants were free to ask the instructors questions about the interface of the network experiment. After the exercise round, control questions were distributed to verify whether the participants had a sufﬁcient understanding of the network experiment to play it independently. Answers to these were

617

Network Horizon

checked and if necessary, some additional instruction was given to the participant before the experiments started. Participants were not allowed to talk to each other and neither was there any help from the instructors during the second and third round, the rounds that were used for data collection. There was a debrieﬁng session at the end of the experiments for educational purposes. The winner, the participant with the highest gross margin at the end of the experiments of a session was awarded with a gift coupon worth h25; the other participants received a ﬁxed fee of h10.

Dependent Variables Strength of bridging position: We measure the strength of the bridging position by counting the number of structural holes of the focal ﬁrm and we use two measures. The ﬁrst measure is effective size and the second measure is constraint. The deﬁnition for effective size is (Burt, 1992, p. 52): ! X X Effective size of firm i ¼ 1 piq mjq ; qai; j j

q

Sum j is the set of partner ﬁrms (alters). piq is the proportion of i ’s network time and energy invested in the relationship with q and mjq is the marginal strength of j’s relation with contact q (Burt, 1992, p. 51). piq ¼ "

ðziq þ zqi Þ P

ðzij þ zji Þ

# ; iaj and mjq ¼

ðzjq þ zqj Þ ; jak max zjk þ zkj

j

zjq is the network variable measuring the strength of relation j to q (Burt, 1992, p. 51). Effective size measures the extent to which ties in the ego network are non-redundant and is therefore a measure of the information beneﬁts of a structural hole. When the degree of a ﬁrm (the number of ties it has) is equal to its effective size it means that every single tie gives access to new information. Effective size can never be greater than the degree of a ﬁrm. When the effective size is not equal to the degree of a ﬁrm than some relationships in ego’s network are redundant. Effective size is an unstandardized measure with a minimum value of one and a maximum value the network size minus one. For the robustness test, we calculate for each ﬁrm the strength of its bridging position by using Burt’s (1992) network constraint measure.

618

DIEDERIK W. VAN LIERE ET AL.

Constraint measures the extent to which a ﬁrm is dependent on its partner ﬁrms and is therefore a measure of the control beneﬁts of a structural hole. When a ﬁrm it highly dependent on its network partners it means that it has little autonomy and therefore little control over the outcome of its activities. Constraint is a normalized measure with values ranging from zero to one where one means that a ﬁrm is very constrained and does not span any structural holes and zero means a ﬁrm is very unconstrained and spans many structural holes. This is in opposite direction of the effective size measure, hence to clarify the interpretation of the network constraint variable we invert this measure by taking 1 – network constraint. Furthermore, we convert it to a percentage by multiplying it by 100. The deﬁnition for network constraint is (Burt, 1992, p. 60): !2 X Network constraint of firm i ¼ pij þ piq pqj ; iaqaj. q

Independent Variables The ﬁrst independent variable, network horizon, is a continuous variable that counts how many ﬁrms the focal ﬁrm can ‘‘see.’’ For ﬁrms with a limited network horizon we count the number of ﬁrms that can be reached within two steps, for ﬁrms that have an extended (or full) network horizon the number of ﬁrms the focal ﬁrm sees is equal to the network size (depending on the scenario 14 or 15 ﬁrms). This variable is calculated as a percentage of the total network and is updated after each decision point. The second independent variable is the rate at which the bridging position is strengthened. We calculate this by multiplying network horizon and time and use this variable to test hypothesis 2 (Singer & Willett, 2003). The third independent variable is the network horizon heterogeneity. Assigning individual ﬁrms a network horizon is an implicit conﬁguration of the network horizon heterogeneity. The network horizon heterogeneity was determined using the ratio of ﬁrms with a limited network horizon and ﬁrms with a complete network horizon. Figs. 4 and 5 illustrate the effect of having a limited or a full network horizon in the network experiment environment. Different ratios of individual network horizons translate into different levels of network horizon heterogeneity. There are multiple ways to conceptualize this heterogeneity, each with different underlying theoretical rationales (see Klein & Harrison (2007) for more detail on the different ways). The

Network Horizon

619

Fig. 4.

Screenshot Network Experiment with Full Network Horizon (Focal Firm Phoenix).

Fig. 5.

Screenshot Network Experiment with Limited Network Horizon (Focal Firm Hermes).

620

DIEDERIK W. VAN LIERE ET AL.

underlying attribute that is heterogeneous (network horizon) represents a socially valued asset or resource of the ﬁrms, the appropriate conceptualization is of the disparity-type, which can be measured by the inequality or relative concentration measures such as the Gini-coefﬁcient or the coefﬁcient of variation (Klein & Harrison, 2007). We opt for the Ginicoefﬁcient (Weisstein, 2006), which is calculated as the mean difference of network horizon between every possible pair of ﬁrms, divided by the mean. A network horizon heterogeneity value of zero indicates that there is perfect equality of the network horizons among the different ﬁrms. A value of one indicates that there is maximum inequality: one ﬁrm has information about the whole network structure and the other ﬁrms do not have any information. The fourth independent variable is the network horizon heterogeneity multiplied by time (Singer & Willett, 2003), this variable is used to measure the rate of change of the strength of a bridging position due to the network horizon heterogeneity. The ﬁfth and ﬁnal independent variable is network horizon squared; we use this variable to measure the diminishing returns of network horizon.

Control Variables We add the following control variables to rule out alternative explanations for the strengthening of a ﬁrm’s bridging position. Density. A denser network indicates the existence of dense cliques. Brokering between such cliques has a positive effect on the strength of a bridging position and thus networks with a higher level of density will have ﬁrms with stronger bridging positions. Density is calculated as: P xij =nðn 1Þ; iaj where xij represents the value of a relationship between i and j (assuming that xij only takes values of 0 or 1) and n is the number of nodes in the network. Effective size at t–1/network constraint at t–1. A ﬁrm that spans many structural holes at t–1 will span many structural holes at t because a network position correlates strongly between two observations but also because a bridging position creates a vision advantage (Burt, 2005) that results in discovering new opportunities for spanning new structural holes. Outdegree. A ﬁrm with a greater outdegree will span more structural holes compared with P a ﬁrm with a lower outdegree. Outdegree centrality is calculated as: i xij or the sum of relationships ﬁrm i maintains.

Network Horizon

621

Betweenness centrality. A ﬁrm with a greater betweenness centrality is more often on the shortest path between any two random chosen ﬁrms and will therefore P P gikj span more structural holes. Betweenness centrality is calculated as: i j gij ; iajak gij is the number of geodesic paths from i to j, gikj is the number of geodesic paths that pass along k and a geodesic path is the shortest path connecting two nodes in a network (Wasserman & Faust, 1994). Number of capabilities owned. This variable counts how many capabilities a ﬁrm possesses. A ﬁrm that possesses many capabilities is pursuing a strategy of vertical integration, this implies that it is less dependent on its network partners and therefore is less likely to span structural holes. Attractiveness partner ﬁrms. This variable measures the average number of capabilities possessed by partner ﬁrms. Partner ﬁrms that possess many capabilities are more attractive to partner with compared with partner ﬁrms that possess fewer capabilities because the focal ﬁrm has limited resources to invest in new relationships. A focal ﬁrm is more likely to partner with a ﬁrm that possesses many capabilities (i.e., is more attractive) given the fact that the focal ﬁrm is resource constrained. Hence, we control for this attractiveness of partner ﬁrms. This variable is calculated by summing the total number of capabilities possessed by the partner ﬁrms and divided by the number of partner ﬁrms. Firm performance. Firm performance is deﬁned as the natural log of the net income of a ﬁrm. Net income is calculated by subtracting ﬁxed and variable cost from the turnover. Furthermore, we add dummy variables for each ﬁrm, each type of strategy, a ﬁrm’s initial network position and the size of the network.

ANALYSIS AND RESULTS: NETWORK EXPERIMENTS We use a multi-level model with random coefﬁcients and random slopes to estimate the effect of network horizon and network horizon heterogeneity on the number of structural holes a ﬁrm spans. We use a multi-level model with random effects and random slopes to determine: (a) whether there is a difference in the intercept of the number of structural holes a ﬁrm spans because of a ﬁrm’s network horizon and (b) whether there is a difference in the rate of spanning new structural holes because of a ﬁrm’s network horizon. Furthermore, the observations are nested in multiple levels: each observed ﬁrm has multiple observations which leads to autocorrelation and

622

DIEDERIK W. VAN LIERE ET AL.

implies that the error terms are correlated. Autocorrelation does not impact the coefﬁcient estimates but leads to underestimated standard errors and overestimated t-values (Kennedy, 2003). Furthermore, multiple ﬁrms are nested in a single experiment; this requires a multi-level model to take into account this dependence. We control for the following differences between the ﬁrms that remain constant throughout the experiment such as different strategies (product leadership, customer intimacy, and operational excellence), different initial network position, different cost structures for the different capabilities, whether students or managers were the participants of the experiments and if the scenario consists of 14 or 15 ﬁrms. Before continuing the analysis, we ﬁrst test whether the initial network position of a ﬁrm impacts the proﬁt of a ﬁrm. A one-way ANOVA test conﬁrms that there are no signiﬁcant differences ( p ¼ 0.669) in proﬁt between ﬁrms due to their initial network position. We estimate ﬁve models with effective size as dependent variable. The ﬁrst mode is the unconditional growth model. This model depicts the general trend we observe in the 53 experiments. This model is used as a baseline to calculate to what extent the subsequent models improve upon this model. There are no predictor variables in this model except for time. Table 6 summarizes the means, standard deviation, minimum, and maximum for each variable (excluding the dummy variables). The second model is a baseline model, this model explains the number of structural holes a ﬁrm spans based on the control variables. The third model adds the network horizon main effect to test hypothesis 1 (network horizon leads to more structural holes) and hypothesis 2 (network horizon leads to a faster rate of spanning structural holes). In the fourth model, we test hypothesis 3, we add the effect of network horizon heterogeneity on the number of structural holes a ﬁrm spans and hypothesis 4 that a greater inequality of network horizon among the ﬁrms leads to a faster rate of spanning new structural holes. Finally, the ﬁfth model adds the diminishing effects of network horizon to test hypothesis 5. The estimates of the coefﬁcients are robust across the three models and all three models are signiﬁcant ( p ¼ 0.0000). However, according to the AIC information criterion is model 5 the best ﬁtting model and according to the BIC information criterion is model 3 the best ﬁtting model. We interpret the results based on model 5 because the direction and signiﬁcance of model 3 and model 5 are compatible. The results of these models are presented in Table 7. The effect of network horizon is positive and signiﬁcant, conﬁrming hypothesis 1 that states that ﬁrms with a more extended network horizon are more aware of brokerage opportunities and hence will span more structural

Effective size 2.8343 Effective size at t–1 2.6813 Network constraint 0.4967 Network constraint at t–1 0.5126 Time 2.5642 Density 0.1227 Outdegree 1.771 Betweenness centrality 0.0609 Number of capabilities owned 2.036 Attractiveness partner ﬁrms 0.6198 Network horizon 12.134 Rate of change network 33.1813 horizon Network horizon 0.3968 heterogeneity (nhh) Rate of change NHH 2.9195 Network horizon2 167.3777 Variables (continued) 9 Number of capabilities owned 1 Attractiveness partner ﬁrms 0.1457 Network horizon 0.0777 Rate of change network 0.256 horizon Network horizon 0.1234 heterogeneity Rate of change NHH 0.1311 Network horizon2 0.0605

Mean

0

0 0 0.1 0 0 0.0578 0 0 0 0 0 0

Min

0.0186 0.062

0.0024

1 0.0579 0.039

0.1501 0.9871

1

2

3

0.0881

1

0.4077 0.5058

0.4534 0.1065

1 1 0.1066

1

1 0.8119 0.3062 0.1954 0.2942 0.0215 0.277 0.9177

5

1 0.3637 0.2881 0.2673 0.0358 0.4143 0.8253

6

0.4609 0.2318

0.1295 0.3773

0.1267 0.2958 0.458

1 0.3365 0.4044 0.5547 0.329 0.1288 0.0174 0.3523 0.4475

4

22.0861 0.1697 0.0571 0.1636 0.0341 256 0.3646 0.3451 0.3959 0.3335 12 13 14 15

0.5208 0.1216 0.1437

9.3182 1 9.3182 0.824 1 1 0.7991 0.6437 1 1 0.6481 0.7615 0.701 5 0.3795 0.3704 0.358 0.2311 0.4583 0.4301 0.4582 11 0.8345 0.7522 0.595 0.8465 0.5088 0.4172 0.3912 8 0.1582 0.134 0.1333 1 0.1096 0.0577 0.0695 16 0.3908 0.3602 0.4438 80 0.4939 0.492 0.4916

Max

0.1047 0.2616

1 0.5359

3.0711 0 87.8534 0 10 11

0.347

1.5705 1.492 0.2401 0.2466 1.6623 0.0435 1.8564 0.0997 0.9276 0.338 4.4887 25.1254

S.D.

Descriptive Statistics and Correlations.

1 0.0563 0.0439 0.1847 0.2455

8

0.1437 0.2332

0.0881 0.1663

0.0945 0.0653

1 0.4476 0.1944 0.0642 0.2559 0.3718

7

Correlations W |.11| are signiﬁcant at po0.001, dummy variables excluded from correlation table; number of observations is 3,599. NHH ¼ network horizon heterogeneity

14. 15.

13.

9. 10. 11. 12.

14. 15.

13.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

Variables

Table 6.

