Group Cohesion, Trust and Solidarity (Advances in Group Processes) (Advances in Group Processes)

LIST OF CONTRIBUTORS Casey Borch

Department of Sociology University of South Carolina, USA

Vincent Buskens

Department of Sociology/ICS Utrecht University, The Netherlands

Steve E. Clayman

Department of Sociology University of California, USA

Douglas D. Heckathorn

Department of Sociology Comell University, USA

Jeffrey A. Houser

Department of Sociology Bowling Green State University, USA

Edward J. Lawler

School of Industrial and Labor Relations, and Department of Sociology, Comell University, USA

John M. Levine

Department of Psychology University of Pittsburgh, USA

Michael J. Lovaglia

Department of Sociology University of Iowa, USA

Lynn Smith-Lovin

Department of Sociology, University of Arizona, USA

Miller McPherson

Department of Sociology University of Arizona, USA

Richard L. Moreland

Department of Psychology University of Pittsburgh, USA vii

viii

LIST OF CONTRIBUTORS

Werner Raub

Department of Sociology/ICS Utrecht University, The Netherlands

Judith E. Rosenstein

Department of Sociology, Cornell University, USA

Shane R. Thye

Department of Sociology University of South Carolina, USA

David Willer

Department of Sociology University of South Carolina, USA

Robb Willer

Department of Sociology Cornell University, USA

Jeongkoo Yoon

Program of Organizational Behavior and Human Resources Management Ajou University, South Korea

PREFACE EDITORIAL POLICY Advances in Group Processes publishes theoretical analyses, reviews and theorybased empirical chapters on group phenomena. The series adopts a broad conception of "group processes." This includes work on groups ranging from the very small to the very large, and on classic and contemporary topics such as status, power, exchange, justice, influence, decision making, intergroup relations and social networks. Contributors have included scholars from diverse fields including sociology, psychology, political science, philosophy, mathematics and organizational behavior. The series provides an outlet for papers that may be longer, more theoretical and/or more integrative than those published by standard journals. All papers undergo a peer review process to ensure a consistent standard of quality. We place a premium on the development of testable theories and theory-driven research. Chapters in the following categories are especially apropos: •

•

•

Conventional and unconventional theoretical work, from broad metatheoretical and conceptual analyses to refinements of existing theories and hypotheses. One goal of the series is to advance the field of group processes by promoting theoretical work. Papers that review and integrate programs of research. The current structure of the field often leads to the piecemeal publication of different parts of a program of research. This series offers those engaged in programmatic research on a given topic an opportunity to integrate their published and unpublished work into a single paper. Review articles that transcend the author's own work are also of considerable interest. Papers that develop and apply social psychological theories and research to macrosociological processes. One premise underlying this series is that links between macro and microsociological processes warrant more systematic and testable theorizing. The series encourages development of the macrosociological implications embedded in social psychological work on groups.

In addition the editors are open to submissions that depart from these guidelines. ix

x

PREFACE CONTENTS

OF VOLUME

19

With Volume 19 we begin a new trend in the Advances in Group Processes series. Over the next several years, we plan to publish a set of special volumes that examine core issues or fundamental themes in the group processes tradition. Each special volume will be organized around a particular problem, substantive area, or topic of study, broadly defined to include a range of methodological and theoretical orientations. Volume 19 represents the first volume of that series, offering a compendium of new theory and data on the topic of Group Cohesion, Trust, and Solidarity. The volume opens with a paper by McPherson and Smith-Lovin, entitled "Cohesion and Membership Duration: Linking Groups, Relations and Individuals in an Ecology of Affiliation." After reviewing the classic literature and recent trends in the psychology and sociology of cohesion, McPherson and Smith-Lovin offer a new theory of group cohesion based on principles of homophily and group competition. This work is provocative in that it seamlessly combines psychological and sociological forces acting on groups, ranging from simply dyadic homophily to the relative "niche location" of the group. The benefit of this conception is that it draws together principles of cohesion based on individual characteristics, group properties, and social network configurations. After laying out the theory as a set of formal propositions, the authors report a new simulation that compares the relative strength of homophily vs. group competition in producing cohesion. The second paper also addresses issues of homophily and in-group bias, but in a slightly different manner. In "Group Solidarity as the Product of Collective Action: Creation of Solidarity in a Population of Injection Drug Users," Heckathorn and Rosenstein introduce a new theory of solidarity and accompanying measurement procedures. This work suggests that factors critical to group solidarity, such as monitoring, sanctioning, and dependence are the products of collective action. Heckathorn and Rosenstein use formal theories of collective action to show how the shape of the production function governs the return actors get from their in-group. The authors then illustrate the theoretical principles and measurement procedures using data from 488 injection drug users sampled from three Connecticut cities. The authors conclude by discussing some of the more practical implications of their work, such as how collective action processes can be adapted to prevent the spread of AIDS within this population. The next two papers offer creative theory integrations to resolve problems of solidarity and cohesion. In "Building a Model for Solidarity and Cohesion Using Three Theories," D. Wilier, Borch and R. Wilier integrate principles from

Preface

xi

Elementary Theory, Game Theory, and Status Characteristics Theory to provide a theory of solidarity and group cohesion. The strength of his work is that, by using these three theories together, the authors provide solutions to the first- and second-order free riding problems. Another important dimension of the paper is that group solidarity is modeled as the function of the economic, social, and status structure of the group. The authors provide a number of experimental and historical examples to support the main theoretical claims. Next, Houser and Lovaglia integrate ideas from the sociology of emotion and recent work in status theory in "Status, Emotion, and the Development of Solidarity in Stratified Task Groups." The authors describe an interesting body of contemporary research on the evolution of emotions, and trace the broader implications of this work for modern research on status organizing processes. Houser and Lovaglia show how "status compatible" emotions follow from status differentiated encounters, and provide one basis for solidarity and cohesion in groups. The authors then propose a series of experiments to test the main claim. Scholars interested in social status, power and emotions should find these two papers especially provocative. Chapters five and six review and analyze two programs of research that target perceptions of cohesion and trust. In "The Theory of Relational Cohesion: Review of a Research Program," Thye, Yoon and Lawler review and summarize a decade of theory growth and empirical research on relational cohesion theory. This theory represents a clear departure from traditional exchange-theoretic accounts of cohesion, by theorizing the importance of emotions. After reviewing several early studies that set the stage for the emergence of relational cohesion theory, the authors trace the theory's development through today and summarize the corpus of empirical evidence. Next, Buskens and Raub explore how two theoretical mechanisms, termed learning and control, produce trust in social relationships. After reviewing the theoretical and empirical literature on these two mechanisms, the authors present data from two new experiments in which subjects are presented with hypothetical transactions. That data indicate that learning and control are intertwined at both the dyadic and network level. This research represents an important first step that will likely spur more comprehensive theorizing of trust in social relations. The final two chapters offer creative insights into social processes that often accompany and enhance trust, cohesion and solidarity. In "Group Reactions to Loyalty and Disloyalty," Levine and Moreland analyze the causes and consequences of group loyalty and disloyalty, a topic that has traditionally been neglected in the group processes tradition of sociology. This analysis uncovers important links between loyalty, personal characteristics, group properties, and intergroup relations. Drawing on the ingroup/outgroup imagery of social

xii

PREFACE

identity theory, the authors synthesize factors related to loyalty and propose a number of hypotheses suitable for empirical study. This is perhaps the most comprehensive chapter on loyalty of its kind. Finally, Clayman utilizes a conversation analytic approach to the study of solidarity in "Sequence and Solidarity." His analysis indicates how subtle features of the sequential arrangement of speech can either maintain or disrupt solidarity. Of particular interest, Clayman offers a typology of actions, solidary responses, and discordant responses. An important implication of this work is that the organization of social interaction can become institutionalized and normative in character. The broader significance is to show how micro-patterns of social interaction create, maintain, and solidify larger institutional methods of interaction. Shane R. Thye Edward J. Lawler Volume Co-Editors

COHESION AND MEMBERSHIP DURATION: LINKING GROUPS, RELATIONS AND INDIVIDUALS IN AN ECOLOGY OF AFFILIATION Miller McPherson and Lynn Smith-Lovin

ABSTRACT The study of group cohesion has a rich but confused history. Cohesion was originally a group-level concept, referring to the degree to which a group tends to maintain a stable, committed membership over time. As a largely psychological literature developed, however, an increasing focus on interpersonal attraction translated into the individual-level study of liking and interdependence. Recent advances in both psychology (Hogg, 1992) and sociology (Lawler & Yoon, 1996) usefully reassert the central role of social structure in determining a group's cohesiveness. We argue, however, that current approaches have enriched our understanding of intraindividual processing at the expense of the sociological understanding of the coevolution of groups and their members' networks within a larger community structure. We review the literature on this ecology of affiliation to draw inferences about both group cohesiveness and members' attachment to the group. Then we extend a theoretical simulation of these ecological processes to show how system-level properties of communities can influence group cohesion.

Group Cohesion, Trust and Solidarity, Volume 19, pages 1-36. © 2002 Published by Elsevier Science Ltd. ISBN: 0-7623-0898-2

2

MILLER McPHERSON AND LYNN SMITH-LOVIN

INTRODUCTION In the 1950s and 1960s, the study of group cohesiveness was one of the central topics of social psychology. Lewin's (1948) field theoretic conception of social space and the current events of the day both led to a fascination with the unitary character of social groups and the profound impact that these groups can exert on individual members. After a large body of rather confusing empirical findings accumulated (see an excellent review in Lott & Lott, 1965), both psychology and sociology largely dropped the topic for decades. In the early 1990s, however, theorists in both disciplines revived the important question of how individuals are bound into groups (Hogg, 1992; Lawler & Markovsky, 1994). These researchers used theoretical advances that occurred in other areas (such as social identity theory and social network theory) to tackle the definitional issues that had limited the earlier cumulation of knowledge. In doing so, they focused some attention on the larger social structural environment in which the group operates, in addition to the intra- and inter-individual processes that occurred among group members. In this chapter, we continue this very useful shift in focus toward the larger environment in which the group and its members operate. Rather than focusing on the intraindividual processes like cognition or emotion that have been the basis of recent theoretical innovations, we emphasize the ecology of groups and network ties. We return theoretical attention to issues in the classic literature that recent developments have ignored: the issue of membership continuity and tumover. We use an ecological theory of affiliation developed by McPherson (1983) to link cohesion, membership duration and group half life to structure in the larger community of groups. First, we briefly review the classic work on group cohesion, focusing on its central questions and definitional problems. Then, we summarize the modem theoretical advances that have occurred in both psychology and sociology. We point out some areas that have been largely overlooked in these advances. We suggest that ecological theory can remedy these lacunae. After summarizing earlier research that is relevant to these theoretical issues, we develop a simulation model of ecological processes to examine how networks, competing groups and community structure affect cohesion.

THE CLASSIC L I T E R A T U R E Festinger and his colleagues at the University of Michigan's Research Center for Group Dynamics (Festinger et al., 1950, Festinger, 1950) developed the first formal theory of group cohesiveness to explain how small, face-to-face informal

Cohesion and Membership Duration

3

social groups exerted pressure on their members to uphold social norms. Festinger (1950, p. 274) defined cohesiveness as "the resultant of all the forces acting on members of a group to remain in the group." The attractiveness of the group to its members, the mutual attraction of those members to each other as individuals, and the mediation of individual goals by the group were important determinants of cohesiveness. Membership continuity and adherence to group standards were important outcomes. Although Festinger's classic definition is the benchmark for most later discussions of group cohesion (see Hogg, 1992; Markovsky & Chaffee, 1995), researchers at the time immediately recognized its problems. In effect, the Festinger theory did not define cohesiveness independently of its sources and consequences. As later researchers tried to cope with this inherent ambiguity, psychologists usually retreated from group-level phenomenon and operationalized cohesion as either a summation of group members' attraction to the group as an entity or (more commonly) as the mean level of interpersonal attraction among group members. Both operationalizations had problems: they reduced this group concept to a sum of individual or interpersonal properties, and they focused on liking to the exclusion of other forces that could bind collectivities together into a cohesive unit. Hundreds of studies were completed over the next two decades, but the definitional problems prevented much theoretical development. While researchers cataloged a variety of sources and consequences, cohesiveness was operationalized in many different ways that often had little correlation with one another (see Hogg, 1992, Chap. 3). In such an intellectual field, it was difficult to interpret contradictory findings or to abstract empirical generalizations into a more general theory of the process through which groups formed, stabilized and operated as entities within a larger social structure. Perhaps more importantly, the gradual shift from an emphasis on stability of group membership (the core of Festinger's original definition) to the affective states of individuals (either liking for the group or liking for other members) led researchers to an almost exclusive focus on forces that operated within the group. So, in the comprehensive review of empirical findings in Lott and Lott (1965, summarized in Hogg, 1992, Table 3.1), only one of 29 factors that were discussed refers to the group's external environment. (The one potentially external factor was "frustration/threat", which included both internal negative emotion created by frustration in reaching desired goals and external threat by other actors or groups). By the 1970s, the study of group cohesion had essentially devolved into the study of what aspects of group structure, group activities or group members led people to like one another.

4

MILLER McPHERSON AND LYNN SMITH-LOVIN

Interestingly, sociology had its own classic literature discussing similar topics under a different conceptual label, solidarity. Durkheim (1956/1893), of course, featured the concept in his discussion of mechanical solidarity and organic solidarity. The former was based on similarity of group members' characteristics and experiences, while the latter was based on interdependent differences created by a division of labor. As Markovsky and Lawler (1994) pointed out, Durkheim's formulation shared a basic definitional problem with the classic Festinger work: there was no conceptualization of solidarity independent of its sources in mechanical or organic social relations. Because of the fundamentally different natures of mechanical and organic solidarity, sociologists have tended to concentrate on either an emotional, similarity-based basis for group formation or a utilitarian, rational-choice basis for group formation. These literatures were not connected in a sustained, productive way because there was no clear definition of a common outcome (solidarity or cohesion) that would serve to link them as two alternative, perhaps sufficient, sources of the same endpoint. NEW

DEVELOPMENTS

IN TWO

DISCIPLINES

While an untheorized mass of empirical generalizations cumulated through the second half of the century, both psychology and sociology were undergoing dramatic intellectual changes. The cognitive revolution in psychology turned the attention of that discipline's scholars to the ways in which individuals thought about their social world. In sociology, network theory provided a formalized conceptualization of social structure, exchange theories developed a more sophisticated understanding of how interdependencies held people in stable social arrangements, and emotion was revived as a central concern. As often happens, these theoretical and methodological innovations revived interest in an old topic, cohesion, and infused it with new vigor.

A New Psychology of Cohesion: Social Identity and Self-Categorization Social identity theory (Tajfel & Turner, 1979) and the more recent selfcategorization theory (Turner, 1985) made a fundamental distinction between interpersonal processes and group processes. They argued that group-level phenomena like ethnocentrism, ingroup bias, stereotyping, conformity and ingroup cohesion could not be reduced to person-to-person relations. Social identity theory dealt with self-inclusive social categories (e.g. nationalities, religions, sports teams or work groups) that both constitute a part of the selfconcept and influence perceptions and behavior toward self, in-group members

Cohesion and Membership Duration

5

and out-group members. At its core, the theory is a self-enhancement perspective: groups and their members make intergroup comparisons that favor the ingroup, and thus the self. Self-categorization theory added more insights from the cognitive revolution in psychology. It emphasized the perceptions that maximized differences between ingroup and outgroup and minimized differences within categories. These processes serve to structure the content of stereotyping, prejudice and conformity. Self-categorization theory also developed the concepts of social attraction and personal attraction. The former is defined as attraction among members of a salient subgroup because of their prototypicality and attraction to the group itself. This group-based attraction is seen as quite distinct from true interpersonal attraction, based on specific (non-group determined) characteristics of an individual or one's relationship with him/her. For social identity theorists, then, social cohesion became the study of social attraction (as opposed to interpersonal attraction) (Hogg, 1992). They argued that the valence and magnitude of social attraction was a function of: (1) the social relationship between individuals, which could be intra- or intergroup; (2) the degree to which specific others are perceived to be prototypical of their respective categories; (3) whether or not the social boundaries between the groups are perceived as permeable and/or legitimate (Hogg, 1992, p. 105). Much research in the social identity tradition already provided support for these basic processes (see review in Hogg, 1992, Chap. 7). The key insight is to link the various self-categorization phenomena to the more general question of group cohesiveness. The application of social identity theory to cohesiveness changed the traditional conception in two important ways. First, Hogg and his colleagues explicitly shifted the research question from the operation of groups themselves to the intrapsychic cognitive processes through which people think about their groups. Indeed, they argued that any psychological theory by necessity had to deal with this level of analysis; to do otherwise would leave them with a purely sociological or physiological treatment. They did not see this cognitive thrust as reductionist, however, because of the central position of social structural features in their theory (Hogg, 1992, p. 133). Much of how individuals thought about groups, their own self-categorization as group members, and their social attraction to ingroup and outgroup members was determined by the relative positions, permeability, legitimacy and/or competitiveness of the group in historical and social perspective. Second, shifting the emphasis to cognitive processing of group-related information allowed the social identity researchers to radically expand the scope of their theory. Now, rather than focusing on the small, face-to-face groups that had concerned Festinger and his colleagues,

6

MILLER McPHERSON AND LYNN SMITH-LOVIN

the new theoretical view could encompass any collectivity to which an individual thought that he or she belonged. This phenomenon could include large categories such as nationalities as well as local work groups. Indeed, it could include categories whose members had little or no potential for interaction at all. A group became any two or more individuals who shared a common social self-categorization (Tumer, 1982, p. 15). The major advance in a self-categorization conception of social cohesion, in the social identity theorists' view, was to remove the reductionist emphasis of the earlier empirical literature on interpersonal liking. Hogg and his colleagues started with a heavily cognitive theory, and moved it into the domain of attraction. But they focused on a type of attraction that was based on common group membership (and the characteristics it implied) rather than on interpersonal, individualized affect. Interestingly, parallel developments in sociology also emphasized a new, emotional facet of a theory that previously had been quite affect-free.

A New Sociology of Cohesion: Network Ties, Emotion and Solidarity In the early 1990s, sociologists also noted some of the persistent problems with the concepts of cohesion and solidarity, and began to develop solutions. As one would expect from their disciplinary proclivities, they were less inclined to abandon the concept of group in favor of cognitive self-categorization. Instead, they made use of developments in network theory to define cohesion in structural terms. Markovsky and Lawler (1994) defined cohesion as the level of reachability within a collectivity. Reachability contains both elements of connectedness and structure: members of a collectivity are reachable to the extent that they are tied together by numerous connections without internal divisions. Markovsky and Lawler (1994) were inspired by analogies to physical forms (gas, liquid and solid) to define aggregates, assemblages, and solidarity groups as collectivities with low, medium and high reachability, respectively. They then went on to catalog some of the properties that these collectivities would be expected to exhibit (in terms of their strength, stability, boundary definitions, etc.), and to suggest how emotions evoked by interactions among individuals can cumulate to produce an emotional bond to the group as a whole (Markovsky & Lawler, 1994, pp. 127-131). Markovsky and Chaffe (1995, pp. 262-268) linked the sociological work to social identity theory by extending the network conception to include ties that linked an individual to an identity concept, creating "reachability" through joint identification with a social category. But the most extensive theoretical development in sociology

Cohesion and Membership Duration

7

occurred as Lawler and his colleagues expanded their view of the emotional dynamics that lead to group cohesiveness and solidarity. The theoretical work has gone through several iterations. A good starting point is Lawler's (1992) Choice-Process Theory. This theory linked choice processes with the experience of transitory emotion and the development of enduring affective attachments to nested collectivities (i.e. subgroups within a larger group, organization, or society). According to Lawler, individuals become emotionally attached to groups that strengthen their generalized sense of control, primarily because that sense of control leads to positive emotions. ~ Lawler (1992) found that such positive emotion strengthens attachments to proximal subgroups more than to larger, more encompassing collectivities. The second, more sustained theoretical effort was Lawler's and Yoon's (1993, 1996, 1998; Lawler, Thye & Yoon, 2000) Theory of Relational Cohesion. Here, relational cohesion is defined as a perception of an exchange relation (or set of such relations) as a cohesive unit, while commitment behavior is the willingness to take risks, make sacrifices or to maintain behaviorally the exchange relation (Lawler & Thye, 1999, p. 237). The exogenous elements in the theory are the structure of exchange relations - the total power and relative power of exchange partners - within a power-dependence framework. Most of the empirical work focused on the endogenous processes through which exchange structure is transformed into cohesive, committed social units. There are two endogenous routes to cohesion within the theory, one emotional and the other cognitive. Exchange structures with high total power and equal power create stable patterns of successful exchange, which in turn elicit positive emotions and reduce uncertainty. Positive emotions (pleasure/satisfaction) and the reduction of uncertainty lead to attribution processes that imbue the exchange relation with salience and lead the exchange partner to view the exchange relation or group as a cohesive unit. This perceived cohesion then leads to behavioral commitment to the exchange relationship, even when the structural forces of exchange would not support such behavioral stability. The next step in the research program combined the Theory of Relational Cohesion and Network Exchange Theory (Walker et al., 2000) to create a Theory of Network Cohesion (Thye & Lawler, 1999). This new effort attempted to explain when networks of self-interested actors will evolve into collectively oriented groups. Effectively, the Theory of Relational Cohesion was extended to larger, more complex networks. Thye and Lawler (1999) predicted that a high proportion of equal power relations and high density promote cohesion and solidarity. Equal power promotes satisfying, pleasant exchanges, and high density allows such exchanges with more alters in the network. Together, these conditions foster salient perceptions of a common experience across members

8

MILLER McPHERSON AND LYNN SMITH-LOVIN

in a network, and lead each member to infer that others are feeling the same emotions that s/he is. Perceived cohesion is high when subjects view the group (the exchange network) as "close", "coming together", "cohesive", "cooperative" and "integrated." Behavioral commitment is high when subjects give gifts or distribute points to other group members in a non-contingent situation (illustrating that they wish to help other group members even when they are not sure of receiving anything in return). In the most recent iteration, Lawler (2001) developed the Affect Theory of Social Exchange to focus further on how the emotions produced by social exchange generate stronger or weaker ties to relations, groups or networks. It focused in more detail on the attributional processes through which actors attempt to understand the source of feelings that occur during exchange. The theory suggests that social units like relations or groups are perceived as the source of feelings to a greater degree when the jointness of the task is high: and when they perceive a shared responsibility for success or failure at the task. For the first time, the distinction between positively and negatively connected networks was included explicitly in the theoretical treatment. Positively connected networks will promote the diffusion of positive feelings from one relation to another, while such feelings remain more relation-specific in negatively connected networks. This process should make unitary structures (which Markovsky & Lawler, 1994, require for a solidary group) more difficult to achieve in negatively connected networks, since pockets of cohesive sub-units tend to develop in negatively connected networks. An important strain of work by Noah Friedkin (1998) has resulted in a theory of social influence which captures the dynamics of consensus formation in groups. Following the theoretical formulations of French (1956), Friedkin (2001) models the impact of network on opinion and norm formation, based on a structural analysis of roles and social differentiation. His structural social psychology begins the task of understanding how social structure is played out in social process. In a creative blend of qualitative ethnography and mathematical modeling, Friedkin shows how the production of social consensus - essential to group cohesion - depends on the structure of connections among group members. Friedkin's work ties together the structural features of cohesion (such as those emphasized by Markovsky) with the cognitive and affective agreement (emphasized by Lawler and the psychologists). His model produces a dynamic view of how network patterns, even those that have significant internal cleavages, can produce stable consensus. This decade of intense theoretical and empirical work by network researchers did much to bring the study of cohesion and solidarity back to the forefront of sociological social psychology. While others had attempted to relate

Cohesion and Membership Duration

9

interdependence to conformity to group norms (Hechter, 1987; Coleman, 1986, 1990), Lawler was the first to successfully use theoretical advances in exchange theory and in the sociology of emotions to link the two dominant mechanisms of solidarity - interdependence and emotional response. He was also largely successful at linking the structural network conceptions of cohesion to the cognitive self-categorization conceptions that were becoming dominant in psychology. Given such dramatic progress in two disciplines, it is reasonable at this point to step back and assess what we have achieved, and what has been left behind in our recent progress. Our Assessment

Clearly, social psychologists in both disciplines have made major advances in understanding how people's affective and cognitive responses to social structural environments create: (1) a concept of a unitary group from a set of relations; and (2) an emotional attachment to that concept. Our goal here is not to review comprehensively these developments, or to compare them (although this would be a useful task for other scholars to undertake at some point). We do note a few differences between the disciplinary approaches that will prove salient to our own contributions, however. Sociologists have retained the classic focus on directly connected interactants, while psychologists have widened their scope to consider social categories. 3 Sociologists have focused more on how emotional responses to interaction lead to the development of group commitments, while psychologists have focused primarily on how cognitive self-categorizations drive emotional responses, perceptions and behaviors. Sociologists have concentrated more on the utilitarian, organic, interdependency bases of group affiliation, while psychologists have focused more on the mechanical solidarity that is created by perceived similarity. 4 More important for our current contribution is some similarities in the two theoretical advances. In both cases, the theorists began by bemoaning the definitional problems with cohesion (Hogg, 1992; Markovsky & Lawler, 1994). They propose somewhat different solutions to this problem. The social identity theorists develop a theoretical definition of social (as opposed to individual) attraction, based on the depersonalized aspects of the self and alter as group members. Cohesion then becomes the degree or density of social attraction in the group/category, usually measured at the level of the individual subject. In the sociological tradition, Markovsky and Lawler proposed a structural network definition of cohesivenss as the reachability among group members, with solidarity defined as groups with high reachability and unity of structure. However, as the theoretical ideas developed, structural features became an

10

MILLER McPHERSON AND LYNN SMITH-LOVIN

independent variable (see discussion in Thye & Lawler, 1999) with perceived cohesiveness measured as an individual's sense of the network's groupness (Lawler & Yoon, 1996; Lawler, Thye & Yoon, 1999). Two things are striking about this definitional shift. First, both traditions have moved back to measuring cohesiveness not as a structural group property but as a combination of individual level reactions. For the psychologists, it is another type of interpersonal attraction; for the sociologists, it is a cognition about the degree to which a set of relations constitutes a committed group. This concentration on measurement of individuals is somewhat at odds with the initial critique of the classic cohesion literature. Perhaps more importantly, the new definitions ignore two features that seemed key to the original conception of cohesion. First, most discussions (classic and current) imply that cohesion is a property of a group (representing some sense of how well the group "hangs together" and survives as an entity separate from its social environment). Only in measurement do we retreat to the individual level. Second, the current definitions have lost the common element that underlies both Festinger's (1950) seminal definition and almost all of the variants that came after it - the degree to which members stay in the group, giving it stable membership and a recognizable form (Shaw, 1976, p. 197). Lawler's tradition comes closest to dealing with the issue, when talking about commitment to relationships (see especially Lawler, 1992). But both traditions end up defining the group's cohesivenes in terms of a static set of perceptions, rather than as a dynamic characteristic of its social structure. The second feature that the current theoretical advances emphasized at the outset was the dependence of group cohesion on a larger social structural environment. Interestingly, the social identity theorists were most explicit about this flaw in the classic literature) A chief shortcoming of the interpersonal liking conception of cohesion was that it concentrated attention on the intemal environment of the group, and ignored the larger community of cross-cutting group affiliations (Hogg, 1992, p. 90). For the social identity theorists, the content of the group prototype and degree of intergroup competition are both profoundly affected by the presence of other groups, their characteristics, and the degree to which group membership is open to social change (Hogg, 1992, pp. 92, 131-133). The sociologists were a little less explicit, but Markovsky and Lawler (1994, pp. 124-125) included as defining properties of their aggregate, assemblage and solidary group distinctions the degree of boundary definition and penetrability of the collectivity. Still more implicitly, the entire Lawler theoretical tradition, through all four of its variants, has as its exogenous forces the network exchange

Cohesion and Membership Duration

11

structure of the actors involved in the collectivity, including their level of interdependence and relative power. These factors are, of course, related to the actors' ties to other sources of valued resources outside the group. A network collectivity has higher total power (more interdependence among its actors) if those actors have few or no ties to exchange partners outside the collectivity. And, actors within the network are more likely to have equal power if they have relatively equal levels of connections to outside providers; an actor who has many outside alternatives will almost necessarily be higher power than his or her less connected alters, ceteris paribus. Therefore, the exchange theories of cohesion build in (albeit implicitly) a considerable attention to the larger environment within which the exchange partners operate. While we regard the renewed attention to the larger social structural environment as one of the major advances of the current theoretical efforts, we argue that the central attention to intraindividual processing (cognition in the case of the psychologists, emotion for the sociologists) has led to a lack of concrete progress on this front. Furthermore, we contend that greater attention to the interaction of groups with their environments is what is necessary to bring together the individual-level measurements of cohesion and the more structural group-level conceptions. We appreciate the greater clarity that the above research programs have brought to the ways in which individuals think (and feel) about their group affiliations. We hope to expand this understanding by returning to some of the core questions about cohesion that have been somewhat neglected in recent years. In particular, we hope to do two things. First, we would like to focus on the stability of group affiliation as a core element of cohesion, and to link the individual-level membership duration with the overall membership turnover in the group. Second, we would like to relate the structural definition of cohesion to these dynamic properties, and to explore how these two elements depend jointly on features of the larger social system. This linking of individual, network and group can best be accomplished within the context of a dynamic ecological theory (McPherson, 1983; McPherson & Ranger-Moore, 1991). Luckily, this is another domain in which substantial theoretical strides have been made over the past couple of decades. In developing these new theoretical threads, we do not wish to take away from the accomplishments of the two theoretical research programs reviewed above. Rather we hope to supplement them with a more dynamic, macro-level understanding of how groups are influenced by their wider social structural environment. We hope to speak to questions that have been largely ignored in the past developments, rather than to contest those theorists' answers to the questions that they have raised.

12

MILLER McPHERSON AND LYNN SMITH-LOVIN AN ECOLOGICAL

VIEW

OF GROUPS,

NETWORKS

AND M E M B E R S We begin by describing a general theory that links groups, the networks that coevolve with them, and the members that constitute them. We then address our two key questions: (1) what determines the stability of groups and memberships; and (2) how is the structural cohesion of groups related to other properties of the social system? Before we begin, however, we should specify the scope of our theoretical statement and some of its key definitions. Our ideas apply specifically to face-to-face groups of members that have the capacity to interact with one another. The theory we describe applies to any entity that spreads through networks and competes to some degree with other entities for individuals' time and energy. 6 As such, it may apply to categorical collectivities (like nationality, gender, race, etc.) that do not entail direct interaction. But since that competition would be for salience within a self-structure, rather than a competition for actual time and energy expended through group activity, we leave open that question for future development. Here, we restrict our theoretical statement to interacting face-to-face membership groups. The theory that we develop here operates across a wide variety of groups and kinds of network ties. We define groups as repeated face-to-face assemblies that have a stable character over time. 7 The ties might include exchange relations (both positively and negatively connected), interpersonal attractions (both social and individual), and a wide variety of other connections. In general, we think of a network tie in terms of any positively valenced relationship through which information (including valued goods) can flow. Again, it may be that the theory would be applicable even when ties were of a negative sort (dislike, punishment, etc.), but we leave that potential extension for a later development. Our core question is: how does the structure of networks both within and across groups determine their stability? Since the definition of cohesion has strayed so far from the original Festinger (1950) concern with membership stability, we define cohesion in the structural terms advocated by Markovsky and Lawler (1994), as the teachability of group members. We therefore discuss membership duration and reachability as separate concepts.