Network Horizon 623

Rate of change network horizon Network horizon heterogeneity

Number of capabilities owned Attractiveness partner ﬁrms Network horizon

Betweenness centrality

Outdegree

Density

Effective size at t–1

Time

H3: –

H2: þ

H1: þ

Hypothesis and Expected Sign

0.327 0.014

Model 1 Unconditional Growth

0.009 0.017 0.113 0.014 5.743 0.736 0.516 0.013 2.237 0.142 0.127 0.021 0.017 0.052

Model 2 Baseline Model 0.193 0.035 0.100 0.013 3.392 0.702 0.504 0.013 2.064 0.136 0.105 0.02 0.029 0.05 0.034 0.006 0.015 0.002

Model 3 Main Effect

0.210 0.038 0.101 0.013 2.986 0.794 0.502 0.013 2.047 0.136 0.110 0.02 0.018 0.05 0.037 0.007 0.015 0.003 0.064 0.021

0.225 0.038 0.109 0.013 2.615 0.803 0.502 0.013 2.043 0.136 0.109 0.02 0.021 0.05 0.099 0.02 0.016 0.003 0.063 0.021

Model 4 - Network Model 5 Horizon Diminishing Returns to Network Heterogeneity Effect Horizon

Multi-level Random Intercept Random Coefﬁcient Model of the Relationship Network Horizon: Firm Effective Size.

Dependent Variable is Effective Size

Table 7.

624 DIEDERIK W. VAN LIERE ET AL.

3

759

1.007 0.195

Yes

4.7

67.9

Average

5

75

Maximum

0.2682133 0.2540137 0.2765944 0.2670362 0.0357258 0.0318909 0.0576923 0.0532466 7363.825 7126.955 7574.231 7349.738 3599 3599 240.87

1.167 0.196

Yes

0.2540226 0.2623271 0.0317817 0.0526199 7121.177 7356.336 3599 9.78

1.113 0.198

Yes

0.023 0.009

0.2564837 0.2446364 0.0309005 0.0497562 7113.184 7354.532 3599 9.99

0.947 0.203

0.003 0.001 Yes

0.018 0.009

Control variables include ﬁrm strategy, network position, manager or student, scenario (14 or 15 ﬁrms) and ﬁrm characteristics. po0.05, po0.01, po0.001.

42

53

Number of experiments Number of ﬁrms

Minimum

0.4335237 1.277857 0.1015733 0.0894696 10197.274 10240.590 3599

2.040 0.072

No

Number of observations (per group)

Within ﬁrm In initial status In rate of change Covariance

H5:

H4: þ

Observation type

AIC BIC N D Log-Likelihood

Variance components Level 1 Level 2

Control variables included Initial status intercept

Rate of change network horizon heterogeneity Network horizon2

Network Horizon 625

626

DIEDERIK W. VAN LIERE ET AL.

holes. Hypothesis 2 is conﬁrmed as well; the direction of coefﬁcient is as expected and signiﬁcant. This ﬁnding suggests that ﬁrms that have a more extended network horizon not only span more structural holes but also at a faster rate compared with ﬁrms that have a limited network horizon. The coefﬁcient for the third hypothesis is negative and signiﬁcant as we predicted in hypothesis 3. The focal ﬁrm is worse off because the competition for structural holes intensiﬁes when there is a greater heterogeneity between the ﬁrms about what they know of the network structure. Higher network horizon heterogeneity indicates that a small group of ﬁrms spot valuable brokerage opportunities that lead to a temporal strengthening of the focal ﬁrm’s bridging position but the partnering decisions of the competing ﬁrms are likely to adversely impact the bridging position of the focal ﬁrm. The coefﬁcient of the fourth hypothesis is positive and signiﬁcant, in line with hypothesis 4. This means that as the competition for valuable brokerage positions intensiﬁes, the ﬁrms that have a comparative advantage will be able to span at faster rate new structural holes. This ﬁnding is consistent with the ﬁnding in hypothesis 2. Finally, the ﬁfth hypothesis is signiﬁcant; the direction of the coefﬁcient is coincides with our expectations. This ﬁnding suggests that there might be a point where too much information about the network structure leads to sub-optimal decision-making. This suggests that bounded rationality plays a role during the partnering decision-making process. Participants have increasing difﬁculty to shift their network position and to accurately assess the impact of their partnering decision on the strength of their bridging position as more information about the network structure becomes available. The participants (in the experiments) start to have difﬁculty with analyzing all the information about the network structure and do not recognize the opportunity to span new structural holes while these brokerage opportunities do exist. An alternative explanation could be that there is a limited information processing capacity of the participants due to the short time given to them to make their decisions.

Robustness Tests In order to increase the convergent validity of the previous ﬁndings, we reestimate the previous model using an alternative operationalization.1 First, we estimate four new models with as dependent variable network constraint; see Table 8 for the new models. The coefﬁcients are signiﬁcant and the signs of the coefﬁcients are according to our predictions and thus conﬁrming our

Rate of change network horizon Network horizon heterogeneity

Number of capabilities owned Attractiveness partner ﬁrms Network horizon

Betweenness centrality

Outdegree

Density

Constraint at t1

Time

H3:

H2: þ

H1: þ

Hypothesis and Expected Sign

5.246 0.195

Model 1 Unconditional Growth

0.353 0.33 0.120 0.013 180.835 14.035 1.847 0.239 21.412 2.62 1.725 0.383 0.141 1.035

Model 2 Baseline Model 3.766 0.615 0.118 0.012 105.350 13.038 1.449 0.212 16.257 2.374 0.945 0.337 0.143 0.947 1.450 0.142 0.267 0.043

Model 3 Main Effect

4.847 0.652 0.122 0.012 130.695 14.829 1.395 0.212 15.668 2.36 1.040 0.335 0.073 0.945 1.238 0.157 0.284 0.043 0.447 0.484

7.008 0.635 0.184 0.012 97.296 14.63 1.484 0.196 14.977 2.265 0.949 0.309 0.123 0.894 5.924 0.348 0.448 0.043 0.895 0.466

Model 4 - Network Model 5 Diminishing Horizon Heterogeneity Returns to Network Horizon Effect

Multi-level Random Intercept Random Coefﬁcient Model for the Relationship Network Horizon: Network Constraint.

The Dependent Variable is Network Constraint (Reverse Coded)

Table 8.

Network Horizon 627

53 759

Number of observations (per group)

Within ﬁrm In initial status In rate of change Covariance

42 3

Minimum

1.885 1.443 3.114 0.976 37073.234 37117.89 3599

No 36.975 1.273

Model 1 Unconditional Growth

Yes 19.409 3.5

Model 3 Main Effect

67.9 4.7

Average

75 5

Maximum

0.790 0.516 1.467 1.385 2.929 2.929 1.328 1.620 28770.192 28262.284 28986.79 28491.25 3599 3599 511.91

Yes 29.210 3.689

Model 2 Baseline Model

0.337 1.377 2.924 1.633 28244.08 28485.43 3599 22.20

Yes 19.290 3.547

0.639

0.458 1.219 2.767 1.822 28057.142 28304.68 3599 188.94

0.265 0.018 Yes 8.689 3.286

0.442 0

Model 4 - Network Model 5 Horizon Diminishing Heterogeneity Returns to Network Horizon Effect

Control variables include ﬁrm strategy, network position, manager or student, scenario (14 or 15 ﬁrms) and ﬁrm characteristics. po0.05, po0.01, po0.001.

Number of experiments Number of ﬁrms

Observation type

AIC BIC N D Log-Likelihood

Variance Components Level 1 Level 2

Control variables included Initial status intercept

H4: þ

Rate of change network horizon heterogeneity Network horizon2

H5:

Hypothesis and Expected Sign

The Dependent Variable is Network Constraint (Reverse Coded)

Table 8. (Continued )

628 DIEDERIK W. VAN LIERE ET AL.

Network Horizon

629

hypotheses. The robustness tests conﬁrm the previous ﬁndings, leading to the overall conclusion that network horizon and network horizon heterogeneity are signiﬁcant predictors of the number of structural holes a ﬁrm spans. Network horizon is a robust predictor of the strength of a ﬁrm’s bridging position. In order to assess the magnitude of this effect, we calculate the effect sizes of the network horizon of a ﬁrm. On average, a ﬁrm spanned 2.83 structural holes in our experiments. A ﬁrm with a more extended network horizon is able to expand its effective size with 1.26 ﬁrms, which is an increase of 44.44%. Intense competition for structural holes is expected to decrease the effective size of a ﬁrm by 8.49% or a reduction in its effective size of 0.2405. Fig. 6 visualizes models 3 (panel 1) and 5 (panel 2) of Table 5. Both models are consistent in their prediction that ﬁrms with limited network horizon (deﬁned as average horizon minus one standard deviation which is roughly 5 ﬁrms) will see their effective size decrease over time. What is striking about panel 1 of Fig. 6 is that the slope of the line for full network horizon is very gentle. This suggests that network horizon is a necessary but insufﬁcient condition for a ﬁrm to be able to span structural holes. Even though the information about brokerage opportunities is available, participants in the experiments were, on average, not able to increase their effective size. The implication of this ﬁnding is that participants do not necessarily understand the beneﬁts of a structural hole ex ante of the experiment. This ﬁnding is consistent with Burt and Ronchi’s (2007) ﬁnding that teaching executives about the structure of social capital enhances their performance because of their better understanding of how social capital functions. Panel 2 of Fig. 6 illustrates the effect of incorporating network horizon heterogeneity and shows that even though individuals may not be able to increase their effective size, they can increase their effective size because of the inequality between ﬁrms in their network horizon.

NETWORK HORIZON, BROKERAGE AND FIRM PERFORMANCE There is a wealth of evidence indicating the positive effect of brokerage on performance (Burt, 2005). Our results show the importance of having information about the network structure in order to increase performance by occupying a bridging position. As ‘‘brokerage across structural holes provides a vision of options otherwise unseen’’ (Burt, 2005, p. 59), we can

Fig. 6.

Illustration of Model 3 (panel 1) and Model 5 (panel 2).

630 DIEDERIK W. VAN LIERE ET AL.

631

Network Horizon

extend this argument to include not only information about valuable opportunities, but also information about any network locations where such information may be more likely obtained in the future (i.e., second-order vision). However, this does raise the question of whether network horizon has any effect on performance itself or whether any effect is simply a by-product of occupying a bridging position (even though an extensive network horizon was necessary to start occupying such a position in the ﬁrst place). In other words, is there an effect of network horizon on performance and to what extent is this effect mediated by brokerage? To test for this mediation effect we use the method proposed by Baron and Kenny (1986). Fig. 7 shows the standardize coefﬁcients. As expected, there is strong relationship between bridging position and ﬁrm performance. However, the effect of network horizon on ﬁrm performance is partly mediated by the bridging position. There is a small but signiﬁcant direct relationship between network horizon and ﬁrm performance. Thus, the ﬁndings from the mediation analysis ﬁnd support for Burt’s vision argument and extend it to include second-order vision, that is information about network locations where valuable opportunities are more likely to be found. They also show that this is not the whole story though, as network horizon also has a direct effect on ﬁrm performance, which further highlights the importance of network horizon.

Bridging Position (Effective size) 0.3211***

Network Horizon

0.5277***

0.1273***

Firm Performance (Natural log net income)

Path coefficients are standardized. *** significant at p<0.001.

Fig. 7.

Bridging Position Partly Mediates the Effect of Network Horizon on Firm Performance.

632

DIEDERIK W. VAN LIERE ET AL.

CONCLUSION This study introduces an information based view on the dynamics of network positions. Although other scholars have proposed an informationbased view on the dynamics and evolution of interﬁrm networks (Moldoveanu et al., 2003), our research is one of the ﬁrst empirical studies to demonstrate the importance of this information-based view. Using network experiments we demonstrated that the completeness of information about the network structure (network horizon) and the interﬁrm differences in the distribution of this type of information (network horizon heterogeneity) are signiﬁcant factors that explain why certain ﬁrms have stronger bridging positions than other ﬁrms do. Our results suggest two key ﬁndings. First, organizational decision makers have to recognize the value that a brokerage opportunity represents and shift their network position accordingly, in other words they have to have a network strategy. Our results, although not a deﬁnitive test of this previously untested assumption, are consistent with the notion that network strategies exist and that they do matter: differences in network-strategy-speciﬁc information result in substantially different network positions. We show that only when decision-makers have sufﬁcient information about the network structure and understand the value of a bridging position, then they will be able to span more structural holes and hence strengthen their ﬁrm’s network position. This ﬁnding suggests that the network horizon and the structural properties of a bridging position are not deterministic factors that shape the evolution of an interﬁrm network. Rather, how organizational decisionmakers collect, process and act on the information about the network structure of an interﬁrm network, given their bounded agency, will determine which ﬁrm will acquire a strong bridging position. Furthermore, organizational decision makers who only focus on strengthening their bridging position at the expense of their closed position may ﬁnd their ﬁrm in the long-run to be vulnerable to free-riding and excessive competitive pressures. Burt (2000, p. 345) concluded in an extensive review of the literature about structural holes that ‘‘structural holes are the source of value added, but network closure can be essential to realizing the value buried in the holes.’’ Thus, our study should not be seen as a normative statement that ﬁrms should maximize the strength of their bridging position, nor are we suggesting that organizational decision-makers are spanning structural holes deliberately (Rowley & Baum, 2008), rather, we demonstrate the value of the information-based view proposed in this chapter. Indeed, despite using quite varied approaches (quantitative, game-theoretic,

Network Horizon

633

and qualitative), several other chapters in this volume (Burt, 2008; Doreian, 2008; Hite, 2008) converge on a similar conclusion as our chapter: taking into account what information actors have about the broader network structure they are embedded in, is crucial for understanding network strategy and the dynamics of network positions. Second, our research suggests that the interﬁrm differences in distribution of information about the network structure are as important as the network horizon of a ﬁrm itself. The network horizon heterogeneity is a robust predictor for the difference in the number of structural holes a ﬁrm spans. Networks characterized by a more heterogeneous distribution of information lead to greater differences between ﬁrms in the strength of their bridging positions. In line with recent theorizing that extends the resourcebased view to network-based competitive advantage (Lavie, 2006), our results suggest that network horizon is another driver of ﬁrm heterogeneity. Thus, to understand the differences in ﬁrms’ network strategies it is not sufﬁcient to only look at the traditional resources of the ﬁrm but rather the ﬁrms’ map of the network, as well as differences in these maps between ﬁrms need to be taken into account. As ﬁrms differ in the resources they allocate to monitoring their environment (Daft, Sormunen, & Parks, 1988), this is likely to result in substantial heterogeneity in ﬁrms’ network horizons. In addition to these theoretical contributions, we have shown the usefulness of laboratory experiments in testing these new constructs and rigorously establishing the direction of causality as going from network horizon to bridging position. Even though the use of experiments in strategy research has been limited so far, despite calls for its use in an earlier volume of Advances in Strategic Management (Weigelt, Camerer, & Hanna, 1992), as well as demonstrations of its value (e.g., Bateman & Zeithaml, 1989; Knez & Camerer, 1994; Weber & Camerer, 2003), laboratory experiments have an important, speciﬁc role to ﬁll. The focus on internal validity that experiments bring, forms a natural complement to the external validity that a ﬁeld study brings, but where the direction of causality and alternative explanations are often much more difﬁcult to evaluate. Especially given the rich tradition of experiments with communication networks and exchange networks, conducting lab experiments makes it a natural testing ground for further development of a theory of network strategy. This research raises a new challenge for management: given the fact that resources are scarce and a more expanded network horizon can be beneﬁcial, which area of a network structure should be monitored? Organizational-decision makers must therefore pay much more attention

634

DIEDERIK W. VAN LIERE ET AL.

to their ﬁrm’s position within the network: this requires that they have an accurate view of the network structure; that they have a network ‘‘map’’ charting their position in respect of other ﬁrms in the network. As advances in communication technologies necessitate ﬁrms to operate in much wider and larger networks, more advanced tools and methods are needed to give organizational decision-makers a view of the important parts of their interﬁrm networks. As most networks are invisible, it is not apparent who has ties with whom; an interesting question becomes how to recognize a network structure that has many opportunities to span structural holes without having a complete overview of the network. However, the fact that we found diminishing returns suggests that a ﬁrm does not have to monitor the complete network in order to ﬁnd an efﬁcient network position. Network horizon is a dynamic construct, the information a ﬁrm has of the network can increase or decrease over time. New sources of information are tapped into by establishing new non-redundant ties that in turn will increase the available information about the network structure, while disbanding ties will result in a loss of information about the network structure. Firms can decide to invest more time and energy in monitoring the network and thereby expanding the information they have about the network structure. By bringing in a ﬁrm construct (network horizon) in the explanation of network structure properties and the explanation of network structure emergence, we move away from a pure structuralist perspective (Emirbayer & Goodwin, 1994) and we hope that this research will contribute to a greater understanding of the dynamics of network positions and that the information-based view of interﬁrm networks inspire other scholars as well to study network strategies.

NOTE 1. We also estimated the full model using effective size as the dependent variable for the sub population of the managers and the sub population of the student team. These models also support our previous ﬁndings. These models are available upon request.

ACKNOWLEDGMENTS We are grateful to Lorike Hagdorn, Martijn Hoogeweegen, Niek Hoek, Frank Elion, Gerrit Schipper, Herman Bennema, Dominique Campman, and Ido de Lepper who made this study possible and we are indebted to the

635

Network Horizon

many useful suggestions from the participants of the Network Strategy conference hosted at the Rotman School of Management and in particular Joel Baum and Tim Rowley.

REFERENCES Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative Science Quarterly, 45(3), 425–455. Anderson, J. C., Hakansson, H., & Johanson, J. (1994). Dyadic business relationships within a business network context. Journal of Marketing, 58(4), 1–15. Bae, J. H., & Gargiulo, M. (2004). Partner substitutability, alliance network structure, and ﬁrm proﬁtability in the telecommunications industry. Academy of Management Journal, 47(6), 843–859. Bala, V., & Goyal, S. (2000). A noncooperative model of network formation. Econometrica, 68(5), 1181–1229. Barney, J. B. (1986). Strategic factor markets: Expectations, luck, and business strategy. Management Science, 32(10), 1231–1241. Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173–1182. Bateman, T. S., & Zeithaml, C. P. (1989). The psychological context of strategic decisions: A model and convergent experimental ﬁndings. Strategic Management Journal, 10(1), 59–74. Baum, J. A. C., Rowley, T. J., Shipilov, A. V., & Chuang, Y. T. (2005). Dancing with strangers: Aspiration performance and the search for underwriting syndicate partners. Administrative Science Quarterly, 50(4), 536–575. Baum, J. A. C., Shipilov, A. V., & Rowley, T. J. (2003). Where do small worlds come from? Industrial and Corporate Change, 12(4), 697–725. Bavelas, A. (1950). Communication patterns in task-oriented groups. Journal of the Acoustical Society of America, 22, 723–730. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. (2000). The network structure of social capital. In: R. I. Sutton & S. B. M. (Eds.), Research in organizational behavior (Vol. 22, pp. 345–423). New York, NY: JAI/ Elsevier. Burt, R. S. (2002). Bridge decay. Social Networks, 24(4), 333–363. Burt, R. S. (2005). Brokerage and closure: An introduction to social capital. Oxford, UK: Oxford University Press. Burt, R. S. (2008). Industry performance and industry access to structural holes. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in strategic management (Vol. 25, pp. 315–360). Bingley, UK: JAI/Emerald Group. Burt, R. S., & Ronchi, D. (2007). Teaching executives to see social capital: Results from a ﬁeld experiment. Social Science Research, 36, 1156–1183. Buskens, V., & van de Rijt, A. (2007). Dynamics of networks if everyone strives for structural holes. Working Paper.

636

DIEDERIK W. VAN LIERE ET AL.

Casciaro, T. (1998). Seeing things clearly: Social structure, personality, and accuracy in social network perception. Social Networks, 20(4), 331–351. Chen, M. J. (1996). Competitor analysis and interﬁrm rivalry: Toward a theoretical integration. Academy of Management Review, 21(1), 100–134. Cook, K. S. (1977). Exchange and power in networks of interorganizational relations. Sociological Quarterly, 18, 62–82. Cook, K. S., & Emerson, R. M. (1978). Power, equity and commitment in exchange networks. American Sociological Review, 43(5), 721–739. Cook, K. S., Emerson, R. M., & Gillmore, M. R. (1983). The distribution of power in exchange networks: Theory and experimental results. American Journal of Sociology, 89(2), 275–305. Daft, R. L., Sormunen, J., & Parks, D. (1988). Chief executive scanning, environmental characteristics, and company performance: An empirical-study. Strategic Management Journal, 9(2), 123–139. Daft, R. L., & Weick, K. E. (1984). Toward a model of organizations as interpretation systems. Academy of Management Review, 9(2), 284–295. Denrell, J., Fang, C., & Winter, S. G. (2003). The economics of strategic opportunity. Strategic Management Journal, 24(10), 977–990. Doreian, P. (2008). Actor utilities, strategic action and network evolution. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in strategic management (Vol. 25, pp. 247–271). Bingley, UK: JAI/Emerald Group. Emerson, R. M. (1962). Power-dependence relations. American Sociological Review, 27, 31–40. Emerson, R. M. (1972a). Exchange theory, Part I: A psychological basis for social exchange. In: J. Berger, M. Zelditch, Jr. & B. Anderson (Eds), Sociological theories in progress (Vol. 2, pp. 38–57). Houghton-Mifﬂin. Emerson, R. M. (1972b). Exchange theory, Part Ii: Exchange relations and networks. In: J. Berger, M. Zelditch, Jr. & B. Anderson (Eds), Sociological theories in progress (Vol. 2, pp. 58–87). Houghton-Mifﬂin. Emirbayer, M., & Goodwin, J. (1994). Network analysis, culture, and the problem of agency. American Journal of Sociology, 99(6), 1411–1454. Freeman, L. C. (1992). Filling in the blanks: A theory of cognitive categories and the structure of social afﬁliation. Social Psychology Quarterly, 55(2), 118–127. Friedkin, N. E. (1983). Horizons of observability and limits of informal control in organizations. Social Forces, 62(1), 54–77. Fund, B. R., Pollock, T. G., Baker, T., & Wowak, A. J. (2008). Who’s the new kid? The process of developing centrality in venture capitalist deal networks. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in strategic management (Vol. 25, pp. 563–593). Bingley, UK: JAI/Emerald Group. Gimeno, J. (2004). Competition within and between Networks: The contingent effect of competitive embeddedness on alliance formation. Academy of Management Journal, 47(6), 820–842. Gnyawali, D. R., & Madhavan, R. (2001). Cooperative networks and competitive dynamics: A structural embeddedness perspective. Academy of Management Review, 26(3), 431–445. Gomes-Casseres, B. (1996). The alliance revolution: The new shape of business rivalry. Cambridge, MA: Harvard University Press. Gould, R. V. (2002). The origins of status hierarchies: A formal theory and empirical test. American Journal of Sociology, 107(5), 1143–1178.

Network Horizon

637

Guetzkow, H., & Simon, H. A. (1955). The impact of certain communication nets upon organization and performance in task-oriented groups. Management Science, 1, 233–250. Gulati, R. (1995). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40(4), 619–652. Gulati, R. (1998). Alliances and networks. Strategic Management Journal, 19(4), 293–317. Gulati, R. (1999). Network location and learning: The inﬂuence of network resources and ﬁrm capabilities on alliance formation. Strategic Management Journal, 20(5), 397–420. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1439–1493. Gulati, R., Nohria, N., & Zaheer, A. (2000). Strategic networks. Strategic Management Journal, 21(3), 203–215. Hite, J. M. (2008). The role of dyadic multi-dimensionality in the evolution of strategic network ties. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in strategic management (Vol. 25, pp. 133–170). Bingley, UK: JAI/Emerald Group. Hite, J. M., & Hesterly, W. S. (2001). The evolution of ﬁrm networks: From emergence to early growth of the ﬁrm. Strategic Management Journal, 22(3), 275–286. Hoogeweegen, M. R., van Liere, D. W., Vervest, P. H. M., Hagdorn van der Meijden, L., & de Lepper, I. (2006). Strategizing for mass customization by playing the business networking game. Decision Support Systems, 42(3), 1402–1412. Inkpen, A., & Choudhury, N. (1995). The seeking of strategy where it is not: Towards a theory of strategy absence. Strategic Management Journal, 16(4), 313–323. Kennedy, P. (2003). A guide to econometrics (5th ed.). Cambridge, MA: The MIT Press. Kirzner, I. M. (1997). Entrepreneurial discovery and the competitive market process: An Austrian approach. Journal of Economic Literature, 35(1), 60–85. Klein, K., & Harrison, D. A. (2007). What’s the difference? Diversity constructs as separation, variety or disparity in organizations. Academy of Management Review, 32(4), 1199–1228. Knez, M., & Camerer, C. (1994). Creating expectational assets in the laboratory: Coordination in weakest-link games. Strategic Management Journal, 15, 101–119. Krackhardt, D. (1990). Assessing the political landscape: Structure, cognition, and power in organizations. Administrative Science Quarterly, 35(2), 342–369. Lavie, D. (2006). The competitive advantage of interconnected ﬁrms: An extension of the resource-based view. Academy of Management Review, 31(3), 638–658. Leavitt, H. J. (1951). Some effects of certain communication patterns on group performance. Journal of Abnormal and Social Psychology, 46, 38–50. Leik, R. K. (1992). New directions for network exchange theory: Strategic manipulation of network linkages. Social Networks, 14(3–4), 309–323. Lucas, J. W., Younts, C. W., Lovaglia, M. J., & Markovsky, B. (2001). Lines of power in exchange networks. Social Forces, 80(1), 185–214. Madhavan, R., Koka, B. R., & Prescott, J. E. (1998). Networks in transition: How industry events (re)shape interﬁrm relationships. Strategic Management Journal, 19(5), 439–459. Makadok, R., & Barney, J. B. (2001). Strategic factor market intelligence: An application of information economics to strategy formulation and competitor intelligence. Management Science, 47(12), 1621–1638. Markovsky, B., Willer, D., & Patton, T. (1988). Power relations in exchange networks. American Sociological Review, 53(2), 220–236.

638

DIEDERIK W. VAN LIERE ET AL.

Moldoveanu, M. C., Baum, J. A. C., & Rowley, T. J. (2003). Information regimes, information strategies and the evolution of interﬁrm network topologies. In: F. J. Yammarino & F. Dansereau (Eds), Research in Multi-Level Issues (Vol. 2, pp. 221–264). Oxford, UK: Elsevier. Moore, M. C. (1992). Signals and choices in a competitive interaction: The role of moves and messages. Management Science, 38(4), 483–500. Ozgen, E., & Baron, R. A. (2007). Social sources of information in opportunity recognition: Effects of mentors, industry networks, and professional forums. Journal of Business Venturing, 22(2), 174–192. Pascale, R. T. (1984). Perspectives on strategy: The real story behind Honda’s success. California Management Review, 26, 42–72. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations: A resource dependence perspective. New York, NY: Harper & Row. Pollock, T. G., Porac, J. F., & Wade, J. B. (2004). Constructing deal networks: Brokers as network ‘‘Architects’’ in the US IPO market and other examples. Academy of Management Review, 29(1), 50–72. Powell, W. W., Koput, K. W., Owen-Smith, J., & Smith-Doerr, L. (1999). Network position and ﬁrm performance: Organizational returns to collaboration in the biotechnology industry. In: S. B. Andrews & D. Knoke (Eds), Networks in and around Organizations (Vol. 16, pp. 129–159). Greenwich, CT: JAI Press. Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network dynamics and ﬁeld evolution: The growth of interorganizational collaboration in the life sciences. American Journal of Sociology, 110(4), 1132–1205. Reagans, R., & McEvily, B. (2003). Network structure and knowledge transfer: The effects of cohesion and range. Administrative Science Quarterly, 48(2), 240–267. Rowley, T. J., & Baum, J. A. C. (2004). Sophistication of interﬁrm network strategies in the Canadian investment banking industry. Scandinavian Journal of Management, 20(1–2), 103–124. Rowley, T. J., & Baum, J. A. C. (2008). The dynamics of network strategies and positions. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in Strategic Management (Vol. 25, pp. 641–671). Bingley, UK: JAI/Emerald Group. Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Boston, MA: Houghton Mifﬂin Company. Shaw, M. (1964). Communication networks. In: L. Berkowitz (Ed.), Advances in experimental psychology. New York, NY: Academic Press. Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York, NY: Oxford University Press. Skvoretz, J., & Willer, D. (1993). Exclusion and power: A test of four theories of power in exchange networks. American Sociological Review, 58(6), 801–818. Sorenson, O., & Ryall, M. (2007). Brokers and competitive advantage. Management Science, 53(4 (Special Issue Strategic Dynamics)), 566–583. Stuart, T. E. (1998). Network positions and propensities to collaborate: An investigation of strategic alliance formation in a high-technology industry. Administrative Science Quarterly, 43(3), 668–698. Sutcliffe, K. M. (1994). What executives notice: Accurate perceptions in top management teams. Academy of Management Journal, 37(5), 1360–1378.