The Ecology of Group Affiliation The core idea of McPherson's (1983) ecology of affiliation is that social groups exist and compete in a multidimensional space of socio-demographic characteristics that he calls Blau Space (McPherson & Ranger-Moore, 1991). s Each individual can he thought of as a point in the multi-dimensional property

Cohesion and Membership Duration

13

space, located by his or her socially salient characteristics like age, education, wealth, etc. 9 Blau space is important to groups and their members because of its connection to homophily. Homophily is the positive relationship between similarity (on almost any dimension) and the probability that two people will have a network connection between them. Homophily is one of the best established empirical facts in the sociological literature (see review in McPherson et al., 2001). The fact that homophily operates powerfully in Blau space allows us to translate network distance (which is often hard to measure in a large social system) into distance in the multi-dimensional space of salient social characteristics. In general, the further two people are apart in Blau space, the less likely they are to have a direct connection, the greater the network distance between them, and the greater the probability that one will not be reachable from the other. Since the probability of a connection between any two individuals is a declining function of distance in Blau space, McPherson's ecological theory can use Blau space to represent the structure of the very sparse, complex networks that actually exist in large systems, l° Essentially, distance in Blau space becomes a macro-level proxy for distance in the social network. Because of this fact, activities that require coordination of effort (like social groups) tend to he localized in Blau space (McPherson et al., 1992, Fig. 2), because niches form through contacts between similar others. Any social entity that recruits or spreads through network contacts will form a niche in Blau space. Niches are not widely spread across the social space, but will tend to exist in one more or less specialized region.~l Of course, the width of the niche varies for different entities, and is in fact one of the variables that the ecological theory explains (cf. McPherson & Rotolo, 1996). The ecological theory argues that groups change or stabilize over time according to a simple process of selection and retention. Selection of members into the group occurs through the recruitment of similar people into the group through homophilous network ties between members and non-members. Retention occurs when the group replicates the members' characteristics over time by either keeping current members or replacing lost members with equivalent new recruits. In the absence of other groups, we might expect that a group would slowly expand the range of its niche in Blau space, as recruitment of new members by group members at the edge of the niche gradually expanded its boundaries. But groups do not exist in isolation; instead, they occupy a position within a community of other social entities, most notably other groups. McPherson (1983) introduced the idea of ecological competition: when two groups overlap in Blau space, they are competing for the same kinds of individuals as members. In other words, when the niches of two groups intersect,

14

MILLER McPHERSON AND LYNN SMITH-LOVIN

the overlapping area is a region in which the two groups are competing for the same resource - a particular kind of person, defined by that position on the socio-demographic dimensions. The case is clearest when two groups have activities at the same time; a member literally cannot participate simultaneously in both groups. But, given limited time, energy and other resources, this competition exists even when two groups are not directly in scheduling conflict, or even aware of each others' existence. The social environment in which a group operates - specifically, the other groups that surround it in Blau space - will shape the group's growth or decline, changes in composition, and even its survival (McPherson, 1983, 1990; McPherson & Ranger-Moore, 1991; McPherson et al., 1992; McPherson & Rotolo, 1996). In the original statement of the theory, the niches of voluntary associations were defined as local areas in a space defined by years of age, occupational prestige, years of education, and gender. In that study, professional groups occupied a substantially higher range of education than unions, sports related groups barely overlapped with veterans groups in the age dimension, unions tended to be heavily male while church-related groups were mostly female, and so forth. Each of the groups occupied a distinctive region in the multidimensional property space, showing varying degrees of overlap with its neighbors. Competition in the multidimensional space consists of the extent to which the niches of groups overlap in the space. For instance, taking all four dimensions into account, church related groups and social groups have a very high degree of overlap, while unions and civic groups are almost unrelated. We emphasize that the ecological theory should apply to any social entity that spreads or recruits through homophilous networks, and which requires some time and/or resources for membership maintenance - a precondition for competitive forces to operate. Clearly, the face-to-face groups whose cohesion we wish to study qualify on both scores, since: (1) virtually all observable groups not formed through explicit randomization occur through the operation of social networks; and (2) all groups require some minimal level of energy expenditure on the part of members to persist. What makes the theory particularly useful is that it has precise, powerful implications for the behavior of both individual members and the groups themselves in the context of a social structural environment. We now turn to these implications, and relate them to the theoretical questions about cohesion that we raised above.

Retaining Members: Membership Duration and Network Ties We turn first to that relatively neglected core of the original definition of cohesion: the forces that operate on people to remain in the group (Festinger,

Cohesion and Membership Duration

15

1950; Shaw, 1976, p. 197). As we have noted, most of the classic answers to this question have concentrated on forces internal to the group - mutual liking, interdependence, success at achieving goals, etc. Here, we wish to reinforce the recent theoretical advances in attending to what is going on outside of the group, in its social environment. The ecological theory outlined briefly above makes quite explicit statements about when member will be at high risk of leaving a group and when groups will therefore be at risk of high turnover or decline. We describe the predictions about individual members first, then move to the larger question of the group's stability. The ecological model deals with two key elements: niches and the networks that cause them to form. Both elements have strong implications for the probability that a member will leave a group. Since the theory posits that people join and leave groups because of their network ties, the predictions here are fairly straightforward (McPherson et al., 1992). Groups create ties internally by organizing repeated contacts between members (Feld, 1981). As much of the traditional cohesion literature noted, members who are most heavily connected inside a group are the most integrated into that group. The important element added by the ecological view is that ties with people outside of a group connect one to events and structures that compete with the group for attention. Therefore, we have two basic propositions about network ties and group membership duration: Proposition 1: Network ties to other members of a group increase the duration of membership. Proposition 2: Network ties to non-members decrease the duration of membership. We briefly note the connection of these two propositions to earlier literatures. First, Proposition 1 clearly echoes the traditional literature on interpersonal attraction within group (Lott & Lott, 1965). We suggest that some of the previously uninterpretable inconsistency in those results might have been induced by Proposition 2. High numbers of network ties in general might pull one into many groups, because ties lead to joining (McPherson et al., 1992); but it is only the ties within the group that tend to hold one in (cf. Popielarz & McPherson, 1995). Ties outside the group will tend to pull one toward other activities, and away from the group. Failing to consider the competing ties with people outside the group may have led the non-experimental literature on cohesion to yield confusing results (Hogg, 1992, Chap. 4). The studies which used laboratory groups and which manipulated cohesion through interpersonal

16

MILLER McPHERSON AND LYNN SMITH-LOVIN

attraction often had more consistent results, probably because the groups existed in isolation with no competing ties (Lott & Lott, 1965). Proposition 1 is also related to the sociological literature on organizational commitment (Kanter, 1968; Knoke, 1981; Hall, 1988) which generally finds that members of groups with more friendship ties to other group members feel more likely to remain in their group. People who study cults, communes and other groups that approach the status of "total institutions" have been more sensitive to the operation of Proposition 2. They noted that the most important feature holding members into groups that have very high demands is the absence of continuing ties to family or friends outside of the extreme environment (Snow et al., 1980). McPherson et al. (1992) found that ties inside the group and outside the group had highly significant effects on membership duration in voluntary associations. Each network alter who was a co-member of the group increased the duration of membership by 60%. Non-member ties had a less dramatic impact, reducing the duration of membership by 14%. 12 Propositions 1 and 2 are also implicitly related to the exchange/interdependency approaches that Lawler and his colleagues have developed to cohesion. In the Theory of Relational Cohesion, high total power and equal power are preconditions for the positive, low uncertainty exchanges that lead to cohesion and commitment. In communities of groups, where some exchange ties will be within group and others will be with non-group members, total power within any given group will clearly be higher when few exchanges ties with non-members are available and most exchange ties are to group members. Power equality brings up a slightly more complex issue. When all group members have roughly the same level of exchange connection to external non-members, equal power will result; that level of extemal connection could be either very high or very low. We suggest, however, that such equality is implicitly more likely to result in high cohesiveness when outside connections are low, because of the inherent connection between total power and power equality. If outside connections are high but relatively equal, the group members may have roughly equal power, but the total power will be so low that the network ties may not be activated with great frequency at all. Therefore, at low levels of total power, equality is more or less undefined (or at least not very relevant). Therefore, the absence of outside connections will create both high total power within the group and the possibility of equal power relations within the group. (Of course, power inequality created by the configuration of exchanges within the group could also influence cohesion, but that type of intemal structure is a different issue, about which the ecological theory has little to say and the exchange theory of relational cohesion has a well developed perspective).

Cohesion and Membership Duration

17

Similarly, the Theory of Network Cohesion (Thye & Lawler, 1999) relates to both Propositions 1 and 2, through its propositions about the structures of exchange and how they will lead to cohesive, collectively oriented groups. Thye and Lawler (1999) predicted that a high density of exchange ties would promote group cohesion - a direct analogue to Proposition 1, since these ties are internal to the potential group. The equal power nature of the ties is related to Proposition 2 through the logic outlined above in our discussion of the Theory of Relational Cohesion, since ties outside the group will create less internal dependence and fewer active exchange ties within the group (i.e. lower density).

Retaining Members: Membership Duration and Tie Strength Obviously, not all network ties are equal. While McPherson's ecology of affiliation should apply to any tie that is structured by homophily and through which information (broadly defined) can pass, some thought has been given to the differential effects that strong and weak ties might have on members' attachment to groups (McPherson et al., 1992, pp. 158-159). Granovetter (1973, p. 1361) argued that the strength of an interpersonal tie could vary with the amount of time, the emotional intensity, the intimacy, and the reciprocal services that characterized the relationship. McPherson et al. (1992) suggested that the contact frequency of ego and alter, the density of connections among alters (which indicates the probable intensity, intimacy and importance of the tie), and the distance of the relationship in Blau space were important structural features that should influence how ties pull members in and out of groups. For the stabilizing effect of internal ties, McPherson and his colleagues argued that the stronger the tie, the more likely it was to engender joint activities and shared information and therefore to hold people into a common enterprise like group membership. Given the different ways that ties can be viewed as strong this leads to a series of propositions:

Proposition 3a. High contact frequency ties to other members will increase membership duration more than low contact frequency ties. Proposition 3b: Ties to other members who are themselves tied to one another will increase membership duration more than ties to members who are not tied to one's other alters. Proposition 3c: Ties to other members will increase membership duration to a greater extent when they are close in Blau space than when they are relatively distant in Blau space.

18

MILLER McPHERSON AND LYNN SMITH-LOVIN

Clearly, Proposition 3a is closely related to Lawler's work on exchange networks, where he argued that frequent exchanges lead to positive emotion and reduced uncertainty, which in turn lead to cohesion and commitment. Proposition 3b is implicitly related to Thye's and Lawler's (1999) ideas about network cohesion: when density is higher in a network, more people will have ties to alters who in fact share ties among themselves. As Lawler (2002) pointed out, this effect will operate more strongly in positively connected networks than negatively connected networks, since the preconditions for "jointness" will be stronger without the competitive nature of negatively connected exchange ties. Positive feelings will be more likely to diffuse from relation to relation in positively connected networks, while they will remain more relation-specific in negatively connected networks. Proposition 3c is unique to McPherson's ecological approach, since it concentrates more on the way in which Blau space structures networks and the flow of information in society. It does parallel, however, some of the classic literature in psychology. These studies consistently showed that similarity of group members lead to more cohesion, measured as interpersonal attraction among members (Lott & Lott, 1965). Ties to non-members are quite a different matter. Since weak ties are presumed to carry more new information from people more distant in both in terms of network structure and in terms of Blau space, they should draw members out of a group at a higher rate. McPherson et al. (1992) argued that this would be true for two reasons: (1) weak ties with non-group members will draw a member into more new groups, which compete with the old membership for time and energy; and (2) weak ties with non-members will be bridging ties that expose the member to new information and lower his or her commonality with the existing group, lowering the jointness of activities and information that tend to hold groups together. Therefore, our propositions about ties external to the group are the dual of our propositions about internal ties.

Proposition 4a: High contact frequency ties to non-group members will decrease membership duration less than low contact frequency ties to nongroup members.

Proposition 4b: Ties to non-group members who are themselves tied to one's other alters will decrease membership duration less than ties to members who are not tied to one's other alters.

Proposition 4c: Ties to non-group members will decrease membership duration to a greater extent when they are distant in Blau space than when they are relatively close in Blau space.

Cohesion and Membership Duration

19

These propositions, which come from the image of network ties as conduits of shared information, activity and emotional response in McPherson's ecological theory, are counter intuitive. We generally think of strong ties as implying a strong influence on social processes. But, just as Granovetter (1973) and others (e.g. Lin, 2001) have found that weak ties can be extremely useful forms of social capital in finding jobs and other career achievements, McPherson et al. (1992) found that weak ties (defined as density and as Blau space distance) were more potent forces than strong ties in shaping the joining and leaving of voluntary associations. 13

Retaining Members: Membership Duration and Niche Location Because of its reliance on network structure (in particular, homophily), McPherson's ecological theory gives a central role to the position of a member in a group's niche, and the competitive forces which that position exerts on him or her (Popielarz & McPherson, 1995). If a member is near the center of a group's niche in Blau space, the homophily principle dictates that most of his or her network relations will be with those who are in the group (or others who are similar to them), and there will be few ties outside of the group. Therefore, from Propositions 1 and 2, we can see that members located near the center of a group's niche will have longer membership durations than members nearer the edges of the niche. Effectively, the niche operates as an opportunity structure for the homophilous ties that shape membership. Popielarz and McPherson (1995, p. 703) characterized forces acting on members near the middle of a niche as centripedal, while forces acting on members near the edge of the group are centrifugal.

Proposition 5: The nearer a member is to the center of a group's niche, the longer that member's duration of membership. This proposition is clearly related to the classic research on similarity and cohesion. Lott and Lott (1965) reviewed over 50 studies that showed that similarity on sociodemographic characteristics, attitudes, values and behaviors were related to cohesion (measured as interpersonal attraction among group members). The ecological theory here would argue that groups with high similarity are more cohesive because they have more specialized niches. In effect, most members are closer to the niche center than in a more diverse, generalist organization. Social ties to others in such a specialist group are more likely to be redundant and have multiple kinds of connections, since the density of ties in the local area of the niche is greater than ties farther away in the

20

MILLER McPHERSON AND LYNN SMITH-LOVIN

space. Such packing of members in social space enhances sociation (as the classic studies found), and produces higher stability of membership through those dense and multiplex relationships. In contrast, groups with high variability (generalist) groups have dynamics which lead to greater turnover. This process is closely related to Kanter's (1977) discussion of the pressures that operate on tokens. People who are unlike most others in an organization or group experience predictable liabilities: they are stereotyped, visible and less likely to have network ties that provide social support. While Kanter's (1977) concern focused primarily on dynamics within organizations, McPherson's ecological theory is also oriented toward the external competitive pressures. Tokens not only have fewer positive experiences (or ties) within the group, they also have more competing interests outside the group, because of their location on the periphery. Competition between groups occurs at these niche edges when niches overlap - when groups recruit the same types of members. The more groups that are recruiting in an individual's neighborhood in Blau space, the more organizational competition operating on the individual. The conflicting forces created by intergroup competition lead to shorter membership durations for the individuals who experience them, because they do not have the time or other resources necessary for involvement in all of the activities available to them. Therefore, individuals in areas of niche overlap have shorter membership durations.

Proposition 6: Members in areas of the group's niche in Blau space where more groups overlap will have a shorter duration of membership than members in areas where fewer or no groups overlap. Popielarz and McPherson (1995, pp. 705-706) noted that these two principles are independent of one another, and both operate simultaneously. Indeed, they found support for both the niche edge effect (Proposition 5) and the niche overlap effect (Proposition 6) in their study of voluntary organization memberships. It is generally true that group overlaps occur at the edges. The principle of competitive exclusion (Gauss, 1934) operates in a community of groups (Popielarz & McPherson, 1995, p. 706), such that groups stop growing in directions that compete with other groups, and form edges there. The tendency for groups to grow at their edges eventually comes into equilibrium with their competition with groups at their edge to form the niche. But some edges may have more competition than others (McPherson & Ranger-Moore, 1991), and the two vary somewhat independently within any community of groups that is

Cohesion and Membership Duration

21

not in perfect equilibrium. Therefore, the niche edge and niche overlap effects can be assessed independently of each other, and are not definitionally related.

Retaining Members: Group Persistence The propositions developed above from the ecological theory of affiliation are all couched at the level of the individual membership. We need to develop their implications for group structure to relate Festinger's (1950) classic concem with forces acting on members to remain in the group to the more modem structural definitions of cohesion as a unified, dense network structure that has clear boundaries v i s a vis its environment (Markovsky & Lawler, 1994). An important group feature dependent on membership duration is the group half life - the duration of time until half of the members are expected to leave the group (McPherson et al., 1992, p. 166). Groups with lower half lives are better able to move in social space. They can change their niche position more quickly through selection and retention of different types of members, in response to competitive pressures. On the other hand, the changing composition of the group and the shifting membership makes these adapters less grouplike: their very instability makes them closer to what Markovsky and Lawler (1994) described as aggregates or assemblages than solidary groups. The group half life is a relatively straightforward function of the variables that determine membership duration (see McPherson, 1992, Eq. 3). Since ties to members and to non-members pull people into and out of groups, respectively, the relative preponderance of these ties determines its half life. An analysis of extremes illustrates the effect. A group with all of its ties intemal to the group and no ties outside the group would be projected (given McPherson et al.'s 1992 empirical estimates) to have a half life longer than the human life span. On the other hand, a group with a large preponderance of external ties would have a half life of less than three years. Therefore, we can make a straightforward system-level prediction about group turnover, based on our membership-level propositions.

Proposition 7: Group half life will be positively related to the proportion of member ties internal to the group, and negatively related to the proportion of member ties extemal to the group. Note that this proposition is closely related to the discussions of total power in the Lawler theoretical tradition. If network members have few available sources of valued resources outside of a group, they by definition are highly dependent on other group members for those resources and total power will be high. If

22

MILLER McPHERSON AND LYNN SMITH-LOVIN

there is no highly unequal internal structure (e.g. if there is high unity of structure according to Markovsky's and Lawler's (1994) discussion), then equal, high power will result. These are the most favorable conditions for relational cohesion and network cohesion in the Lawler tradition (Lawler & Yoon, 1996; Lawler & Thye, 1999). The processes described in Proposition 7 are accelerated or retarded by the preponderance of strong and weak ties in the system. Strong ties are stabilizing, tending to hold people into groups more with shared information and joint activity; they work in favor of stabilizing selection and membership retention. Weak ties accelerate change by exposing people to new information further from that held by their close associates; weak ties create disruptive or directional selection. They create new memberships from people who are near the niche edge, which Proposition 5 tells us are likely to be of short duration. Therefore, systems with more strong ties will have longer group half lives, and systems with more weak ties will produce more transitory, unstable groups. We expect this pattern to hold for all of the major ways of defining strength of tie - the contact volume, the density of interconnections among alters, and the distance of a tie in Blau space.

Proposition 8a: Group half life will be positively related to the average contact frequency of ties in a social system.

Proposition 8b: Group half life will be positively related to density of ties in a social system.

Proposition 8c: Group half life will be positively related to the degree of homophily that operates in a system. The persistence of groups, in McPherson's theory, depends heavily on the relationship of each group to the other groups in the system. Groups interact with each other as social entities through the mechanism of competition. Groups consume the time and other resources of the individual through their collective activities. This process places constraints on the number of groups that can exist at any given locus in Blau space. When an individual's resources are consumed by one or more groups, the probability of that individual joining another group is decreased, and the probability of exit from one of the current groups is increased (cf. McPherson, 1981), as is specified in proposition 6 above. Thus, groups feeding in the same region of social space will compete with each other for the ultimately limited resources contained in that region. This leads to a general hypothesis about the half life of groups as it depends on competition.

Cohesion and Membership Duration

23

Proposition 9: Group half life will be negatively related to the amount of competition in the niche of the group. Having moved our discussion of membership duration up to the group and system level, we are now ready to relate our discussion of "the forces acting on members to remain in the group" (Festinger, 1950) to the more modem definitions of cohesion as structural properties of the relationships within groups. In particular, we are interested in how the insights of ecological theory about the importance of system properties and groups' environments will affect both properties. For this purpose, we turn to simulation to see the effects of system-level parameters on groups' characteristics. AN ECOLOGY

OF COHESION:

A SIMULATION

The two forces that are at the heart of the ecological model of affiliation are homophily and competition. Here, we develop a simulation of these two social forces that drive the evolutionary development of communities of groups in order to see their joint effects on cohesion. The simulation shares its basic assumptions with McPherson's ecological theory (McPherson, 1999). First, groups gain and lose adherents through social contacts. People are recruited to groups through network contacts with members who are already associated with the group. In fact, the simulation here uses empirical estimates (McPherson et al., 1992) for the effect of each social contact on the probability of joining and leaving a given group. In effect, we assume that Propositions 1 and 2 are true, and employ empirical studies that have supported them to set their effects in this simulation. The second assumption of the simulation is that new social network ties are created at a higher rate inside groups. In the simulation, common membership in a group increases the probability that two actors will initiate social contact in future rounds of interaction. This assumption is consistent with Feld's (1981) argument that groups act as a focus of activity for network tie formation, and with empirical evidence from McPherson et al. (1992), which is used to set the level of the effect in the simulation.

Inputs and Outputs The two input parameters which vary in the model are the degree of homophily bias in the system's tie creation and the amount of competitive interaction among the groups. These two factors represent the core of the ecological theory,

24

MILLER McPHERSON AND LYNN SMITH-LOVIN

which argues that groups and networks co-evolve in Blau space because of the recruitment of members through homophilous networks and the competition with other groups. These two forces create the niches that groups occupy in Blau space, through the competitive exclusion that keeps groups from expanding indefinitely at their niche edges. We assess two features of the system that we have argued above mediate between these basic properties of the system and the cohesion of groups within it. Both these features indicate what types of relationships dominate the system - strong or weak ties. First, we measure the dyadic similarity of the ties in the system, which affects how strong, intimate, high contact frequency and densely interconnected those ties are likely to be. This dyadic similarity (which can be thought of as observed cross-sectional homophily) depends strongly, but not completely, on the homophily bias that we describe above, because it is generated by the formation of ties within groups, the competition of groups for members and other features of the system. As we will see in the simulation results, only about fifteen percent of the variance in observed homophily is ultimately account for by homophily bias. The remainder is due to the direct and indirect effects of competition on group and network structure, and the fact that homophily bias is probabilistic, rather than deterministic. Second, we assess the stability of ties - how long they endure across time. Again, we suggest that longer lasting ties are more likely to be strong ties: intimate, high frequency contact and representing high levels of shared information. We investigate two aspects of group cohesion, following the logic of prior work. First, we trace the reachability of group members from each other (Wasserman & Faust, 1994). This aspect of cohesiveness derives directly from Markovsky and Lawler (1994). The pattern of ties within the group creates the possibility of information flow between members. If all members are reachable from all other members, our reachability measure attains a value of 1.0. If, on the other hand, no connections among members exist, reachability is zero. More formally, our measure is based on the reachability matrix (Harary et al., 1965), which is an N by N matrix containing entries of unity for the case in which the ith member is reachable from the jth member, and zero otherwise. In the simulation, ties are symmetric, so that if i is reachable from j, the converse is also true. Thus, there are N(N-1)/2 independent entries in both the ties matrix and the teachability matrix. Our measure of reachability sums the entries in the reachability matrix, and divides by the number of independent entries, producing a range from 0 (low cohesiveness) to 1 (perfect cohesiveness). The second aspect of cohesiveness, more related to the classic definition of cohesiveness in the psychological literature (Festinger, 1950; Shaw, 1976, p. 197), is the duration of the life for the group itself. When all members of a

Cohesion and Membership Duration

25

group leave the group, it ceases to exist, and may be said to have lost all cohesiveness. Group age effectively captures the totality of forces that hold t[ae group together, in the context of its environment.

The Program The core device in the simulation is the N x N ties matrix, which is a matrix of zeros and ones that indicates whether or not each person in the social system has a tie to each other person. When a tie is present between person i and person j, the corresponding entry in the matrix is 1; if not, the entry is 0. For the purposes of this simulation, the tie matrix is symmetric with a diagonal of zeros. The ordering of individuals in the rows and columns of the tie matrix represents values on a Blau dimension (that is, some socially salient distinction). Therefore, rows near each other in the matrix indicate individuals similar to one another on the Blau characteristic. The second major component of the simulation is the person-by-groups matrix, which models the relationships between actors and groups. In each simulation, the number of persons is fixed; in this case, we model systems of size 50.14 The fact that groups form and dissolve during the evolution of the system means that the number of columns in this matrix can vary from iteration to iteration. An entry of 1 indicates that the ith person is a member of the jth group. A group dies when it no longer has at least two members (i.e. when the column corresponding to that group has one or no entries of 1). Groups form when social contact perseveres. When a social contact survives more than one round of interaction, it forms a group of size two. That group can either grow, stabilize as a dyad, or die. Growth occurs through the addition of new group members, as a probabalistic result of ties between current group members and non-group members. Groups decline when current members leave the group, either probabilistically or through their ties to non-members. Social contacts are generated probabilistically as a function of the homophily principle and common membership in groups. When the homophily bias parameter is high, persons nearby in Blau space are more likely to be connected, while those distant in Blau space are less likely to be connected. When the ties matrix is ordered strongly by homophily, there is a clustering of ones near the main diagonal of the matrix. When the homophily bias parameter is set at a low level, the ties are probabilistically generated with less clustering near the main diagonal. At its logical minimum of zero, ties would be generated randomly throughout the matrix. Common membership in groups also heightens the probability of a tie being generated probabalistically, to represent the fact that groups are foci for tie formation.

26

MILLER McPHERSON AND LYNN SMITH-LOVIN

The second key feature of the ecological theory, the competition among groups, is varied in the simulation as well. When competitive forces among groups are higher, group memberships in the person-by-group matrix die at a faster rate if there are more non-zero entries in the person row (i.e. if the person is a member of multiple groups). At the logical extreme of zero competition, group memberships would have no affect on one another: memberships would die only at the base rate and the rate at which people are pulled out of the group by ties with non-members. At the logical extreme of high competition, only one group membership in each person row could survive an iteration.

Results of the Simulation Study Varying both homophily bias and competition over most of their range, ~5 with 20 values of homophily bias and 22 values of competition estimated, produces a ~lata set of 440 simulated societies. We evolve each society for 10 iterations to produce the patterns of network ties and groups that we analyze here. First, we examine how homophily bias (hombias) and competition (comp) affect the basic structure of relationships in a society - the dyadic similarity or observed homophily (homm) and the proportion of ties within groups (pwin). These relationships are shown in Fig. 1, which gives the results of a regression analysis of output from 440 runs of the simulation, j6 The parameter estimates are standardized regression coefficients produced by full information maximum likelihood. The correlation between the error terms of observed homophily (homm) and proportion of ties within the group (pwin) may be regarded as the partial correlation of these two variables, controlling for competition and homophily bias. As we note above, observed homophily is related toi but not completely determined by, the homophily bias displayed by individuals as they form ties within the system. The effect of homophily bias on observed homophily is positive and significant, but competition also has an important effect. The level of competition between groups in the system (effectively, the degree to which maintaining multiple memberships is low probability) has a negative effect on observed homophily. Basically, increased competition between groups tends to drive people out of group memberships at a higher rate. Since groups are typically less diverse than the general population in Blau space, and since groups form the principal foci for the formation of ties, observed homophily is increased when people belong to more groups and form more of their ties within those environments. Restricting group memberships by making them more competitiCe for members limited resources tends to increase the range of Blau space that one is exposed to through one's personal contacts. As McPherson and

27

Cohesion and Membership Duration

homm •

1.00-~

[.~-0.861

0.3

hombias

f

0.00

0.22

pwin Fig. 1.

0.91

System Level Sources of Observed Homophily and the Proportion of Ties Within the Group.

Smith-Lovin (1986, 1987) discovered, ties formed within groups are even more homophilious than ties formed in other contexts like work, family or neighborhood; rather than increasing the integration of a society through crosscutting ties, voluntary groups tend to reinforce its social divisions. Competition also has a strong negative impact on the proportion of ties within the group, for many of the same reasons. When people are forced to drop competing group memberships and belong to fewer groups, they also over time drop network ties that developed in the context of those groups. Groups and networks reinforce each other, so if the competition among groups is increased, the commitment to specific relationships tends to decrease as well. Interestingly, the dynamics of group competition has a much larger effect on ties within the group than dyadic similarity. Basically, the proportion of ties within the group in our simulated societies is more determined by the ecology of groups than the network formation of individuals. Figure 2 shows how our system parameters and the tie structure of our simulated societies affect the average cohesion of groups, using the Markovsky and Lawler (1994) definition of reachability (mreach). Again, we find that competition among groups for members is a much more powerful force than dyadic similarity. Competition has a positive direct effect on reachability within groups, primarily because it makes groups smaller and more homogeneous, with less overlap. This direct effect is counteracted, however, by the negative impact that operates through the proportion of ties within the group. Competition

28

MILLER McPHERSON AND LYNN SMITH-LOVIN

cA

.o

..=

O

[-

eq

Cohesion and Membership Duration

29

tends to reduce ties within, but ties within have a strong positive impact on reachability within groups. If ties are prevalent, the ties that form within groups tend to cumulate over time, tying the group together into a more cohesive unit. As Lawler and his colleagues have noted, behavioral commitment to particular relationships tends to create stable, cohesive collectivities that have a stable, grouplike character. Therefore, lower competition tends to both tie groups together internally and to cut them off from the disrupting influences of information outside of their boundaries. As Snow et al. (1980) noted, groups with this characteristic have a very tight hold on their members. Systems with high competition among groups (i.e. where multiple group membership is difficult or impossible) are characterized by such groups. Homophily bias at the level of individual tie formation actually has a small negative effect on reachability within groups. This effect occurs because high levels of homophily bias, combined with the tendency for ties to form within groups, leads to the creation of subcliques within groups. High levels of what McPherson and Smith-Lovin (1987) called "choice homophily" (the tendency to choose network ties with similar others within groups) creates substructures that break down the unitary character of groups. These groups are less solidary, in Markovsky and Lawler (1994) terms. Figure 3 provides an interesting contrast to Fig. 2. Here, we look at the effects of homophily bias, group competition, and tie characteristics on the average age of a group in the system (mage). This focus on the survival of the group as a unit within the larger system is closer to the Festinger (1950) emphasis on the forces that tend to keep the group together over time. We see that reachability (the Markovsky & Lawler [1994] conceptualization of cohesion) and the actual survival of the group over time have somewhat different relationships to the community structure and the patterns of ties within our simulated societies. Again, we find that competition has a much larger effect on group survival than does homophily bias in individual tie formation. Not surprisingly, competition among groups reduces the mean age of groups. Groups are less likely to survive for long periods if it is difficult to belong to many groups at once. This negative impact of competition on group survival operates both directly and indirectly through its negative impact on ties within the group. Ties within the group tend to increase group survival and stability over time, and competition reduces the ties within by removing the co-memberships that sustain interpersonal relations. Homophily bias has a smaller, positive effect on group survival. This effect operates through a number of paths - there is a small direct effect, there is a weak

30

MILLER M c P H E R S O N A N D L Y N N SMITH-LOVIN

~5

0.)

<

I::I C)

O¢3

c5

¢J I"~

c5

"O

cD

~5 aj

,,,q

Cohesion and Membership Duration

31

indirect effect through the observed homophily of the system's ties, and there is a weak indirect effect through the positive influence of homophily bias on ties within the group. These cumulate to make systems where people are more likely to form ties with similar others more likely to have longer lasting groups. Taking the simulation results together, we see two general conclusions. First, the system-level properties of an ecological community have powerful impacts on the character of ties within that system and the cohesion of groups in the society. The level of homophily in a system's tie formation leads to a more homophilous observed tie structure, and to somewhat more stable personal relationships in the system. Both of these forces contribute to group cohesion, through the development of a higher proportion of stable ties within the groups of the system. But homophily bias has some negative impact as well, since it leads people within groups to form cliques with those who are more similar to them than other group members, reducing reachability. The competition of groups for members has a much larger impact than homophily, however. Competition reduces both observed homophily and ties inside the group, leading to less group cohesion. The negative effect on reachability within the group is moderated by the fact that competition tends to push groups apart in social space, leaving group members to form ties with one another without much competition from outside the group. But the negative effect of competition on group survival is profound, operating through both directly and through the structure of relationships within the society. This distinction leads us to our second major conclusion: the two facets of group cohesion - reachability and survival - have different relationships to community structure. They clearly tap different aspects of the groups, and are not reducible to interpersonal tie structure. Since both of these aspects of cohesion have been incorporated into our more abstract discussions of cohesion in the theoretical literature, this difference is important. Relationship strength, membership commitment, reachability and group survival are all related, partially because they are all dependent on the some of the same ecological community parameters. But they are distinct theoretical concepts, with somewhat different relations to antecedents and outcomes. Theorists would be well advised to be specific in their discussions of these phenomena. CONCLUSIONS We began this theoretical effort with two goals: (1) to redirect attention in the study of cohesion to the impact of the larger social environment in which groups operate; and (2) to encompass within our consideration of cohesion the

32

MILLER McPHERSON AND LYNN SMITH-LOVIN

traditional concem with the continuity of both membership and the group itself over time. To accomplish these goals, we have drawn on a well-developed literature on the evolutionary ecology of affiliation. We have suggested that the duration of memberships, the internal coherence of groups, and the survival of groups over time all depend importantly on the community structure within which groups exist. Concentrating only on dynamics that occur within groups was a persistent problem in the cohesion literature of the 1950s and 1960s; we hope that our new understanding of the system-level processes in which both networks and groups are embedded can help us avoid the same pitfalls in the new round of theoretical insights. In exploring the dependence of membership commitment and group cohesion upon community structure, we have again illustrated the point made by many other critics of the cohesion literature (see especially Hogg, 1992; Markovsky & Lawler, 1994) - that a concrete, theoretically based definition of cohesion is necessary for theoretical advancement. Our analyses show that different aspects of "groupness" that can easily be embraced under a common sense notion of cohesion have quite different relationships to the larger social system in which memberships and groups are embedded. Clearly, theorists need clarity about whether they are talking about membership duration, internal reachability, group continuity of membership, group survival or some more affective or cognitive response of individual members to these system features. We believe that attention to the community level of analysis for group processes is long overdue.

NOTES 1. Lawler (1992) explicitly states that a parallel set of processes work in the opposite direction: lack of control leads to negative emotion, which leads to disaffection from larger identification. 2. Jointness is high when individual actors find it hard to distinguish their individual effects toward solving an exchange task. 3. This is not to say that sociologists have ignored the parallel developments in ps3}chology. Markovsky and Chaffee (1995) incorporate these ideas by allowing categories as nodes in a sociological network formulation. 4. We note an odd twist in intellectual connections here. Typically, mechanical solidarity is thought to be more based on emotional responses. It is interesting that the Lawler theoretical research program has brought emotions to such a central role in explaining how groups form as the result of Individual interest-serving utilitarian exchanges. 5. Indeed, they argued that one of the reasons for the lack of progress in our understanding of cohesion was that psychological social psychology became increasingly separate from sociology as the disciplines developed (Hogg, 1992, pp. 52-54).