Network Horizon

639

Treacy, M., & Wiersema, F. (1993). Customer intimacy and other value disciplines. Harvard Business Review, 71(1), 84–93. Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42(1), 35–67. van de Ven, A. H. (1976). On the nature, formation, and maintenance of relations among organizations. The Academy of Management Review, 1(4), 24–36. van Liere, D. W., Hagdorn, L., Hoogeweegen, M. R., & Vervest, P. H. M. (2004). Embedded coordination in a business network. Journal of Information Technology, 19(4), 261–269. Wasserman, F., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge, UK: Cambridge University Press. Weber, R. A., & Camerer, C. F. (2003). Cultural conﬂict and merger failure: An experimental approach. Management Science, 49(4), 400–415. Weigelt, K., Camerer, C. F., & Hanna, M. (1992). The use of experimental economics in strategy research. In: P. Shrivastava, J. E. Dutton & A. Huff (Eds), Advances in strategic management (Vol. 8, pp. 163–200). Greenwich, CT: JAI Press. Weisstein, E. W. (2006). Convex hull. From MathWorld: A Wolfram web resource. Wilcox King, A. (2007). Disentangling inter- and intra-ﬁrm causal ambiguity: A conceptual model of causal ambiguity and sustainable competitive advantage. Academy of Management Review, 32(1), 156–178. Willer, R., & Willer, D. (2000). Exploring dynamic networks: Hypotheses and conjectures. Social Networks, 22(3), 251–272.

THE DYNAMICS OF NETWORK STRATEGIES AND POSITIONS Timothy J. Rowley and Joel A. C. Baum ABSTRACT In this study, we seek to broaden the research focus in the strategic alliance literature from a ﬁrm’s ‘‘partner strategy’’ to its ‘‘network strategy’’ by linking a ﬁrm’s partnering choices to changes in its network position over time. Using data on all underwriting syndicates in Canada over nearly 40 years, we conceptualize and model the interplay between an investment bank’s own and its partners’ syndicate participation. Our ﬁndings indicate that the lead banks, which have greater discretion in choosing syndicate partners than co-lead banks, are more likely to make partner selections that create bridging positions that provide access to timely and non-redundant information as well as opportunities to play a broker role across unconnected others. We also ﬁnd, however, that lead banks’ bridging positions deteriorate when they form ties with other lead banks. Network-based competitive advantages are thus inﬂuenced by network opportunities and constraints as well as partner-speciﬁc concerns, suggesting that new insights into the dynamics of interﬁrm networks and competitive advantage of ﬁrms are possible within this broader view.

Network Strategy Advances in Strategic Management, Volume 25, 641–671 Copyright r 2008 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0742-3322/doi:10.1016/S0742-3322(08)25018-7

641

642

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

INTRODUCTION The strategic alliance and networks literature is dominated by the idea that ﬁrms’ positions in interﬁrm networks can be a powerful source of competitive advantage. The web of partnerships that ﬁrms spin embed them in various network positions, offering network-based competitive advantages in terms of differential access to and control of scarce information, skills, knowledge, and other resources ﬂowing within the network (Powell, Koput, & Smith-Doerr, 1996). To date, however, empirical research has not focused on ﬁrms’ strategic use of collaborative arrangements to improve their network positions, but rather on their partner strategies – their use of partnering to create dyadic competitive advantages, primarily based on more efﬁcient access to scarce, complementary resources (Hennart, 1988; Kogut, 1988; Williamson, 1991). This research emphasizes how uncertainty about the reliability, trustworthiness and capabilities of potential partners makes selection of more familiar, past partners and partners’ partners more likely, bringing stability and closure to emergent network structures and positions (Anand & Piskorski, 2000; Chung, Singh, & Lee, 2000; Gulati, 1995; Li & Rowley, 2002; but see Baum, Shipilov, & Rowley, 2003; Baum, Rowley, Shipilov, & Chuang, 2005). From this perspective, ﬁrms’ network positions take shape as they enter into collaborative agreements based on information obtained from their own and their partners’ prior relationships, resulting in self-reinforcing patterns of ‘‘cliquey’’ relations among ﬁrms characterized by strong ties and closure (Baum & Ingram, 2002), and distinct relational proﬁles (e.g., core and periphery positions) (Gulati & Gargiulo, 1999). In this study, in contrast, we are concerned with ﬁrms’ network strategies – their use of partner selection to improve their access to information and brokerage advantages in the network (e.g., Burt, 1992; Doreian, 2002; Walker, Kogut, & Shan, 1997). Network strategies move beyond the dyadic concerns of the quality of particular partners as ﬁrms pursing such strategies are also attentive to relational advantages – for example, the affect a new tie will have on its ability to access and broker timely information among its existing partners (Burt, 1992; Emerson, 1962; Pfeffer & Salancik, 1978). Although research on ﬁrms’ partner strategies provides robust evidence of managers’ preference for past partners and partners’ partners for new alliances, this does not mean that managers inevitably reproduce their ﬁrms’ past ties (Baker, Faulkner, & Fisher, 1998; Palmer, 1983). Indeed, partner strategy studies show that as two ﬁrms continue to accumulate priorties, or

The Dynamics of Network Strategies and Positions

643

time elapses since they last allied, they become less likely to partner again (e.g., Hagedoorn & Frankort, 2008; Gulati, 1995). Managers’ attention to partner attributes does not necessarily imply that they ignore the network position consequences (information access and brokerage opportunities) of their partner selections (Burt, 1992; Chung et al., 2000; Doreian, 2002; Podolny, 1993). While managers tend to maintain ties with a core set of ﬁrms, stabilizing the network, they also partner with new ﬁrms, altering the network structure and their ﬁrm’s position within it (Hite & Hesterly, 2001; Leik, 1992). As Fund, Pollock, Baker, and Wowak (2008) show, for example, venture capital ﬁrms use partner selection to migrate from the periphery to the core of their industry network. Drawing on theory and research identifying advantages of particular network positions (e.g., Ahuja, 2000; Burt, 1992; Rowley, Behrens, & Krackhardt, 2000) and suggesting that ﬁrms’ partnering behavior is inﬂuenced by information and brokerage opportunities and constraints (e.g., Burt, 1992; Walker et al., 1997), we examine how ﬁrms’ partner choices affect their network positions, improving or worsening them over time. Thus, while taking seriously the role of partner-speciﬁc concerns in partner selection (partner capabilities and reliability), which provides fundamental insights into the creation of dyadic advantages, we seek to broaden research from its focus on ﬁrms’ partner strategies to include consideration of their network strategies by linking ﬁrms’ partner choices to changes in their network positions over time. Few researchers have yet to examine the mechanisms underlying persistence and change in interﬁrm networks and ﬁrms’ positions within them (Nohria, 1992). By focusing on ﬁrms’ network strategies and changing network positions over time, our study necessarily departs from static conceptions in which networks are viewed ‘‘as given contexts for action, rather than as being subject to deliberate design’’ (Madhavan, Koka, & Prescott, 1998, p. 439). In a dynamic conception, partnering decisions produce, reproduce and transform the network as ﬁrms construct and reconstruct their network positions. A dynamic perspective also highlights interﬁrm rivalry for scarce preferred positions; how a ﬁrm’s partner selection inﬂuences its network position depends on the choices its partners make. Our focus on such endogenous drivers of network dynamics – ﬁrms’ partner and network strategies within the network – complements research on exogenous factors reinforcing and loosening network structures (Madhavan et al., 1998; Neuman, Davis, & Mizruchi, 2008), as well as micro work on the dynamics of ﬁrms’ ego network composition emphasizing managers’ and entrepreneurs’ proactive (re)shaping of their ﬁrms’ relationships over time to

644

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

match changing demands (Fund et al., 2008; Hite, 2008; Hite & Hesterly, 2001; Steier & Greenwood, 2000). We test our ideas in a study of underwriting syndicates formed by investment banks in Canada between 1952 and 1990. This empirical setting provides an opportunity to examine how these ﬁrms’ partnering behavior shapes their network positions over a timeframe of almost 40 years. The history of investment banks’ syndicate ties is a meaningful setting for our analysis of network strategies. In the investment banking industry, the ability to attract partners is essential to survival and success: Underwriting syndicate relationships formed among multiple investment banks are essential for marketing deals and spreading risk (Eccles & Crane, 1988; Podolny, 1993). Not only is partnering a common practice vital to effectively placing issues, but these relationships also act as conduits to underwriting opportunities and contribute greatly to banks’ reputations and performance (Pollock, Porac, & Wade, 2004). Below, we brieﬂy describe two types of network advantages – bridging and embeddedness – and why bridging ties are especially advantageous in our empirical context. We then derive hypotheses relating a bank’s own partnering decisions and the decisions of its partners to changes in the bank’s network position and advantages.

NETWORK ADVANTAGES Embedded Positions Meyer and Rowan (1977) and Oliver (1991) argue that densely interconnected ties among actors facilitate the diffusion of norms across the network constituted by those ties. As a result, ﬁrms embedded in highly interconnected networks develop shared behavioral expectations leading to cooperative behavior (Coleman, 1988; Rowley, 1997). ‘‘If all ﬁrms in an industry had relationships with each other, interﬁrm information ﬂows would lead quickly to established norms of cooperation. In such a dense network, information on deviant behavior would be readily disseminated and the behavior sanctioned’’ (Walker, et al., 1997, p. 111). Consequently, the advantage of embedded positions resides in network density, which creates governance and reputation effects that mitigate partner uncertainty – that is, a ﬁrm is less likely to cheat an alliance partner, that is tied to the same third parties, because these mutual contacts will be aware of and able to punish non-cooperative behavior (Burt & Knez, 1995; Kreps, 1990).

The Dynamics of Network Strategies and Positions

645

Researchers have not only articulated advantages associated with embeddedness, but also the process by which ﬁrms come to occupy these positions. Local search behavior in partner selection decisions is the central mechanism creating embedded positions; and this behavior is triggered when ﬁrms face high levels of partnering uncertainty (Anand & Piskorski, 2000; Chung et al., 2000; Gulati, 1995; Gulati & Gargiulo, 1999; Li & Rowley, 2002; Uzzi, 1997). When ﬁrms are unfamiliar and/or concerned about their (potential) partners’ capabilities and willingness to cooperate, they tend to choose their past partners and partners’ partners because evaluating potential partners’ reliability and capabilities is problematic beyond their local networks (Baum & Ingram, 2002; Cyert & March, 1963; Gulati & Gargiulo, 1999; Walker et al., 1997). For example, Ahuja (2000) ﬁnds embeddedness is especially important for R&D alliances among U.S. chemical ﬁrms because high levels of partnering uncertainty would otherwise impede partners’ willingness to share proprietary knowledge.

Bridging Positions Despite evidence that interorganizational partnering behavior propels ﬁrms into embedded positions, embeddedness is not universally advantageous as ﬁrms may beneﬁt from bridging positions. Indeed, researchers have observed ﬁrms forming ties with partners outside their local networks, reducing their embeddedness (Baum et al., 2003, 2005; Gulati, Nohria, & Zaheer, 2000; Hite & Hesterly, 2001; Leik, 1992; Walker, 2008). Firms occupy bridging positions when they are connected to partners who are not connected to one another. By bridging these so-called ‘‘structural holes’’ ﬁrms reduce their number of redundant ties (relative to embedded positions) because a greater proportion of its partners represent novel information source (Burt, 1992). Further, ﬁrms in bridging positions are awarded control/power beneﬁts because they act as intermediaries between disconnected partners, who rely on the ﬁrm to facilitate exchange ﬂows across the network (Burt, 1992). The downside of a sparsely interconnected network is that the structure does not act as a governance mechanism to impede opportunism. Consequently, bridging positions are especially advantageous in information-intensive environments in which access and control of information are keys to ﬁrm survival and success (Burt, 1992). And ﬁrms are better able to forego the relative certainty of choosing local, embedded partners to forge less familiar linkages when alternative means to embeddedness exist to reduce partnering uncertainty (Rowley et al., 2000). When these conditions

646

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

are present ﬁrms gain more from bridging positions than in other situations. However, the existence of these conditions will not necessarily mean ﬁrms will come to occupy bridging positions. Not all ﬁrms can occupy bridging positions within a network – structural hole advantages are diminished as more ﬁrms span them (Burt, 1992). So, researchers need to consider what types of ﬁrms are able to gain these positions and how the interplay between a ﬁrm and its partners’ alliance choices inﬂuence network positions.

Investment Banking We examine the process through which investment banks come to occupy bridging positions in the network of underwriting syndicates because the conditions making bridging positions especially valuable are present. Investment banking is characterized as a relatively small and stable community of players (Pollock et al., 2004): The smallness of the community creates reputation effects as all bankers are aware of one another, and its stability creates a long shadow of the future for banks engaged in partnerships with one another. The long shadow of the future decreases the value of (short-term) non-cooperative behavior, because the long term advantage of future collaborations with that partner (and its partners) would be diminished (Axelrod, 1984). Research has demonstrated the importance of reputation in this industry for attracting partners and serving as a governance mechanism encouraging collaboration (Li & Rowley, 2002). Industry characteristics thus create incentives for cooperative behavior and reduce partnering uncertainty. Consequently, investment banks rely less on governance beneﬁts of embeddedness relative to other relationship-oriented settings. Underwriting is also characterized as information oriented. With underwriting fees generally constant across the industry (Ljungqvist & Wilhelm, 2003), performance is strongly tied to market share (volume), which is achieved by increasing the bank’s access to underwriting opportunities – either through the bank’s issuing ﬁrms or its syndicate partners. Information regarding potential underwriting deals, other banks’ relationships, and investor interests is essential to success; and obtaining diverse, high quality information sooner rather than later is a vital source of competitive advantage. These characteristics are consistent with Burt’s (1992, p. 116) description of bridging position advantages: Firms in bridging positions ‘‘are the players who know about, have a hand in, and exercise control over more rewarding opportunities. They have broader access to information

The Dynamics of Network Strategies and Positions

647

because of their diverse contacts (the information access beneﬁt of holes). That means they are often aware of new opportunities and aware earlier than their peers (information timing beneﬁts).’’ Thus, by spanning structural holes within the industry network investment banks are able to maximize the number of underwriting deals to which they have access and control, while minimizing their costs in terms of the number of relationships managed. And the governance beneﬁts of embeddedness are partly redundant to the advantages gained from the nature of underwriting relationships: Banks create repeated short-term ties among a relatively stable set of ﬁrms, which is instrumental in aligning syndicate members’ goals and limiting incentives for opportunism (Ritter, 1991). Supporting this view, Rowley and Baum (2004) and Shipilov (2006) found positive relationships between investment banks’ market share and the degree to which they occupied bridging rather than embedded positions. Because banks are positioned unevenly across networks, however, only a subset tends to occupy bridging positions (Burt, 1992). What types of banks tend to occupy and reap the beneﬁts of bridging positions? And, are managers aware of network opportunities and do they pursue them?

THEORY AND HYPOTHESES1 Partner versus Network Awareness To date, research on ﬁrms’ choice of partners has focused on partner strategy – the creation of dyadic advantages through the strategic use of partnering to obtain efﬁcient, certain or timely access to scarce, complementary resources (Hennart, 1988; Kogut, 1988; Williamson, 1991). Interﬁrm alliance studies consistently ﬁnd that ﬁrms have a propensity to select their more familiar past partners as well as their partners’ partners when forming new alliances (Anand & Piskorski, 2000; Chung et al., 2000; Gulati, 1995; Gulati & Gargiulo, 1999; Li & Rowley, 2002; Uzzi, 1997). Because evaluating these qualities beyond their direct relationships is problematic, their partnering choices tend to congeal around the existing network structure, fostering densely interconnected and cliquey network structures (Baum & Ingram, 2002). The network strategy perspective, however, suggests a different logic, in which ﬁrms’ partner strategically to gain network-based advantage – superior access to timely and non-redundant information as well as the ability to broker between unconnected others (Burt, 1992; Leik, 1992;

648

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

Walker et al., 1997). Structural hole theory directs ﬁrms to embed themselves in sparse networks of disconnected partners, portraying managers as actively seeking network opportunities: ‘‘The task for a strategic player building an efﬁcient-effective network is to focus resources on the maintenance of bridge ties’’ (Burt, 1992, p. 30), and s/he must thus be alert to the locations of structural holes and their beneﬁts. ‘‘The social beneﬁts of bridges,’’ Burt asserts, ‘‘are an incentive to build them’’ (2002, p. 340). Doreian’s ‘‘ﬁrst principle’’ of network evolution, which states that ‘‘networks have instrumental character for network members as these members have structured goals and some goals are achieved through network choices’’ (2002, p. 95), encapsulates the network strategy view. The network perspective is based on an assumption that managers are, to some degree, aware of network positions and (dis)advantages. Pollock et al. (2004), for example, suggest that in relationship-intensive industries such as investment banking, managers behave as ‘‘network architects,’’ intentionally designing relationship structures to create advantage. However, network cognition is a relatively understudied and thus contentious factor in models of partnering behavior and network dynamics. Are ﬁrms’ managers capable of visualizing their networks in sufﬁcient detail to see network positions; and do they pursue network positions that provide advantage, such as bridging tie opportunities in the underwriting industry? While ‘‘structural holes’’ may be an abstract network concept beyond the experience of many managers, the beneﬁts of occupying positions between two disconnected partners are highly tangible. Accessing timely information and brokering relationships are well-known fundamentals of rent generation. Exchange and resource dependence theory research provides numerous empirical examples of managers searching for relationships that provide brokering positions that increase their organizations’ relative power (Emerson, 1962; Pfeffer & Salancik, 1978). To pursue a bridging strategy, investment bank managers need not comprehend the industry network in its entirety. Indeed, they face two far less demanding requirements: (1) knowledge of their ego networks; that is, the identity of their bank’s partners and their partners’ partners, and (2) an understanding of the value of brokering or mediating the ﬂow of information between its unconnected partners.2 Regarding network knowledge, Friedkin’s (1983) work on ‘‘network horizons’’ suggests that actors typically possess knowledge of their networks within two geodesic steps. Thus, while lacking knowledge of their entire network, managers’ cognitive network maps plausibly include their own ﬁrm’s ego networks and the bridging opportunities surrounding its partners’

The Dynamics of Network Strategies and Positions

649

relationships. And, regarding the value of brokering, in an experimental setting, van Liere, Koppius and Vervest (2008) showed that subjects possessing greater information about network structure made partnering decisions that moved them into positions richer in structural holes. Together, these studies support the idea that managers may possess sufﬁcient awareness of their ego network structure and the value of brokering relationships to pursue advantageous network positions. For investment bank managers, in particular, these requirements are well within the scope of their regular underwriting syndicate activities because they have both the opportunity and incentive to be knowledgeable about whom their partners are connected to as well as the value of playing a brokering role among them. Underwriting relationships among investment banks become public knowledge because published ‘‘tombstones’’ comprehensively announce the lead and co-lead banks involved in every syndicate. At minimal cost banks are aware of the relationships among their partners as well as those banks not linked to their partners. Investment bank managers also understand the value of being a bridge between unconnected banks as brokering is their core revenue-generating activity: The basis of underwriting success is having knowledge and control of information regarding issuing corporations, investors and other banks. Consequently, we expect investment bank managers to be able and motivated to pursue a network strategy in which their partnering decisions serve to increase the number of structural holes in their bank’s ego network. Below, we identify bank characteristics and competitive conditions under which we expect banks to be more or less able to engage in such a strategy.

Firm Characteristics While it is plausible that investment bank managers understand their local networks and the value of spanning structural holes, there are two types of partnering decisions that affect the extent to which bank managers can pursue network strategies. Acting as a lead bank for an issue, a bank must select co-lead banks to form the underwriting syndicate. And, as a potential co-lead, it must decide whether or not to accept an invitation to join a given underwriting syndicate. As initiators of underwriting syndicates, lead banks can inﬂuence their network positions through their selection of co-leads, which, in turn, must choose to accept or decline their invitations. Although prior research on partner strategies suggests that lead banks will tend to select past partners (and partners’ partners), lead banks have substantial

650

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

discretion in syndicate member selection, potentially inviting any bank in the industry to create advantageous brokering opportunities. Acting as co-leads, in contrast, banks are by deﬁnition more restricted in their choices, accepting or rejecting invitations they receive, rather than actively initiating ties. Nevertheless, it is still possible for co-lead banks to pursue network strategies, refusing invitations that worsen their network positions, and accepting those that improve them. Indeed, it is possible that as lead banks seek brokering opportunities for themselves, they create similar opportunities for co-lead banks positioned at the opposite ends of the structural holes they endeavor to span. In sum, if investment bank managers are pursuing network strategy logic – searching for ties that provide information and brokerage advantages – one would expect that investment banks’ partnering decisions will increase the number of structural holes they span, but to a greater extent when acting as lead bank than as co-lead bank. Hypothesis 1 (H1). The larger the number of lead roles a bank plays, the greater the number of structural holes it spans. Hypothesis 2 (H2). The larger the number of co-lead roles a bank plays, the greater the number of structural holes it spans; however, the increase is smaller than for lead roles.

Competitive Dynamics of Network Position Although a bank’s own partnering decisions are vital to its pursuit of network advantage, so are those of its partners. A bank’s ability to access non-redundant information and brokerage opportunities depends on with whom its partners partner. Just as competitive advantages based on product differentiation and low cost production are often short-lived as ﬁrms imitate and improve on one another’s products and processes, information and brokerage advantages are similarly eroded through interﬁrm competition to span structural holes. Once a bank has established a position between unconnected partners, others may follow. As more and more banks maneuver to secure these positions, the density of interﬁrm network ties increases and the structural hole disappears along with the incumbent’s initial advantage. Bridging positions are thus not only a highly valuable resource – they are also a highly contested resource. Ultimately, this

The Dynamics of Network Strategies and Positions

651

competitive interaction weakens bridging position advantages for all banks as the density of the network increases. As noted, however, not all banks are equally capable of pursuing network strategies. In contrast to banks acting predominately as co-leads, banks acting mostly as leads should thus be more effective in securing bridging positions because they are initiators of syndicate relationships. This differential ability suggests that the intensity of network ‘‘competition’’ a bank experiences from its partners’ will depend on the frequency with which it and its partners act as lead banks. Competition for bridging positions between a bank and its partners should be greatest when both the bank and its partners frequently act as syndicate leads, and so have frequent opportunities to initiate partnerships that reduce their dependence on the existing partners and span structural holes. So, while we expect that a bank frequently acting as lead will be able to improve its network position (H1), we also expect that when such banks select each other to co-lead syndicates, this will have a crowding effect: Because lead banks are likely to occupy structurally similar positions and to have similar discretion and interests with respect to network positions (Doreian, 2008), when they partner they are likely to compete with one another to bridge the same structural holes, reducing each partner’s ability to span structural holes. Competition is more lopsided for a bank that acts in the lead role frequently, while its partners typically act as co-leads, enabling the focal bank to ‘‘outcompete’’ its partners in the struggle for autonomy and information and brokerage advantages that accompany non-redundant ties. In the opposite situation – when the focal bank is frequently a co-lead and its partners are frequently leads – the focal bank should suffer a similar fate. The foregoing discussion of network competitive dynamics suggests that Hypothesis 3 (H3). The larger the number of lead roles a bank and its partners play, the smaller the number of structural holes it spans. Hypothesis 4 (H4). The larger the number of lead (co-lead) roles a bank plays and co-lead (lead) roles its partners play, the greater (smaller) the number of structural holes it spans. It is also possible for a syndicate to be formed and comprised of banks that typically act as co-lead, and only infrequently as lead banks. Such syndicates seem destined for disadvantaged positions in the network, with neither the acting lead nor co-leads in the syndicate having access to signiﬁcant brokering opportunities. Comparatively, while a bank that

652

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

frequently co-leads for banks that frequently lead syndicates is competitively disadvantaged (H4), it can at least potentially gain access to information ‘‘secondhand’’ through the structural holes its syndicate leaders span (Burt 2008). Thus, we expect syndicates formed by and comprised of banks that commonly act as co-leads to provide the most limited opportunities to establish brokering relationships: Hypothesis 5 (H5). The larger the number of co-lead roles a bank and its partners play, the smaller the number of structural holes it spans.

DATA AND METHODS To study the relationship between partner choices and network positions over time in the Canadian investment banking industry, we collected data comprising all securities offerings on all Canadian stock exchanges – Vancouver, Alberta, Winnipeg, Montreal ,and Toronto – from 1952 to 1990, inclusively. The Record of New Issues, published yearly since 1952 by Financial Data Group in Toronto, was the main data source and provided the following information: issuer’s name, date of issue, name of lead issuing investment bank and co-lead investment banks, security type (debt, bond, preferred share, common share, right and unit), value of issue proceeds, and issue price and volume. In 1952, 19 investment banks participated in 27 underwriting syndicates in Canada. The life histories of this subset of banks, representing 5% of the 403 banks in the sample; are left-censored. Over the observation period, the number of syndicates grew rapidly, and by 1990, 152 banks participated in 422 syndicates. Thus, although life-histories for a small number of banks are left-censored, our data covers the period during which raising equity in the capital markets superceded bank debt as the dominant mode of corporate ﬁnancing (Davis & Mizruchi, 1999). Our comprehensive longitudinal data permit us to avoid the common problem of establishing boundaries in network analysis (Doreian & Woodard, 1992), and to model network dynamics over a period of time sufﬁcient to yield meaningful variation in network structure and composition. Our research design avoids the common sample selection problem of over-representing currently successful ﬁrms or particular points in time that can undermine inferences about factors producing organizational behavior and success (Berk, 1983).

The Dynamics of Network Strategies and Positions

653

NETWORK DEFINITION AND MEASUREMENT Following earlier research on interﬁrm networks in the investment banking industry, we operationalized the interﬁrm network based on banks’ memberships in underwriting syndicates (e.g., Podolny, 1993). Underwriting syndicate ties represent formal economic relationships that bind banks together in economic transactions. Although formed by individuals (i.e., bank managers), these ties are embedded within strongly institutionalized patterns of interﬁrm cooperation that have emerged from, but also extend beyond, ties among speciﬁc individuals across ﬁrm boundaries. Interﬁrm networks, deﬁned by memberships in underwriting syndicates, were constructed for two-year moving periods (i.e., 1952–1953, 1953–1954, 1954–1955, etc.).3 We used two-year periods for three reasons. First, syndicate ties represent only the visible manifestation of relationships; banks participating in syndicates together in any given year are also likely to interact in other ways with each other in years proximate to the syndicate. Second, because syndicates can remain intact for nine or more prior to the date of the offering, syndicates that conclude in any given year may have been formed, but unobserved, in prior years. And third, the two-year period permits us to gauge more accurately the strength of network ties by incorporating information on repeated ties over two years. Constructing the network based on two-year moving periods should thus represent the network more reliably and accurately. These networks were used to compute measures characterizing each bank’s network position in the second year of each two-year period. Thus, the 1952–1953 network was used to measure banks’ network positions for 1953, the 1953–1954 network for positions in 1954, etc. Ego and overall network variables were coded from these matrices and organized in ﬁrm-byyear panels. In addition to information about a bank’s network position, each record included independent and control variables described below.