Cohesion and Membership Duration

33

6. The theory has been successfully applied to musical tastes (Mark, 1998a), occupations (Rotolo & McPherson, 2001), religious affiliations (Chaves & Giesel, 2001) and cultural traits (Mark, 1998b), as well as voluntary membership groups. 7. In the survey literature on voluntary groups and associations, groups are often identified by giving the respondent a list of examples. Mentions of previously uncognized groups may be elicited by closeness to the prototype groups mentioned. For example, one of McPherson's Nebraska respondents might not have thought of the eight people who meet every Tuesday night to play bridge as a "group" until "bridge club" was mentioned as an example of a hobby group (McPherson et al., 1992). Clearly, assemblies vary in terms of their "groupness" or cohesion. This is one of the factors that we would like the ecological theory to explain. In this sense, our conceptualization is similar to Markovsky's and Lawler's (1994) where group cohesion can vary from zero (aggregate) to maximum density of strong ties (solidary group). We simply add the element of time, in keeping with the dynamic nature of the evolutionary ecological theory. 8. McPherson chose the label "Blau Space" to honor Peter Blau's (1977) contribution to sociology's understanding of how homophily along these socio-demographic variables structures societies. He believed that honoring our foundational ideas through this naming process would confer some of the prestige that other disciplines give to their core ideas (e.g. Laffer curves). Some journal editors have disagreed: therefore, Blau Space has been called the more generic "social space" in some publications (see: McPherson, Popielarz & Drobnic, 1992, footnote 2; Popielarz & McPherrson, 1995) and "sociodemographic space" in others (Rotolo & McPherson, 2001). 9. The degree of correlation of the salient characteristics in the space affects its dimensionality. If age, wealth, education and other status variables are correlated more or less perfectly, then the social space in which affiliative structures operate is effectively unidimensional. If such characteristics are essentially uncorrelated, then social differentiation is more pronounced and multidimensional (see McPherson & Ramger-Moore, 1991, pp. 20-21, for a discussion of how this unfolding occurred over time, and its relationship to Durkheim's concepts of mechanical and organic solidarity). 10. As McPherson and Ranger-Moore (1991, p. 22) noted, homophily reduces the complexity of the network by translating N individuals with R possible types of relations into N individuals with K values on metric dimensions. 11. We wish to emphasize that this concentration is not an explicit, thought-out strategy of either the group or any of its members. It occurs quite naturally as a result of members being drawn into group membership through network ties. 12. Although the negative effect of non-member ties is less dramatic, its practical importance is still quite strong: there are many more non-member ties, even in McPherson's data, which is limited to five ties total. 13. It is interesting to think briefly about how these propositions relate to the network exchange literature. If a non-group alter is not connected to any of one's other exchange partners, one might have more power in the relationship and it might be more likely to evolve into a committed relationship that would threaten group membership. This reveals some tension in the definition of a "strong tie" for exchange relationships, however, since the feature that is positioned to make a tie weak within the McPherson formulation tends to make a tie become strong over time in the exchange case. 14. Experimentation with larger system sizes, which require more computing time, indicate that the patterns we find here are not particularly sensitive to system size.

34

MILLER M c P H E R S O N AND LYNN SMITH-LOVIN

15. At some extremes of homophily or competition, the system produces degenerate solutions. When homophily is extremely high, dyads of near Blau-space neighbors form and groups larger than size two cannot form. When competition is very low, groups tend to grow to encompass the entire system so that all people belong to all groups. 16. Standard errors and tests of significance are a function of sample size, so we do not use them here, since increasing the number of computer runs could make any difference from zero arbitrarily significant. The parameter estimates are stable enough at 440 runs to interpret the qualitative nature of the relationships among the variables.

ACKNOWLEDGMENTS The authors gratefully acknowledge the support of National Science Foundation Grants SES-9905149 (to McPherson) and SES-01110599 (to Smith-Lovin). The authors bear full responsibility for any interpretation of results contained in this document.

REFERENCES Blau, P. M. (1977). Inequality and Heterogeneity. New York: Free Press. Chaves, M., & Giesel, H. (2001). How should we study religious competition? Paper presented at the meetings of the Society for the Scientific Study of Religion, Columbus, OH. Coleman, J. S. (1986). Individual Interests and Collective Action. New York: Free Press. Coleman, J. S. (1990). Foundations of Social Theory. Cambridge, MA: Harvard University Press. Festinger, L. (1950). Informal social communication. Psychological Review, 57, 271-282. Festinger, L., Schacter, S., & Back, K. (1950). Social Pressures in Informal Groups. New York: Harper and Row. Feld, S. L. (1981). The Focused Organization of Social Ties. American Journal of Sociology, 86, 1015-1035. Friedkin, N. E. (1998). A Structural Theory of Social Influence. Cambridge University Press. Friedkin, N. E. (2001). Norm Formation in the Bank Wiring Observation Room. Social Networks, 23, 167-189. Gause, G. F. (1934). The Struggle for Existence. Williams and Wilkins. Granovetter, M. S. (1973). The Strength of Weak Ties. American Journal of Sociology, 78, 1360-1380. Hall, J. R. (1988). Social Organization and Pathways of Commitment: Types of Communal Groups, Rational Choice Theory, and the Kanter Thesis. American Sociological Review, 53, 679-692. Harary, F., Norman, R. Z., & Cartwright, D. (1965). Structural Models. New York: Wiley. Hechter, M. (1987). Principles of Group Solidarity. Berkeley: University of California Press. Hogg, M. A. (1992). The Social Psychology of Group Cohesiveness. New York: Harvester Wheatsheaf. Kanter, R. M. (1968). Commitment and Social Organizations: A Study of Commitment Mechanisms in Utopian Communities. American Sociological Review, 33, 499-517. Kanter, R. M. (1977). Men and Women of the Corporation. New York: Free Press. Knoke, D. (1981). Commitment and detachment in voluntary associations. American Sociological Review, 46, 141-158.

Cohesion and Membership Duration

35

Lawler, E. J. (1992). Choice Processes and Affective Attachments to Nested Groups. American Sociological Review, 57, 327-339. Lawler, E. J. (2001). An Affect Theory of Social Exchange. American Journal of Sociology, 107, 321-352. Lawler, E. J., & Thye, S. (1999). Bringing Emotions into Social Exchange Theory. Annual Review of Sociology, 25, 217-244. Lawler, E. J., Thye, S. R., & Yoon, J. (2000). Emotion and Group Cohesion in Productive Exchange. American Journal of Sociology, 106, 616-657. Lawler, E. J., & Yoon, J. (1993). Power and the Emergence of Commitment Behavior in Negotiated Exchange. American Sociological Review, 58, 465-481. Lawler, E. J., & Yoon, J. (1996). Commitment in Exchange Relations: Test of a Theory of Relational Cohesion. American Sociological Review, 61, 89-108. Lawler, E. J., & Yoon, J. (1998). Network Structure and Emotion in Exchange Relations. American Sociological Review, 63, 871-894. Lewin, K. (1951). Field Theory in Social Science. New York: Harper. Lin, N. (2001). Social Capital: A Theory of Social Structure and Action. London: Cambridge University Press. Lott, A. J., & Lott, B. E. (1965). Group cohesiveness as interpersonal attraction. Psychological Bulletin, 64, 259-309. Mark, N. (1998a). Birds of a Feather Sing Together. Social Forces, 77, 455-485. Mark, N. (1998b). An Ecology of Culture. Unpublished doctoral dissertation, University of Arizona. Markovsky, B., & Chaffee, M. (1992). Social Identification and Solidarity: A Reformulation. In: B. Markovsky, K. Heimer, J. O'Brien & E. J. Lawler (Eds), Advances in Group Processes (Vol. 12, pp. 249-270). Greenwich, CT: JAI Press. Markovsky, B., & Lawler, E. J. (1994). A New Theory of Group Solidarity. In: B. Markovsky, K. Heimer, J. O'Brein & E. J. Lawler (Eds), Advances in Group Processes (Vol. 11, pp. 113-137). Greenwich, CT: JAI Press. McPherson, J. M. (1983). An Ecology of Affiliation. American Sociological Review, 48, 519-532. McPherson, J. M. (1981). A Dynamic Model of Voluntary Affiliation. Social Forces, 59, 705-728. McPherson, J. M. (1999). Modeling Change in Fields of Organizations: Some Simulation Results. In: D. llgen & C. Hulin (Eds), Computational Modeling of Behavioral Processes in Organizations. American Psychological Association Press. McPherson, M., Popielarz, P. A., & Drobnic S. (1992). Social Networks and Organizational Dynamics. American Sociological Review, 57, 153-170. McPherson, J. M., & Ranger-Moore, J. R. (1991). Evolution on a Dancing Landscape: Organizations and Networks in Dynamic Blau Space. Social Forces, 70, 19-42. McPherson, J. M., & Rotolo, T. (1996). Testing a Dynamic Model of Social Composition: Diversity and Change in Voluntary Groups. American Sociological Review, 61, 179-202. McPherson, M., & Smith-Lovin, L. (1986). Sex Segregation in Voluntary Associations. American Sociological Review, 51, 61-79. McPherson, M., & Smith-Lovin, L. (1987). Homophily in Voluntary Organizations: Status Distance and the Composition of Face-to-Face Groups. American Sociological Review, 52, 370-379. McPherson, M., Smith-Lovin, L., & Cook, J. (2001). Birds of a Feather: Homophily in Social Networks. Annual Review of Sociology, 27, 415-444. Popielarz, P., & McPherson, M. (1995). On the Edge or In Between: Niche Position, Niche Overlap and the Duration of Voluntary Association Memberships. American Journal of Sociology, 101, 698-720.

36

MILLER M c P H E R S O N AND LYNN SMITH-LOVIN

Rotolo, T., & McPherson, J. M. (2001). The System of Occupations: Modeling Occupations in Sociodemographic Space. Social Forces, 79, 1095-1130. Shaw, M. E. (1976). Group Dynamics: The Psychology of Small Group Behavior (2nd ed.). New York: McGraw Hill. Snow, D. A., Zurcher, L. A,, Jr., & Eckland-Olson, S. (1980). Social Networks and Movements: A Microstructural Approach to Differential Recruitment. American Sociological Review, 45, 787-801. Tajfel, H., & Turner, J. C. (1979). An Integrative Theo~ of Intergroup Conflict. In: W. G. Austin & S. Worchel (Eds), The Social Psychology oflntergroup Relations (pp. 33--47). Monterey CA: Brooks-Cole. Thye, S., & Lawler, E. (1999). Collaborative Research on Social Exchange and Network Cohesion. Proposal to the National Science Foundation (funded, August 15, 1999). Turner, J. C. (1985). Social Categorization and the Self-Concept: A Social Cognitive Theory of Group Behavior. In: E. J. Lawler (Ed.), Advances in Group Processes (Vol. 2, pp. 77-122). Greenwich, CT: JAI Press. Walker, H. A., Thye, S. R., Simpson, B., Lovaglia, M. J., Willer, D., & Markovsky, B. (2000). Network Exchange Theory: Recent Developments and New Directions. Social Psychology Quarterly, 63, 324-338. Wasserman, S., & Faust, K. (1994). Social Network Analysis: Methods and Applications. Cambridge University Press.

GROUP SOLIDARITY AS THE PRODUCT OF COLLECTIVE ACTION: CREATION OF SOLIDARITY IN A POPULATION OF INJECTION DRUG USERS Douglas D. Heckathorn and Judith E. Rosenstein

ABSTRACT This paper introduces a theory of group solidarity and a method for measuring it. Solidary groups are characterized by strong internal monitoring and sanctioning systems, strong intra-group ties, high exit costs, and lack of information about resources outside the group. This analysis suggests that all these attributes derive from the choice to invest differentially in social relationships within the group rather than forming cross-cutting ties. To explain variations in solidarity across groups thus requires an account of the conditions that favor intra-group ties. Drawing on a formal theory of collective action, the analysis shows how the return from investments in intra-group ties varies based on the shape of the production function for the collective goods produced by the group. The proposed measure of group solidarity is based on the degree to which the proportion of intra-group ties exceeds that which would be expected

Group Cohesion, Trust and Solidarity, Volume 19, pages 37-66. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2 37

38

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN were ties formed randomly. The theory of group solidarity and the measurement procedures are illustrated using data from a survey of 488 injection drug users in Connecticut. INTRODUCTION

Solidarity is one of the venerable social science concepts for which there are almost as many definitions as theorists who have analyzed it. Following Homans (1950), Fararo and Doreian (1998, p. 16) distinguish four distinct ways to analyze solidarity. Analyses can focus on the affective bonds that unite members of solidary groups; the norms defining group obligations; the collectively oriented activity patterns characteristic of these groups, especially a preponderance of prosocial behavior; and the interaction patterns in which ties within the group are denser than ties across groups. Diverse approaches to analyzing solidarity have flourished because any of these four elements can be taken as the fundamental characteristic of solidarity, and thereby be used to explain the other elements. For instance, Widegren (1997) views solidarity as deriving from altruistic sentiments directed toward those who reciprocate those sentiments. Altruism then is invoked to explain prosocial activity, strong group norms, and dense intra-group ties. In contrast, Lawler and Yoon (1998) view positive sentiments not as fundamental but as deriving from repeated interactions, especially mutually satisfactory interactions among status equals. Given that all four elements of solidarity generally occur together, analysts face a problem analogous to multicolinearity in regression analysis, when highly correlated terms make it difficult to disentangle causal linkages. This paper has two purposes. The first is to propose an operational definition of solidarity that is applicable in contexts where corporate obligations lack clarity and the resources allocated on their behalf do not produce a documentary record. The second is to propose a theoretic account of the factors governing choice of intra-group vs. out-group ties. The aim is to extend Hechter's theoretic account of solidarity, which focuses on structural attributes of solidary groups, with an analysis of the conditions under which these attributes will be created as products of collective action. Part I reviews Hechter's theory of group solidarity, extending his analysis of the structural determinants of solidarity to identify the form of collective action by which these structures are produced. Part II introduces a method for measuring solidarity, based on the argument that the strength of preference for intra-group ties, homophily, provides a quantitative way to measure solidarity that is applicable in a wide range of structured and non-structured settings.

Group Solidarity as the Product of Collective Action

39

Part III proposes a model to explain the relative preference for intra-group vs. out-group ties, which is based on an expansion of Heckathorn's (1988, 1989, 1990, 1993, 1996) theory of collective action. As in Hechter's analysis, the model focuses on the nature of the collective goods produced by solidary groups. The analysis differs by shifting the focus from whether the goods are fungible or imminent, to the forms of social dilemmas that arise when goods are produced within the group. The analyses are illustrated using a study of variations in group solidarity among a total of 488 injection drug users from three Connecticut cities, Meriden, Middletown, and New London. The conclusion discusses methods by which group solidarity can be increased and describes how those methods have been employed in the design of AIDS-prevention interventions.

I: HECHTER'S THEORY OF GROUP SOLIDARITY An important step toward a coherent and comprehensive analysis of solidarity is Hechter's (1987) extension of dependence theory to solidarity. Noting that some analysts discuss solidarity without offering a definition, and others use definitions that cannot be operationalized, one of Hechter's aims was to define solidarity in a manner that would make the concept measurable. A second aim was to offer a general theoretic model that could be used to explain variations in solidarity across groups and over time. Hechter (1987) defines group solidarity as the proportion of member resources employed to fulfill corporate obligations. Solidarity therefore depends on two factors, the rate of compliance with corporate obligations and the extensiveness of these obligations. His illustrations of how this definition makes solidarity measurable include a study of party solidarity in a legislature with roll-call votes. The corporate obligation involves legislators voting in the manner supported by party leaders. The extensiveness (E) of the group's corporate obligations is the proportion of votes on which party leaders take a position, and compliance (C) with those obligations involves party members voting consistently with that position. In this case, solidarity (S) refers to the proportion of votes consistent with party positions (i.e. S = E * C).~ In a legislature, the essential resource is the ability to vote. In other groups, the relevant resource involves time spent on group rather than individual activities. Hechter's substantive theory focuses on the factors that determine compliance with and extensiveness of corporate obligations. Compliance, he argues, depends on the sanctioning capacity of the group and the group's monitoring capacity. Thus, when violations of corporate obligations are detectable and can be effectively sanctioned, compliance is high. This is consistent with Hechter's view of behavior as incentive driven. However, it ignores research demonstrating the

40

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

paradoxical effects of incentives. For example, overly severe sanctions can backfire. For this reason, Ostrom (1990) emphasizes the importance of graduated sanctions, sanctions that are both proportionate to the offense and consistent with group norms, and Heckathorn (1990) shows that overly severe sanctions can provoke resistance rather than compliance. Similarly, research on the "hidden cost of reward" (Deci & Ryan, 1985) shows that rewards can weaken intrinsic motivation. The implication is that, paradoxically, strengthening sanctions or improving monitoring can reduce rather than increase compliance with corporate obligations. Hechter's analysis of the extensiveness of corporate obligations seeks to identify the upper and lower bounds within which this may vary. The upper bound, he argues, is determined by the actor's dependence on the group. This determines the maximum demands the group can place upon the individual. Dependence is based on four factors: (1) lack of close substitutes outside the group for resources provided within the group; (2) lack of information about alternatives outside the group; (3) cost of exiting the group; and (4) commitments to personal ties within the group. Thus, the greater the individual's dependence upon the group, the greater may be the corporate demands on the individual. Corporate obligations may be less extensive than the upper limit established by dependence, but they cannot be greater, otherwise an individual would have incentives to exit the group, abandon within-group ties, and seek alternatives for that which the group provided. Therefore, dependence defines the upper bound of corporate obligations. The lower bound for corporate obligation is determined by the cost of the joint activities undertaken by the group (Hechter, 1987). Given that solidarity refers to the extent to which individual resources are used in a manner consistent with corporate dictates, when solidarity increases, so too must the individual resources employed to fulfill corporate obligations. This follows directly from the definition of solidarity and from Hechter's assumption that the resources required to fulfill corporate obligations come from group members. Hechter's theoretic contribution includes a measurable definition of solidarity and a consistent theoretic account of factors upon which solidarity depends. Two remaining problems provide the opportunity for further theorizing. One problem concerns the measurability of solidarity. When applied to a legislature, as in the above analysis, the measure is clear as long as one knows individual votes, party affiliations, and the positions of party leaders. In less structured contexts, measurement becomes less straightforward because corporate obligations lack clarity, and the resources allocated to fulfill them - such as time spent on alternative activities - do not leave a paper trail. This problem is apparent in occupational studies of time allocation. The variance in time spent

Group Solidarity as the Product of Collective Action

41

on occupational pursuits is highest for university faculty members, ranging from as little as a dozen to more than 60 hours per week. This variation results, in part, from ambiguity in the definition of academic work. If it is restricted to student contact time and structured research activities, investments of time are far lower than if the definition is expanded to include thinking about writing projects when in the shower, when preparing to sleep, and the like. Similarly, in family and peer groups, corporate obligations include mutual assistance, participating in joint recreational activities, and avoiding activities harmful to the group. For a compliant individual, satisfying the latter obligation could consume the entirety of available time. One aim of this paper is to propose an operational definition of solidarity that is applicable in contexts where corporate obligations lack clarity and the resources allocated on their behalf do not produce a documentary record. The second problem - and the second area of potential further theoretic development - is a failure to specify how the structural features upon which solidarity depends are themselves created. According to Hechter, solidary groups are characterized by extensive systems for monitoring and sanctioning, which provide the tools by which the behavior of group members can be controlled. The question that remains unresolved is, Why would group members invest resources to create extensive systems of monitoring and sanctioning? The factors leading to this investment decision (the proximate causes of group solidarity) can then provide the basis for explaining the intervening causes of solidarity that is, the enforcement resources that secure group control of members' behavior and restrict information about alternatives outside the group. When the determinants of solidarity are examined, they can all be seen as deriving from a single proximate (i.e. originating) cause. Consider again the first determinant of solidarity, compliance with corporate obligations. In a solidary group, intra-group ties are denser than out-group ties, so group members necessarily have greater abilities to monitor one another's performance than do out-group members. Similarly, owing to the density of intra-group ties, group members have enhanced abilities to mediate rewards and punishments, thereby strengthening their sanctioning capacity. Therefore, greater compliance can be seen as resulting from a structural feature of solidarity systems, their greater density of intra-group ties. The question then becomes, under what conditions do group members invest differentially in intra-group ties? Dependence can also be seen as a product of differential investments in intra-group ties. Consider, for example, the determinants of dependence identified by Hechter. In a solidary group, intra-group ties are denser than out-group ties, so individuals tend to have less information about substitutes for their intra-group ties and less information about alternative ties, and the

42

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

costs of exiting the group are greater because more ties would need to be reestablished. Furthermore, intra-group ties tend to be stronger than the out-group ties with which they could be replaced because newly formed ties tend to be weak and gain strength only over time if interactions prove rewarding. Therefore, all four determinants of dependence depend directly on the greater investment in intra-group ties in solidary groups. Once again, the question arises, What motivates the preference for intra-group over out-group ties? Hechter provides a partial answer to this question in his discussion of the differences between groups producing fungible (salable) goods vs. imminent goods, such as participating in a ritual or enjoying the company of friends, which are consumed instantaneously with their production. The solidarity of groups producing fungible goods tends to be lower because the output can be accumulated and used for transactions outside the group, thus encouraging interactions with non-group members. In contrast, imminent goods can neither be hoarded nor used in extra-group exchanges. Yet imminent goods are produced in all social systems because whether solidarity is high or low, individuals enjoy one another's company and engage in the myriad other forms of interactions in which positive affect is exchanged. Thus, dense intra-group networks are not a prerequisite to the production of imminent goods, and some high-solidarity groups (e.g. charismatically organized religious sects) produce fungible goods. In sum, monitoring and sanctioning capacities, and dependence on the group, are all products of collective action - products created when group members invest differentially in intra-group rather than out-group ties. Therefore, a fundamental issue in the explanation of solidarity is identification of the factors that motivate individuals to invest differentially in intra-group rather than out-group ties. Some of the factors relevant to this question are revealed in Burt's (1998) analysis of mobility patterns within a large investment banking firm. He found that the optimal network configuration varied based on duration of employment, ethnicity, and gender. For white males employed more than one year, advancement occurred more rapidly if they cultivated what Burt termed "entrepreneurial networks." These are networks that avoid transitive closure. That is, cultivating far-flung relationships rather than cultivating relationships with the friends of friends (an association that would produce transitive systems of social relationships, because A knowing B, and B knowing C, would lead to A knowing C). The optimal network structure for these men involved associations with many individuals who were socially distant from one another, many of whom did not even know one another. This structure optimized information gathering by filling what he termed "structural holes," in which poorly connected networks provide opportunities for brokers to gain from

Group Solidarity as the Product of Collective Action

43

facilitating the flow of information. In contrast, for women, minorities, and white men employed for less than one year, advancement was most rapid if they cultivated a dense network structure, a structure consistent with the relatively closed character of solidary groups. Burt interpreted this difference as resulting from variations in trust. Initially constructing a dense network ensured that those monitoring the employee's performance could readily share information. For white males who performed well, this led to acceptance as trusted organizational members, and they could then fashion entrepreneurial networks. In contrast, minorities and women were blocked from taking this additional step. What is relevant here is that a variety of situational factors may affect the relative preference for intra-group vs. out-group ties. This is consistent with the observation that groups vary greatly in solidarity, reflecting heavy investments in intra-group ties in some contexts and minor investments in others. II: CONCEPTUALIZING

SOLIDARITY

If, as was argued above, the mechanisms underlying solidarity derive from disproportionate investments in intra-group over out-group ties, then a conceptualization based on the distribution of ties may provide the basis for a quantitative measure of solidarity. Much prior research relevant to such an approach has been done. Building on Simmel's (1955) approach to analyzing social structure in terms of affiliation patterns, several formal models have been proposed (Fararo & Sunshine, 1964; Blau, 1977, 1994; Rapoport, 1979), all of which are based on the idea that if a social system lacked structure, ties would be formed randomly. Each group's ties would therefore reflect an identical pattern, in which the number of ties directed to a specific group would be exactly proportional to that group's size. Of course, there are no social systems in which ties are formed without respect to gender, age, ethnicity, social status, and myriad other factors; hence the universality of social structure. One form of departure from randomness has received special attention in the literature on affiliation patterns. As Galton recognized more than a century ago, affiliations tend to form among those who are similar in age, education, prestige, social cl'ass, and race and ethnicity (McPherson & Smith-Lovin, 1987). This bias toward self-affiliation is termed homophily, and it is a fundamental element of social structure. In Durkheim's (1956) terms, this corresponds to mechanical solidarity, in which solidarity is based on similarity. To the extent that solidarity derives from differential investments in intra-group rather than out-group ties, then homophily provides the network-structural basis for solidarity and implies that homophily can provide a structurally based indicator of solidarity.

44

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

Fararo and Sunshine's (1964) "inbreeding bias" parameter provides the basis for a quantitative measure of homophily (Heckathorn, 1997). Perfect homophily, in which all ties are formed within the group, is assigned the value + 1; and no homophily, in which ties are formed without regard to group membership, is assigned the value zero. When the level of homophily is intermediate, such as 1/3, ties are formed one third of the time within the group, and two thirds of the time randomly, without regard to group membership. Tie formation need not favor the in-group. It can also favor the out-group, as in social systems with exogamous marriage norms, competitive career tracks in which upward mobility depends on the ability to cultivate ties to those of higher status, and musical groups in which different types of instruments are represented. Therefore, negative homophily, termed heterophily, is also possible. This corresponds to Durkheim's (1956) concept of organic solidarity, in which solidarity is based on interdependence. Encompassing this possibility requires a straightforward expansion of the definition of homophily (Heckathorn, 2002). Perfect heterophily exists when no intra-group ties are formed, so homophily is -1. Intermediate levels of negative homophily can be defined in a similar way; for example, if homophily is -1/4, then 25% of the time ties are formed with out-group members, and the other 75% of the time ties are formed randomly, irrespective of intra- or out-group membership. As defined in this way, homophily is negative if and only if the proportion of ties that are internal to the group is less than the proportional size of the group. More formally, where Pa is the proportional size of group A, and S a is the proportion of A's ties that are within the group, the homophily of group A homophily, H a, can be defined as:

H-

Pa- S aa (~_~

if

Saa >

Pa (1)

S a -- Pa

H - - P~

ifP>S

For example, surveys of 190 injection drug users in Meriden, Connecticut (see Table IA) showed that the proportion of intra-group ties among Hispanic injectors was substantially larger (0.452) than this group's estimated size (0.198) so homophily was substantial (0.198-0.452)/(0.198-1)=0.317. Thus, these injectors created networks as though 31.7% of the time, they formed ties to other Hispanic injectors, and the other 68.3% of the time, they formed ties

Group Solidarity as the Product of Collective Action

45

randomly, irrespective of ethnicity. Homophily was also substantial for nonHispanic black injectors (35.7%), and non-Hispanic whites (36.2%). Only the small group in the "other" category, consisting of Native and Asian Americans, was not homophilous. When affiliation by gender was examined a different pattern was found (see Table IB). The estimated 41% of female injectors formed within-gender ties 42% of the time, so homophily was near zero (1.8%). In contrast, the estimated 59% of male injectors formed within-gender ties 73.6% of the time, so homophily was substantial (35.5%). This difference may reflect the relative status of males in injector populations. As thus conceived, positive and negative homophily can coexist in the same system. This can occur when a division of labor emergences in a group of otherwise similar individuals. For example, jazz musicians perform in groups that bring together performers with similar musical preferences, such as a preference for classical jazz, contemporary jazz, or fusion. But groups also include performers specializing in diverse types of instruments, so homophily coexists with heterophily (Heckathom & Jeffri, 2001). The implication is that the measure for homophily and heterophily are not rivals, because in complex systems patterns of differential association generally include affiliation based on both similarity and on difference. Therefore, the appropriate indicator for solidarity is the absolute value of the homophily measure, because positive and negative values reflect solidarity based on similarity vs. solidarity based on interdependence. This approach to measuring homophily in terms of network structure resembles Markovsky and Lawler's (1994) proposal that solidarity be conceptualized in terms of the network property of reachability, which is defined as the number of links required to go from any group member to any other group member. Reachability increases, as does homophily, when intra-group ties are substituted for out-group ties, so the two measures are positively related. An advantage of the latter is that it facilitates comparisons among groups of different sizes because homophily depends only on the proportion of in-group vs. out-group ties. Homophily provides only an indirect measure when solidarity is conceived in terms of resource allocation. In this respect, it resembles other measures of solidarity. For example, in the legislature analyzed by Hechter (1987, p. 79), solidarity was measured by counting votes consistent with party dictates. This provides an indirect measure of resource allocation, because from the standpoint of party leaders, not all issues are of equal value: votes on bills viewed as crucial by party leaders and votes that provide the party with the margin needed for victory, for example, have high value. When that margin has been fulfilled, party members whose votes are not needed may be released

46

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

Table 1.

Recruitment by Race/Ethnicity and Gender.

Table 1A: Race and Ethnicity (RecruitmentCount; Selection Proportion, S)

Race and Ethnicity of Person who Recruited

Race and Ethnicity of Recruit White

Non-HispanicWhite Hispanic

Non-Hispanic Black Other Total Distributionof Recruits, Sample Distribution,SD Equilibrium,E Mean Network Size, N Homophily, H Population Estimate, P

Hispanic Black

102.000 0.810 18.000 0.429 7.000 0.500 3.000 0.375

10.000 0.079 19.000 0.452 2.000 0.143 5.000 0.625

8.000 0.063 4.000 0.095 5.000 0.357 0.000 0.000

130.000 0.684 0.700 55.200 0.372 0.702

36.000 0.189 0.169 38.400 0.317 0.198

Other 6.000 0.048 1.000 0.024 0.000 0.000 0.000 0.000

17.000 7.000 0.089 0.037 0.094 0.037 63.300 76.700 0.301 - 1.000 0.080 0.020

Total 126.000 1.000 42.000 1.000

14.000 1.000

8.000 1.000

190.000 1.000

Table 1B: Gender (RecruitmentCount; Selection Proportion, S)

Gender of Recruiter

Gender of Recruit Female Male Total

Female

21.000

Male

0.420 37.000 0.264

29.000 0.580 103.000 0.736

50.000

Total Distributionof Recruits Sample Distribution,SD Equilibrium,E Mean Network Size, N Homophily, H Population Estimate, P

58.000 0.305 0.313 37.500 0.018 0.410

132.000 190.000 0.695 1.000 0.687 57.100 0.355 0.590

1.000

140.000 1.000

from party obligations. When supporting the party line would entail a political cost, legislators who are facing strong electoral opposition may be asked to support the party position only when that is essential for the party to prevail. For this reason, controversial bills, such as tax increases, generally pass by only small margins. Because of these complexities in the voting process, merely counting the proportion of a legislator's votes that are consistent with party

Group Solidarity as the Product of Collective Action

47

dictates provides an imperfect measure of the extent to which the resources possessed by the legislator fulfill corporate obligations. However, because the right to vote is a legislator's principal resource, counting votes remains the most feasible measure of resource allocation. Similarly, the ability to create and dissolve network ties in less structured groups is not the only resource available to individuals; however, these choices about ties determine an individual's dependence upon the group and the enforcement resources available to the group. Thus, these choices serve as a reasonable measure of solidarity, since each time an individual creates or dissolves a network tie, he or she thereby votes to increase or decrease the group's solidarity. Another advantage of conceptualizing solidarity in terms of homophily is that analysis can be extended beyond formally constituted groups to any category by which individuals can be differentiated. It thereby provides a way of determining what attributes are socially significant. An example of a socially irrelevant differentiation is whether one is born in an odd or even month. No group of which we are aware considers this socially significant. Given that this factor has no effect on tie formation, social systems lack structure with respect to this factor. That is, homophily is zero. In contrast, a host of other factors affect tie formation, such as the demographic factors upon which much social science analysis depends, including age, gender, race, ethnicity, and social status. A consideration in using homophily as a measure of solidarity is that homophily can result from multiple processes. For example, McPherson and Smith-Lovin (1987) distinguish between choice homophily and induced homophily; the former arises when choices among alternative ties are based on similarity, and latter results when opportunities to form out-group ties are limited, as when those living in ethnically homogeneous neighborhoods have no choice but to form intra-ethnic ties. Similarly, Yamaguchi and Kandel (1993) distinguish among homophily based on selection (i.e. choice homophily), homophily based on socialization in which affiliated persons influence one another in ways that make them become more similar, and homophily resulting from common responses to shared social influences. As these examples illustrate, homophily can result from multiple processes. Irrespective of its origin, greater homophily reflects greater investment of ties among persons of the same status, so dependence and enforcement resources are increased. Consider again the analogy with solidarity as measured by counting votes. Many processes can affect voting decisions. A principled legislator may vote his or her conscience, and conformity with party dictates depends on the extent to which the two converge. Alternatively, a pragmatic legislator may respond to compliance incentives from the party when deciding upon a vote. Campaign

48

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

promises and past voting practices also can affect voting decisions. Therefore, a vote consistent with party obligations can result from many considerations, some involving the control capabilities of the party, and others for which party power is irrelevant. Nonetheless, counting votes remains a plausible measure of solidarity. But irrespective of the reasons why intra-group ties are formed, they nonetheless can validly be seen as contributing to group solidarity, just as votes consistent with party obligations strengthen the party's position, irrespective of the complex and frequently conflicting motivations that govern the votes of party members. The solidarity literature has focused on groups, which, in Hechter's definition as in most standard definitions, include clear boundaries defined by criteria for membership. Yet, solidary processes can occur in collectivities that are more diffuse. For example, in a study of attitudes and affiliation patterns among university undergraduates in three large Midwestern public universities, preferred forms of recreation structured interactions. Three forms were prominent, one oriented toward heavy alcohol use, as defined by an orientation toward "getting drunk at parties"; a second oriented toward soft drugs, particularly marijuana; and a third oriented toward non-drug related forms of recreation including studying together (Heckathorn & Lucas, 1982). The boundaries among these sets of students were too diffuse to include membership criteria, yet each group was in many respects distinct. Thus, solidary processes can occur among sets of individuals even though they do not reach the level of organization required to count as a group in the standard restrictive sense. Racial and ethnic groups provide another example. Although these categories are frequently important determinants of affiliation patterns, they generally do not constitute groups in the restrictive sense. Similarly, the greater the extent to which intra-ethnic affiliations predominate over cross-ethnic affiliations, the greater is the indication that some form of solidary process is reflected in these affiliation patterns, such as the presence of a cohesive ethnic subculture.