Dependent Variable and Model Speciﬁcation For the analysis, we pooled the yearly data and estimate a single model on the pooled cross-sections using time series regression models. Each bank is represented in the sample for the years in which it participated in underwriting deals. Pooling repeated observations on the same organizations is likely to violate the assumption of independence from observation to observation and result in the model’s residuals being autocorrelated.

654

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

First-order autocorrelation occurs when the disturbances in one time period are correlated with those in the previous time period, resulting in incorrect variance estimates. This renders ordinary least squares (OLS) estimates inefﬁcient. Therefore, to obtain unbiased and efﬁcient estimates for the pooled data we estimated random-effects generalized least squares (GLS) models, which correct for autocorrelation of disturbances (Greene, 1993). We measured the dependent variable using Burt’s (1992) effective size, which captures the degree to which the bank occupies a bridging position as it is a measure of the number of structural holes a bank spans in its ego network – ties among the bank (ego) and its direct partners. Speciﬁcally, effective size is the number of partners a bank has minus the average number of ties its partners have (not counting ties to the ego) within the ego network (Borgatti, Everett, & Freeman, 2002). High effective size means a bank’s partners have few ties to one another and the bank enjoys the beneﬁts of bridging many structural holes among them.4 In all models we include lagged values of the dependent variable to predict the current year’s dependent variable. This helps to account for the possibility that, even though well speciﬁed, our empirical models likely still suffer from speciﬁcation bias due to unobserved heterogeneity (Jacobson, 1990), and permits us greater conﬁdence when inferring causal relationships between independent and dependent variables. In particular, if banks’ market share or network positions are themselves a result of unobserved factors affecting performance or network position, controlling for the lagged dependent variable will help eliminate spurious effects resulting from such endogeneity. Inclusion of the lagged dependent variables also helps control for ‘‘inertia’’ in bank performance or network position. Controlling for inertia is particularly important for our models of effective size given empirical evidence of ﬁrms’ tendency to repeat past ties, stabilizing network structures and positions (Chung et al., 2000; Gulati, 1995; Gulati & Gargiulo, 1999).

Independent Variables To test H1 and H2, we used directional degree centrality measures. Because lead banks initiate relationships with co-lead banks, we coded the number of syndicate ties a bank initiated as a lead bank as outdegree centrality and the number of syndicate ties it accepted as a co-lead bank as indegree centrality. We used the number of underwriting syndicates in which each bank acted as

The Dynamics of Network Strategies and Positions

655

the lead – own lead count (outdegree) – to test H1, and co-lead – own co-lead count (indegree) – to test H2. To test H3, H4, and H5, we constructed two analogous variables, partners’ lead count, deﬁned as the number of syndicates a bank’s partners’ had led, and partners’ co-lead count, deﬁned as the number of syndicates a bank’s partners’ had co-led. We interacted these partner counts with the bank-speciﬁc counts to test H3 (own lead count partners’ lead count), H4 (own lead count partners’ co-lead count; own co-lead count partners’ lead count), and H5 (own co-lead count partners’ co-lead count). These count variables were each computed for the two-year moving periods used to construct the interﬁrm networks, with the count values for each two-year period being assigned to the second year of the period (i.e., the 1952–1953 count was used for 1953, etc.). The variables were lagged one year for model estimation to avoid simultaneity problems.

Control Variables Many other factors may inﬂuence the performance and propensity of investment banks to span and compete for structural holes. Accordingly, to account for important alternative explanations for our hypotheses, our analyses control for detailed baseline models that include a range of additional characteristics of banks, their partners, and industry-speciﬁc environmental factors. Unless indicated otherwise, control variables were updated annually and lagged one year in the analysis. Organizational factors. Our ﬁrm-level controls include the dollar value of the offerings in which a bank acted as lead and co-lead (value of own deals as lead and value of own deals as co-lead), the dollar value of its non-syndicate deals (value of deals alone), and its overall market share (own market share), logged to normalize the variable’s distribution. These variables control for the possibility that a bank’s size and recent performance affect the deals and partners to which it gains access, and thus its network position. We also controlled for two aspects of a bank’s network position that might affect the deals it participates in and number of structural hole opportunities it can pursue. Closeness centrality, measured as the average geodesic distance a bank is from all other banks in the network, captures a bank’s access to others banks and information ﬂowing across the network (Wasserman & Faust, 1994). High closeness means a bank has independent access (few intermediaries) to different points in the network. We constructed closeness measures for both lead and co-lead bank roles (own

656

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

lead closeness and own co-lead closeness). Betweenness centrality, measured as the proportion of geodesic paths between other banks that pass through a given bank reﬂects that bank’s power or control over information in the network (Wasserman & Faust, 1994). The bank with the highest betweenness is the bank on the shortest path between all pairs of other banks in the network. As banks’ managers gain experience occupying bridging positions, they may become better acquainted with their beneﬁts, and more adept at identifying and bridging them as they form syndicates or accept invitations to join them (Baum et al., 2003). Therefore, we also controlled for each bank’s prior experience with structural holes – own average effective size in past ﬁve years – measured as the ﬁve-year moving average effective size, logged to normalize its distribution. We use a ﬁve-year moving average to account for possible antiquation (due to environmental change) or decay (due to forgetting) of knowledge gained from experiences in the past (e.g., Stuart & Podolny, 1996).5 Lastly, we created a left-censored dummy variable – bank operating in 1952 – coded one for banks already operating in 1952, and zero otherwise, to control for the possibility that banks founded prior to 1952 exhibited systematically different performance and/or partnering behavior. Partners. Main effects for partners’ lead manager count (partners’ lead count) and partners’ co-lead manager count (partners’ co-lead count) were include as controls, because a bank’s partners’ network strategies can inﬂuence its network position, and because we need to include these main effects to obtain unbiased estimates for the interaction terms including these variables (Cohen & Cohen, 1983). As before, these variables were computed for the two-year moving windows used to construct the interﬁrm networks, with values for each two-year period being assigned to the second year of the period (i.e., 1952–1953 value assigned to 1953, etc.), and then lagged one year. Environmental factors. The ﬁrst environmental control variable was the density of the interﬁrm network; that is the actual number of banks linked by syndicate ties relative to the total number of possible links among the banks. Density serves both as a check on our contingency view of sources of network advantage, and to control for possibility that the number of available structural holes decreases as density of network ties increases. The next three environmental controls emphasized economic factors that might inﬂuence issuer and/or investor behavior, the underwriting process itself, or serve to reinforce or loosen the network structure (Madhavan et al., 1998). These included the prime interest rate (interest rate), the volume of

The Dynamics of Network Strategies and Positions

657

shares traded on the Toronto Stock Exchange (TSE share volume) and the Toronto Stock Exchange 300 index value (TSE300 Index), the largest and most inﬂuential Canadian exchange and index. TSE share volume was logged to normalize its distribution.6 We also included controls speciﬁc to underwriting activity that might affect bank performance and the number and type of partnering opportunities available to banks. These included the total values of all public offerings (Total offering value) and total number of public offerings (Total offering count). These variables were also logged to normalize their distributions. The models included the total number of syndicates (Total syndicate count) as well. These three variables were computed for the twoyear moving windows used to construct the interﬁrm networks, and, as before, the variable values for each two-year period assigned to the second year of the period, and then lagged one year for the analysis. Lastly, we created dummy variables to control for whether a particular deal took place in the 1950s, 1960s, 1970s, or 1980s (i.e., deal in 1950s) to account for possible additional period-speciﬁc effects on bank performance and network position. Table 1 gives descriptive statistics and correlations among the study variables. The correlations are generally small in magnitude (i.e., o.20, or 4% shared variance), although, as would be expected, correlations among the several of the environmental control variables are larger. Such levels of multicollinearity among explanatory variables can result in less precise parameter estimates (i.e., larger standard errors) for correlated variables but will not bias parameter estimates (Greene, 1993; Kennedy, 1992). So, although this does not pose a serious estimation problem, it can make it more difﬁcult to draw inferences about the effects of adding particular variables to the models. Therefore, when estimating our models, we followed a strategy of estimating hierarchically nested models to check that multicollinearity was not causing less precise parameter estimates, and suppressing the signiﬁcance of some variables (Kmenta, 1971). We detected no evidence of multicollinearity affecting our estimation.

RESULTS Table 2 reports random effect model estimates of changes in investment banks’ effective size (Models 1–3). Model 1 presents estimates for the baseline model. Each of the environmental control variables is signiﬁcant in the direction that would be expected. However, contrary to expectations the

DV. Market Share .01 Interest rate 7.76 Total offering value 10.02 Total offer count 2.43 Total syndicate count 119.43 Ln TSE share volume 20.66 TSE 300 index 1582.01 Deal in 1950s .15 Deal in 1960s .26 Deal in 1970s .21 Deal in 1980s .38 Bank operating in 1952 .24 Value of own deals as 2.6Eþ08 lead 1.9Eþ08 Value of own deals as co-lead Value of deals alone 8.8Eþ10 (non-syndicate) Own market share .01 Density .05 Own average effective 1.77 size in past 5 years Own lead closeness 2.00 Own co-lead closeness 1.61 Betweenness centrality 1.72 Own lead count 8.54 Own co-lead count 7.11 Partner’s lead count 13.67 Partner’s co-lead count 12.02 Effective size 4.32

N ¼ 1259

19 20 21 22 23 24 25 26

16 17 18

15

14

1 2 3 4 5 6 7 8 9 10 11 12 13

Mean

1.27 .74 3.11 26.25 14.88 23.28 18.11 4.32

.02 .04 1.38

1.7Eþ11

2.4Eþ09

.02 3.11 .83 .39 127.57 1.66 993.71 .36 .44 .41 .48 .42 3.3Eþ09

Std. Dev.

2

3

4

5

6

.23

.06 .87

.17 .80

.15 .86

.17 .65

.13

.16 .12 .42 .19 .15 .24 .14 .32

.24 .13 .10 .10 .08 .09 .05 .26 .04 .10 .06 .04 .02 .02 .04 .12 .13 .09 .13 .12 .16 .18 .12 .18 .16 .20 .31 .21 .27 .26 .22 .36 .25 .31 .30 .24 .33 .26 .31 .30

.84 .09 .18 .20 .16. .17 .10 .56 .58 .42 .61 .57 .36 .19 .24 .22 .21 .22

.15

.07

9

10

.09 .01 .01 .06 .09 .15 .17 .17

.04

.60 .20

.11

.00

12

.23 .05 .04 .15 .18 .21 .23 .30

.27 .08 .04

.22

.14

13

15

16

17

.19 .15 .43 .20 .16 .25 .13 .31

.60 .48 .10 .22 .26 .24 .25 .32

.10 .15 .08 .43 .08 .08 .15 .35 .14

.17

14

.05. .04 .01 .02 .07 .06 .01 .10 .27 .06 .10 .01 .11 .05 .08 .08 .06 .00 .07 .13 .19 .09 .13 .26 .09 .01 .09 .30 .20 .08 .15 .26

.12 .15 .40 .05 .67 .21 .00 .21 .37

.36 .19 .34 .31 .10 .05 .12 .01 .17 .05 .21 .10 .23 .12 .19 .01

.18 .05 .02 .67 .29 .56 .20 .23 .04

.76 .18 .30 .24

.18 .01 .02 .14 .18 .25 .29 .31

11

.25 .22 .31 .32 .47 .40 .19 .10 .00 .23 .04 .06 .04 .12

8

.15 .04 .06 .03

.34 .50 .29 .91 .24 .16

7

Descriptive Statistics and Correlations.

.08 .14 .63 .15 .41 .80 .13 .40 .83 .91 .14 .78 .87 .79 .78 .15 .67 .92 .84 .84 .86 .05 .55 .38 .26 .26 .37 .02 .46 .46 .19. .34 .47 .09 .04 .13 .43 .25 ..18 .13 .79 .81 .73 .71 .86 .40 .24 .25 .19 .20 .24 .19 .05 .18 .16 .18 .14

1

Table 1.

.33 .18 .45 .35 .34 .55 .44 .66

18

.64 .29 .31 .29 .38 .22 .42

19

.14 .13 .24 .08 .14 .13

20

.53 .34 .51 .32 .75

21

23

24

25

.13 .56 .32 .14 .63 .55 .59 .49 .79 .61

22

659

The Dynamics of Network Strategies and Positions

Table 2.