Measuring Homophily Using Respondent-Driven Sampling As defined by Eq. 1, measuring homophily requires information on affiliation patterns and information on group size. Both types of information are readily available for a small number of variables, such as census data on intermarriage by race and ethnicity (e.g. see Blau, 1977). Gathering such data in other settings, such as McPherson and Smith-Lovin's (1987) study of voluntary associations, requires substantial resources. Therefore, studies of homophily have been characterized by McPherson (1998) as "a theory starved for data." This changed with the advent of a new means for gathering information on

Group Solidarity as the Product of Collective Action

49

affiliation patterns and group size, termed respondent-driven sampling (RDS) (Heckathorn, 1997, 2002). RDS, a form of chain-referral sampling, was developed to improve methods for sampling hard-to-reach populations, such as injection drug users, by overcoming the biases traditionally associated with chain-referral methods, including volunteerism, masking, and oversampling of groups with large networks (see Heckathorn, 1997). RDS has also served as a mechanism for recruiting active injection drug users for HIV prevention education and services (Broadhead et al., 1998; Heckathoru et al., 1999). When it plays that role, RDS is termed participant-driven recruitment (PDR). This article suggests that RDS can also serve as a means for studying affiliation patterns. A comprehensive discussion of RDS would exceed the scope of this paper (see Heckathorn, 2002); however, essential features of the method can be summarized. RDS employs recruitment incentives to produce a highly robust system of chain referrals, such that over the course of more than a dozen recruitment waves, a single initial subject can yield hundreds of additional subjects (see Fig. 1). Long referral chains ensure that the sample will have considerable sociodemographic depth and that initial subjects will constitute only a minor portion of the total sample. A counterintuitive feature of this sampling procedure is the demonstration that as the sample expands from wave to wave, the effects of the choice of initial subjects become progressively weaker and the sample composition ultimately attains an equilibrium that is independent of the starting point. The manner in which this occurs is illustrated in Fig. 2, which depicts the results of two simulations showing how the composition of each wave would have changed had recruitment begun from either one or more Hispanic intravenous drug users (Fig. 2A) or one or more non-Hispanic white intravenous drug users (Fig. 2B), based on projections from Table IA's recruitment patterns. The vertical axes represent the percentage of each type, and the horizontal axes represent the number of recruitment waves, where wave 0 refers to the seed or seeds, which in this exercise were assumed to be ethnically homogeneous. Wave 1 refers to the seeds' recruits, wave 2 refers to the recruits' recruits, and so forth. If recruitment had begun with only Hispanic seeds, the percentage of Hispanics in each wave would have decreased from the initial value of 100%, to 45% in the first wave and 27% in the second wave, and eventually stabilized at 17% after five waves. This equilibrium does not change with later waves. The implication is that when referral chains are long enough for equilibrium to be attained, RDS is a reliable sampling method. In RDS, information on affiliation patterns is derived from a recruitment process in which respondents recruit those with whom they have a pre-existing

50

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

•~

~e~ ..~e~ II

II

II

~ m ~

\

•

~

~°~ ~ II

II

II

~-~ ~ II

~

II

II

o~-

/ ~D cD

E

~

O

.=.

° B

q'J

o--..~,,//~

~'-

~_~

0 eD

Z

~

2

~/~

e4/~ ,.,4 N

E

Group Solidarity as the Product of Collective Action

51

A: Startlng Point = All Hispanic Seeds

0.9 0.8 tO

0.7

-I

~-

0.6

---,---B--'-"

0. tO

~ O.4 0

~ o.3

a.

0.2 0.1

0

2

1

3

4

5

6

7

Recruitment Wave '--II,-Non-Hispanic White J H i - s ~ n i c =~Non-HispanicRiack " ~ O t l l e r

B: Starling Point = All While Seeds 1 0.9 0.8 0.7 0.6

-

,,

-

...

_

.t

.L

.IL,

5

6

7

0.5

0.4. 0.3. .

.

.

.

0.2"

.

.

.

. J. .

.

.

.

. .L

.

.

,,,.

0.1 0 1

Fig. 2.

2

3

4

Race and Ethnicity of Recruits in a Respondent-Driven Sample, Beginning with Only White or Hispanic Sets.

......

52

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

relationship, and the link between recruiter and recruit is established using recruitment coupons, each of which has a unique serial number. This provides a behavioral measure of affiliation patterns, a measure that correlates strongly with self-reported network composition (see Heckathorn et al., 2002). Using a behavioral measure of network affiliation has an advantage, for when self-reports are used as the measure of affiliation, analysis is limited to matters of public knowledge, such as race, ethnicity, and gender. In contrast, when the measure of affiliation is behavioral, patterns can be identified, even those of which the respondents are unaware. For example, even though injectors generally do not share information about their HIV status, affiliation by HIV status was studied using the RDS method. In this study, HIV status was determined through an optional HIV test offered to respondents following their interviews. Based on the test results, affiliations were determined behaviorally, by who recruited whom (Heckathorn et al., 1999). The results showed that homophily by HIV status was near zero, indicating that the HIV infected individuals are thoroughly dissolved into the larger population of injection drug users.

llI: SOLIDARITY AS THE PRODUCT OF COLLECTIVE ACTION Insightful analyses of alternative types of homophily have been proposed (McPherson & Smith-Lovin, 1987; Yamaguchi & Kandel, 1993), yet they have not led to an explanatory theory of homophily. A step toward a theory of homophily derives from the recognition that homophily is a form of collective action. The formation of a reciprocal relationship requires mutual consent, so it requires action by at least two individuals. For example, choice and selection homophily are based on the decision of with whom to build ties, whereas socialization and common response homophily depend on the choice to maintain ties. Thus, homophily is ultimately based on choices to create or dissolve ties. An apparent exception is induced homophily, in which ties reflect opportunities for contact; however, recall that homophily refers not to a preponderance of intra-group ties but rather to a greater proportion of such ties than would be produced by random affiliation. Therefore, large groups will possess many intra-group affiliations even in the absence of homophily (i.e. a number in exact proportion to the group's size). The situation with induced homophily is more complex in systems with multiple groups, as in McPherson and Smith-Lovin's (1987) study of voluntary association. They found that similar people tended to affiliate, not because they sought one another out in associations with diverse membership, but because voluntary associations tended to be homogeneous.

Group Solidarity as the Product of Collective Action

53

From this they concluded that induced homophily was more prevalent than choice homophily. This distinction is clear so long as one focuses on affiliations within organizations; however, the decision to join a voluntary association is also a choice. In such cases, the line between choice and induced homophily depends on the level of decision. For instance, if an individual chooses to join a homogeneous voluntary association, such as a racially exclusionary country club rather than an association that is more diverse, choice homophily at the organizational level leads to induced homophily at the level of individual ties. This suggests that choice homophily has priority over induced homophily, because induced homophily is a consequence of earlier choice homophily. At whichever level it occurs, homophily involves the pulling together of similar people through the joint work of individuals to build and maintain relationships. In this way, homophily is a form of collective action. Heterophily involves pulling together people who are complementary, such as persons from different clans or of different genders, so it too is a form of collective action. The implication is that theories of collective action may contribute to an understanding of homophily. A number of formal theories of collective action have been proposed (Heckathorn, 1988, 1989, 1990, 1993, 1996; Marwell & Oliver, 1993; Macy, 1990), any of which could provide a basis for a theory of homophily. One approach would be to detail the implications of each for an explanation of homophily, compare those conclusions with what is known about homophily, and make a principled decision regarding which theory or combination of theories could best explain homophily. That would exceed the scope of this paper. An important insight from recent studies of collective action is that all collective action produces some form of collective good (i.e. some joint benefit), and that the shape of the production function importantly affects the emergence, stability, and dynamics of collective action (Marwell & Oliver, 1993; Macy, 1990; Heckathorn, 1996). Production functions for collective goods, like production functions for many other goods, often reflect startup costs when production begins, and diminishing marginal returns above an upper level of production. When viewed graphically, the result is an S-shaped curve linking contribution and production levels (Fig. 3). This is the standard function assumed in much economic analysis, and it fits other cases as well, such as social movements (Heckathorn, 1996). Thus, when production begins, the production function is accelerating, it then becomes approximately linear, and it eventually decelerates. Given this production function, the marginal gains from each additional unit of contribution (e.g. from the addition of each contributor) are initially low, but eventually they increase and then decline, thereby defining a

54

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

o

\

w o

\

o

\ \

_m O.

\ \

C

\

0

\

\

/

o i

0

0 o

o

o

~

o

o

o

o

peonpoJd poo9 o!lqnd to uo!NodOJd

o

.~

Group Solidarity as the Product of Collective Action

55

bell-shaped curve (Fig. 4). The curve defines the net gains attainable from participating in a collective endeavor, including collective action. To see how this is related to homophily, consider the hypothetical case of someone who wants to participate in a joint endeavor, such as playing volleyball, protesting government policy, or participating in a drug-oriented subculture. If the number of suitably oriented persons is too small, the endeavor is not feasible, so homophily based on the endeavor is low and collective action fails. If the number of suitably oriented persons is larger, then by banding together they make the endeavor possible, and homophily is substantial. Finally, if the number of suitably oriented persons is so large that no special effort needs to be made to reach them, homophily is again low. Thus, a curvilinear relationship can theoretically be expected between the abundance of persons suitable for the collective endeavor and homophily. One of us observed the emergence of a solidary group that illustrated some of the above-described processes. Two first-year graduate students learned that they shared a love of opera. They searched the graduate student community in other departments and located several other opera buffs, and then established a routine. They car-pooled to New York City, attended performances, and returned very late the same night. By sharing transportation, they reduced the burden of driving and the cost of gas; by returning the same night they avoided hotel bills; and by attending jointly, they shared the performances with others who appreciated them. When Wagner's Ring cycle was presented, the group showed remarkable stamina by commuting to New York each day of the series and returning early the next morning. Had the number of opera fans been smaller, such a group could never have formed, and had it been far larger, devotees would not have had to seek out one another, so the impact on affiliation patterns would have been minimal. Support for the hypothesis that homophily depends on group size is provided by an analysis of homophily by age. Figure 5 shows the relationship between homophily and the proportion of injectors aged 18 to 25 in RDS samples from New London, Middletown, and Meriden, Connecticut. When the percentage of younger injectors is small (9.6%), so too is homophily (7.2%). When the percentage of younger injectors is greater (18.7%), homophily becomes very substantial (56%). Finally, when the percentage of younger injectors is large (48.5%), homophily falls dramatically, to 13.6%. This pattern becomes comprehensible if associations among younger injectors are seen as a form of collective good. If there are not enough younger injectors, forming a youth subculture is impossible, so younger injectors are dissolved into the general injector population, and homophily is low. When younger injectors are more common, a critical mass exists for creation of a youth

56

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN 0 o T.-

0

m. 0

/o

/

0

O O

/

t~

Y

c)

i

\

O

I

-I-

O

\

o

J\ O O

O

o. v~. o

o

0

0

u!e~ 7,eN

0

o

O

o

.~

Group Solidarity as the Product of Collective Action

57

o

o

9

/

~A

it) 0

/

•

=

o

/

ei C

~

>,-

*

0

c

°

~

N 0

,

°

u~

,

I

I I

0 0 0

~

0

o

o

o

sJoz:)e.l'ulJe6unoA to/~l!qdotUOH

58

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

subculture, so homophily is substantial. Finally, when younger injectors are abundant, establishing a youth subculture requires no special selectivity in forming ties. In this instance, younger injectors' taste for associating with other younger injectors is amply satisfied without the need to seek out other younger injectors, resulting in low homophily. In addition to illustrating how changes in homophily are theoretically linked to collective action, this analysis has substantive implications. Life course studies have routinely emphasized the changes in social positions and relations as individuals age. Understanding this process is complicated by the recognition that these changes are context-dependent; even within a given society, growing older has quite different implications depending on socioeconomic status and other factors. The analysis of homophily among younger injectors illustrates the way in which RDS can be used to study context dependence. The finding reported above shows that whether a younger intravenous drug user becomes part of a homophilous subgroup of younger injectors depends on the proportion of younger injectors in the system. This has theoretic implications for understanding the contexts under which age-specific peer norms will emerge, and practical implications for interventions targeting this group for HIV-prevention services, as illustrated by the relationship between group size and homophily. The bell curve also characterizes syringe sharing. When the proportion of those who do not share syringes is moderate (50% in Meriden) or large (75% in Middletown), their homophily is near zero (0% and 6%, respectively), but when the proportion is intermediate (64% in New London), homophily is substantial (50%). This indicates both that HIV-risk behavior can be salient enough to affect affiliation patterns, and that whether this occurs is context dependent. Though the production function for interactions by age and syringe sharing may be consistent with the standard S shape, not all production functions can be theoretically expected to take this form, for two reasons. First, because of technological or other constraints, the curve may be truncated, so only a segment of the curve is empirically realized. Collective sanction systems (Heckathorn, 1988) provide an example. These are systems in which responsibility is joint, in the sense that everyone is responsible for one another's behavior. The group may be promised a collective reward or threatened with a collective punishment based on each individual's compliance. Such systems are characterized by an accelerating production function (Heckathorn, 1996), so they reflect only the left part of the S curve. In contrast, in the brokerage systems analyzed by Burt's studies of structural holes, exclusive access to information confers benefits. The value of information degrades as it becomes more broadly disseminated, so the production function is decelerating. Such systems therefore reflect only the right part of the S curve.

Group Solidarity as the Product of Collective Action

59

More complex production functions are identified by contemporary theories of collective action. For example, Heckathorn (1996) analyzed the essential features of both collective action and social cooperation as described in a wide range of literatures and found that: (1) social cooperation must yield some form of benefit - a collective good, whose valuation can be low or high or even negative; (2) contribution to production of the collective good entails a cost that varies across individuals and systems; and (3) contributions to production of the collective good are linked to the ultimate amount of production by a production function. The aim of the analysis was to identify the range of social dilemmas that can arise given these three conditions, where a social dilemma is defined as a situation in which individual actions can aggregate into a collectively irrational outcome. By this definition, collective action is concemed with resolving social dilemmas. Based on an analysis of these three essential features of collective action systems, a theorem was derived showing that collective action systems can give rise to exactly five possible forms of social dilemmas (Heckathorn, 1996). Distinct issues are problematic in each of the five dilemmas. Although much of the formal-theory literature has focused on the prisoner's dilemma, in which trust is problematic, other dilemmas were identified. These include coordination problems, which arise when production functions are accelerating and hence interdependence is high; bargaining problems, which arise when production functions are decelerating and hence interdependence is low; invisible-hand systems when the gains from cooperation are high; and the altruist's dilemma in which autonomous action yields greater benefits than does collective action. Given that each of these five dilemmas is characterized by a distinct production function, and that RDS can help to identify the shape of production functions empirically, RDS potentially provides a means for identifying the type of dilemma underlying collective action systems. Power Homophily and Social Inequality An important form of homophily results from a tendency for ties to form among status equals. One means by which this can occur is illustrated by Axelrod's (1984) account of the emergence of cooperation in very large groups. He argued that a prerequisite for the emergence of cooperation under these circumstances is that those who are disposed toward cooperation are able to identify one another. By selectively interacting, cooperators form homophilous clusters. The non-cooperators are thereby left only with the opportunity to interact with one another. The emergence of cooperation thereby entails the emergence of a

60

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

stratification system. This is consistent with the deviance literature, in which the principal social sanction involves allocation of status. Norm adherence increases status; deviance decreases it. Norm adherents therefore form homophilous clusters. This may be termed power homophily because homophily reflects this group's superior social status. The non-elite group is also homophilous because of members' inability to form ties with the elite. This may be termed exclusion homophily because homophily results from exclusion from groups of higher status. This differentiation between power and exclusion homophily is illustrated by studies of the network structure of U.S. schools (Farmer and Farmer 1996). When each student was asked to list the other students he or she most liked, the most popular students were often named by both the most and the least popular. However, this does not mean that the less popular students are isolates, for they associate with one another. Thus homophily may reflect not only social differentiation (i.e. separate but equal) but also social inequality (i.e. separate but unequal). A process of this type was found in the study of Connecticut drug injectors. Heavy injectors, defined as injectors whose injection frequency exceeds 20 during the last 30 days, appeared to have differing status depending on their numbers (see Fig. 6). The relationship has a sharp negative slope. In New London the percentage of heavy injectors was small (22%), and their homophily was substantial (42%). In Middletown the percentage of heavy injectors was greater (31%), and their homophily was lower (36%). In Meriden the percentage of heavy injectors was still larger (47%), and homophily fell to near zero (-1%). To make sense of this finding, it is useful to consider the resources relevant to an injector population. One resource is a reliable source of drugs. This requires ties to multiple sources, because any single source could run dry or be shut down by police, thereby producing an interruption in drug access. A second major concern is police activity. Knowledge of recent arrests is valuable, as those who are arrested are frequently pressured to inform on their associates. Similarly valuable is knowledge of where enforcement efforts are concentrated, because enforcement intensity varies by area. These resources are relevant to understanding the relationship between homophily and injection frequency because heavy injectors buy more product and hence tend to have better contacts with suppliers and purchase their drugs at lower prices. Through these contacts, they are also likely to have better information about police. When heavy injectors are modest in number, the information they possess is scarce and therefore valuable. To obtain further knowledge, heavy injectors may tend to seek out one another, thereby exhibiting power homophily. They occupy a privileged status. However, as in Burt's (1998) brokerage systems, contacts with additional heavy injectors have diminishing marginal value. At some

Group Solidarity as the Product of Collective Action

I

i

/

/

/

/

Y

61 t/) o

u~

O

e-, o

"E' I

i

O tO

"

~2 8. el

T.-

© ©

~Z

I::= O

O

~6

O

O O

i

sJo),oe|ul A^eeH jo ~l!qdowoH

O

62

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

point, additional information about dealers and police becomes redundant, so when heavy injectors are abundant, an individual's knowledge is not in high demand and the homophily of heavy injectors approaches zero. They are neither differentially sought out by themselves (power homophily) nor do they thereby induce homophily in others (exclusion homophily). The system thereby ceases to be stratified. CONCLUSION Solidary groups are united by a shared attribute that shapes affiliation patterns. The proposed means for conceptualizing and measuring homophily provides an approach for studying these patterns. Socially irrelevant factors, such as whether one's surname has an odd or an even number of letters, do not affect affiliation, so homophily is zero. By contrast, socially salient factors, such as race and ethnicity, affect affiliation; hence, they have non-zero homophily, as do factors associated with them, such as in parts of the United States where it matters whether one's surname ends in a vowel or a consonant. The essential hypothesis is that solidary groups emerge through collective action to produce a collective good or goods. In the case of injection drug-using communities, these collective goods involve information about enforcement activities, such as where police surveillance is most intense; and information about sources of drugs, including assessments of the quality of drugs available from alternative dealers, and the pricing and reliability of those dealers. In addition, the group provides other goods, such as companionship, a sense of common purpose derived from collective participation in a non-conventional lifestyle, and access to potential sexual partners. A high level of cooperation therefore characterizes groups of injection drug users. Thus, lonely people may join not just for the drugs but to become a part of a supportive community. This also illustrates one of the reasons it is important to foster alternative focal activities around which solidary groups can congeal. The creation of solidary groups can be facilitated through bringing together people who share common interests or problems. This possibility is illustrated by interventions implemented to combat AIDS that were developed and implemented by Broadhead and Heckathorn (1994). The interventions are based on what are termed secondary sanctions - rewards given not for the individual's behavior but for the positive behaviors the individual elicits from peers. These incentives provide a means for harnessing peer pressure. They also provide a means for initiating the development of prosocial norms, by strengthening the incentives of individuals to regulate one another's behavior. The original intervention of this type (Broadhead et al., 1998) targeted active injection drug

Group Solidarity as the Product of Collective Action

63

users for HIV prevention education and services. The aim was to create and strengthen prevention norms and thereby reduce HIV-associated risk behavior. The design was highly participatory, in that participants in the intervention carried out two of the most important tasks, education of their peers and recruitment using a form of RDS. The intervention met with sufficient success so as to be described in a National Academy of Sciences report as the "state of the art of preventive intervention" (Institute of Medicine, 1995). To emphasize the extent to which the intervention design is based on participants providing services for their peers, the intervention is termed a peer-driven, or equivalently, a participant-driven intervention (PDI). An extension of the PDI design was intended to increase adherence to antiretroviral therapy among HIV-infected active injection drug users (Broadhead et al., in press). Through a combination of standard medical screening and participant-driven recruitment, a set of subjects were identified. These subjects were then provided the opportunity to participate in the intervention, in which each would play the roles of Health Advocate and Peer. As Health Advocate, a subject regularly monitored the adherence of his or her Peer. Very modest financial rewards were provided based on the ability to elicit positive responses from the Peer. The aim was to harness peer pressure on behalf of the intervention's public health objective (increasing adherence to therapy), as well as reducing HIV risk behavior. What is notable in this context is that the subjects bonded to the intervention, welcoming the opportunity to serve as a vital part of the intervention and strengthening their ties to other intervention participants. In essence, the intervention created a structure in which prosocial behavior flourished, a structure that promoted the development of a solidary system. Its aims were to harness peer pressure, create pro-health norms, and increase adherence to the exacting demands of primary care for HIV disease. Given that the altematives to those seeking solidary associations in some communities include joining the injection drug-using community or a gang, the development of a means for fostering safer and healthier forms of solidary association is a worthwhile objective.

NOTE 1. Extensiveness and compliance are therefore not the causes of solidarity, rather they are the terms by which solidarity is defined and measured. As Markovsky and Lawler (1994) have noted, this distinction is not consistently maintained by Hechter (1987), who at various points discusses extensiveness and compliance as both the causes and definitions of solidarity, but an examination of the internal logic of the model makes clear that these serve as definitional terms.

64

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

ACKNOWLEDGMENTS A n earlier v e r s i o n o f this p a p e r w a s p r e s e n t e d at the c o n f e r e n c e S o l i d a r i t y a n d Prosocial Dispositions held in Schiermonnikoog, The Netherlands, March 2 7 - 2 9 , 2001. T h i s r e s e a r c h w a s m a d e p o s s i b l e b y g r a n t s f r o m the C e n t e r s for Disease Control and Prevention (U62/CCUl14816-01) a n d the N a t i o n a l Institute on Drug Abuse (RO1 DA08014).

REFERENCES Axelrod, R. (1984). The Evolution of Cooperation. New York: Basic Books. Blau, P. M. (1977). Inequality and Heterogeneity. New York: Free Press. Blau, P. M. (1994). Structural Context of Opportunities. Chicago: University of Chicago Press. Broadhead, R. S., & Heckathorn, D. D. (1994). AIDS Prevention Outreach among Injection Drug Users: Agency Problems and New Approaches. Social Problems, 41,473-495. Broadhead, R. S., Heckathorn, D. D., Weakliem, D., Anthony, D., Madray, H., Mills, R., & Hughes, J. (1998). Harnessing Peer Networks as an Instrument for AIDS Prevention: Results from a Peer-Driven Intervention. Public Health Reports, 113 (Supplement 1), 42-57. Broadhead, R. S., Heckathorn, D. D., Altice, F. L., van Hulst, Y., Carbone, M., Friedland, G., O'Connor, P. G., & Selwyn, P. A. (in press). Increasing Drug Users' Adherence to HIV Therapeutics: Feasibility Results of a Peer Driven Intervention Study. Social Science and Medicine. Burt, R. (1998). The Gender of Social Capital. Rationality and Society, 10, 5-46. Deci, E. L., & Ryan, R. M. (1985). Intrinsic Motivation and Self-Determination in Human Behavior. New York: Plenum. Durkheim, E. (1956). The Division of Labor in Society. New York: Free Press. Fararo, T. J., & Doreian, P. (1998). The Theory of Solidarity: An Agenda of Problems. In: P. Doreian & T. Fararo (Eds), The Problem of Solidarity: Theories and Models (pp. 1-33). Amsterdam: Gordon and Breach Publishers. Fararo, T. J., & Sunshine, M. H. (1964). A Study of a Biased Friendship Net. Syracuse NY: Syracuse University Press. Farmer, T. W., & Farmer, E. M. Z. (1996). Social Relationships of Students with Exceptionalities in Mainstream Classrooms: Social Networks and Homophily. Exceptional Children (March/April). Hechter, M. (1987). Principles of Group Solidarity. Berkeley, CA: University of California Press. Heckathorn, D. D. (1988). Collective Sanctions and the Emergence of Prisoner's Dilemma Norms. American Journal of Sociology, 94, 535-562. Heckathoru, D. D. (1989). Collective Action and the Second-Order Free-Rider Problem. Rationality and Society, 1, 78-100. Heckathorn, D. D. (1990). Collective Sanctions and Compliance Norms: A Formal Theory of GroupMediated Social Control. American Sociological Review, 55 (June), 366-384. Heckathoru, D. D. (1993). Collective Action and Group Heterogeneity: Voluntary Provision vs. Selective Incentives. American Sociological Review, 58, 329-350. Heckathorn, D. D. (1996). Dynamics and Dilemmas of Collective Action. American Sociological Review, 61, 250-277.

Group Solidarity as the Product o f Collective Action

65

Heckathorn, D. D. (1997). Respondent Driven Sampling: A New Approach to the Study of Hidden Populations. Social Problems, 44, 174-199. Heckathorn, D. D. (2002). Respondent-Driven Sampling II: Deriving Statistically Valid Population Estimates from Samples of Hidden Populations. Social Problems, 49, 11-34. Heckathorn, D. D., & Jeffri, J. (2001). Finding the Beat: Using Respondent-Driven Sampling to Study Jazz Musicians. Poetics, 28, 307-329. Heckathorn, D. D., Broadhead, R. S., Anthony, D. L., & Wealdiem, D. L. (1999). AIDS and Social Networks: Prevention through Network Mobilization. Sociological Focus, 32, 159-179. Heckathorn, D. D., & Lucas, W, L. (1982). Bridging Consensus and Conflict Theories in Drug Use Research: A 'Unified' Theoretic Perspective. Journal of Drug Issues, 12, 443-455. Heckathorn, D. D., Semaan, S., Broadhead, R. S., & Hughes, J. J. (2002). Extensions of RespondentDriven Sampling: A New Approach to the Study of Injection Drug Users Aged 18-25. AIDS and Behavior, 6, 55-67. Homans, G. C. (1950). The Human Group. New York: Harcourt, Brace and World. Institute of Medicine (1995). Assessing the Social and Behavioral Science Base for HIV/AIDS Prevention and Intervention. Washington, D.C.: National Academy Press. Lawler, E. J., & Yoon, J. (1998). Network Structure and Emotion in Exchange Relations. American Sociological Review, 63, 871-894 (need page numbers). Macy, M. W. (1990). Learning Theory and the Logic of Critical Mass. American Sociological Review, 55, 809-826. Markovsky, B., & Lawler, E. J. (1994) A New Theory of Group Solidarity. Advances in Group Processes, 11, 113-137. Marwell, G. & Oliver, P. (1993). The Critical Mass in Collective Action: A Micro-Social Theol. Cambridge U.K.: Cambridge University Press. McPherson, J. M. (1998). Personal communication. San Francisco. McPherson, J. M., & Smith-Lovin, L. (1987). Homophily in Voluntary Organizations: Status Distance and the Composition of Face-to-Face Groups. American Sociological Review, 52, 370-379. Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. New York: Cambridge University Press. Rapoport, A. (1979). A Probabilistic Approach to Networks. Social Networks, 2, 1-18. Simmel, G. (1955). Conflict: The Web of Group Affiliations. K. H. Wolff & R. Bendix (Trans.). New York: Free Press. Widegren, O. (1997). Social Solidarity and Social Exchange. Journal of the British Sociological Association, 31,755-773. Yamaguchi, K., & Kandel, D. (1993). Marital Homophily on Illicit Drug Use among Young Adults: Assortative Mating or Marital Influence? Social Forces, 72, 505-528.

APPENDIX: COMPUTING HOMOPHILY This a p p e n d i x describes the procedures for c o m p u t i n g h o m o p h i l y based on R D S data. A c h a l l e n g e is that whereas information on affiliation patterns is available through r e c o r d i n g patterns o f recruitment, c o m p u t i n g h o m o p h i l y also requires i n f o r m a t i o n on group size. This is p r o b l e m a t i c w h e n studying populations such as injection drug users (Heckathorn et al., in press) and j a z z musicians ( H e c k a t h o r n & Jeffri, 2001), for w h i c h no sampling frame exists. F o r e x a m p l e , there are no m e m b e r s h i p records for these groups. Therefore, m e t h o d s w e r e

66

DOUGLAS D. HECKATHORN AND JUDITH E. ROSENSTEIN

developed for estimating subgroup sizes based on available data, specifically affiliation patterns and network sizes. This appendix summarizes these procedures for the two-category case. For a more detailed presentation that includes analysis of the general n-category case, see Heckathorn (2002). Estimates of group size using RDS are based on ana analysis of the structure of networks of reciprocal ties. In such networks, relative group sizes can be computed based on two types of information, affiliation patterns and network sizes (Heckathorn, 2002). This method for estimating population size is suitable for analysis of RDS data because respondents recruit almost exclusively that is, more than 97% of the time - from those with whom they have preexisting ties, generally friends, acquaintances, or relatives. Such ties are reciprocal because a link from any individual x to y implies that a link also exists from y to x. Hence for two groups a and b, the number of links from a to b (Tab) will equal the number from b to a (Tba), SO Tab = Tba. Furthermore, the number of ties from any group x to y is the production of three terms, the size of the group (Px), the mean network size of group members (N×), and the proportion of ties from that go and from x to y (S×y), so, Txy = P N x Sxy. Hence, for two groups, a and b, Pa Na Sab = Pb Nb Sba" Without loss of generality group size can be expressed as a proportion, so 1-Pa can be substituted for Pb' and this expression can be solved for group a's size, Pa as follows: -

ea m

S,,oN + So N This is the estimate of group size based on the reciprocity model (Heckathorn, 2002), and it provides the means for estimating subgroup sizes based on both affiliation data, and on self-reported network size. For example, consider again the case of gender. From Fig. 1B, where females are group a and males are group b, the proportion of males recruited by females was Sab = 0.58, and the proportion of females recruited by males was Sba = 0.264. The mean network size for females was N a = 37.5, and N b = 57.1 for males. Substituting these values into the above expression yields the estimated proportion of females in the population as: 0.264 0.264

57.1

57.1 +0.58

= 0.41 37.5

Finally, homophily is computed based on the population estimate derived from the reciprocity model, and from the recruitment selection proportions in the manner described above.

BUILDING A MODEL FOR SOLIDARITY AND COHESION USING THREE THEORIES David Willer, Casey Borch and Robb Willer

ABSTRACT Drawing on Game Theory, Elementary Theory, and Status Characteristics Theory, this paper offers a theoretical model for a social group that is solidary and cohesive. The group has an economy, a social structure and a cultural structure. Applying Game Theory, economic conditions for solidarity are found. Within those conditions are groups that can be solidary if their social dilemma is resolved. Applying Elementary Theory shows how decentralized norm enforcement solves that social dilemma, but at the cost of second- and higher order free rider problems. Applying Status Characteristics Theory resolves higher order problems. Experiments and ethnographic examples support the analysis.

INTRODUCTION This paper builds a theoretical model for a 'group' that can be 'solidary' and

'cohesive.' A group is solidary if and only if members produce collective goods to their mutual advantage. Cohesion is the degree to which group members stick together; it is measured by the relative attraction of each member to the group together with that member's cost of exit. Some level of cohesion is a

Group Cohesion, Trust and Solidarity, Volume 19, pages 67-107. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2

67

68

DAVID WILLER, CASEY BORCH AND ROBB WlLLER

necessary condition for the continuance of any group. Therefore, cohesion is a necessary condition of solidarity. As will be seen later, a level of solidarity higher than alternative groups is decisive in retaining members; thus solidarity can be sufficient for cohesion. The modeled group to be built here is a theoretic construct. As such it will have no properties but those endowed by the theories we deploy. Nevertheless, with explanation and prediction in mind, we intend to build a modeled group (hereafter 'm-group') that corresponds in certain important regards to real groups that can be observed in the field or laboratory. For example, the m-group is constructed of three components: an economic structure that forms its base, a social structure of norm enforcement that regulates the economy resolving its social dilemma, and a cultural structure that resolves the second and higher order free rider problems of the social structure. The m-group is these three components, the relations among them and nothing more. To build these three components, we call upon Game Theory, Elementary Theory and Status Characteristics Theory respectively.

BACKGROUND Though Durkheim founded its study in sociology, in more than 400 pages he failed to formulate a definition for solidarity ([1893] 1960, passim). Durkheim's tradition continues today. Johnsen (1999) lists ten major references in sociology which deal extensively with solidarity and offer no definition for it. More recently, however, definitions have been offered and, following Markovsky (1999), we find that the best of them are embedded in explanatory theory. The definition for a solidary group that we offered above is a case in point. Its value can be no more than the value of the theoretic model following from it. Markovsky and Lawler (1994) identify two main approaches to the study of solidarity in sociology: utilitarian and emotional. "The former begins with the assumption that social order is created and maintained because (and only if) interdependence makes cooperation a valued commodity" and the latter "begins with the assumption that social order is created and maintained by the affecfive ties of individuals to groups" (1994, pp. 115-116). There is an extensive literature on emotional theories of solidarity (Parsons, 1951; Fireman & Gamson, 1979; Scheff, 1990; Lawler, 1992). The theoretical model that we develop is largely but not wholly of the utilitarian - or rational choice - kind. While not rejecting an emotional approach, we make no contribution to it. Hechter defines solidarity as a "joint product" of "two independent factors: first, the extensiveness of [the group's] corporate obligations, and, second, the degree to which individual members actually comply with these obligations"

Building A Model for Solidarity and Cohesion Using Three Theories

69

(1987, p. 18; see also p. 52). Corporate obligations are a function of a member's material dependence on the group. Members will comply with these obligations to the degree that the group can monitor and sanction each member's behavior, termed "control capacity." A group's control capacity is determined by: (1) the "measurability of the individual's contribution," (2) the group's ability "to limit the privacy of its members," and (3) the group's sanctioning capacity (1987, p. 53). To monitor both contribution to, and distribution of, collective goods, groups use 'agents,' group members who are "compensated for providing control and are motivated to do it on this account" (1987, p. 52). By extension, agent free riding is curtailed by compensatory sanctions. For Hechter, the differentiation of agents is prior to social control and implies centralized authority. By contrast, we see solidarity as more fundamental than centralization and seek modes of decentralized social control that are logically and historically prior to such differentiation. In Coleman's (1990) approach, low solidarity is evidenced by extensive free riding in producing collective goods.~ Free riding causes actors to face harmful consequences, termed 'negative externalities.' Because of these negative externalities, it becomes rational to increase solidarity by reducing free-riding, which leads to the creation of a normative system. Unlike Hechter, positive sanctions are used to reinforce desired behavior that Coleman suggests initiates a cycle of increasing conformity. As a result, group members can become 'zealots' in pursuit of collective goods. Whereas we agree that solidarity occurs when collective goods are produced, we hold that, to suppress free riders with payments, is to reward deviance over conformity. Given such rewards, rational choice theory clearly asserts that all will become deviants. Just as in market societies where contracts are enforced with penalties not payments, our theory of solidarity will find free riding suppressed with negative sanctions. Still, following Coleman and Hechter, our theoretic actors are rational egoists. In a series of papers, Markovsky and colleagues (Markovsky & Lawler, 1994; Markovsky & Chaffee, 1995; Markovsky, 1999) advance a structural theory of group solidarity. Their theory is perhaps the first to assert that 'cohesion' is a key part of solidarity. 2 Markovsky and Lawler define cohesion as reachability, which "refers to the strength and directness of relations among members of a set of actors" (1994, p. 121). Then reachability is combined with 'unity of structure,' which is "the absence of substructures, that is, subsets of actors with higher reachability among one another than with those in other subsets." They assert that a group has 'solidarity' when it has "a set of actors with high cohesion and unity of structure" (1994, p. 123). Then a decrease in solidarity is "attributable to the decreased average reachability (cohesion) between members" (1994, p. 126).

70

DAVID WILLER, CASEY BORCH AND ROBB WILLER

In an extension of the earlier papers, Markovsky offers "a way of thinking about solidarity that, compared to alternatives, is more about groups and their structures, and less about people and their feelings" (1999, p. 344). He goes on to introduce a method for determining a network's reachability and homogeneity. Lower values indicate greater heterogeneity and lower solidarity, while higher values indicate greater homogeneity and higher solidarity (1999, p. 368). The extension carries with it many of the concepts put forward in earlier work, but now solidarity is defined in terms of a 'referent network.' A referent network is "a pattern of actors and ties that identifies a class of networks," and the new definition of solidarity is "the reachability and homogeneity of a network relative to its referent network" (Markovsky, 1999, pp. 346-347). When the researcher offers a referent network, it may set a baseline "above which solidarity increases and below which solidarity is zero" (p. 348). The m-group constructed in this paper is itself a referent network with prefect reachability and a mean distance between actors at or near 1.0. Therefore, it has no subgroups and has uniform density throughout.

FROM SOLIDARITY TO COHESION AND BACK Since Durkheim, cohesion and solidarity have often been used interchangeably. 3 Here we treat them as distinct but related. As defined in the introduction, an m-group is solidary if and only if its members produce collective goods to their mutual advantage. And the membership of any m-group is the set of those benefited when collective goods are produced. We define solidarity by the production of collective goods, not individual goods, because only in their production and/or distribution is cooperation necessary. As we use the term, solidary m-groups actually produce collective goods and their members are advantaged by that production. The solidarity of m-groups varies; the more effective and valuable the production of collective goods, the higher the solidarity. The m-group will have an economic structure analyzed as a game and, for that game, two and only two conditions determine the value of collective goods produced. The effectiveness of collective good production is the proportion of members who cooperate. When group size is n and number of cooperators n , then effectiveness varies with ne/n. Said somewhat differently, effectiveness varies inversely with the proportion who, by choosing to defect, free ride. The value of collective goods produced is G i which is the ratio of i's payoff, were all m-group members to cooperate, to the payoff were all to act individually. Gi is large when the value of goods produced by each cooperator is large relative to that produced when defecting and acting individually. Then the strength of solidarity, 'S,' is

Building A Model for Solidarity and Cohesion Using Three Theories

71

S = G i n/n As already defined, cohesion is the extent to which group members stick together. The force of cohesion is q, the benefits lost and gained by the actor i when exiting the group together with the cost of exiting. The actor i is attracted by benefits of its group and by benefits of other groups that i could join. 4 Importantly, cohesion is a property of actors who are group members, not a property of the group itself. We define cohesion actor-by-actor because a group for which cohesion varies across actors is not like a group where cohesion is uniformly high, and both differ from a group where cohesion is uniformly low. Later the case of the Danish artisans traces some consequences when cohesion varies across members. 5 In our theory, solidarity and cohesion are linked in a feedback process. Since some level of cohesion is necessary for the continued existence of the m-group, and solidarity is a property of the m-group and not its members as individuals, cohesion is necessary for solidarity. Furthermore, q, the force of cohesion, is

q=G i n /n-G

n /n-c

where Gj is the G for i's best alternative, n / n the proportion of cooperators by group and c is the cost of exit. In words, for i, the force of cohesion, q is the collective good received by i in i's group (G i n/n), less the collective good i could receive in i's best alternative group ( G n/n), minus c, the cost of the exit itself. For m-groups, there is a straightforward relation from solidarity to cohesion. The force of cohesion is Gj nc/n, that is the strength of solidarity of i's group less G i nc/n, the strength of solidarity of i's best alternative, less the cost of exit. The term 'c' is extent of circumscription - the extent to which there are barriers impeding the exit of i. When there are barriers, c is negative and, as it increases in a negative direction, q, the force of cohesion, increases. 6 For example, when G i n / n < G n / n there is at least one other group to which members will move if the size of c does not impede that movement. If c then approaches zero, the group has zero or negative cohesion. As people exit, fewer collective goods are produced and solidarity goes to zero. When Gi nc/n > G n / n - c, however, the group is cohesive and, as seen below, if G i > 1, it may be solidary. More generally, as G i increases, the prospects for both solidarity and cohesion increase. Furthermore, proofs will be offered that, for very high values of G i , no one will free ride and, with production of collective goods optimal (i.e. n / n = 1), solidarity is high thus maximizing cohesion relative to alternative m-groups.

72

DAVID WILLER, CASEY BORCH AND ROBB WILLER

This investigation of solidarity and cohesion is divided into four further sections. In the first we introduce the game-theoretic economy of the m-group and examine two boundary conditions for cooperation. Below the lower bound, cooperation is impossible. Any m-group below the lower bound for which members have alternatives will dissolve if exit is not impeded. Above the upper bound, cooperation is assured. Any m-group above the upper bound will be perfectly solidary on economic grounds alone and perfectly cohesive relative to any group below the bound. These m-groups are utopias that need no social structure to govern their economy. The most theoretically interesting m-groups fall between the upper and lower bound and it is upon their qualities that our exposition focuses. For those groups, cooperation is possible but not assured by the economy alone. For these m-groups cooperation is produced by a social structure of norm enforcement. In turn, the m-group's cultural structure resolves the second and higher order free rider problems of its social structure. Focusing on m-groups where cooperation is possible but not but not assured, the second section finds norms embedded in the game-theoretic economy. Three example cases add meat to the bones of the theoretical analysis. The third section introduces the social structure of norm enforcement. The analysis shows that a decentralized norm enforcement system will produce cooperation in m-groups, but at the cost of generating second and higher order free rider problems. The third section carries forward the analysis of the three example cases introduced in the second section to show how norms are enforced. The second and higher order free riding problems of the m-group's social structure are resolved in the fourth section which offers a new approach linking social control to status. With status resolving higher order problems, a fully articulated theoretical model for a group for which solidary and cohesion can vary has been offered.

ECONOMIC CONDITIONS FOR SOLIDARITY This section investigates conditions for cooperation by developing formulations for a simple economy. The economy is modeled as game that m-group members play. The economy replicates elements of a familiar experimental design where investment in collective and individual goods are exclusive alternatives (Marwell & Ames, 1979, 1980). When any collective goods are produced, all m-group members and no outsiders receive an equal share and that reception is completely independent from whether the receiver contributed collectively or not. Goods produced individually are retained by the individual producer.

Building A Model for Solidarity and Cohesion Using Three Theories

73

For the model economy in a given time period t, m-group members each make a unit labor input selecting either 'C' to cooperate and produce collective goods or 'D' to defect and produce individually. For any member i, G is the ratio of (1) the payoff to i when all group members cooperate to (2) the unit payoff when all act individually. Said somewhat differently, G is the productivity when cooperative goods are generated relative to individual ones. Unlike experiments, successful social groups are ongoing. Thus, when solidary and cohesive, the m-group's game is iterated. Departing from these conditions, we seek to identify the games played by group members, some but not all of which contain dilemmas. This analysis owes much to Komorita's model for n-person dilemma games (1976). C,,: and D,,c are outcomes (payoffs) from C and D choices, respectively where n designates the number of is choosing C. Cn,' and Dnc are linear functions of the proportion choosing C. Whereas Komorita's focus was on dilemma-type games only, we seek to find the full spectrum of games defined by values of G. By finding the spectrum of games, we locate: (1) the boundary distinguishing games where cooperation is impossible from games where cooperation is possible, and (2) the boundary distinguishing games where cooperation is possible from games where cooperation is prefect - at least for rational egotists. Call these the lower bound and upper bound respectively. This analysis will show that between the upper and lower bounds the games are n-person Prisoner's Dilemma (hereafter NPD), which Komorita and Parks define as (1) Each of N-persons has two choices, cooperate (C) or defect (D); (2) The outcomes for each choice increase monotonically with the proportion of members who make the cooperative C-choice; (3) The D-choice yields a higher outcome than the C-choice no matter how many members choose C; and (4) The outcome if everyone chooses C is greater than the outcome if everyone chooses D (1996, pp. 40-41). The third point above might be misunderstood. It does not assert that every D payoff is larger than every C payoff. Instead, it asserts that, for each player taken one at a time - for the choice between defection and cooperation, the payoff to defection is always greater. ( D ~> C,,,. for any n < n.) In that specific sense, defection dominates cooperation. Basic Formulations

The possibility of cooperation is found by analysis of the game played by group members. Let group size be n, C n the payoff when all cooperate and D o the

74

DAVID WILLER, CASEY BORCH AND ROBB WILLER

payoff when all defect. Then ' G ' is the ratio of the payoff from cooperative activity to individual activity, G -Cn Do

(1)

To simplify the analysis, let the payoff to i from a unit individual activity be 1. 7 Then Do = 1

(2)

C n -- G

(3)

and

Since each m-group member receives 1/n part of the collective good, when the number of cooperators is n c, and C i is the payoff to any cooperator, G • nc Ci -

(4)

n

Furthermore, since the payoff is 1 from individual activity, the payoff to any defector when the number of cooperators is n c is Di= 1 +

Groups Where Cooperation

G . nc

(5)

n

is I m p o s s i b l e

a n d S o l i d a r i t y is Z e r o

When G = 1 cooperation gives no advantage in production. Cooperative goods are generated with exactly the same productivity as are individual goods. Nevertheless, the two are distributed quite differently. Cooperative goods are distributed equally in 1/n quantities to all group members whereas defectors keep all of the good they produce. Figure 1 gives two G = 1 games that capture those conditions. To describe the 2-person game we use the terms R = reward for cooperation, P = penalty for defection, T = temptation to defect and S = the sucker's payoff that were first coined for the 2-person PD game. Because T > R when the other player cooperates, ego should defect. Because P > S when the other defects, ego should defect. Thus defection is the dominant strategy. To see how defection dominates the 3-person game of Fig. lb, imagine any m-group member, 'ego' deciding between cooperation and defection. If two others cooperate and ego chooses cooperation, the payoff is C3=1. If two others cooperate and ego chooses defection, the payoff is D 2 = 1.67. Because D 2 > C 3

75

Building A Model for Solidarity and Cohesion Using Three Theories

a.

G=I

n=2

0 0

R

D R

S

T

1

.5

1.5

T

S

P

P

1.5

.5

1

1

1

D

G=I

b.

n=3

G

D

12

1

D

.67

C

1.67

D

1.33 1.33 .33

1

1

0 1.67 .67 .67

.67

.67

.67

1.67

.33 1.33 1.33 1.33 .33

1.33

D Fig. 1.

1

1

1

Games Where Cooperation is Impossible.

ego should defect. But it is also true that D~ = 1.33 > C 2 = 0.67 and D 0 = 1 > C~ = 0.33. Therefore, for all choices between D and C, defection always pays more. Since defection always dominates cooperation, and C 3 = D 0, cooperation is always disadvantageous. Call these G = 1 games, 'defection games.' Appendix offers proof that, when G = 1, what was found above is true for any n; all are defection games. The proof shows that always C,, = D o and

76

DAVID WILLER, CASEY BORCH AND ROBB WlLLER

D -I > C~c" Since cooperation is always disadvantageous, it is impossible. Since cooperation is impossible, when G = 1, it is also impossible when G < 1. In the economies of groups for which G _< 1, all action will be individual and none cooperative. In a previous section, the strength of solidarity, was defined as S = G i n/n. When G < 1 there will be no cooperation. Therefore n c = 0 and it follows immediately that S = 0. Thus when G < 1, n individuals are not a 'group.' Instead, they are like n potatoes in a sack of potatoes. They have no solidarity and their cohesion can be positive only if there are substantial costs in leaving. 8

Finding the Lower Bound for Cooperation and Solidarity We now show that cooperation is possible when G is larger than 1. Now when group members act collectively, because G > 1 each produces more value than when each defects to produce individual goods. For values of G in the neighborhood of one, however, the equal distribution of collective goods to all means that each cooperator gains a higher payoff by defecting. Figure 2 gives two games illustrating these effects. In Fig. 2a, G = 1.1 and, because T > R > P > S, this is a PD game. Nevertheless, as in the defection game above, defection is the dominant strategy in PD. As above, for the 3-person game here, D 2 > C 3, D~ > C 2 , and D O> C~ such that defection always dominates cooperation. Unlike the G = 1 game, here G > 1 and therefore, C > Do: the 3-person game is NPD. More generally, the Appendix shows that the lower bound for NPD games is G = 1 + A where A is just larger than zero (1 >>> A > 0 ) for thenDnc_~ > C c a n d C n > D o. For m-groups where G = I + A, cooperation is marginally possible because: (1) the cooperation of all has a higher joint payoff than when all defect (i.e. C > Do); and (2) as mentioned above the m-group's game is iterated. More accurately the m-group's game is iterated when it is solidary and cohesive. Since cooperation is possible, it is possible for n c > 0 and, with G = 1 + A, it is also possible for S > 0 and similarly for cohesion. Therefore, G = 1 + A is the lower bound for both cooperation and solidarity.

Groups Where Cooperation is Perfect and Solidarity is Assured Cooperation is perfect when C is the dominant strategy equilibrium and it is Pareto Optimal. 9 Given in Fig. 3 is a 3-person game where cooperation is perfect. For that game, C i and Di values are calculated from G = 3.1. That cooperation is assured can be seen by tracing paths of cooperation and defection through the matrix. At C j, the single cooperating player receives 1.03 with is greater

77

Building A Model for Solidarity and Cohesion Using Three Theories

a.

G=I.1

n=2

G G

D

R

R

S

T

1.1

1.1

.55

1.55

T

S

P

P

.55

1

1

13 1.55 b.

G=2

n=3

G C

2

2

D 2

1.33 1.33 2.33

1.33

.67 1.67 1.67

1.33 1.33

1.67 .67 1.67

G D

1.33 2.33

C

2.33

13 D Fig. 2.

1.67 1.67 .67

1

1

1

Games Where Cooperation is Possible.

than D O= 1 were that player to defect. Furthermore, at Dj the defecting player receives 2.03 which is smaller than C 2 = 2.07 which that player would receive were she to cooperate. Finally, at D 2 the defecting player receives 3.07 which is smaller than C 3 = 3.1, which would be received were that last defector to cooperate. Because C 3 > D 2, C 2 > D v and CI > D o cooperation always dominates defection and the three players should always cooperate. Call this a 3-person Privileged game and games like it N-Person Privileged (hereafter NPV) games, Note that, for the game to be NPV, it is not necessary for all C payoffs to be larger

78

DAVID WlLLER, CASEY BORCH AND ROBB WILLER

G=3.1

n=3

C C

3.1

3.1

D 3.1

2.07 2.07 3.07

D

2.07 3.07 2.07

1.03 2.03 2.03

C

3.07 2.07 2.07

2.03

D

2.03

13

1.03 2.03

D 2.03

1.03

1

1

1

Fig. 3. A Game of Perfect Cooperation.

than all D payoffs for they are not. For example, D 2 = 3.07 > C 2 = 2.07. For cooperation to dominate defection, it is only necessary that C c > D c ~ as in C 2 = 2.07 > Dj = 2.03. For the game to be NPV we need add only that C n > D 0. In other words, the payoff to ego increases when ego changes from defection to cooperation; and when all cooperate the payoff is larger than when all defect. The Appendix gives the proof that, when G > n, C > D Oand C,,,. > Dn,. _~ such that the game is always NPV. Thinking about the distribution of collective goods helps link the condition G > n to the preference for cooperation. When ego chooses to work cooperatively, the product of that work is equally divided among all n members of the m-group. When G > n, however, the payoff that ego receives from that distribution is larger than the payoff from defection and individual production (see Eqs 4 and 5 above). Therefore, all members always choose C independently from the choices of others; and n = n. Finally, given that the strength of solidarity is S = G i n/n, it follows that S = G > n: solidarity is high and, because n = n, it is optimal for the G value. It also follows that the prospects for cohesion are excellent. It must be emphasized that, when G > n and cohesion is greater than zero, both solidarity and cooperation are assured by the qualities of the economy's game alone. Since Olson (1965), it has been known that cooperation can be assured in NPD games by adding sanctions contingent on player's selection of C or D (see below). When G > n, however, cooperation is assured without

Building A Model for Solidarity and Cohesion Using Three Theories

79

sanctions. In fact, any m-group that had devised a social structure with sanctioning to produce cooperation when G was smaller would never need to employ it once G > n. It follows that there are no free rider problems at the first, second, or higher order. Thus, the only threat to cooperation and solidarity, given G > n, is the existence of an alternative group for which G is higher.

The Upper Bound: Where Cooperation and Solidarity Possible The Appendix shows that the upper bound for NPD games - for games where cooperation is possible, but not perfect - is G = n - A. We have just seen that, when G > n, the game is not NPD but NPV where cooperation dominates defection. In fact, when G = n all m-group members are indifferent between cooperation and defection and the game is neither NPD nor NPV but at the boundary between the two. At G = n - A , however, the game is NPD: defection dominates and when all cooperate the payoff to each is more than when all defect. Previously it was found that the lower bound for NPD is G = 1+ A and here that G = n - A is the upper bound for NPD games. Therefore, the game is NPD for any n > G > 1 and within that range cooperation and solidarity are possible. The section to follow examines some little- known conditions which favor both.

The Conditions for Cooperation and Solidarity when n > G > 1 When n > G > 1 the m-group economy is NPD and, like all NPD games, contains a social dilemma. In this section, we investigate conditions that favor the resolution of that dilemma focusing on the effects of coalitions. Whereas cooperation is normal in iterated PD games with only two players (Axelrod, 1984), cooperation is not normal in NPD games (Komorita & Parks, 1996) even when iterated. Therefore, cooperation is not assured even when m-group members have ongoing relations. Nevertheless, since ongoing relations are possible, so are coalitions. Whereas defection dominates for the individual, shifting the perspective from individuals to coalitions finds three important conditions where cooperation is favored. First, we show that, in NPD, a coalition formed to support cooperation will have no opposition. That is to say, if a coalition forms: (1) it will support cooperation, never defection; and (2) its members will prefer to increase its numbers. Looking to Eq. 4, because payoff increases with n C, if there is a coalition of cooperators, each member would like the coalition to be as large as possible. That is to say, the members of any coalition of cooperators will prefer to add members until n c = n.

80

DAVID WILLER, CASEY BORCH AND ROBB WILLER

By contrast, no coalition will form to support defection; if, by chance alone, a defection-based coalition came to be, its members would prefer to decrease its numbers. Looking to Eq. 5, as the number of defectors increases, n c decreases; thus the payoff to each defector decreases. It follows that there is no motive for a set of defectors to form a coalition encouraging defection. If there were a coalition of defectors, each member would want all others to quit and cooperate. Said somewhat differently, since D i is maximal at D _ 1, each defecting m-group member prefers to be the only defector. Therefore, a coalition f o r m e d to support cooperation will have no opposition.

Second, to achieve cooperation throughout the m-group requires, not that all initially cooperate, but only that a "critical mass" act together. The critical mass in question is the number that can shift from D to C without loss; call its size M c. When nct is the number of cooperators at to and net + M c is the number at t I, applying Eqs 4 and 5, G ( n c t + M c)

G.nt

-

n

1+ - -

n

n

thus,

Mc-

G

and the proportion of critical mass is I/G of the m-group. Interestingly, as G, the productivity of cooperative action, approaches one, the size of the critical mass approaches the size of the m-group. Alternatively, as G increases the proportion of the critical mass decreases. For example, when G = 2, a coalition of 50% can change from D to C without loss, when G = 4 a coalition of 25% can change from D to C without loss and similarly. Therefore, the greater the relative productivity of cooperation, the larger the G and the smaller the critical mass for any m-group size. Remember that coalitions support cooperation and never defection. It follows that if there is a critical mass coalition, it will support cooperation. Furthermore, the critical mass coalition is a force for cooperation because it can affect other m-group members. Just as the critical mass can change from D to C without loss, it can also change from C to D without loss; but others in the group will suffer a loss. Thus any critical mass coalition has a threat which can be used to induce others to cooperate. Thus once a critical mass coalition is formed, all others should join them in cooperation. Third, in NPD, cooperation can be assured by introducing sanctions extemal to the game. Call S u the value of the sanction needed to induce cooperation. Then that value is just slightly more than the difference between the payoff at defection and the payoff at cooperation. SN =

1+

G.n c

G ( n c + 1)

n

n

+ A

(6)

Building A Model f o r Solidarity and Cohesion Using Three Theories

81

and S N = 1 - G / n + A. Given social sanctions as large as S N or larger, cooperation can be assured, because any actor will gain more from cooperation than defection. Since the payoffs of cooperators and defectors alike increase with n c, not just cooperators but defectors as well have an interest in using social sanctions on (other) defectors. More accurately they have an interest in using social sanctions if the cost of use is less than the benefits gained from more cooperation. Therefore, for any given cost of sanction use, the interest in using sanctions increases with G. Furthermore, since a coalition using sanctions to promote cooperation will have no organized opposition, its minimum size is the smallest number needed for norm enforcement. As seen below that number is two. This section has shown that, were a coalition to form, it would be a coalition of cooperators and its minimum size is no larger than the critical mass which is 1/Gth of the group. Once that proportion is achieved, the coalition can induce others to cooperate by threatening to jointly defect at no cost. It was also found that defectors will not form coalitions. Thus, if a coalition forms, any defector stands as an individual opposed to that coalition. Further, when sanctions external to the game can be used, they will only be used to induce cooperation and the minimum size of the coalition needed to enforce norms is two. Although these conditions favor cooperation, they do not assure cooperation, but, instead, systematically narrow the problem of cooperation to problems of first and higher order free riding. For example, while a critical mass has an efficacious threat, that threat is empty unless the critical mass is an organized coalition. Since organizing in a coalition is not costless, the process of organizing must resolve a first order free riding problem. Even if organization occurs, carrying out the threat to change from cooperation to defection affects payoffs of all outside the coalition and not defectors alone. Thus the threat cannot be targeted to defectors. By contrast social sanctions can be targeted to individuals and, due in part to the absence of organized opposition, can be employed by as few as two members. But the employment of social sanctions: (1) requires at least minimal organization the costs of which, as above, produce first order free riding; and (2) the subsequent employment of the sanctions requires that second and higher order free riding problems be resolved. In the sections to follow we resolve the problems of free riding for m-groups in two steps. First order free riding problems are solved by tracing the material conditions for norms from the economic structure to the m-group's decentralized social structure of norm enforcement. A subsequent section explains how the

82

DAVID W1LLER, CASEY BORCH AND ROBB WILLER

second and higher order free riding problems of that social structure are resolved by the m-group's cultural structure. 1°

THREE ECONOMIES AND THEIR NORMS In Elementary Theory (hereafter ET), norms are rules for behavior which occur when "the actor in a system has one set of interests in its own behavior and a different or opposed set of interests in the behavior of others" (Southard, 1981, p. 60). When n > G > 1, the group's economy is NPD and the analysis just above showed that, while each member prefers to defect, all prefer that no other defects. ET places equal emphasis on both preferences. Thus, when n > G > 1, m-group members are understood as having an individual interest in defection (free riding) and an opposed interest that all other m-group members cooperate. For this opposition of interest, it is not necessary that actors be opposed in distinct classes or status groups. Nor need they occupy opposed positions in economic or coercive relations. Actors in identical positions will come in conflict with each other when some act on their individual interest and defect, while others act on their collective interest and cooperate. This section finds norms in three groups. The first group is a set of subjects in a network exchange experiment who are forming a coalition in opposition to those who are high in power. We demonstrate how to find the NPD game embedded in the power structure, derive a value for G, and show why there is a norm supporting cooperation. The second group is a Pygmy band where a norm for cooperative hunting is violated by one group member. Though an exact value cannot be given to G, there is strong ethnographic evidence that it is substantially greater than one. Since there is deviance, the analysis above suggest that G is smaller than n. The third group is composed of Danish artisans engaged in a system of social exchange where a norm supports contributions triggered by other's expressed needs. Again ethnographic evidence strongly suggests that G > 1. Norms and Free Riding in a Network Exchange Structure

Here we analyze a network exchange structure applying strategic analysis (Wilier & Skvoretz, 1997a) to find the game played by those low in power. Figure 4 gives the NET43 structure with three A positions each connected to four B positions and conversely. The As are not connected to each other, nor are the Bs. Let each position be allocated one exchange opportunity for each time period t. Following conventions of network exchange experiments, a pool of 24 valued resources can be divided in each relation. Since there are three A positions and four Bs, one B will be excluded

Building A Model for Solidarity and Cohesion Using Three Theories

83

B~

B2

1

3

Ba

64 Fig. 4.

The NET43 Experimental Network.

each time period. Since no A is necessarily excluded, this is a strong power structure (Simpson & Wilier, 1999). Both theory and research indicate that the As will gain better and better resource divisions as the Bs, seeking to avoid exclusion, make better and better offers (Cook et al., 1983; Markovsky et al., 1988; Wilier & Skvoretz, 1997). Let the "critical mass coalition" be the smallest coalition that will necessarily receive resources in exchanging (Willer & Skvoretz, 1997). Since there is just one exclusion, the critical mass for the low power Bs is two. We now find the game played by the Bs by attributing the following parametrically rational strategy to the As: "Let each A accept its best offer and, when offers are tied, select randomly." Since the critical mass is two, Figure 5 displays B l and B 2 playing B 3 and B 4, but gives the payoffs individually by each player. At t = 0, let all Bs initially make the equipower offer of 12-12 to the As. Then three of the four offers are accepted and the expected payoff to each B is 12 × 3/4 = 9. Figure 5 at t = 0 displays the 9, 9 payoff in the C, C cell. When any two Bs defect, they make the 11-13 offer to the

84

DAVID WILLER, CASEY BORCH AND ROBB WILLER

As, and both offers are accepted leaving one exchange for the remaining two Bs. Still offering 12-12, one of the two Bs' offers is accepted and the expected payoff is 12 × 1/2 = 6. The 11 and 6 payoffs are displayed in the C, D and D, C cells on the off-diagonal. Finally, when all four Bs defect to offer 11 - 13, three of the four offers are accepted and the expected payoff is 11 × 3/4 = 8.25 as displayed in the D, D cell of the t = 0 game matrix. Since C = 9 > D 0 = 8 . 2 5 , and D3= 11 >C4=9 and D~ = 11 > C 2 = 6 , the game is NPD. j j The game found above is the first step of a process of power development traced in Fig. 5. Since defection is the dominant strategy, at t = 0, all Bs offer 11-13 favoring the As which constitutes the NPD game at t = 1 where the 11-13 offer is cooperation, 10-14 is defection and again the game is NPD. Now the Bs defect to the better offer of 10-14 such that the game at t = 2 is constituted and similarly to t = 8 where the game is not NPD but defection is the best response. At t = 9, cooperation (C) is the best response and the power process ends. Now G is calculated from C 4 of the first game and D O of the t = 8 game which is the last where the Bs defect. G=

c4 Do

=

9

=4

2.25

Because the Bs are embedded in a larger power structure, their economic base is constituted differently from the m-group above. Instead of a single game, there is a series of linked games which, taken together, inflate the value of G. Thus here G = n, which indicates, not an NPV game, but a series of related NPD games. For each game of the series, 1 < G < 2: all are NPD. As in all NPD games, the individual interest in free riding is opposed by the collective interest in cooperation. Thus the B's economic base produces a norm supporting cooperation. Whether experimental subjects recognize that norm and, if recognized, form a coalition and act is discussed in a following section.

The Crime of Cephu Here we examine the hunting economy of a Pygmy band reported by Turnbull (1961) to differentiate cooperation from defection and estimate the value of G. In the communal hunt, each adult male is armed with net and spear. The nets, as described by Turnbull, are approximately 300 feet long and about four feet high; they are placed in a semicircle as shown to the left of Fig. 6. In the hunt there is a division of labor by gender and age. After the nets are set, it is the

Building A Model for Solidarity and Cohesion Using Three Theories t= 0

85

B3&B4

C

D

C

9

9

D

1~ 6

6

11

BI&B2 8.25 8.25

(12- 12)

t= I

B3&B4

C B~ & B~

t=2

Bt&B2

C

8.25 8.25

D

10 5.5

5.5 10

7.5 7.5

(II-13)

B3 & B4

C

D

D

C

7.5 7.5

D

9

5

5

9

6.75 6.75

(10- 14)

B3&B4

t= 7

C C

D

3.75 3.75

2.5

4

3

3

BI&B2

D t=8

B3&B4

C BI&B2

2.5

( 5 - 19)

D

C

3

3

2

D

3

2

2.25 2.25

(4 - 20)

4

3

Not a P/D game. t=9

B3&B4

C

BI&B2

D

(3 - 2D

Fig. 5.

C

D

2,25 2.25

1.5 2

2

1.5

1.5 1.5

Defection Chain for Low Power Actors in NET43.

task of women and children to take up positions some distance away and 'beat' their way toward the nets as illustrated by the arrows pointing to the left in Fig. 6. The effect of beating is to drive animals into the nets where they are speared by the men. Because this is a communal hunt, any game caught is communal property; it is usually divided when all have returned to camp.

86

DAVID WILLER, CASEY BORCH AND ROBB WlLLER

.

i

D

Fig. 6. A Diagram for the Pygmy Hunt.

For norms to support the communal hunt it must be more effective than individual hunting such that G is greater than one. Communal hunting organized like that of the Pygmies is frequently judged to be more efficient than individual hunting. For example, the Plains Indians, whose open environment, which could hardly be more unlike the Pygmy's dense forest, preferred the communal hunt. If G is greater than 1 for the Pygmy band and elsewhere, the effectiveness of the communal hunt would not have gone unnoticed. In fact, Sahlins (1972) finds bands and tribes that hunt to be affluent, as measured by leisure, not accumulation; he finds that time spent at work averaged between three and four hours per day. Furthermore, Turnbull reports that the communal hunt was of short duration and that a very large number of animals were killed. The alternative, running down animals in the dense rain forest, would require far more energy per person while not giving as high a yield per person. Estimating G conservatively, let the communal hunt be 1.5 times as productive as individual hunting. Then the economy is NPD exactly as modeled for the m-group above. The communal hunt is an NPD game and the norm at issue for the Pygmies is the placement of nets. For any given density of animal biomass, the size of catch is proportional to the distance across the net semicircle. Nets placed in front of the semicircle add nothing to the catch. Because it reduces the total catch and results in a higher individual payoff, any net moved from the semi-

Building A Model for Solidarity and Cohesion Using Three Theories

87

circle to, for example, the point D in Fig. 6, is a defection and also a norm violation. Since only one of 11 Pygmies is defecting this is Dlo and its payoff should be very large. To verify that there is such a norm, the ethnographer must have observed its violation. The crime of Cephu (Tumbull's term) was to place his net in front of others of the tribe as at D in Fig. 6. After making his catch which was very large, Cephu left before being observed. There was no food shortage. Cephu's motivation to defect was not to eat more, but to gain status by giving more food to others. Below we return to the crime to trace norm enforcement and its effects. The Deviant Dane This story concerns a small group of Danish artisans who exchanged services as friend's favors. These service exchanges did not follow a rule of finely calculated or balanced reciprocity. According to Hansen (1981) those who expressed a need for services received them. And those services were sent by the group member whose artisan skills best qualified him. A service received did not call for a service returned. Instead, all participants understood that contributions should follow the other's expressed need for help. Not all services were equally valued. The value of a service was the multiple of time expended and skill applied. More highly skilled members were understood as making larger contributions per time spent than those with lower skills. Though time and skill were not being measured by group members, all shared an understanding of who was making larger and who smaller contributions. Furthermore, there was a 'balancing' mechanism that governed contributions. When contributions were unequal, those who made larger contributions gained higher status. For example, "Erling," the group leader, was making the largest ongoing services to others. ~2 We assert that the economy of skilled service exchange was NPD and that the rules governing labor contributions were norms for the following reasons. When comparing the work of a skilled person to one who is unskilled two differences stand out: (1) skilled work is more productive per unit time; and (2) of higher quality. Because skills are specializations, working outside his specialization, the skilled worker is unskilled. No single individual could have the array of skills exercised by the artisan group. It follows that, by pooling skills, and matching them to work, G > 1 for the service exchange system. Thus C > D 0. Furthermore, Dcn - l > Cc,' because, for a given no, any who receive services but do not contribute gain higher net payoffs than those who are contributing. Thus the service economy is NPD and the rules governing labor contributions are norms.