Random Effects Models of Investments Banks’ Effective Size. Model 1

Interest rate Ln Total offerings value Ln Total offer count Total syndicate count Ln TSE share volume TSE 300 index Deal in 1950s Deal in 1960s Deal in 1970s Deal in 1980s Bank operating in 1952 Value of own deals as lead Value of own deals as co-lead Value of deals alone (non-syndicate) Own market share Density Own average effective size in past 5 years Own lead closeness Own co-lead closeness Betweenness centrality Own lead count Own co-lead count

0.7485 (0.0829) 2.0171 (0.4251) 4.5812 (0.7024) 0.0237 (0.0022) 0.3309 (0.1353) 0.0007 (0.0003) 12.2136 (4.0850) 12.1715 (3.9750) 10.6580 (4.0534) 9.1756 (4.0706) 0.3148 (0.1829) 3.08E-11 (0.0000) 4.17E-11 (0.0000) 1.73E-11 (0.0000) 25.6380 (4.7459) 57.6276 (5.8288) 0.0580 (0.0806) 0.2157 (0.0940) 0.2678 (0.1529) 0.0021 (0.0370)

Model 2 0.7501 (0.0826) 1.9442 (0.4260) 4.5818 (0.7011) 0.0236 (0.0022) 0.3346 (0.1349) 0.0007 (0.0003) 11.6448 (4.0830) 11.6148 (3.9728) 10.0852 (4.0508) 8.5858 (4.0684) 0.3434 (0.1828) 3.04E-11 (0.0000) 4.27E-11 (0.0000) 1.72E-11 (0.0000) 25.9130 (4.7348) 56.5382 (5.8333) 0.0513 (0.0804) 0.2301 (0.0939) 0.2889 (0.1528) 0.0126 (0.0373) 0.0101 (0.0033) 0.0168 (0.0058)

Model 3 0.7582 (0.0820) 1.9762 (0.4232) 4.3577 (0.6973) 0.0229 (0.0022) 0.3450 (0.1339) 0.0079 (0.0003) 2.9980 (0.8421) 3.0540 (0.7287) 1.4874 (0.5115) 9.8046 (4.0482) 0.2938 (0.1824) 2.54E-11 (0.0000) 4.15E-11 (0.0000) 1.77E-11 (0.0000) 19.3441 (4.8846) 57.738 (5.8215) 0.06369 (0.0814) 0.1473 (0.0967) 0.3858 (0.1539) 0.0240 (0.0385) 0.0428 (0.0092) 0.0279 (0.0116)

660

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

Table 2. (Continued ) Model 1 Partners’ lead count Partners’ co-lead count Effective size

0.0139 (0.0047) 0.01512 (0.0049) 0.6778 (0.0414)

Own lead count Partners’ lead count Own lead count Partners’ co-lead count Own co-lead count Partners’ lead count Own co-lead count Partners’ co-lead count Intercept w2 D w2 (df)

(dropped) 6349.45

Model 2

Model 3

0.0093 (0.0050) 0.0206 (0.0060) 0.6500 (0.0425)

0.0174 (0.0063) 0.0206 (0.0069) 0.6077 (0.0438) 0.0003 (0.0001) 0.0005 (0.0002) 0.0003 (0.0002) 0.0003 (0.0002) (dropped) (dropped) 6370.44 6419.27 20.99 (2) 48.83 (4)

po.05, po.01, po.001; N ¼ 1259; Standard errors in parentheses

coefﬁcient for density in the prior year is positive, suggesting that ﬁrms increasingly spanned structural holes as network density increased. We found no evidence of a quadratic effect for density. Network density was thus likely too low (mean ¼ .05; S.D. ¼ .04) to have the anticipated limiting effect on the number of structural holes. The ﬁrm level control variables are generally signiﬁcant and in the expected direction other than own average effective size in past ﬁve years and betweenness centrality, which are not signiﬁcant. Closeness through lead ties are signiﬁcant and positive, but for co-lead ties signiﬁcant and negative, indicating that access to other lead banks in the network increased a bank’s likelihood of forming bridging ties, while access to other co-lead banks diminished it. Similarly, partners’ lead count and partners’ co-lead count are both signiﬁcant and positive, suggesting that having active partners helped improve a bank’s bridging position. The coefﬁcient for lagged effective size is positive and highly signiﬁcant, consistent with research on partner strategies indicating that network structures and positions tend to be reproduced from one period to the next as a result of ﬁrms’ tendencies to repeat past ties (e.g., Gulati & Gargiulo, 1999). In Model 2, which provides a signiﬁcant improvement over Model 1 (Dw2 ¼ 20.99, 2 df, po.001), we add counts of a bank’s lead and co-lead

The Dynamics of Network Strategies and Positions

661

roles to test H1 and H2. Own lead count is signiﬁcant and positive, providing preliminary support for H1, that in their roles as initiators of syndicate ties, lead banks exercise their ability to seek out independence and information and brokerage advantages, which result in bridging positions. The coefﬁcient for own co-lead count is also signiﬁcant and positive, providing partial support for H2, which predicted that co-lead banks would accept and reject invitations to join syndicates based, in part, on the independence, information and brokerages opportunities they provide. Thus, contrary to the general prediction from partner selection research that ﬁrms choose their partners based on evaluations of their reliability and capabilities, our results indicate that as investment banks form ties – as lead and co-lead – they tend to surround themselves with structural holes. This suggests that, in our empirical setting, the stability of network structures and positions implied by the positive lagged effective size coefﬁcient may reﬂect managers tending to reproduce their banks’ ties that create bridging opportunities. In Model 3, we add four interaction terms among a bank and its partners’ lead and co-lead counts to fully specify the model and test H3, H4, and H5. Model 3 provides a further signiﬁcant improvement over Model 2 (Dw2 ¼ 48.83, 4 df, po.001), with coefﬁcients for three of the four interactions signiﬁcant. Introducing the interactions did not affect substantive interpretations for the main effects comprising, although the magnitudes of some of their effects change, reﬂecting speciﬁcation error corrected by inclusion of the interactions, as well as conversion of the main effects from unconditional to conditional marginal effects in the presence of the interactions (Jaccard, Turrisi, & Wan, 1990, pp. 24–26). The coefﬁcient for own lead count partners’ lead count is signiﬁcant and negative, as predicted by H3, highlighting that bridging positions are crowded out when partners both frequently act in the lead role. The coefﬁcient for own lead count partners’ co-lead count is signiﬁcant and positive, as predicted by H4; however, the estimate for own co-lead count partners’ lead count is not signiﬁcant. Thus, in contrast to our original hypothesis, banks that frequently lead syndicates do not appear to ‘‘prey’’ on syndicate partners that frequently co-lead, reaping network beneﬁts while limiting those available to their co-leads. Instead, an asymmetric, or partial mutualism results; while a bank that leads syndicates frequently spans more structural holes when its partners are frequently coleads, the co-leads’ network positions are unaffected by the association. H5 is supported by the signiﬁcant negative coefﬁcient for own co-lead count partners’ co-lead count. As we predicted, a bank that commonly acts as a co-lead in underwriting deals formed by and comprised of banks that

662

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

also commonly co-lead is availed few opportunities to establish brokering relationships, and the bank’s network position is worsened. In this full model, we can also complete our test of H2, which predicted not only that co-lead banks would accept and reject syndicate invitations based on structural hole opportunities, but also that the own lead count coefﬁcient would be larger than for own co-lead count because lead banks have greater latitude than co-lead banks to choose partners that improve their bridging positions. Consistent with this prediction, in Model 3, the coefﬁcient for the main effect of own lead count (b ¼ 0.0428) is more than 50% larger than the coefﬁcient for own co-lead count (b ¼ 0.0279). And, we conﬁrmed the signiﬁcance of this difference using Wonnacott and Wonnacott’s (1970, p. 275) comparison-of-means test (test statistic ¼ 47.12, po.001). To illustrate the magnitudes of the effects associated with the hypotheses, we present the ﬁndings graphically in Figs. 1 and 2 using estimates from Model 3 in Table 2. The ﬁgures help illustrate how the interaction effects associated with a focal bank’s and its partners’ lead and co-lead activities inﬂuence the bank’s network position. In the ﬁgures, we consider a focal

1

Estimated Effect Size

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2/18

4/16

6/14

8/12

10/10

12/8

14/6

16/4

18/2

My Lead/ My Colead Line 1: Partners' Lead 0, Colead 0

Line 2: Partners' Lead 5; Colead 0

Line 3: Partners' Lead 10; Colead 0

Line 4: Partners' Lead 0; Colead 5

Line 5: Partners' Lead 0, Colead 10

Fig. 1.

Estimated Effect Size (H3-H4).

663

The Dynamics of Network Strategies and Positions 1

Estimated Effect Size

0.9 0.8 0.7 0.6 0.5 0.4 2/18

4/16

6/14

8/12 10/10 12/8 My Lead/My Colead

14/6

16/4

18/2

Line 4: Partners' Lead 0; Colead 5

Line 5: Partners' Lead 0, Colead 10

Line 4a: Partners' Lead 0, Colead 5 (H6)

Line 5a: Partners' Lead 0; Colead 10 (H6)

Fig. 2.

Estimated Effect Size (H5).

bank that participates in 10 underwriting syndicates, which may be comprised of various mixes of lead/co-lead roles – 9/1 to 1/9. Line 1 in Fig. 1 provides a baseline for comparison, showing the focal bank’s effective size for each combination of lead/co-lead roles before accounting for its partners’ activities. To illustrate the own lead count partners’ lead count (H3) effect, we construct lines 2 and 3, which set partners’ lead count to 5 and 10, respectively. To isolate this effect, partners’ co-lead count is set to zero. Clearly, a bank’s effective size falls when it and its partners act increasingly as lead. For example, effective size for a bank with 10 lead and 10 co-lead roles falls as its partners’ lead count increases from 0 (Line 1) to 5 (Line 2) to 10 (Line 3). And, the decline increases as the bank’s number of leads increases. Thus, consistent with H3, a bank that frequently leads worsens its bridging position (i.e., lowers its effective size) when its partners also frequently lead. The own lead count partners’ co-lead count (H4) effect is shown in lines 4 and 5, which set partners’ co-lead count to 5 and 10, respectively. Again, to isolate the effect, partners’ lead count is set to 0. Comparing lines 1, 4, and 5 at any given point – for example, where the focal bank has 10 leads and 10 co-leads – shows how the bank’s effective size increases as its partners’

664

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

co-lead count increases from 0 (line 1) to 5 (line 4) to 10 (line 5). And, as a result of the interaction, the increase is ampliﬁed by the bank’s own lead count. Thus, corroborating H4, a bank that frequently leads syndicates improves its bridging position (i.e., raises its effective size) when its partners frequently co-lead. Fig. 2 illustrates the co-lead interaction effect – own co-lead count partners’ co-lead count (H5). Lines 4 and 5, reproduced from Fig. 2, show the effect of partners’ co-lead count without the co-lead interaction effect. Lines 4a and 5a incorporate the interaction effect with partners’ co-lead count set to 5 and 10, respectively. Adding the interaction effect shows how the network position of a bank that co-leads frequently is hampered by joining syndicates formed by banks that also frequently act as co-lead. As the ﬁgure shows, consistent with H5, the penalty for partnering with other regular co-leads increases with the number of co-lead roles in the focal bank’s portfolio of deals.

DISCUSSION AND CONCLUSION The literature on interﬁrm networks is founded on the premise that network structures and positions inﬂuence competitive advantage. Although the mechanisms leading ﬁrms to occupy embedded positions and the associated beneﬁts are well-known, the path to bridging position advantages has been understudied. To investigate the types of ﬁrms and partnerships that create bridging positions we analyze all underwriting syndicates in Canada over nearly 40 years, a context in which bridging positions are especially advantageous. Our study is among the ﬁrst to examine mechanisms underlying change over time in ﬁrms’ network positions, and our focus on ﬁrms’ role in these dynamics complements recent work on environmental factors reinforcing and loosening network structures (Madhavan et al., 1998). Past research on ﬁrms’ partner strategies has emphasized their strategic use of partnering to create dyadic competitive advantages, and emphasized how uncertainty about the reliability and capabilities of potential partners makes selection of more familiar, past partners more likely, bringing stability to emergent network structures and positions. We too ﬁnd evidence of network stability over time; but it is a different kind of stability. In the context of richly speciﬁed empirical models accounting for a wide range of alternative explanations for spanning structural holes, our results show that when forming underwriting syndicates as lead banks, and to a lesser degree

The Dynamics of Network Strategies and Positions

665

accepting invitations to serve as co-leads, investment banks’ partner selections tend to improve their bridging positions by increasing the number of structural holes they spanned (H1 and H2). Our ﬁndings also contrast with the prediction that ﬁrms prefer the comfort of dense, cohesive ties with past partners. Instead, lead investment banks in our sample show a preference for ties with co-lead banks that create information and brokerage advantages – that is, ties that bridge structural holes. Some of our more intriguing ﬁndings relate to the interdependence between a bank’s and its partners’ roles (i.e., lead and co-lead) in the industry. Our results reveal a competition for brokering opportunities among banks that frequently act as syndicate leaders (H3). They also reveal partial mutualism between banks that frequently act as leads and banks that frequently act as co-leads (H4): While a bank that frequently leads syndicates is likely to ﬁnd itself spanning more structural holes when its partners frequently co-lead, the co-leads’ network positions are not affected by the association. Thus, contrary to our expectation, lead banks do not ‘‘prey’’ on the network opportunities of their co-leads. Lastly, joining syndicates initiated by banks that also frequently act as co-leads undermines network positions for banks that also frequently co-lead (H5). Overall, our ﬁndings suggest that in settings where information and brokerage are a source of competitive advantage and cooperative behavior is governed by non-network mechanisms (e.g., a long shadow of the future), building ties with partners pursuing different, perhaps complementary, roles and strategies – lead and co-lead strategies – can improve a ﬁrm’s position. Forming ties with partners pursuing similar roles and strategies worsen it. The information and brokerage advantages of ties that bridge structural holes are not sustainable through lead-to-lead ties (see Fig. 2, Line 5): Competition for brokering opportunities acts to reduce those opportunities. Thus, as with most sources of competitive advantage, the interplay among similar ﬁrms changes the network structure, making advantageous positions difﬁcult to sustain. Nor is network advantage achievable through co-lead-to-co-lead ties either (see Fig. 1, Line 3): Frequent co-leads, because of their limited inﬂuence over the network structure, occupy peripheral, less advantageous positions, which do not afford them timely access to information on new opportunities. While supporting a contingency view of Burt’s (1992) theory of structural holes, our ﬁndings also suggest that role coordination and specialization may be a necessary condition for achieving network advantage in industry settings where spanning structural holes is beneﬁcial. A bank that frequently acts as a lead will improve its network position when partnering with others willing to routinely act as co-lead, despite the limitations this place on the