88

DAVID WlLLER, CASEY BORCH AND ROBB WILLER

"Mads," though a contributing member of the artisan group, was the exception for he was not skilled. As result, for Mads to fulfill the norm for labor contributions, it was necessary that he work longer than others. Said somewhat differently, because Mads was unskilled, for him alone, it was not true that G > 1. Problems of cooperation should soon arise and they did. Problems stemmed from the point at which Erling decided to help Mads build a house. In fact, all artisans made major contributions and, when the house was completed, Mads' debt to the group was substantial. As explained by Hansen, "Mads became indentured to the group, not free to decide when and where he would give his services" (1981, p. 78). The house, though built under the service exchange system, was recognized by the state as Mads' private property. By cooperating and continuing his labor contributions, Mads would have the enjoyment of his house and the ongoing friendship of group members. By defecting, Mads would sell the house and appropriate the value of the group's invested labor. His choice and attempts at social control are found below. Danish artisans display a condition not anticipated by theory introduced earlier, but one for which the theory can be adapted. The theory allows cohesion to vary across m-group members, but solidarity, a group quality, was treated as constant across its membership. For the Danish artisans, however, G > 1 for all but Mads because only he was unskilled. For contributions to Mads G > 1. For Mads' work G = 1. Since a contribution was counted as work time multiplied by skill, when the subscript m is for Mads and g for group, the demand on Mads was G m tm = G

t

which explains his very long work time obligations. To equal the contributions of others, Mads had to work longer. Mads did not make unit labor contributions, assumed in the theory above, but contributions G i G m larger than the unit. Therefore, cooperation held no advantage over defection and Mads was playing a defection game. NORMATIVE SOCIAL STRUCTURES AND SOLIDARITY In this section we first explain how a decentralized social structure can enforce norms without central authority. Using procedures for game transformation introduced below, we show that enforcement transforms the games of the economies just discussed from NPD to NPV - from N-person Prisoners' Dilemma to

Building A Model for Solidarity and Cohesion Using Three Theories

89

N-person Privileged - thus assuring cooperation. With cooperation assured, if cohesion is high enough, the group is solidary. The theory of normative social structure is then applied to the three cases of the previous section. Two of the three groups are solidary due to norm enforcement. The third group provides an important contrast. For one member it is not solidary and lacks cohesion. Figure 7 drawn from Southard (1981, p. 57) gives three types of enforcement that have increasing levels of severity. In shunning, which is given as System 1, in the network to the left d is engaged in the social exchange system. When shunned, as in the network to the right, d is disconnected. Shunned actors, though excluded from normal intercourse, are not expelled from the group. In negative convergence, displayed as System 2, again d is first engaged in social exchange as displayed in the network to the left. In the network to the right, d is disconnected from those exchanges and negatively sanctioned by multiple others. Negative sanctions can range in severity from symbolics, such as insults and threats, to material sanctions such as beating and stabbing. Expulsion given as System 3 in Fig. 7 is importantly related to cohesion. In the network to the left, d is engaged in ongoing social exchanges while, in the network to the right, d has been expelled from the group. The severity of expulsion is related to the values of G and c n, compared to the quality of the environment outside the group. All else being equal, the higher the cohesion of the group, the larger the expulsion penalty. When the group member is not free to join another band or tribe, the effect of expulsion is at least - Gnfn which is a very severe sanction. By contrast, for groups embedded in larger societies, those expelled may profit. We now turn to the effect of norm enforcement on the game played by group members using a game transformation procedure new to this paper. Figure 8 illustrates the procedure. Given in Fig. 8a is a 2-person PD game where T > R > P > S. Adding two points to the payoffs of any cooperative actor produces the Fig. 8b game for which R > T > S > P. The transformed game is PV; and because the dominant strategy equilibrium is Pareto optimal, cooperation is perfect. A second transformation is also shown in Fig. 8. The Figure 8a game is transformed into the Fig. 8c game by subtracting two points from the payoff of the defecting actor. Again the PD game is transformed into the PV game where the dominant strategy equilibrium is Pareto optimal and cooperation is perfect. The same kinds of transformations occur in NPD games as just shown for PD, and, because norm enforcement systems use negatives, the effects will be like the transformation of 8a to 8c. The examples to follow show that a social structure of norm enforcement will produce cooperation and, with one exception, solidarity. Said somewhat differently, a social structure of norm enforcement

90

DAVID WILLER, CASEY BORCH AND ROBB WILLER System State A

41+

System State B

4+

+lb ~ ' ~

+lld )

) System 1: Shunning

4+

+1~

@.

C)

Q

0 System 2: Convergence

4+

@,+

+ID

.© System 3: Expulsion

Fig. 7.

Normative Enforcement Systems of Social Control (Southard, 1981).

Building A Model for Solidarity and Cohesion Using Three Theories

91

a.

P~

C C

D R

S

T

1

-1

2

S

P

P

-1 0 T>R>P>S

0

Pl T

D 2

+2 for C -2 for D P2

b: e

C PI D

R

R

S

T

3

3

1

2

T

S

P

2

1 0 R>T>S>P

P2

C:

Pl

D

C

D

R

R

S

T

1

1

-1

0

S

P

P

-1 -2 R>T>S>P

-2

T

0

Fig. 8.

E T ' s G a m e Transformation Procedure.

92

DAVID WILLER, CASEY BORCH AND ROBB WlLLER

will resolve the first order free rider problem of the economy. We recognize that these social structures may, in turn, produce second order problems which they do not resolve. Second order problems are taken up in the section to follow] 3

Sanctioning and Cooperation in the Network Exchange Experiment Figure 9a displays the t = 0 game of the low power Bs first seen in Fig. 5. In Fig. 9b, B l and B 2 have formed a coalition while B 3 and B4, who have not, are acting independently. Coalitions in this experiment alter two conditions relevant here. First the Bs who join share equally all payoffs gained when exchanging with the high power As. Second, emulating the model for negative convergence, any Bs who do not join can be negatively sanctioned by those who have. The cost of transmitting those sanctions is zero, but the effect of reception is to reduce the B's payoff to zero, an effect much larger than needed. That effect is shown in Fig. 9b. Now if B 3 and B 4 believe that the coalition of B j and B 2 will cooperate, they should cooperate and join the coalition. As soon as a critical mass coalition of two Bs is formed, the full coalition of four Bs will follow. When B 3 and B 4 join the coalition, all share payoffs equally from all exchanges, and the full coalition produces the Fig. 9c Privileged game. Because resources are shared, game payoffs in each cell are always equal for all Bs and payoffs are highest when no one defects. Therefore, the coalition is solidary and will produce the collective good, which is equipower payoffs when exchanging with those high in power. Experimental results support this inference (Borch & Wilier, 2001). We add that, for the experiment, cohesion was never in question because escape meant leaving the experiment and forfeiting all payoffs.

The Punishment of Cephu Recall that Cephu's crime was to set his net between the beaters and the line of nets. While he initially received more than his share of the day's kill, retribution in the form of negative sanctions was forthcoming. Cephu was sanctioned in two ways: (1) members of the tribe took away the meat that he caught when defecting; and (2) after negative convergence the tribe expelled him for a few hours. Evaluating the effect of a few hours expulsion, Turnbull asserts that Cephu "must have suffered the equivalent of as many days of solitary confinement for anybody else" (1961, p. 109). It was already determined that the Pygmy hunting economy is NPD. Since all but Cephu cooperated, we understand the motive of his defection because D,,C > C n. The effect of his food being taken away, his loss of status and expulsion

Building A Model for Solidarity and Cohesion Using Three Theories

93

a. B 3 &: B 4

C C

D

R

R

S

T

9

9

6

11

T

S

P

P

B1 & B 2

D 6 8.25 T>R>P>S

11

8.25

- 11 forD

by B 3 & B 4

B 3 & B4

b:

C

Coalition Bt & Bz

D

R

R

S

T

9

9

6

0

T

S

P

P

11

6

8.25

8.25

Coalition B3 & B4

C:

C C

Coalition

D

R

R

S

T

9

9

8.5

8.5

B1 & B2 O

T

S

P

P

8.5

8.5

8.25

8.25

R>T=S>P

Fig. 9.

Norm Enforcement and Cooperation in the Network Exchange Experiment.

94

DAVID WILLER, CASEY BORCH AND ROBB WILLER

was to reverse the inequality such that C n > Dnc_l. More generally, for Cephu and any other tribe member, given negative convergence and expulsion, cooperation always dominates defection (i.e. Cn, > D e_] ). Since cooperation is assured, it is reasonable to ask why the crime was committed. Following Malinowski (1959), we suggest that those who deviate in the face of a highly effective normative system do so in the belief that they will not be found out.

The Flight of Mads Recall that the Danish artisan group studied by Hansen (1981) exchanged services and helped to build Mads' house. By accepting the labor of the others, Mads obligated himself to all group members. This obligation was large because Mads' labor was unskilled and less valuable than the labor of others. After some attempt to balance his obligations, Mads defected, refused to offer further services, and, as a result, the group members shunned him. Though every day for two years Mads had obtained a fide to work from one of the group, suddenly "he was passed while waiting for the bus" (p. 82). As explained by Hansen, "Mads sold his house and land and moved to a village near work" (1981, p. 83). Unlike Cephu, Mads was able to escape the group members' attempted norm enforcement. Whereas cohesion is high for bands and tribes, with few exceptions it is low for any group embedded in a large contemporary state. Whereas today, the political state regulates who comes and goes across its borders, groups embedded in the state generally lack that power. The case of Mads, however, is extreme in that cohesion was negative because, for him, the collective goods of the group did not have a higher payoff than his individual work. Or, more accurately, they would not have had higher payoffs but for the fact that, in escaping, Mads appropriated their value. Mads was impelled from the group by his labor debt and drawn by the profits gained by selling the house. The profit realized by Mads included value added by the labor of the artisan group. It included that value because the property system of the state recognized only Mads' claim that the house was private property. It did not recognize value added by the artisan group. Only the group, and not the state, would enforce the group's property fights over Mads and the house they built. In the absence of cohesion, however, the group had no enforcement capability.

STATUS AS A CULTURAL STRUCTURE THAT SOLVES SECOND-ORDER FREE RIDING In this section we apply formulations for status to solve second order free rider problems. Having shown above how groups sanction free riders, the second

Building A Model for Solidarity and Cohesion Using Three Theories

95

order problem that remains asks: Why does any individual participate in the sanctioning, when she could free ride leaving the sanctioning to others? Applying theory and referencing ethnographic examples, we show how supportive status orderings resolve the second-order free rider problem without producing higher order problems. Briefly, the theory to be offered is the following. Normative sanctioning to enforce contributions to the collective good benefits each group member. Those who sanction free riders, because they further cooperation, will gain higher status than those who do not. Thus, over time, the status order will come to place those who enforce norms above those on whom norms are enforced. Those who neither enforce nor break norms fall in the middle of the order. It follows that, when standing in the status order is valued, norms will be enforced. For status orders to resolve the second order free rider problem, standing in the order must be continually achieved. Therefore, those with high status will sanction free riders or they will fall down the status order below those who do. With group members reordering status hierarchies as needed to correspond to members' activity, the administration of second order incentives is either free or nearly so and always distributed throughout the group. Therefore, this solution to the second order free rider problem bypasses the infinite regress to higher level problems while asserting that status orders offer ongoing coordination for group activity.

The Problems of Norm Enforcement First and second order free rider problems are linked in the following way. As shown above, when first order sanctions that punish defectors change an individual group member's incentives, cooperation dominates defection. Because the administration of sanctions by others enforces first-order contributions by all such that n = n, payoffs to group members become optimal (see also Yamagishi, 1986; Heckathorn, 1989; Coleman, 1990). Though the first order cooperation problem is solved by sanctioning free-riders (Olson, 1965), when the administration of those sanctions is a cost to the sanctioner (Oliver, 1980), the solution is incomplete. As Hechter points out, "the provision of control is itself subject to the free-rider dilemma" (1987, p. 51). There is a second order free rider problem when all actors have an interest in avoiding the costs of administering sanctions, but maintain an interest in other group members administering sanctions. The second order problem can lead to an infinite regress in the following way. Let second order free riding be avoided by sanctioning those who fail to sanction first order free riders. Again, all are tempted to defect leaving second order sanctioning to others.

96

DAVID WILLER, CASEY BORCH AND ROBB WILLER

But the second order problem is solved by sanctioning those who fail to sanction at the second-order. But now there is a third order free riding problem and resolving it with sanctioning produces a fourth order and similarly without end. Status as a Reward f o r Cooperation

If there is an infinite regress of free rider problems, why is it that groups actually administer normative sanctioning systems? Moreover, since sanctioning while others free ride appears to be the sucker's choice, why do the most prestigious group members either carry out the sanctioning or manage those who do? In several ethnographic examples, we find that those who fulfill norms are granted high status, while those who do not lose status. In reference to norm compliance Malinowski notes, "the positive aspect of compliance to primitive custom, the fact that obedience to rules is baited with premiums, that it is rewarded by counterservices, is as important, in my opinion, as the study of punitive sanctions" (1959, p. 36). For an example of status loss, Malinowski offers a case where a tribesman remained after being exiled, "[a] man who would try to brave the dreadful insult involved in [being exiled] and remain in spite of [being exiled], would be dishonored for ever. In fact, anything but immediate compliance with a ritual request is unthinkable for a Trobriand Islander" (1959, p. 104). Here remaining in the group assures low status for the unfortunate tribesman, but what of those who passed down the decree to exile: how did they gain the status to pass judgement? Service states that when "reinforcement is a function of particular persons, it is very informal and largely a matter of social status rather than true authority" (1975, p. 55). t4 The primary responsibility of chiefs, the highest status position in simple communal societies, is to ensure that all contribute to the collective good. Hoebel (1960) suggests that chiefs must exemplify the cooperative norms set by the tribe. "The personal requirements for a tribal chief, reiterated again and again by the Cheyennes, are an even-tempered good nature, energy, wisdom, kindliness, concern for the well-being of others, courage, generosity, and altruism. These traits express the epitome of the Cheyenne ideal personality" (1960, p. 37). Lizot cites similar traits for Yanomani leaders. "He is sensitive on questions of honor. In his adopted community, he is esteemed and respected for his qualities as a good hunter and courageous warrior. He is still young and has the stuff of a future leader. His undertakings are always well thought out.

Building A Model for Solidarity and Cohesion Using Three Theories

97

He is a man on his way up, a fiery and prolific orator, a worker who knows how to be generous" (1985, p. 87). Beyond exemplary living, the chief also disciplines those who deviate from group mandated norms. Malinowski states that a chief "exposes laggards to public censure and so exerts control on their behavior" (1967, p. 194). When public exposure is not enough to eliminate deviance, chiefs can take more drastic measures. "Here was an insult delivered with every aggravating circumstance. It was addressed to a chief, it was said aloud and in public, and the personal name was added, the form in which the insult is deadliest. After the war, when peace was concluded and all other enmities forgotten, Si'ulobubu (the deviant) was openly speared in broad daylight by a few men sent by To'uluwa (the chief) for that purpose" (Malinowski, 1967, p. 196). Lowie concurs adding that, "the police acted, at least in a number of tribes, under the direction of the tribal chief or council" (1967, p. 81). With such extreme sanctioning power at their disposal chiefs are granted great deference. ~5 A modern example of status enhancement is reported in a critique of the Hawthome studies found in an investigation of that study' s documents by Carey (1967). In the Relay Testing Room, the production rates of five 'girls,' separated from the main work force, was recorded over a period of two years. There was a group piece rate such that the pay to each was based on the collective output of the group. Following Roethlisberger and Dickson (1939) most reports attribute production increases to the enhanced sense of community brought on by the study, the so called 'Hawthorn Effect.' To the contrary, Carey found status differences supporting a norm enhancing pay. The young woman who became the work group leader was the fastest worker. Carey states that, "it is doubtful whether any operator could have secured this position unless she had been the fastest worker" (1967, p. 414). Furthermore, when the production of several members fell, the leader was reported to have said, "Oh! what's the matter with those other girls. I'll kill them" (1967, p. 414). Production subsequently increased. In the examples just presented there is a consistent relation between high status and sanctioning norm violators. There is also a consistent relation between being found to violate a norm and low status. In game theoretic terms, those with high status cooperate at the first order to ensure production of the collective good, and also cooperate at the second order to sanction first order deviants. It follows that the m-group that solves its social dilemma with a social structure of sanctioning will have a linked status structure that solves second and higher order free rider problems. We now look more closely at Status Characteristics Theory's understanding of the working of the status structure.

98

DAVID WILLER, CASEY BORCH AND ROBB WILLER

Status and Sanctioning We define status as "an actor's relative standing in a group based on prestige, honor, and deference" (Thye, 2000). Status orders rest on shared understandings produced in the status-organizing process (Berger et al., 1977). Following Ridgeway (1982), the group's status order changes as a result of actors' past behavior. That is to say, status is achieved in an ongoing process. Applied to the normative group, sanctioning to enforce cooperation at the first order either maintains the actor's high status or produces status gains. Similarly, failure to sanction in the face of first order free riding produces status loss. Status gains and losses will, under conditions specified later, transform the game faced by potential sanctioners to one where cooperation and the sanctioning of first order free riders dominates defection thus eliminating second order free riding. Research indicates that, the higher the levels of contribution to group goals, the greater the increase in an actor's status (Berger et al., 1985; Wilier et al., 2001). More specifically, research by Ridgeway (1982) indicates that contributions to group goals will increase the contributor's status if those contributions are presented in a group-motivated way. In fact, normative sanctioning promotes the acquisition of collective goods and improves payoffs to all group members. We assert that sanctioning to enforce such a normative system is a groupmotivated contribution to a group goal. When the status-organizing process takes into account all status-relevant information of actors, sanctioning at the first order either increases status or maintains high status for sanctioners while failure to sanction has the reverse effect. For our m-group, placement in the status order is an ongoing payoff, either positive or negative depending upon the actor's standing. This assertion is by no means new. According to Berger et al. (1972) high status actors who benefit from a greater share of group participation are more influential over other group members and are accorded more deference. Research and theory also indicates that higher status actors may also enjoy material rewards as a consequence of their standing (Berger et al., 1985; Thye, 2000). Thus, we assert that actors will sanction first order free riders when the cost of administering sanctions is exceeded by the value of the status gain - or status loss avoided - that is expected to result from the sanctioning behavior. Since status orders are continuously reordered, no special cost is associated with these status alterations and, therefore, infinite regress is avoided. That there is no special cost has long been supported by Status Characteristics Theory. According to Berger et al. (1977), the ordering and reordering of status hierarchies may be an unconscious or semiconscious process. In either case, since no cost is experienced, cost calculations cannot be associated with that

Building A Model for Solidarity and Cohesion Using Three Theories

99

ordering and reordering. Furthermore, research indicates that groups will voluntarily reorder status orders based on new status information (Ridgeway, 1982; Markovsky et al., 1984). No special payoffs are needed to institute that reordering process. It follows that, when status is the incentive at the second order, sanctioning becomes individually rational and resolves the first order free rider problem without producing higher-order problems. ~6 Nevertheless, status can operate to resolve second order free riding only when status hierarchies are fluid. Ascribed or legitimated status hierarchies, such as caste and feudal systems, impede movement in the status order and weaken the working of status in furthering norm enforcement. Even in societies with substantial social mobility, when status is gained from sources outside the group, as it can be in any embedded group, the efficacy of status is weakened allowing second and n-order problems to reemerge. Of course, groups embedded in states are regulated by centralized authority systems. For norm enforcement, however, that regulation is no substitute for the social control mechanisms of the group. As seen, in the case of the Danish artisans, when the rules of the state are contrary to the norms of the group, norm enforcement becomes impossible.

CONCLUDING REMARKS The theory construction of this paper is now complete. The goal of that work was to produce a modeled group, called an 'm-group,' in which solidarity and cohesion are linked to cooperation. An m-group is cohesive when members do not leave to join another group. Given cohesion, an m-group is solidary when its members prefer to cooperate. Taken together, the conditions for solidarity and cohesion are the conditions for cooperation. When solidary and cohesive, m-group members will cooperate. The model building for the m-group links three theories: Game Theory, Elementary Theory and Status Characteristics Theory. Working with Game Theory, formulations for the economic base are developed. We define G as the ratio of the payoff to i when all cooperate to the payoff when all defect (G = Cn/D0). To simplify derivations we set D o = 1. When G < 1 the m-group cannot be solidary and, if lacking cohesion, will dissolve. Only if escape is blocked can a G < 1 group be cohesive and continue in existence. But if it is cohesive, having no solidarity, it will have no norms and life will be nasty, brutish and short as in Turnbull's mountain people (1972). By contrast, when G > n, if cohesive, the economy is NPV. When NPV, the payoff to each m-group member is higher when choosing cooperation than when

100

DAVID WILLER, CASEY BORCH AND ROBB WILLER

choosing defection regardless of what others chose. Thus cooperation and solidarity are assured by the economy alone. These m-groups are utopias for which no links can be drawn from the economy to the society and culture. In fact, no society or culture are needed for solidarity. Not unlike, small societies observed in the field, the most theoretically interesting kind of m-group has three linked components; they are economic, social and cultural structures. An economic structure forms the base and, for that economy n > G > 1, such that members play the NPD game. A decentralized social structure resolves the dilemma of that game, but produces higher order dilemmas. The higher order dilemmas are, in turn, resolved by the status order of a linked cultural structure. It could be said that this kind of m-group has a social structure because its economy has an economic dilemma and has a cultural structure because its social structure has a social dilemma. It is in developing our understanding of m-groups for which n > G > 1 that the three theories are linked. The Game Theoretic economy produces both the dilemma and the motivations necessary to resolve it. For that resolution we apply Elementary Theory to show how a decentralized social structure of norm enforcement regulates economic activity. Given that the economy is NPD, m-group members are motivated to apply norms because increased cooperation benefits all. Since the defection of others lowers every individual's payoff, Elementary Theory asserts that defectors, having no basis for solidarity, stand as individuals in opposition to the collective interests of others. The last step in theory building employs Status Characteristics Theory. That theory builds, a status order which is a cultural system that resolves the social dilemmas of norm enforcement. That theory asserts that status hierarchies can be continually and costlessly in flux. Therefore, by linking norm enforcement to status enhancement and norm violation to status decrease, the social dilemma is resolved without producing second and higher order free rider problems. Now that the m-group model has been developed, future work can investigate its explanatory power. An important direction considers the effects of embedding on solidarity and cohesion. For example, embedded groups are not free to negatively converge on norm violators because the modern state claims the right to the legitimate use of force. The status orders of the larger society can impact on the embedded group interfering with the link between norm enforcement and status enhancement. If so, first, second and higher order free rider problems will emerge that cannot be resolved and solidarity will go to zero. Furthermore, the larger economy, within which m-groups are embedded, offer manifold opportunities to leave that may sharply limit cohesion and thus solidarity. A possible exception of this general weakening of solidarity may be

Building A Model f o r Solidarity and Cohesion Using Three Theories

101

the groups of the privileged classes. If so, that exception helps to explain the stability of large stratification structures.

NOTES 1. Free riding occurs when collective interests and individual interests are opposed, and individual group members abstain from the contributing to the collective good, while still receiving collective benefits. Second and higher order free rider problems occur when the sanctioning, which suppresses first order free riding, produces second- and higher-order dilemmas (Oliver, 1980; Yamagishi, 1986; Heckathorn, 1989). 2. Although never formally defined, Johnsen includes in his theory "cohesion as an important component of solidarity" (1999: 264). 3. See, for example, Cartwright (1968); Schachter et al. (1968); Shaw (1976); Kellerman (Ed., 1981); Drescher et al. (1985). 4. Markovsky and Lawler's cohesion, "the strength and directness of relations among members of a set of actors" (1994, p. 121) is, in our approach, one determinant of the 'degree to which group members stick together' (i.e. cohesion). 5. Cohesion as defined here may be analogous to the binding force between atoms holding molecules together. See Silber (1981) and Sudarshan (1981). 6. We adopt the term 'circumscription' from Carneiro (1970) who employed it in his theory of the origin of the state. 7. Setting D 0 = 1 does not mean a loss of generality. For example, to apply the equations for any D O> 1, rescale D Oby adding D - 1 and rescale C, using Eq. 1. 8. For Turnbull's Mountain People (1972), cooperative hunting was disallowed by the state. Thus S = 0. Because the state forced the tribe to remain on their barren mountain c was very large in a negative direction. That is to say, there was no solidarity, but forced cohesion. These conditions together with extreme poverty are undoubtedly implicated in the social collapse and decay he observed. 9. We adopt the following definition by Heap and Varoufakis, "A 'dominant strategy equilibrium' is one which emerges when the existence of a dominant strategy for at least one of the two players provides a reason for each player to choose a particular strategy" (1995, p. 45). The dominant strategy equilibrium has also been termed the "Nash equilibrium." Rational players will not change from the 'Nash' equilibrium strategy. A feasible solution x is called 'Pareto Optimal' if no other feasible solution y beats or matches x on all objectives and beats x on at least one objective (see Bailey, 1972; Bewley, 1981). 10. Komorita (1976) introduced K', a generalization of Rapoport's K index (1967), to identify NPD games and indicate levels of cooperation. 0 < K' < 1 which is the range for NPD. The prospects for cooperation increase as K' increases. The G analysis agrees with and extends K'. Kt_

G-1 1 + G- 2G/n

As G varies between 1 and n, K varies between 0 and 1. Thus both agree on the range of NPD. Furthermore, as G increases, K' increases both indicating that the prospects

102

DAVID WILLER, CASEY BORCH AND ROBB WILLER

for cooperation are increasing. Not considered by Komorita are K' > 1, but for those values G > n and the game is NPV. On the other hand, Komorita asserted that as K' = > 0 games approach the limits of cooperation, also agreeing the analysis offered. When G < 1 (and K' is negative) cooperation is impossible. l 1. To be complete, though not displayed D 2 = 11 > C 3 = 8. 12. Sahlins (1972) finds case after case of status as the balancing mechanism in reciprocal exchange relations. Those who contribute most relative to services received have the highest status while, for lower status, the relation is reversed. 13. Coleman's idea of norm enforcement by positives requires a transformation like that from 8a to 8b. His idea is not carried further because we know of no example where positives are used to enforce norms. 14. Additionally, many social scientists suggest that status is granted through generosity. Sahlins writes, "that person who does not work and must be given food sinks in esteem, while . . . the generous person rises in esteem" (1972, p. 247). Malinowski notes, "nothing has a greater sway over the Melanesian's mind than ambition and vanity associated with a display of food and wealth. In the giving of gifts, in the distribution of their surplus, they feel a manifestation of power, and an enhancement of personality . . . [t]he desire for the display, the ambition to appear munificent, the extreme esteem for wealth and for the accumulation of food" (1959, p. 31), and "[s]ince again, largesse is a matter of honor and praise, the average native will strain all his resources to be lavish in his measure" (p. 29). 15. Chiefs also gain status as 'gatekeepers' for much needed goods and supplies. In his investigation of Tahiti, Service states, "[t]he chief was a redistributing agent and maintained his household, retainers, and craftsmen by withholding some of the people's surplus production" (1975, p. 158). 16. Status as a solution should be compared to Heckathorn's'hypocritical cooperation' that "can potentially serve as a bridge spanning the chasm from collective inaction to full cooperation" (1989, p. 97, italics removed). Hypocritical cooperators enforce the cooperation of others while avoiding cooperation themselves. Compared to status as a solution, hypocritical cooperation leaves cooperation at the first order suboptimal. Thus we suggest that hypocritical cooperation is called forth when status orders fail to resolve second order problems, as when status structures are fixed.

ACKNOWLEDGMENT The authors thank the National Science Foundation for grants supporting the development of theory found in this paper.

REFERENCES Axelrod, R. (1984). The Evolution of Cooperation. New York: Basic Books. Banks, J., Plott, C., & Porter, D. (1988). An Experimental Analysis of Unanimity in Public Goods Provision Mechanisms. Review of Economic Studies, 55, 301-322. Berger, J., Cohen, B. P., & Zelditch, M. Jr. (1972). Status Characteristics and Social Interaction. American Sociological Review, 37, 241-255.

Building A Model f o r Solidarity and Cohesion Using Three Theories

103

Berger, J., Fisek, M. H., Norman, R. Z., & Wagner, D. G. (1985). Formation of Reward Expectations in Status Situations. In: J. Berger & M. Zelditch, Jr. (Eds), Status, Rewards, and Influence (pp. 215-261). San Francisco CA: Jossey-Bass. Berger, J., Fisek, M. H., Norman, R. Z., & Zelditch, M., Jr. (1977). Status Characteristics and Social Interaction: An Expectation States Approach. New York: Elsevier. Borch, C., & Willer, D. (2001). Battling Coalitions. Paper presented at the American Sociological Association annual meetings. Anaheim, CA. Carey, A. (1967). The Hawthorne Studies: A Radical Criticism. American Sociological Review, 32, 403-416. Carneiro, R. L. (1970). A Theory of the Origin of the State. Science, 169, 733-738. Cartwright, D. (1968). The Nature of Group Cohesiveness. In: D. Cartwright & A. Zander (Eds), Group Dynamics Research and Theory (3rd ed., pp. 91-109). New York: Harper & Row. Coleman, J. S. (1990). Foundations of Social Theory. Cambridge, MA: Harvard University Press. Drescher, S., Budingame, G., & Fuhriman, A. (1985). Cohesion: An Odyssey in Empirical Understanding. Small Group Behavior, 16, 3-30. Durkheim, E. (196011893]). The Division of Labor in Society. G. Simpson (Trans.). Glencoe, IL: The Free Press. Fireman, B., & Gamson, W. A. (1979). Utilitarianism Logic in the Resource Mobilization Perspective. In: M. Zald & J. McCarthy (Eds), The Dynamics of Social Movements. Cambridge, MA: Winthrop. Hansen, K. (1981). 'Black' Exchange and its System of Social Control. In: D. Wilier & B. Anderson (Eds), Networks, Exchange and Coercion (pp. 70-83). New York: Elsevier/Greenwood. Heckathorn, D. (1989). Collective Action and the Second-Order Free-Rider Problem. Rationality and Society, 1, 78-100. Heckathorn, D. (1996). The Dynamics and Dilemmas of Collective Action. American Sociological Review, 61, 250-277. Hechter, M. (1987). Principles of Group Solidarity. Berkeley, CA: University of California Press. Hoebel, E. A. (1967). The Law of Primitive Man. Cambridge, MA: Harvard University Press. Hoebel, E. A. (1960). The Cheyennes Indians of the Great Plains. New York: Henry Holt & Company. Isaac, R. M., Walker, J. M., & Thomas, S. (1984). Divergent Evidence on Free Riding: An Experimental Examination of Possible Explanations. Public Choice, 43, 113-149. Isaac, R. M., McCue, K., & Plott, C. R. (1985). Public Goods Provision in an Experimental Environment. Journal of Public Economics, 26, 51-74. Johnsen, E. C. (1999), Structures and Processes of Solidarity: An Initial Formalization. In: P. Doreian & T. Fararo (Eds), The Problem of Solidarity Theories and Models (pp. 263-302). Australia: Gordon and Breach Publishers. Kagel, J., & Roth, A. E. (Eds) (1995). The Handbook of Experimental Economics. Princeton, NJ: Princeton University Press. Kellerman, H. (Ed.) (1981). Group Cohesion Theoretical and Clinical Perspectives. New York: Grune & Stratton. Komorita, S. S. (1976). A Model of the n-person Dilemma-type game. Journal of Experimental Social Psychology, 12, 357-373. Komorita, S. S., & Parks, C. D. (1996). Social Dilemmas. Boulder, CO: Westview Press. Lawler, E. J. (1992). Choice Processes and Affective Attachments to Nested Groups: A Theoretical Analysis. American Sociological Review, 57, 327-339.

104

DAVID WILLER, CASEY BORCH AND ROBB WILLER

Lizot, J. (1985). Tales of the Yanomami. E. Simon (Trans.). Cambridge, U.K.: Cambridge University Press. Lowie, R. H. (196711948]). Some Aspects of Political Organization Among the American Aborigines. In: R. Cohen & J. Middleton (Eds), Comparative Political Systems (pp. 63-87). Garden City, NY: The Natural History Press. Macy, M. (1991). Chains of Cooperation: Threshold Effects in Collective Action. American Sociological Review, 56, 730-747. Malinowski, B. (1959). Crime and Custom in Savage Society. Paterson, NJ: Littlefield, Adams & Co. Markovsky, B. (1999). Social Network Conceptions of Group Solidarity. In: P. Doreian & T. Fararo (Eds), The Problem of Solidarity Theories and Models (pp. 343-372). Australia: Gordon and Breach Publishers. Markovsky, B., Smith, L. F., & Berger, J. (1984). Do Status Interventions Persist? American Sociological Review, 49, 373-382. Markovsky, B., Willer, D., & Patton, T. (1988). Power Relations in Exchange Networks. American Sociological Review, 53, 220-236. Markovsky, B., & Lawler, E. J. (1994). A New Theory of Group Solidarity. In: B. Markovsky, J. O'Brien & K. Heimer (Eds), Advances in Group Processes (Vol. 11, pp. 113-137). Greenwich, CT: JAI Press. Markovsky, B., & Chaffee, M. (1995). Social Identification and Group Solidarity: A Theoretical Extension. In: B. Markovsky, K. Heimer & J. O'Brien (Eds), Advances in Group Processes, (Vol. 12, pp. 249-270). Greenwich, CT: JAI Press. Marwell, G., & Ames, R. (1979). Experiments on the Provision of Public Goods. II. Resources, Interest, Group Size, and the Free-Rider Problem. American Journal of Sociology, 84, 1335-1360. Marwell, G. (1980). Experiments on the Provision of Public Goods. I. Provision Points, Stakes, Experience, and the Free-Rider Problem. American Journal of Sociology, 85, 926-937. Oliver, P. (1980). Rewards and Punishments as Selective Incentives for Collective Action: Theoretical Investigations. American Journal of Sociology, 85, 1356-1375. Olson, M. (1965). The Logic of Collective Action. Cambridge, MA.: Harvard University Press. Palfrey, T., & Rosenthal, H. (1984). Participation and the Provision of Discrete Public Goods: A Strategic Analysis. Journal of Public Economics, 24, 171-193. Parsons, T. (1951). The Social System. New York: Free Press. Rapoport. A. (1967). A Note on the Index of Cooperation for Prisoner's Dilemma. Journal of Conflict Resolution, 11, 101-103. Ridgeway, C. L. (1982). Status in Groups: The Importance of Motivation. American Sociological Review, 47, 76-88. Roethlisberger, F. J., & Dickson, W. J. (1939). Management andMorale. Cambridge, MA: Cambridge University Press. Roth, A. E., & Murnighan, J. K. (1982). The Role of Information in Bargaining: An Experimental Study. Econometrica, 50, 1123-1142. Roth, A. E., & Schoumaker, F. (1983). Expectations and Reputations in Bargaining: An Experimental Study. American Economic Review, 73, 362-372. Sahlins, M. (1972). Stone Age Economics. Chicago: Aldine Publishing Company. Schachter, S., Ellertson, N., McBride, D., & Gregory, D. (1968). An Experimental Study of Cohesiveness and Productivity. In: D. Cartwright & A. Zander (Eds), Group Dynamics Research and Theory (3rd ed., pp. 192-198). New York: Harper & Row.

Building A Model for Solidarity and Cohesion Using Three Theories

105

Scheff, T. J. (1990). Microsociology: Discourse, Emotion, and Social Structure. Chicago: University of Chicago Press. Service, E. R. (1975). Origins of the State and Civilization. New York: W. W. Norton & Co. Shaw, M. E. (1976). Group Dynamics: The Psychology of Small Group Behavior. New York: McGraw-Hill Book Co. Silber, L. (1981). Group Interactions and Group Behavior in Physics. In: H. Kellerman (Ed.), Group Cohesion Theoretical and Clinical Perspectives (pp. 57-68). New York: Grune & Stratton. Simpson, B., & Macy, M. (2001). Collective Action and Power Inequality: Coalitions in Exchange Networks. Social Psychology Quarterly, 64, 88-100. Simpson, B., & Willer, D. (1999). Part 2: A New Method for Finding Power Structures. In: D. Willer (Ed.), Network Exchange Theory. (pp. 270-284). Westport, CT: Praeger. Skvoretz, J., & Lovaglia, M. (1995). Who Exchanges with Whom: Structural Determinants of Exchange Frequency in Negotiated Exchange Networks. Social Psychology Quarterly, 58, 163-177. Southard, F. (1981). Normatively Controlled Social Exchange Systems. In: D. Willer & B. Anderson (Eds), Networks, Exchange and Coercion (pp. 52~69). New York: Elsevier/Greenwood. Sudarshan, E. C. G. (1981). Cohesion and Physical Structure. In: H. Kellerman (Ed.), Group Cohesion Theoretical and Clinical Perspectives (pp. 123-132). New York: Grune & Stratton. Thye, S. R. (2000). A Status Value Theory of Power in Exchange Relations. American Sociological Review, 65, 407-432. Thye, S., Lovaglia, M., & Markovsky, B. (1997). Responses to Social Exchange and Social Exclusion in Networks. Social Forces, 75, 1031-1049. Turnbull, C. M. (1961). The Forest People. New York: Clarion. Turnbull, C. M. (1972). The Mountain People. New York: Touchstone. Walker, J. M., Gardner, R., & Ostrom, E. (1990). Rent Dissipation in Limited Access Commonpool Resource Environments: Experimental Evidence. Journal of Environmental Economics and Management, 19, 203-211. Willer, D. (1987). Theory. and the Experimental Investigation of Social Structures. New York: Gordon and Breach. Willer, D. (Ed.) (1999). Network Exchange Theory. Westport, CT: Praeger. Willer, D., & Skvoretz, J. (1997). Games and Structures. Rationality and Society, 9, 5-35. Willer, R., Troyer, L., & Lovaglia, M. (2001). Power, Status, and Philanthropy. Paper presented at the American Sociological Association Meetings. Anaheim, CA. Yamagishi, T. (1986). The Provision of a Sanctioning System as a Public Good. Journal of Personality and Social Psychology, 51, 110-116.

106

DAVID WILLER, CASEY BORCH AND ROBB WILLER APPENDIX

This Appendix offers proofs mentioned in the text. First we show that when G = 1 cooperation is impossible. Cooperation is not possible when cooperating gives no advantage, C = D O and when defection always pays more than cooperation, D c -1 > C . The first immediately follows because, when G = 1, C = Do. The proof for Dnc_l > Cnc follows. If there is a D >0 payoff, there is at least one cooperator. Thus G. n c - -

G .n

> 0,

so

1+

c> 1

H

thus always

D~c>o > 1.

n

Furthermore, if there is a C . . . . . . payoff there is at least one defector; thus, n > n c and G.n

c < 1,

thus,

C

< 1.

Therefore, when G = 1, for any group size, there is no possibility of cooperation. If cooperation is impossible when G = 1, then it is also impossible when G < 1. In the economies of any group for which G _< 1, all action will be individual and none cooperative. Then n individuals are a 'group' like n potatoes in a sack are a sack of potatoes. W e now show that the lower bound for NPD games is G = 1 + A, where A is just larger than zero (1 >>> A > 0). If the game is NPD, then C n > D O and D 1 > Cnc for all i. In other words, defection always gives the larger payoff, but, if all cooperate the payoff is greater than when all defect. The proof is that, since C = G and D o = 1, it immediately follows that C,, > D o. For D,c -i > Cnc

1+

(l+A)(n //

-1)

>

n (I+A) " ///

thus

I+A 1 - - ?/

> 0

which is true because n > 2 and A is just larger than zero. Therefore, the lower bound for PD games is G = 1 + A. W e now find the upper bound for NPD games - for games where cooperation is possible, but not perfect. This is shown in two steps. First we show that when G = n - A the game is NPD. As before, to prove are C n > D o and D c 1 > C , . Again, since C n = G and D O= 1 = > C n > D 0. For Dnc _~ > C c,

Building A Model for Solidarity and Cohesion Using Three Theories 1+

G . ( n c- 1)

G •n

>

substituting, 1 +

n

(n-A)(n

-1)

107 >

(n-A) n

n

n

thus A/n > 0 The second step begins with the observation that just below it is shown that when G > n the game is not NPD but N-Person Privileged Game (NPV) where cooperation is the dominant strategy. Comparing that proof to the one just given it is transparent that when G = n the game has no dominant strategy and cannot be either NPD or NPV. Thus the boundary between NPD and NPV is at G - n and the upper bound for NPD is G = n - A. Since it was already found that the lower bound for NPD is G = 1 + A, the game is NPD for any G for which n > G > 1. An N-Person Privileged G a m e (NPV) is a game in which all players always prefer cooperation to defection - that is to say, a game when any player will have a higher payoff when switching from D to C and conversely. W e propose that, when G > n the game is NPV. The proof is as follows. The NPV game is satisfied by C,, > D o and C,,C> D ~ with the latter indicating that any p l a y e r ' s payoff is greater when switching to D from C and conversely. Again, since C , , = G a n d D o = I = > C , ~ > D O. F o r C , , , >D,~, i,

G. n H

' > 1+

G . ( n - 1) n

subtracting one, multiplying - -

G. n

- 1> - -

G. n

G

/1

//

/1

which is true because G > n. Therefore, when G is larger than n, the game is privileged and all players always cooperate.

STATUS, EMOTION, AND THE DEVELOPMENT OF SOLIDARITY IN STRATIFIED TASK GROUPS Jeffrey A. Houser and Michael J. Lovaglia

ABSTRACT Explaining the development of group solidarity in status-differentiated groups is an interesting theoretical problem because solidarity is usually considered to stem from positive affect in groups of status equals. Analysis of the evolution of human emotions allows the development of social theory focused on the functions of emotions. Human emotions evolved in tandem with the development of status hierarchies as a dominant form of social organization that function to coordinate the work of individuals in groups. Conflicting emotions generated by inequality and intragroup competition interact in status hierarchies to maintain group solidarity. An experimental test using a direct indicator of group solidarity is proposed.

INTRODUCTION The c o m m o n observation that task groups form stable status hierarchies poses a difficult theoretical question: How does group solidarity develop in a status hierarchy? If the emotional basis for group solidarity is positive affect and

Group Cohesion, Trust and Solidarity, Volume 19, pages 109-137. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2

109

110

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

fellow feeling stemming from equality among group members, then the inherent inequality and competition of a status hierarchy might limit group solidarity. Yet solidarity is common in task groups characterized by status differences and internal competition. For example, think of a winning sports team or the research and development work groups that created the Macintosh computer. High-status members expected to contribute to the group's success have more opportunities, reap more rewards, and command more influence than do lower-ranked members. Thus high-status members may well feel the positive emotions conducive to group solidarity, whereas low-status members might resent their own lack of opportunity, rewards, and influence. Why don't conflicting emotions produced by inequalities in status-differentiated groups prevent the development of group solidarity? To restate the question more concretely in a way conducive to empirical test, why doesn't the anger and frustration of low-status members build to the point where they leave the group? Working at the intersection of three research areas in group processes emotion, status, and group solidarity - complicates theoretical development. To understand how emotions relate to social organization (status hierarchies and group solidarity) we first need a theory of emotion that goes beyond the common dichotomy of emotion as positive or negative. It became apparent that two negative emotions could have quite different effects on group solidarity. For example, anger might split a group apart, but fear might push group members together. And some emotions are difficult to categorize as positive or negative. Pride, for example, can be considered positive when felt by an individual but negative in terms of its impact on relationships among group members. Analyzing the evolution of emotion allows the development of social theories of emotion that emphasize the particular functions of different emotions. The following sections present the evolutionary basis for emotions that ties them to the development of human social organization. Parallel to the evolutionary analysis of emotion, we propose that the developing cognitive capacity of humans, especially the cognitive control of emotional expression and the ability to attribute the source of their emotional reactions, allowed status hierarchies to develop as a functionally effective form of social organization. The functional presuppositions of status characteristics theory are validated in that status hierarchies form to effectively coordinate the work of groups of relatively autonomous individuals. Then, having explained how emotions evolved to promote group solidarity in status hierarchies, we propose an empirical test of our theory that includes a direct measure of group solidarity.

Status, Emotion, and Development of Solidarity in Stratified Task Groups

111

AN EVOLUTIONARY FOUNDATION FOR THEORIES OF EMOTION IN GROUPS Recent research in the social psychology of emotions has progressed through categorizing emotion as positive or negative. This essentially psychological view of emotions as feelings that are positive or negative to the individual neglects the fundamentally social functions of emotions in humans. In two extraordinary books, Jonathan Turner establishes the evolutionary basis for human emotions that emphasizes the role of emotions to promote social organization (Maryanski & Turner, 1992; Turner, 2000). Building on Turner's work, we explain how basic human emotions, some positive and some negative, interact to stabilize status hierarchies and promote group solidarity among members of different social ranks. Turner (2000) proposes that distinctively human emotions and human control of emotional expression evolved in response to competition for survival produced by climate changes in the African continent during the period when human groups first separated from their more ape-like ancestors. The evolutionary split between humans and apes occurred 5-6 million years ago when the evolutionary line that produced homo sapiens split from the line that produced chimpanzees. The fossil record shows human evolutionary change proceeded quickly from that point but that chimpanzee evolution was slow. Thus the last common ancestor (LCA) of apes and humans resembled a modern chimpanzee. Maryanski and Turner (2000) studied apes to form a picture of the emotional make-up of our last common ancestor. Chimpanzees as well as the other apes exhibit a narrow range of emotions compared to humans. Ape adults are autonomous individuals with few strong social bonds except between mothers and their offspring. Perhaps eight million years ago when our ancestors lived primarily in the protected environment of large trees, there was little need for strong social bonds or group solidarity. Rapid climate change 6-8 million years ago dried out much of Africa. Dense forests yielded to grassy plains dotted by patches of woodland. To survive, the LCA adapted to living on the ground and foraging more widely for food. Individuals on the ground were more exposed to large predators than individuals had been when living in the trees. Banding together afforded some protection, but the LCA lacked the emotional repertoire needed to form solidary social groups (Turner, 2000). Two major changes would facilitate human survival in a life spent foraging for food on the ground. First, emotions would come under cognitive control

112

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

(Turner 2000). Uncontrolled emotional outbursts would alert predators. On the ground, some distance from the haven of a large tree, a noisy ape would be noticed and eaten. Cognitive control of emotions separated two important aspects of emotion in humans: Emotions expressed can be different from emotions experienced. A human can feel fear but display anger or even happiness (see Hochshild, 1983). Second, the human emotional repertoire would expand to facilitate the formation of strong social bonds needed for group solidarity. Humans probably inherited the primary emotions from the LCA: fear-aversion, assertion-anger, happiness-satisfaction, and sadness-disappointment (Turner, 2000). Recent research suggests that these primary emotions are hardwired into the older parts of human neuroanatomy. Evolution through natural selection has to work on variations that already exist in a species' genetic makeup. Turner (2000) proposes that a wide range of human emotions evolved by combining the primary human emotions. For example, pride combines happiness over personal accomplishment and fear of failing to maintain that level of accomplishment in the future. Shame is disappointment and anger with self combined with fear of the consequences to self. Turner (2000) supplies the theoretical foundation for long-standing social psychological theories of emotions wherein a variety of emotions are made from combining primary emotions. (Kemper, 1978, 1987; Plutchik, 1962, 1980). In Turner's (2000) evolutionary theory of the social psychology of emotions, humans evolved the ability to control emotional expression and to use emotions to build group solidarity for protection. We further propose that by controlling emotions, humans evolved the ability to increase the productivity of human groups through elaborated social hierarchies. EMOTIONS

AND HUMAN

SOCIAL

ORGANIZATION

Early humans faced two fundamental social problems: (1) autonomous individuals had to tolerate the frequent social interaction required when banding together in groups for protection; and (2) social interaction had to be organized to coordinate effectively the actions of individuals living in groups. According to Turner (2000), human emotions evolved to facilitate affective bonds between individuals that promote group solidarity. Human emotions may also function to facilitate status hierarchies, a fundamental form of human social organization.

Emotions and Group Solidarity Early humans, exposed on the woodlands and more open terrain of the African savanna, had to band together in groups for protection. Isolated individuals were

Status, Emotion, and Development of Solidarity in Stratified Task Groups

113

easy prey for large predators and bands of rival humans (Wrangham, 1996). Happiness-satisfaction is a positive emotion not only from the individual's point of view, but also in terms of promoting group solidarity. Kemper (1984) sees positive emotions as integrating, motivating behaviors that bind people together in a group. In contrast, negative emotions are differentiating, motivating behaviors that increase the separation of group members. Turner (2000) proposes that the expansion in human capacity to feel happiness-satisfaction allows humans to feel more pleasure in the company of others than is possible for apes. Humans developed the capacity to form a wide array of close social bonds, countering the autonomous, individualistic nature of the LCA. Mutual exchange evolved as a universal characteristic of human society in part because it" promotes positive affective bonds between individuals (Turner, 2000). The mutual benefit provided by exchange produces satisfaction that in an ongoing exchange relationship encourages loyalty (Kollock, 1993; Lawler & Yoon, 1993, 1996, 1998). The emotional benefits of exchange, however, occur primarily in the context of social equality between trading partners. Lawler and Yoon (1993, 1996, 1998) demonstrate that unequal power among trading partners can reduce the frequency of exchange thus reducing positive affect derived from it. Wilier, Lovaglia and Markovsky (1997) showed that low-power trading partners experience negative emotion when they have no choice but to exchange at a disadvantage with a high-power trading partner. Positive emotions conducive to group solidarity are more likely to occur under conditions of social equality in homogeneous groups. Theorists have proposed that social equality promotes the positive emotion necessary for group solidarity. The more homogeneous and less stratified the group, the more solidary it will be (Markovsky & Lawler, t994). Status hierarchies developed as a fundamental mechanism of human organization despite the social inequalities that result, perhaps because status differences allow more efficient coordination of work necessary for individual survival. The repertoire of human emotions may have expanded in part to enhance the solidarity of stratified groups. Turner (2000) proposes that humans developed a wide range of emotions to counter the disruptive effects of the primary negative emotions, fear-aversion and anger-assertion. When group members disagree, perhaps competing over resources or mates, the anger of one produces fear and perhaps counter-anger in the other. The behavioral response to the negative primary emotions of anger and fear is to attack or flee. Either response disrupts the group. More subtle emotions such as remorse may have evolved to help repair relationships broken by anger and fear. Remorse following an angry outburst promotes conciliatory behavior conducive to renewed social relations.

114

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

The cognitive control of emotions that Turner (2000) proposes evolved along with the wide range of human emotional expression may also have contributed to social solidarity in status hierarchies. Cognitive control of emotions allows individuals to choose emotional responses that further group goals. Status hierarchies are highly stable in task groups with a common goal (Berger, Fisek, Norman & Zelditch, 1977; Ridgeway & Berger, 1988). Lowstatus individuals, while less satisfied than those with high-status, generally accept the status hierarchy of the group. Through a complex choreography of emotional response and expression, human individuals in stratified groups balance emotional reactions to maintain solidarity in status hierarchies (Houser & Lovaglia, 1994; Lovaglia & Houser, 1996). 1

Evolution and the Development of Status Hierarchies Status hierarchies are a fundamental form of human social organization that facilitates the coordination of individual effort to accomplish a common goal. Status hierarchies are an elaboration of the dominance orders found in many species from cockroaches to chickens to macaques and baboons (Cloutier & Newberry, 2000; Moore, Reagan & Haynes, 1995; Suzuki, Hill & Sprague, 1998; Virgin & Sapolsky 1997). A dominance order forms when individuals compete for resources needed to survive: food and opportunities to mate. Striving for high status, then, could have developed as a natural outgrowth of the desire to survive and propagate. A strong, fast, aggressive individual can intimidate others into relinquishing resources. (A commonly used indicator of dominance rank is the ability of one individual to force another away from its food.) In dominance orders, individuals use aggressive displays to establish their place in the pecking order. Status hierarchies are subtler, however, requiring a sense of common purpose. In status hierarchies, rather than attempting to intimidate each other, individuals vie to create the impression of competence. High status is achieved when group members expect that an individual will make extraordinary contributions to group goals (Berger, Cohen & Zelditch, 1972; Berger, Fisek, Norman & Zelditch, 1977). 2 Other group members defer to high status individuals expecting the contributions made by competent members will benefit all. Status hierarchies limit the effectiveness of dominance tactics as the group resists the dominance attempts of low-status members (Ridgeway & Berger, 1986; Ridgeway, 1987; Ridgeway, Johnson & Diekema, 1994). Status hierarchies require more mental capacity than do dominance hierarchies. Among more recently evolved primates, high status members

Status, Emotion, and Development of Solidarity in Stratified Task Groups

115

perform service and control functions for other members and for the group as a whole (Mazur, 1973). The emergence with increased mental capacity of true status hierarchies from the foundations of dominance orders can be seen in the social behavior of baboons. Baboons are widespread and populous large monkeys. Male baboons compete for opportunities to mate with female baboons. A dominance hierarchy forms characterized by displays of aggression and submission and occasional fights between individuals (Smuts, 1985). Actual fights may be uncommon because fighting individuals are vulnerable to the opportunisitic attacks of third parties. Baboons use the strategy of Tertius Gaudens proposed by Simmel. At the end of a fight between two high-ranking males, a lower-ranked male can sometimes step in to finish off the exhausted victor. Baboons also exhibit cooperative behavior that foreshadows human status hierarchies. Successful courtship behavior for males often includes establishing a friendly relationship with and providing services for eligible female baboons (Smuts 1985). Males may also solicit other males as allies in their courtship behavior, perhaps in exchange for reciprocal help in the future. When a male attempts to consort with a female who has been with a highranking male, the interloper may bring along an ally or two to act as lookouts and interfere with the high-ranking male should he discover the consorting pair. High-ranking chimpanzees control food distributions sharing out portions for low-status group members (De Waal, 1996). Among bonobos (close relatives of chimpanzees), females achieve higher status than males through their food sharing strategies (Hohman & Fruth, 1996). Thus, in species with relatively large mental capacity, the ability to form relationships with others and provide services becomes important for reproductive success and a component of status rank. At some point in human evolution, increased mental capacity allowed humans to realize that individual survival depends on group success, paving the way for stable status hierarchies. 3 Along with mental capacity, stable status hierarchies require that relatively autonomous individuals realize more benefits from staying in the group than from leaving it. The increasingly dangerous environment of the LCA provided some incentive toward group cohesion. Individuals leaving a group were vulnerable to predators. A sub-group, however, could leave with relative impunity, weakening the group. Thus it is in the interest of high-ranking members to keep low-ranking members from becoming disaffected and leaving. Positive emotion expressed by high-status group members can help to increase the satisfaction and happiness of potentially disaffected low-status group members. High status and low status, then, became associated with different characteristic emotional experiences and expressions.

116

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

Emotional Experience in Status Hierarchies We propose that primary emotions as well as more subtle human emotions have become adapted to enhance group solidarity within social hierarchies. To bind low-status members to the group, it is in the interest of high-status members to display their positive emotions and attempt to mollify the resentment and other negative emotions typically felt by disaffected low-status members. High rank is associated with positive emotion and low rank with negative emotion in most primates (Mazur, 1973, 1985). Symptoms of stress in low ranking members have been observed in a wide range of species from chickens (the pecking order) to baboons (Mazur 1973, Virgin & Sapolsky, 1997). Among bonobos, high status is associated with being groomed more often by others and other forms of social support (Vervaecke, De Vries & Van Elsacker, 2000). Increased stress and poorer health have also been associated with low social status in humans (Brunner, 1997; Clark, Anderson, Clark & Williams, 1999; Cooper, 1993). High social status is conducive to feelings of satisfaction and happiness while low-status is conducive to feelings of fear, anxiety, resentment and anger (Lovaglia & Houser, 1996, Stets, 1997, Stets & Tsushima, 2001). Note, however, that while high-status individuals may typically feel more positive emotion, social norms may also allow high-status individuals to express more negative emotion, especially anger (Ridgeway & Johnson, 1990). Lovaglia and Houser (1996) used status characteristics theory to develop their theory of status-compatible emotions in task groups. In status characteristics theory, high status confers benefits on group members who are given more opportunities to perform, perform more, receive higher evaluations for their performances, and have more influence over group decisions. 4 Lovaglia and Houser reasoned that given the advantages of high status, high-status group members would typically feel more positive emotion. In contrast, low-status group members whose occasional opportunities to contribute were ignored or devalued and who had little influence would typically feel more negative emotion, primarily anger and resentment. 5 Positive emotion, then, is compatible with high-status while negative emotion is compatible with low status. 6

Status-Compatible Emotions Theory Using Kemper's (1984) insight that positive emotions are integrating while negative emotions are differentiating, Lovaglia and Houser (1996) proposed that the positive emotion compatible with high status interacts with the negative emotion compatible with low status to stabilize the degree of status stratification in a group.

117

Status, Emotion, and Development of Solidarity in Stratified Task Groups

In the absence of emotions, status characteristics theory implies that status differences are self-reinforcing. Because high-status group members have more opportunities to perform and receive higher evaluations for their performances than do low-status group members, over time expectations for the contributions of high-status members should increase while expectations for the contributions of low-status members should decrease. That is, behavior interchange patterns in status-differentiated groups should exacerbate existing inequalities of opportunity and reward (Fisek, Berger & Norman, 1991). Figure 1 represents this process. Between Time 1 and Time 2, the opportunities and rewards of a high-status group member are shown to increase while the rewards and opportunities of a low-status group member decrease. Status-compatible emotions theory proposes that typical emotional reactions of high-status and low-status group members moderate the effects of behavior interchange patterns to stabilize inequalities of opportunity and reward. High-status group members feeling positive emotion will typically act to bind the group together, encouraging low-status members to remain in the group. In Opportunites and Influence High HiStat HiStat

"

HiStat

\ I

\

HiStat

LoStat LoStat -..x..x

/

/

LoStat ~x

LoStat

HiStat

/

"x'~

LoStat

Low

Time 1

Time 2

Time 3

Time 4

Time 5

Fig. 1. Relative Status of Group Members Over Time as Emotional Reactions Become More or Less Status-Compatible.

118

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

contrast, low-status group members feeling negative emotion will typically act to differentiate the group, separating themselves from other group members. One way for high-status group members to encourage those with low-status to stay would be for high-status group members to provide more opportunities and defer more to those of low status than they otherwise would. One way for lowstatus group members to distance themselves from the group would be to decrease the opportunities and deference usually accorded those of high-status. Figure 1 represents this process. Between Time 2 and Time 3, the opportunities and rewards of a high-status group member are shown to decrease while the rewards and opportunities of a low-status group member increase. A key indicator of status differences is the deference accorded high status individuals by those of lower status. When group members expect that an individual is highly competent to contribute to the successful solution of a group task, then they generally defer to that member's suggestions for group action. Thus, the more influence that a person has over others, the higher that person's status. 7 The net result, then, of high-status group members increasing the deference they accord low-status members and of low-status group members decreasing the deference they accord high-status members would be to reduce status differences in the group. Earlier research suggests that emotion can affect deference, the willingness of a person to accept the influence of another person. Baron (1987) found that interviewers in a negative mood evaluated job applicants more negatively. And, Mackie and Worth (1989) found that research participants in a positive mood were less critical of persuasion attempts than were participants in a neutral mood. Thus a low-status group member may have more influence over a highstatus member who feels positive emotion compatible with her high status. And, a high-status person may have less influence over a low-status person who feels negative emotion compatible with her low status. The net result would be the reduction in the difference in influence between the two group members proposed by Lovaglia and Houser (1996). If emotions remained constant during group interaction, with high-status members feeling more positive emotion and low-status members feeling more negative emotion, then we might expect that influence differences would continue to shrink. Emotions change, however, as group members relative influence changes. Kemper (1991) found that status loss was typically accompanied by negative emotion, resentment and anger, while status gain was typically accompanied by positive emotion, joy or happiness. We would expect that high-status group members might resent the loss of deference accorded them due to the negative emotion typical of low-status members. And, we would expect that low-status members might feel happiness

Status, Emotion, and Development of Solidarity in Stratified Task Groups

119

and satisfaction' due to the increased deference accorded them due to the positive emotion typical of high-status members. The net result of these statusincompatible emotions would be a re-widening of the influence gap between high-status and low-status group members. Figure 1 represents this process. Between Time 3 and Time 4, the opportunities and rewards of a high-status group member are shown to increase while the rewards and opportunities of a low-status group member decrease. The widening status gap would then produce status-compatible emotions that would serve to re-narrow the status gap as Fig. 1 shows between Time 4 and Time 5. The result of this reciprocal interplay between status-compatible and status-incompatible emotion within the group would be a stable status hierarchy.

Evidence for Status-Compatible Emotions Lovaglia and Houser (1996) tested status-compatible emotions theory by inducing positive or negative emotion in a face-to-face classroom setting where students worked together in task groups, and in controlled, laboratory setting. In the first study, students were randomly assigned to a high-status, leadership position in their work group by giving them an unusually high score on an aptitude test. As predicted, students randomly assigned high status reported more positive emotion than did students not assigned high status. Then, in a laboratory study, Lovaglia and Houser (1996) induced positive emotion by having a partner give participants a small gift of candy with a friendly note. Negative emotion was induced when a partner refused to give participants a gift, sending only a dismissive note. Participant and partner then participated in group task in which the participant had repeated opportunities to accept the influence of the partner. Partners consistently had more influence when the participant reported positive emotion than they did when the participant reported negative emotion. Lovaglia and Houser (1996) also varied the initial status of participant and partner. Participants were undergraduate women randomly assigned to partners portrayed as graduate students in biomedical engineering extremely high gradepoint averages, or to partners portrayed as high school students with extremely low grade-point averages. Thus the participant was high status relative to her high-school student partner or low status relative to her graduate student partner. Results supported the proposition that emotions combine with status characteristics to produce large difference in influence when emotions are incompatible with status. The partner's high status combined with the participant's positive emotion gave the partner the most influence over the participant. The partner's low status combined with the participant's negative emotion gave the partner

120

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

the least influence. Also as predicted, no differences in influence were found when emotions were compatible with status. The influence of a high-status partner over a participant experiencing negative emotion was similar to the influence of a low-status partner over a participant experiencing positive emotion. In a replication that also extended the theory, Lovaglia and Houser (1996) obtained similar although predictably weaker results when positive or negative emotion was directed at someone other than the partner in the group task. We conclude that status-compatible emotion reduces influence differences between high status and low status members of work groups; and, status-incompatible emotion increases influence differences. Because deference is a key aspect of status, we conclude that decreased influence differences would decrease status differences. 8 SOLIDARITY

IN STATUS

DIFFERENTIATED

GROUPS

Status-compatible emotions theory can be extended to show how status and emotion interact in task groups to produce group solidarity. Recall that equality among group members has been proposed as one of the key factors promoting positive emotion and group solidarity (Markovsky & Lawler, 1994). How can solidarity develop in status-differentiated groups? Given the choice to leave, why would low-status group members stay in a group where their contributions are devalued and ignored? Table 1 represents the expected outcomes of status differentiation when only task relevant (non-affective) dimensions of group interaction are considered. We propose that the attributions actors make for their relative status position affect their assessment of the legitimacy of the status structure. Left unchecked, the legitimation of the group's status stratification may erode its solidarity, in that legitimacy is likely to exacerbate the differences between group members. Essentially we assume that greater differentiation between group members weakens the group's solidarity. Kollock's (1993) findings that homogeneous groups display greater cohesion than heterogeneous groups support this assumption. Assessment of the legitimacy of a group member's initial status position is dependent upon the type of attribution made to support the status differentiation. High status actors who make internal attributions for their relative success in the group's power and prestige order are likely consider their high status as a sign of validation for those characteristics which differentiation them from other group members. For example, if a high status male considers gender to be a legitimate dimension upon which expectations of competency can be

Status, Emotion, and Development of Solidarity in Stratified Task Groups

~

~1~. ~

~

i

o r~

~

.>

z 0

~

°

co

e~ o

,&

z

©

0

0

<~ ,.~

<~

o "c9

°

©

G

~-q

cq

0 o

r'r.

o

e. o

"N

.o

>

o

<

"4

.1

O~

<

0

t e.