666

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

partners’ abilities to ﬁnd network advantage. In the absence of banks willing to ‘‘specialize’’ in the co-lead role, syndicate formation may degenerate into a competitive search for autonomy, diminishing the number of structural hole opportunities available to all players. Such rivalry may also destabilize the network structure as banks compete to recapture advantages destroyed by their partners’ actions. Lead banks are thus better off individually – and collectively – if their co-leads cut ties with partners that are also lead banks, and so may beneﬁt from negotiating restrictions on the roles their partners play. This ﬁnding contrasts with the view that banks should more indiscriminately admit others to its deals to gain access to a broader range of partners’ underwriting syndicates (Ljungqvist & Wilhelm, 2003). Our primary goal in this chapter was to examine how bridging positions emerge and are allocated across ﬁrms and in particular what types of ﬁrms are likeliest to occupy these positions. While our empirical ﬁndings are consistent with the view that managers have sufﬁcient knowledge of the surrounding network structure and value of brokering relationships to pursue network advantage, they do not provide direct evidence of banks’ managers pursuing structural hole advantages. Instead, we treat structural hole spanning as the outcome of banks attempting to reduce their resource dependence on partners by ﬁnding alternative sources of information and brokerage opportunities (Emerson, 1962; Pfeffer & Salancik, 1978). Nevertheless, our empirical ﬁndings point to the need for future research designed to examine more directly investment bank managers’ mental models and cognitive abilities with respect to their interﬁrm networks (e.g., Fund et al., 2008; van Liere et al., 2008). A further limitation of our study, relates to our empirical setting. Investment banking has features that do not characterize all industries, and as our contingency perspective suggests, this likely affects – potentially greatly – the logic of bridging position dynamics. Generalizations should therefore be bounded to settings where structural holes and sparse network positions are a source of competitive advantage. Bridging positions are most advantageous when information access and control are essential components of ﬁrms’ value propositions to customers, that information is accessed through relationships and partnerships occur under a long shadow of the future (i.e., a stable set of ﬁrms). Further research is needed in settings comparable to ours, but where, in contrast to investment banking, managers’ incentives and/or opportunities to pursue bridging positions are weaker, for example, because information on interﬁrm relationships is not publicly available. In sum, while acknowledging the importance of partner-speciﬁc concerns about trustworthiness and reliability in partner selection, we sought to

The Dynamics of Network Strategies and Positions

667

broaden partner selection research. Partner selection research has predominately focused on ﬁrms’ partner strategies – reducing partner uncertainty by ﬁnding the best partner and using the local network to govern the tie. Our analysis includes the consideration of their network strategies – building interorganizational independence and brokerage opportunities – by linking ﬁrms’ partner choices to changes in their network positions over time. Our ﬁndings suggest that new insights into the dynamics of interﬁrm networks and competitive advantage of ﬁrms are possible within this broader view.

NOTES 1. Given the complexity of the investment banking partnerships, we generate hypotheses speciﬁc to the industry so that we are able to more precisely deﬁne and link the theoretical and operational variables. In the discussion section, we address the generalizability of the results by describing the relevant industry conditions. 2. Although Krackhardt (1990) found that employees in a small entrepreneurial ﬁrm had limited knowledge of their ﬁrm’s overall interpersonal network, knowing with whom one’s partners have formed ties is a far less demanding requirement. Moreover, opportunities and incentives for the employees to attend to the network were unclear. 3. A dyadic tie was coded for each underwriting relationships between a lead and co-lead bank; ties were not coded between co-lead banks participating in the same syndicate because they interact infrequently with one another. 4. We use effective size rather than efﬁciency (see Burt, 1992) because our hypotheses implicate the magnitude of a bank’s positional advantage – how many structural holes it spans – rather than how efﬁciently a bank converts relationships into structural holes. In a supplementary analysis, we re-estimated our models with efﬁciency as a control variable. The results do not differ from those reported below, and efﬁciency was not signiﬁcant. 5. For banks that had operated fewer than ﬁve years, we computed experience over the number of years the bank operated. For the 19 banks already active in 1952, we do not have information on network position for the prior ﬁve years. Consequently, the ﬁve-year moving average experience variable is underestimated for these banks until 1957. Re-estimating our models after dropping all observations for these banks between 1952 and 1956 did not substantively alter the ﬁndings from those we report below. 6. The Canadian Socioeconomic Information and Management Database was our source for these variables.

ACKNOWLEDGMENTS This research is supported by a grant from the Social Sciences and Humanities Research Council of Canada. For helpful comments and

668

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

conversations that improved this chapter, we are grateful to Terry Amburgey, Don Bergh, Kristina Dahlin, Henrich Greve, Walid Hejazi, Ravi Madhavan, Joel Podolny, and Huggy Rao as well as workshop participants at Carnegie Mellon University, Penn State University, University of Alberta, and the Strategy Research Forum, Boston MA. We are indebted to Stan Li and Andrew Shipilov for their help with data collection and coding.

REFERENCES Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Anand, B. N., & Piskorski, M. J. (2000). (Why) are networks important? Working paper. Harvard Business School, Harvard University. Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books. Baker, W. E., Faulkner, R. R., & Fisher, G. A. (1998). Hazards of the market: The continuity and dissolution of interorganizational market relationships. American Sociological Review, 63, 147–177. Baum, J. A. C., & Ingram, P. (2002). Interorganizational learning and network organizations: Toward a behavioral theory of the ‘interﬁrm’. In: M. Augier & J. G. March (Eds), The economics of choice, change, and organization: essays in memory of Richard M. Cyert (pp. 191–218). Cheltenham, UK: Edward Elgar. Baum, J. A. C., Rowley, T. J., Shipilov, A. V., & Chuang, Y.-T. (2005). Dancing with strangers: Aspiration performance and the search for underwriting syndicate partners. Administrative Science Quarterly, 50(4), 536–575. Baum, J. A. C., Shipilov, A. V., & Rowley, T. J. (2003). Where do small worlds come from? Industrial and Corporate Change, 12, 597–725. Special Issue in Honor of James G. March. Berk, R. A. (1983). An introduction to sample selection bias in sociological data. American Sociological Review, 48, 386–398. Borgatti, S., Everett, M. G., & Freeman, L. C. (2002). Ucinet for windows: Software for social network analysis. Harvard, MA: Analytic Technologies. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. (2008). Returns to secondhand brokerage in industry networks: Spillover effects on price cost margins in American manufacturing. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 315–360). Bingley, UK: JAI/Emerald Group. Burt, R. S., & Knez, M. (1995). Kinds of third-party effects on trust. Rationality and Society, 7, 255–292. Chung, S. A., Singh, H., & Lee, K. (2000). Complementarity, status similarity and social capital as drivers of alliance formation. Strategic Management Journal, 2, 1–22. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed). Hillsdale, NJ: Erlbaum.

The Dynamics of Network Strategies and Positions

669

Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Cyert, R. M., & March, J. G. (1963). A behavioral theory of the ﬁrm. Englewood Cliffs, NJ: Prentice-Hall. Davis, G. F., & Mizruchi, M. S. (1999). The money center cannot hold: Commercial Banks in the U.S. system of corporate governance. Administrative Science Quarterly, 44, 215–239. Doreian, P. (2002). Event sequences as generators of social network evolution. Social Networks, 24, 93–119. Doreian, P. (2008). Actor utilities, strategic action and network evolution. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 247–271). Bingley, UK: JAI/Emerald Group. Doreian, P., & Woodard, K. L. (1992). Fixed list versus snowball selection of social networks. Social Science Research, 21, 216–233. Eccles, R. G., & Crane, D. B. (1988). Doing deals: Investment banks at work. Boston, MA: Harvard Business School Press. Emerson, R. M. (1962). Power-dependence relations. American Sociological Review, 27, 31–40. Friedkin, N. (1983). Horizons of observability and limits of informal control in organizations. Social Forces, 62, 54–77. Fund, B., Pollock, T., Baker, T., & Wowak, A. (2008). Who’s the new kid? The process of developing network centrality and embeddedness in venture capital deal networks. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 563–593). Bingley, UK: JAI/Emerald Group. Greene, W. H. (1993). Econometric analysis (3rd ed). Upper Saddle River, NJ: Prentice Hall. Gulati, R. (1995). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1439–1494. Gulati, R., Nohria, N., & Zaheer, A. (2000). Strategic networks. Strategic Management Journal, 21, 203–215. Hagedoorn, J., & Frankort, H. T. W. (2008). The gloomy side of embeddedness: The effects of overembeddedness on inter-ﬁrm partnership formation. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 503–530). Bingley, UK: JAI/Emerald Group. Hennart, J. F. (1988). A transaction costs theory of equity joint ventures. Strategic Management Journal, 9, 361–374. Hite, J. M. (2008). The evolution of strategic dyadic network ties: Strategically navigating bounded agency within multi-dimensional and dynamic relationships. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 133–170). Bingley, UK: JAI/Emerald Group. Hite, J. M., & Hesterly, W. S. (2001). The evolution of ﬁrm networks: From emergence to early growth of the ﬁrm. Strategic Management Journal, 22, 275–286. Jaccard, J., Turrisi, R., & Wan, C. K. (1990). Interaction effects in multiple regression. Thousand Oaks, CA: Sage Publications. Jacobson, R. (1990). Unobservable effects and business performance. Marketing Science, 9, 74–85. Kennedy, P. (1992). A guide to econometric methods (3rd ed). Cambridge, MA: MIT Press. Kmenta, J. (1971). Elements of econometrics. New York: Macmillan.

670

TIMOTHY J. ROWLEY AND JOEL A.C. BAUM

Kogut, B. (1988). Joint ventures: Theoretical and empirical perspectives. Strategic Management Journal, 9, 319–332. Krackhardt, D. (1990). Assessing the political landscape: Structure, cognition, and power in organizations. Administrative Science Quarterly, 35, 342–369. Kreps, D. (1990). Game theory and economic modeling. Oxford UK: Oxford University Press. Leik, R. K. (1992). New directions for network exchange theory: Strategic manipulation of network linkages. Social Networks, 14, 309–323. Li, S. X., & Rowley, T. (2002). Inertia and evaluation mechanisms in interorganizational partner selection. Academy of Management Journal (Forthcoming). Ljungqvist, A. P., & Wilhelm, W. J., Jr.. (2003). IPO pricing in the do-com bubble. Journal of Finance, 58, 723–752. Madhavan, R., Koka, B. R., & Prescott, J. E. (1998). Networks in transition: How industry events (re)shape interﬁrm relationships. Strategic Management Journal, 19, 439–459. Meyer, J. W., & Rowan, B. (1977). Institutionalized organizations: formal structure as myth and ceremony. American Journal of Sociology, 83, 431–450. Neuman, E., Davis, G., & Mizruchi, M. (2008). Networks and industry consolidation in U.S. global banking, 1986–2004. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 213–248). Bingley, UK: JAI/Emerald Group. Nohria, N. (1992). Is a network perspective a useful way of studying organizations? In: N. N. & R. C. Eccles, (Eds.), Networks and organizations: Structure, form, and action: 1–22. Boston, MA: Harvard Business School Press. Oliver, C. (1991). Strategic responses to institutional processes. Academy of Management Review, 16, 145–179. Palmer, D. (1983). Broken ties: Interlocking directorates and intercorporate coordination. Administrative Science Quarterly, 28, 40–55. Pfeffer, J., & Salancik, G. R. (1978). The external control of organizations. New York: Harper & Row. Podolny, J. M. (1993). A status-based model of market competition. American Journal of Sociology, 98(4), 829–872. Pollock, T. G., Porac, J. F., & Wade, J. B. (2004). Constructing deal networks: Brokers as network ‘architects’ in the U.S. IPO market and other examples. Academy of Management Review, 29(1), 50–71. Powell, W. W., Koput, K., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in Biotechnology. Administrative Science Quarterly, 41, 116–145. Ritter, J. R. (1991). The long-run performance of initial public offerings. Journal of Finance, 46(1), 3–27. Rowley, T. J. (1997). Moving beyond dyadic ties: A network theory of stakeholder inﬂuences. Academy of Management Review, 22(4), 887–910. Rowley, T. J., & Baum, J. A. C. (2004). Sophistication of interﬁrm network strategies in the Canadian investment banking industry. Scandinavian Journal of Management, 20, 103– 124Special Issue on Interﬁrm Competition and Cooperation. Rowley, T. J., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21, 369–386.

The Dynamics of Network Strategies and Positions

671

Shipilov, A. V. (2006). Network strategies and performance of Canadian investment banks. Academy of Management Journal, 49, 590–604. Steier, L., & Greenwood, R. (2000). Entrepreneurship and the evolution of angel ﬁnancial networks. Organization Studies, 21, 163–192. Stuart, T. E., & Podolny, J. M. (1996). Local search and the evolution of technological capabilities. Strategic Management Journal (Summer), 21–38. Uzzi, B. (1997). Social structure and competition in interﬁrm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. van Liere, D., Koppius, O., & Vervest, P. (2008). Strengthening of bridging positions: Network horizon and network horizon heterogeneity. In: J. A. C. Baum & T. J. Rowley (Eds), Network Strategy: Advances in Strategic Management (Vol. 25, pp. 597–641). Bingley, UK: JAI/Emerald Group. Walker, G. (2008). The rise of ecommerce as an epidemic in the small world of venture capital. In: J. A. C. Baum & T. J. Rowley (Eds), Network strategy: Advances in strategic management (Vol. 25, pp. 3–29). Bingley, UK: JAI/Emerald Group. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of industry networks. Organization Science, 8(2), 109–125. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge, UK: Cambridge University Press. Williamson, O. (1991). Comparative economic organization: The analysis of discrete structural alternatives. Administrative Science Quarterly, 36, 269–296. Wonnacott, T. J., & Wonnacott, R. H. (1970). Econometrics. NY: Wiley.

Network Strategy (Advances in Strategic Management)