,v

o

121

122

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

accurately formed, then he will consider his relative advantage over women in the group as valid. As such, his first order expectations (I'm a man, and this is a male-oriented task therefore I should be in charge) should lead to greater resistance to influence attempts by female members of the group due to the legitimation of gender. His gender-oriented resiliency is likely to disaffect and distance female members of the group. Given few opportunities to perform, and fewer positive evaluations for their performance, female group members are likely to exit the group in search of more egalitarian social settings. Alternatively, high status members who make external attributions for their relative advantage may realize that their elevated position in the group's power hierarchy is not valid. In this situation first-order expectations are incongruent with the reflected appraisals of the other members of the group (second-order expectations). Research by Lisa Troyer and Wes Younts (1998) found that is situations in which second order expectations conflict with self-perceived firstorder expectations, women will act as if the second order expectations are legitimate. That is, low-status female subjects, who were led to believe their high-status partner erroneously transposed their relative status positions, acted as if the expectations of the partner were more legitimate then those they held for themselves. Low-status female subjects took on the mantle of leadership within the group and were more likely to resist the influence of the high-status partner than were low-status females whose first and second order expectations did not conflict. McGovern and Houser's (2001) replication of Troyer and Younts' experiment found these results generalize to males as well, provided the level of risk in "passing" (Goffman, 1959) is low. Walker et al. (1984) identify this type of legitimation as an endorsement of the status structure. Here the evaluation of "perceived" subordinate peers within the group reinforces the group's collective assessment of its social structure. Endorsement should lead high status actor's to increase their resistance to influence attempts from other group members. Once endorsed, the high status actor may behave in a manner that attempts to validate their status rank. Endorsement should then lead to greater separation between status positions and threaten the solidarity of the group. High status group members may attempt to "live up to" the elevated expectations of their peers and in so doing alienate the very people who endorsed them in the first place. Here again the actions of those high in the group's status structure increase the differences between group members and threaten the solidarity of the group. Low status actors who make internal attributions for their low social standing in the group's hierarchy are likely to accept the social structure as "proper." Walker and Zelditch (1993) define propriety as the acceptance of social order because it is perceived as being morally correct. Walker and Zelditch assume

Status, Emotion, and Development of Solidarity in Stratified Task Groups

123

that low status actors who accept a social order on these terms will voluntarily comply with others who are higher up in the group's social hierarchy. As such the status gap between low status and high status group members will once again widen, however in this situation it is the actions of the low status actor that drive group members apart. The situation for low status group members who make extemal attributions for their position in the group is markedly different from low-status actors who have made dispositional attributions for their subordination in the group's status structure. Here the situation is akin to high status actors who have made extemal attributions for their success. That is, the first and second order expectations of these group members do not coincide. In this situation the low status group member has higher expectations for her/his own ability than those shared by other members of the group. Once again research by Troyer and Younts (1998) and McGovern and Houser (2001) provide evidence that in situations where self-appraisals conflict with reflected appraisals, group members who should be higher in status acquiesce with the expectations of their peers. In so doing they are more likely to accept the legitimacy of the group's status structure due to its authority (Walker & Zelditch, 1993). Authority legitimates a social hierarchy when high status members impose order on subordinates. Authorized legitimacy of the status structure could likely lead to greater differentiation between group members and perhaps the disintegration of the group. Without of an affective reaction to their situation low status group members who have made external attributions for their relative subordination are unlikely to challenge the legitimacy of the group even when it is in their own and the group's best interest (Markovsky, 1986). In each of these four situations the response of a group member distance him or her from others in the group. The exacerbation of status differences should threaten the group's solidarity unless some other factor intervenes in this process. One possibility is that there are utilitarian reasons for low-status group members to maintain an affiliation with those of high status. 9 Status hierarchies are found in task groups where all members, including those with low status, benefit from achieving group goals. Because status is based on expectations that members have for each other's contributions to achieving group goals, highstatus members are valuable assets. Some of the solidarity and stability of status hierarchies can be attributed to the desire of low-status group members to belong to a successful group. Depending on the potential reward for success, low status in a group that achieves its goals may be preferable to high status in a group that fails (Frank, 1985). Altematively, the interplay of status-compatible and status-incompatible emotion in task groups may also bond low-status members to the group. Houser

124

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

and Lovaglia (1994) used status-compatible emotions theory to extend Markovsky and Lawler's (1994) theory of group solidarity to explain solidarity in status-differentiated groups. Houser and Lovaglia propose that statuscompatible emotions limit the degree of negative emotion felt by low-status members, reducing their incentive to leave the group. Recall that the many benefits of high status produce the typically positive emotions experienced by those in high-status positions. Status, however, is relative. Without low-status group members to offer the prestige, honor and deference characteristic of high status, the benefits of high status vanish. Thus it is in the interest of high-status group members to increase the happiness and satisfaction of low-status members, encouraging them to remain in the group. One way to encourage the continued presence of low-status members is to allow them increased influence over group decisions. But because status is relative, the increased influence of low-status group members inevitably results in the decreased influence of high-status members. Such loss of influence would produce negative emotion in high-status members, potentially damaging group solidarity. High status members, however, have strong utilitarian reasons to remain in the group and to encourage positive emotion in low-status members. Thus, in status-differentiated groups, a reduction in positive emotion felt by high-status members does not reduce group solidarity. The potentially disrupting effect of negative emotion typically felt by lowstatus group members is also mitigated in status-differentiated groups. Anger and resentment expressed by an individual can produce counter-anger in others (Turner, 2000). The resulting escalation of negative emotional responses, as an angry outburst is met with an even angrier response, could easily disrupt group solidarity. Houser and Lovaglia (1994) propose that such counter-anger in high-status group members to resentment expressed by low-status group members is unlikely. The attenuation principle of status characteristics theory suggests that high-status group members may discount the negative emotion expressed by low-status group members. Attenuation occurs when information about an additional low-status characteristic is combined with previously held expectations for that individual's low capacity to contribute to the group. An additional status characteristic consistent with existing expectations has less impact on influence than would an additional inconsistent status characteristic (Berger, Fisek, Norman & Zelditch, 1977; Webster & Foschi, 1988). Because Lovaglia and Houser (1996) showed that emotional reactions combine with status characteristics to determine influence, we can predict that the negative emotion expressed by lowstatus group members would have less effect on their influence than would positive emotion expressed by low-status members. If the negative emotion

Status, Emotion, and Development of Solidarity in Stratified Task Groups

125

expressed by low-status group members has little negative effect on their influence, then the negative emotion of low-status members would have little disruptive effect on group solidarity (Houser and Lovaglia 1994). These predictions, as well as their implications for the legitimacy of the group's status structure and the resultant solidarity of the group, are summarized in Table 2. By broadening status-compatible emotions theory to incorporate the manner in which the attributions group members make for their status position intertwine with their emotional reactions we are able to trace how and when emotions either foster or fracture group solidarity. We assume that high status group members who make internal attributions for their status advantage feel statuscompatible positive emotional reactions such as pleasure or satisfaction. If this is the case then it is likely that these group members over time will become satiated with the rewards and demands of high status and ease down their authority within the group. As Walker and Zelditch (1993), validity-based legitimacy provides high status group members with the luxury of not having to demand compliance in each and every group decision. As noted earlier, the combination of high status and positive emotion opens the door to allowing low status partners greater influence in the group's decision making. High status group members who experience positive emotions should decrease their resistance to their partner's influence. High status group members who make external attributions for their status advantage should experience a status-incompatible reaction to their elevated position in the group's status structure. Recall these are individuals whose second order expectation (those of other group members) conflict with their first-order expectations (those held for self). In such situations high status group members should experience guilt (Homans, 1961; Kemper, 1984) due to an over-reward in the group's status hierarchy, or anxiety and fear of having their true competency exposed (Goffman, 1959). In an effort to maintain the deception and pass themselves off as truly deserving of the status they have been accorded, they are likely to engage in actions that support their social ranking. That is, we would expect that these status-incompatible negative emotions to increase their resistance to the suggestions of other group members. Having already achieved high status in the group, their emotional reaction to the situation could exacerbate the status gap within the group and jeopardize the group's solidarity. Low status group members who make internal attributions for their relatively low levels of prestige in the group could respond in one of two manners. Either they could feel status-incompatible positive emotion of satisfaction or statuscompatible negative emotions of disappointment and shame. Either emotional reaction is likely to increase the stratification between group members. Those

126

J E F F R E Y A. H O U S E R A N D M I C H A E L J. L O V A G L I A e'~

"0 ¢9

2

o

r~

z

O

0

t~

o ¢9

i° "o

o

O

OD

1::

"r" O

.o

o

,e

e~

<

Z

.~

~9

"~ ~

O

~9

z

<

o

~z

T

z

T .g o ,..3

z

Status, Emotion, and Development of Solidarity in Stratified Task Groups

127

who are satisfied with their subordinate position are likely to free ride on the successes of the group. Having to compensate for the free-riders will likely alienate high status group members who are forced to carry the load. Those low status group members who experience shame in their stanchion may withdraw from the group and in so doing compromise its solidarity. Finally, low status group members who make external attributions for their social status are likely to experience status-compatible negative emotions such as anger or disgust. As discussed previously, the effect of these reactions is mitigated within the solidarity of the group in two ways. First, high status group members are likely to attenuate the negative emotional reaction of their low status counterparts due to its consistency with their expectations for lower status positions. That is, high status members expect low status members to be upset and as such discount the negative emotion from their less privileged counterparts. Alternatively, the reactions of the low status group members themselves may decrease the status differentiation within the group by increasing their resistance to the influence of their superiors. In each case the status-compatible negative emotion fosters group solidarity.

Influence and Solidarity in Groups with Status-Compatible Emotions To test for attenuation in the effects of negative emotion experienced by lowstatus group members, Houser and Lovaglia (1994) replicated and extended the gift-giving laboratory study conducted by Lovaglia and Houser (1996). If attenuation does occur when low-status group members express negative emotion, then a low-status member's influence would drop little after the lowstatus member has angered a high status member. In contrast, a low-status group member's influence should increase after the low-status member makes a high-status member happy. In accepting the influence of a low-status group member, would a high status group member discount the anger she feels when insulted by the low-status member and augment the happiness she feels when given a gift? Results from three experimental conditions in Houser and Lovaglia's (1994) study directly test for the attenuation effect on the influence of low-status group member who express negative, neutral or positive emotion toward a high-status group member.

Status-Compatible Emotion Condition: A high-status participant is confronted with the negative emotion of a low-status partner who refuses to give a gift.

128

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

Neutral Emotion Condition: A high-status participant has no indication of the emotion of the partner. Status-Incompatible Condition: A high-status participant is given a gift by a low-status partner. The participant then works with the partner in the standard Contrast Sensitivity experimental setting designed to measure the influence of the partner on the participant) ° Results of the Contrast Sensitivity setting are expressed as a proportion of trials on which the participant resisted the influence of the partner, P(s). Ranging from zero to one, a high P(s) statistic indicates that the partner had little influence over the participant, while a low P(s) statistic indicates more influence. Preliminary results gave some support to the prediction that a high-status group member would discount the negative emotion expressed by a low-status group member in the status-compatible emotion condition. With 10 undergraduate women participating in each condition, mean P(s) for the status-compatible emotion conditions was 0.70 (SD =0.10). While high, this result was quite similar to mean P(s)=0.68 (SD=0.10) in the neutral emotion condition. Resistance to influence in both these conditions, however, was substantially higher than it was in the status incompatible condition in which the participant experiences positive emotion expressed by the partner, mean P(s)=0.59 (SD = 0.16). A second study failed to replicate the effect, however. Predicted differences in influence are small, and the standard experimental setting used in status characteristics research may not be the best way to investigate the effects of status-compatible or status-incompatible emotions on solidarity.

DESIGNING A DIRECT TEST OF SOLIDARITY IN STATUS-DIFFERENTIATED GROUPS The next step is a test of the proposition that status-compatible emotions enhance solidarity in status-differentiated groups. We first examine how emotions are managed in status-differentiated groups to promote solidarity, and then describe the design of a new experimental setting capable of assessing the attraction of groups to members. While the attractiveness of a group may be only a proxy for a direct measure of group solidarity, it brings us closer to our goal and may provide evidence that justifies a more elaborate research design. The fundamental emotions of fear and anger have basic survival functions for individuals, promoting escape from danger and personal defense for example, that can interfere with group solidarity (Turner, 2000). Happiness and

Status, Emotion, and Development of Solidarity in Stratified Task Groups

129

satisfaction also serve as a reward to encourage survival-enhancing activities. Stable, status-differentiated groups must manage these fundamental emotions in ways that promote group solidarity. Status hierarchies facilitate the coordination of individual group members working together to accomplish goals that no one member could accomplish alone. Thus, status hierarchies would have survival value for early humans as they sought to hunt large animals and protect themselves from rival human groups. The complex relationships in status-differentiated groups can also produce strong emotions capable of disrupting group solidarity. And, as noted above, low-status members whose contributions are typically ignored or devalued can feel anger and resentment. High-status members threatened with status loss may feel fear. Status-differentiated groups must manage fear and anger to promote group solidarity. One task of every group is survival as a group, that is, boundary maintenance and group solidarity. High-status members are expected to be group-motivated while low-status group members are expected to be selfishly motivated (Ridgeway, 1982). Thus, high status group members are expected to use emotions in ways that bind the group together, while low status members are expected to be less competent in accomplishing this goal, expressing emotions designed to help meet their personal needs such as fear and resentment. These expectations constrain the expression of emotion (Ridgeway & Johnson, 1990). For example, high-status members may express anger (when necessary to motivate the group) but may not express fear. Conversely, lowstatus members may express fear (attempting to satisfy their individual needs) and resentment but not anger. The anger of a low-status member could promote unacceptable fear on the part of high-status members. The anger of high-status members, however, can produce acceptable fear in low-status members. We would expect high-status members to promote happiness and satisfaction in low-status members to assuage the negative emotions typically felt by low-status members. Thus, expressions of negative emotion by high-status group members would be constrained to the extent that low-status members can decide to leave the group. It follows that a group wherein members expressed status-compatible emotions would show more solidarity than would a group where members expressed status-incompatible emotions. We propose that group solidarity is promoted when high-status members express happiness and satisfaction while low-status members express fear and resentment. ]~ In contrast, solidarity is undermined when high status members express fear and resentment while low-status members express happiness and satisfaction.

130

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

An Indicator of Solidarity in an Experimental Setting Testing the proposition that status-compatible emotions promote group solidarity while status-incompatible emotions undermine group solidarity requires an indicator of solidarity. Ideally, we might want to assemble long-term work groups with stable status hierarchies, then observe which ones remain stable, retaining all their members. The resources required for such a project are daunting, however, and a simpler proxy for group cohesion could serve as an indicator of solidarity. It seems reasonable to assume that groups with more solidarity would be more attractive to group members and prospective group members than would groups with less solidarity. We have designed a study in which prospective group members observe groups then rate their attractiveness, choosing to join a group or opting to continue looking for an attractive group to join. The attractiveness of the group to a prospective member will be our indicator of group solidarity. In the proposed experiment, participants will watch videos of prospective groups, then choose whether to join a group or to keep looking. They will rate the attractiveness of the group, then make their choice. Choosing to keep looking will entail a cost, however, a portion of their pay for participating in the study. Participant ratings of the attractiveness of the group will constitute a selfreported indicator of group solidarity while participant choice to join the group constitutes a behavioral indicator of solidarity. An initial experiment requires three emotion conditions: a condition in which participants observe group members expressing status-compatible emotions and a condition in which participants observe group members expressing statusincompatible emotions, plus a baseline condition in which participants observe group members expressing little emotion. '2 Status-Compatible Emotion Condition: Participants observe a high-status group member expressing happiness and satisfaction while a low-status group member expresses fear and resentment. Neutral Emotion Condition: Participants observe high-status and low-status group members working together but expressing little emotion.

Status-Incompatible Emotion Condition: Participants observe a high-status group member expressing fear and resentment while a low-status group member expresses happiness and satisfaction. The study will test two hypotheses:

Hypothesis 1: Participants will find the group in the Status-Compatible Emotion Condition more attractive than the group in the Status-incompatible

Status, Emotion, and Development of Solidarity in Stratified Task Groups

131

Emotion Condition. Participants will also choose the group in the Status-Compatible Condition more often than will participants in the StatusIncompatible Condition.

Hypothesis 2: Status-compatible emotions enhance the attractiveness of a group while status-incompatible emotions detract from the attractiveness of a group such that self-report ratings of attractiveness will be highest and behavioral choices to join a group will be most common in the StatusCompatible Emotion Condition; the group in the Neutral Emotion Condition will be next most attractive and frequently chosen, followed by the StatusIncompatible Emotion Condition. That is attractiveness ratings and group choice will be ranked. Status-Compatible Emotion > Neutral Emotion > Status-Incompatible Emotion Videos designed to implement experimental conditions. Videos for all conditions will portray two actors working together in a corporate setting perhaps putting together an advertising campaign for a new product. Several status indicators will make clear that one actor is a portraying a high-status supervisor while the other is portraying a low-status subordinate. The supervisor will be better and more formally dressed, speak in a more directive way, and mention a master's degree from a prestigious university. In contrast, the subordinate will be less formally dressed, express uncertainty about what to do next referring to a lack of formal education, and ask for advice. As they discuss their work, a memo will be delivered giving bad news. They have received negative evaluations on the previous phase of the project. The company may lose a client, which could result in their losing their jobs. The high-status and low-status actors then react with status-compatible emotion, neutral emotion, or status-incompatible emotion. In the Status-Compatible Condition, the low-status subordinate reacts to the bad news with resentment and fear. "That was a good project. They don't know how hard we worked on it. Management just doesn't care about us. What will we do? They will lay us off for sure." The high-status supervisor responds with calm satisfaction, encouraging the the low-status subordinate to feel more positive about the situation. "Don't worry. It was good work and they will realize that eventually. Look at the bright side, a lot of companies are looking for people with our experience. We could get much better jobs than we have now." In the Status-Incompatible Condition, the reactions to the bad news are reversed. The high-status supervisor reacts with resentment and fear while the low-status subordinate responds with calm satisfaction and encouragement.

132

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

The high-status supervisor says, "That was a good project. They don't know how hard we worked on it. Management just doesn't care about us. What will we do? They will lay us off for sure." The low-status subordinate responds, "Don't worry. It was good work and they will realize that eventually. Look at the bright side, a lot of companies are looking for people with our experience. We could get much better jobs than we have now." In the Neutral-Status Condition, both actors' reactions to the bad news will be non-committal. One will say "Not the best news we've gotten this week." The other will respond, "No, not the best."

Prospective status of the participant. One complication would be the expectations a participant has for the status position that the participant would occupy within the group. In the first study using undergraduate participants, it would be appropriate for the participant to expect to join the group in a subordinate position. Subsequent studies could explore the ramifications of participants expecting to join the group in high-status compared to low-status positions. Participants would be randomly assigned to watch one of the three videos, then respond to a series of items asking about their impressions of the group and its attractiveness to them. Then they would choose whether to join that group or pay to view another group. After making a choice, the participant will be informed that the study is over and debriefed. CONCLUSION Jonathan's Turner's (2000) work on the origin of human emotions provides a solid foundation for a theory of compatible emotions in status-differentiated groups (Lovaglia & Houser, 1996). Building on that theory, we propose that status-compatible emotions developed to maintain solidarity in status hierarchies. All groups face the task of maintaining group solidarity to discourage members from abandoning the group. Status hierarchies, while effective in coordinating complex work, produce emotions capable of disrupting group solidarity. Thus, managing the complex emotional relationships present in status hierarchies is one necessary task faced by all work groups. When emotions are status-compatible, high-status group members express positive emotion encouraging low-status members to remain in the group. Lowstatus members, meanwhile, may express negative emotion, resenting the neglect of their contributions. We propose that high-status group members, because of the many benefits of high status, have ample reason to overlook the negative emotion expressed by low-status members. If so, then an attenuation effect should be found in influence rates when low-status members express negative

Status, Emotion, and Development of Solidarity in Stratified Task Groups

133

emotion toward high-status members. That is, the influence of a low-status member who expresses negative emotion should be nearly as high as the influence of a low-status member who expresses neutral emotion. In contrast, the influence of a low-status m e m b e r who expresses positive emotion should be substantially higher. Houser and Lovaglia (1994) found some evidence for this, although their result has not been replicated. In one test, however, high status participants overlooked negative emotion expressed by low-status partners. Our proposed new study provides a direct test of the theory that statuscompatible emotions increase solidarity in status differentiated groups while status-incompatible emotions decrease solidarity. In a group with high solidarity, low-status group members may react with potentially damaging fear and resentment so long as high-status group members respond with calm assurance and express positive emotion to encourage low-status members. In a contrasting group with low solidarity, high-status group members may react with potentially damaging fear and resentment that cannot be repaired by low-status members responding with calm assurance and the expression of positive emotion.

NOTES 1. Status hierarchies can also limit the disruption of group solidarity that results from unequal exchange relations. Thye (2000) showed that resources controlled by high-status individuals are perceived to be more valuable than those controlled by low-status individuals. Thus paying more to a high-status individual may not produce dissatisfaction. 2. Status rank is correlated with ability to dominate in human groups because fighting ability is necessary to promote group defense. Defense against external threat of predators and other human groups is necessary for the survival of human groups in the environment faced by early humans on the savanna. 3. Evolutionary pressure to form social hierarchies comes from intra-species competition for food and mates. High-ranking individuals have greater access to food and mates. Individuals capable of rising in the hierarchy will have more opportunities for food and mates. Thus high rank in a status hierarchy comes to have value. To efficiently find a suitable mate, individuals must avoid losing contests by being aware of their rank, and seeking the opportunities appropriate to that rank. Thye (2000) has demonstrated the value placed on high status by undergraduates. 4. For the development and overviews of status characteristics theory see Berger, Cohen and Zelditch (1972), Berger, Fisek, Norman and Zelditch (1977), Berger, Rosenholtz and Zelditch (1980), Ridgeway (1982), Berger, Wagner and Zelditch (1985), Ridgeway and Berger (1986), Webster and Foschi (1988), Cohen and Zhou (1991), Berger, Norman, Balkwell and Smith (1992), Ridgeway, Boyle, Kuipers and Robinson (1998), Berger, Ridgeway, Fisek and Norman (1998). 5. The evolutionary approach to the study of emotions suggests that fear, while a negative emotion, be separated from anger in explaining the effects of emotions on group solidarity.

134

JEFFREY A. HOUSER AND MICHAEL J. LOVAGLIA

6. The idea that high-status individuals are happy and that low-status individuals are resentful may not seem to follow from the stereotypical picture of a fawning, flattering subordinate interacting with a stem-faced boss. It is important to note that in such a situation, the emotions expressed may not coincide with the emotions felt. The forced show of positive emotion by a subordinate my represent the emotions she is experiencing. In addition, recent research has demonstrated the necessity of separating power from status in attempting to analyze social interaction (Lovaglia, 1994, 1995; Willer, Lovaglia & Markovsky, 1997; Willer, Troyer & Lovaglia, 2001). The stern expression of a high-status individual and the forced expression of positive emotion not felt by a low-status individual may be due to power differences between the two rather than to their relative status. 7. A ceteris paribus clause is implicit here because of the necessity of separating power from status in the analysis. Power differences can create deference that mimics the deference produced by status differences. To avoid punishment, an individual might defer to a powerful person while at the same time harboring low expectations for the powerful person's ability or willingness to contribute to the group task. Thus, influence is a good indicator of a status difference between two people only to the extent that their power is equal (Lovaglia, 1995; Willer, Lovaglia & Markovsky, 1997). 8. It is difficult to argue that a person accorded high deference (in the absence of power differences) does not also have high status in a task group. High deference is assumed to result from expectations of group members that the influential person will make major contributions to the successful attainment of group goals. 9. See the discussion of utilitarian vs. emotional bases of solidarity in Markovsky and Lawler (1994). 10. TASK software used in to implement the Contrast Sensitivity experimental setting is available to researchers from Professor Martha Foschi, Department of Sociology, University of British Columbia, Vancouver, BC, Canada V6T 2b2 (mfoschi@ unixg.ubc.ca). 11. While it might be that more solidarity would be found in groups where all members expressed happiness and satisfaction, consistently positive emotion might not be conducive to accomplishing group goals. 12. The experiment will end and debriefing will begin after the participant makes a decision to choose or not to choose the first group.

REFERENCES Baron, R. A. (1987). Interviewer's Moods and Reactions to Job Applicants: The Influence of Affective States on Applied Social Judgments. Journal of Applied Social Psychology, 17, 911-926. Berger, J., Cohen, B. P., & Zelditch, M., Jr. (1972). Status Characteristics and Social Interaction. American Sociological Review, 37, 241-255. Berger, J., Fisek, M. H., Norman, R. Z., & Zelditch, M., Jr. (1977). Status Characteristics and Social Interaction: An Expectations States Approach. New York: Elsevier. Berger, J., Norman, R. Z., Balkwell, J., & Smith, R. F. (1992). Status Inconsistency in Task Situations: a Test of Four Status Processing Principles. American Sociological Review, 57, 843-855. Berger, J., Ridgeway, C. L., Fisek, M. H., & Norman, R. Z. (1998). The Legitimation and Delegitimation of Power and Prestige Orders. American Sociological Review, 63, 379-405.

Status, Emotion, and Development of Solidarity in Stratified Task Groups

135

Berger, J., Rosenholtz, S. J., & Zelditch, M., Jr. (1980). Status Organizing Processes. Annual Review of Sociology, 6, 470-508. Berger, J., Wagner, D. G., & Zelditch, M., Jr. (1985). Expectation states theory: review and assessment. In: J. Berger & M. Zeldtich, Jr (Eds), Status, Rewards and Influence: How Expectations Organize Behavior (pp. 1-72). San Francisco, Jossey-Bass. Brunner, E. (1997). Socioeconomic Determinants of Health: Stress and the Biology of Inequality. British Medical Journal, 314, 1472-1476. Clark, R., Anderson, N. B., Clark, V. R., & Williams, D. R. (1999). Racism as a Stressor for African Americans: A Biopsychosocial Model. American Psychologist, 54, 805-816. Cloutier, S., & Newberry, R. C. (2000). Recent Social Experience, Body Weight and Initial Patterns of Attack Predict the Social Status Attained by Unfamiliar hens in a New Group. Behavior, 137, 705-726. Cohen, B. P., & Zhou, X. (1991). Status Processes in Enduring Work Groups. American Sociological Review, 56, 179-188. Cooper, R. S. (1993). Health and the Social Status of Blacks in the United States. Annals of Epidemiology, 3, 137-144. Fisek, M. H., Berger, J., & Norman, R. Z. (1991). Participation in Heterogeneous and Homogeneous Groups: A Theoretical Integration. American Journal of Sociology, 97, 114-142. Frank, R. H. (1985). Choosing the Right Pond. New York: Oxford University Press. Hochshild, A. R. (1983). The Managed Heart. Berkeley, CA: University of California Press. Hohman, G., & Fruth, B. (1996). Food sharing and status in unprovisioned bonobos. In: P. Wiessner & W. Schiefenhovel (Eds), Food and the Status Quest (pp. 47-68). Providence, RI: Berghahn Books. Homans, G. C. (1961). Social Behavior." Its Elementary Forms. New York: Harcourt, Brace & World. Houser, J. A., & Lovaglia, M. J. (1994). Compatible emotions and group solidarity: an expectation states approach to the ties that bind groups. Presented to the Annual Meeting of the American Sociological Association Meeting, August, Los Angeles, CA. Kemper, T. D. (1978). A Social lnteractional Theory of Emotions. New York: Wiley. Kemper, T. D. (1984). Power, status, and emotions: a sociological contribution to a psychophysiological dilemma. In: K. R. Scherer & P. Ekman (Eds), Approaches to Emotion (pp. 369-383). Hillsdale, NJ: Lawrence Erlbaum. Kemper, T. D. (1987). How Many Emotions are There? American Journal of Sociology, 93, 263-289. Kemper, T. D. (1991). Predicting Emotions from Social Relations. Social Psychology Quarterly, 54, 330-342. Kollock, P. (1994). The Emergence of Exchange Structures: An Experimental Study of Uncertainty, Commitment, and Trust. American Journal of Sociology, 100, 315-345. Lawler, E. J., & Yoon, J. (1993). Power and the Emergence of Commitment Behavior in Negotiated Exchange. American Sociological Review, 58, 465-481. Lawler, E. J., & Yoon, J. (1996). Commitment in Exchange Relations: Test of a Theory of Relational Cohesion. American Sociological Review, 61, 89-108. Lawler, E. J., & Yoon, J. (1998). Network Structure and Emotion in Exchange Relations. American Sociological Review, 63, 871-894. Lovaglia, M. J. (1994). Relating power to status. In: B. Markovsky, K. Heimer, J. O'Brien & E. J. Lawler (Eds), Advances in Group Processes (Vol. 11, pp. 87-111). Greenwich, CT: JAI Press. Lovaglia, M. J. (1995). Power and Status: Exchange, Attribution and Expectation States. Small Group Research, 26, 400426.

136

J E F F R E Y A. H O U S E R A N D M I C H A E L J. L O V A G L I A

Lovaglia, M. J., & Houser, J. A. (1996). Emotional Reactions and Status in Groups. American Sociological Review, 61, 867-883. Mackie, D. M., & Worth, L. T. (1989). Processing Deficits and the Mediation of Positive Affect in Persuasion. Journal of Personality and Social Psychology, 57, 27-40. Markovsky, B., & Lawler, E. J. (1994). A new theory of group solidarity. In: B. Markovsky, K. Heimer, J. O'Brien & E. J. Lawler (Eds), Advances in Group Processes (Vol. 11, pp. 113-138). Greenwich, CT: JAI Press. Maryanski, A., & Turner, J. H. (1992). The Social Cage: Human Nature and the Evolution of Society. Stanford, CA: Stanford University Press. Mazur, A. (1973). A Cross-Species Comparison of Status in Small Established Groups. American Sociological Review, 38, 513-530. Mazur, A. (1985). A Biosocial Model of Status in Face-to-Face Primate Groups. Social Forces, 64, 377-402. McGovem, V., & Houser, J. A. (2001). Gender and the role of second-order expectations: do reflected appraisals influence male behavior? Presented to the Annual Meeting of the American Sociological Association Meeting, August, Anaheim, CA. Moore, A. J., Reagan, N. L., & Haynes, K. F. (1995). Conditional Signaling Strategies: Effects of Ontogeny, Social Experience and Social-Status on the Pheromonal Signal of Male Cockroaches. Animal Behavior, 50, 191-202. Plutchik, R. (1962). The Emotions: Facts, Theories and a New Model. New York: Random House. Plutchik, R. (1980). Emotion: A Psychoevolutionary Synthesis. New York: Harper and Row. Ridgeway, C. L. (1982). Status in Groups: The Importance of Motivation. American Sociological Review, 47, 76-88. Ridgeway, C. L. (1987). Nonverbal Behavior, Dominance, and the Basis of Status in Task Groups. American Sociological Review, 5, 683-694. Ridgeway, C. L., & Berger, J. (1986). Expectation, Legitimation, and Dominance Behavior in Task Groups. American Sociological Review, 51, 603-617. Ridgeway, C. L., & Berger, J. (1988). The legitimation of power and prestige orders in task groups. In: M. Webster, Jr. & M. Foschi (Eds), Status Generalization: New Theory and Research (pp. 207-231). Stanford, CA: Stanford University Press. Ridgeway, C. L., Boyle, E. H., Kuipers, K. J., & Robinson, D. T. (1998). Resources and Interaction in the Development of Status Beliefs. American Sociological Review, 63, 331-350. Ridgeway, C. L., & Johnson, C. (1990). What is the Relationship between Socio-emotional Behavior and Status in Task Groups? American Journal of Sociology, 95, 1189-1212. Ridgeway, C. L., Johnson, C., & Diekema, D. (1994). External Status, Legitimacy, and Compliance in Male and Female Groups. Social Forces, 72, 1051-1077. Smuts, B. B. (1985). Sex and Friendship in Baboons. New York: Aldine. Stets, J. E. (1997). Status and Identity in Marital Interaction. Social Psychology Quarterly, 60, 185-217. Stets, J. E., & Tsushima, T. M. (2001). Negative Emotion and Coping Responses within Identity Control Theory. Social Psychology Quarterly, 64, 283-295. Suzuki, S., Hill, D. A., & Sprague, D. S. (1998). Intertroop Transfer and Dominance Rank Structure of Nonnatal Male Japanese Macaques in Yakushima, Japan. International Journal of Primatology, 19, 703-722. Thye, S. R. (2000). A Status Value Theory of Power in Exchange Relations. American Sociological Review, 65, 407-432. Troyer, L., & Younts, C. W. (1997). Whose Expectations Matter? The Relative Power of Firstand Second-Order Expectations in Determining Social Influence. American Journal of Sociology, 103, 692-732.

Status, Emotion, and Development o f Solidarity in Stratified Task Groups

137

Turner, J. H. (2000). On the Origin of Human Emotions. Stanford, CA: Stanford University Press. Vervaecke, H., De Vries, H., & Van Elsacker, L. (2000). The Pivotal Role of Rank in Grooming and Support Behavior in a Captive Group of Bonobos (Pan paniscus). Behaviour, 137, 1463-1485. Virgin, C. E., & Sapolsky, R. M. (1997). Styles of Male Social Behavior and Their Endocrine Correlates Among Low-Ranking Baboons. American Journal of Primatology, 42, 25-39. de Waal, F. (1996). Good Natured: The Origins of Right and Wrong in Humans and Other Animals. Cambridge, MA: Harvard University Press. Walker, H. A., Thomas, G., & Zelditch, Jr., M. (1986). Legitimacy, Endorsement, and Stability. Social Forces, 67, 216-228. Walker, H., & Zelditch, Jr., M. (1993). Power, legitimacy, and the stability of authority: a theoretical research program. In: J. Berger & M. Zelditch, Jr. (Eds), Theoretical Research Programs: Studies in the Growth of Theo~ (pp. 364-381). Stanford, CA: Stanford University Press. Webster, M., Jr. and Foschi, M. (1988). Overview of status generalization. In: M. Webster, Jr. & M. Foschi (Eds), Status Generalization: New Theory and Research (pp. 1-22). Stanford, CA: Stanford University Press. Wilier, D., Lovaglia, M. J., & Markovsky, B. (1997). Power and Influence: A Theoretical Bridge. Social Forces, 76, 571-603. Wilier, R. B., Troyer, L., & Lovaglia, M. J. (2001). Power, group motivation, and status attainment. Presented to the Annual Meeting of the American Sociological Association, August, Anaheim, CA. Wrangham, R. W., & Peterson, D. (1996). Demonic Males: Apes and the Evolution of Human Aggression. Boston: Houghton-Mifflin.

THE THEORY OF RELATIONAL COHESION: REVIEW OF A RESEARCH PROGRAM Shane R. Thye, Jeongkoo Yoon and Edward J. Lawler

ABSTRACT In this paper we analyze and review the theory of relational cohesion and attendant program of research. Since the early 1990s, the theory has evolved to answer a number of basic questions regarding cohesion and commitment in social exchange relations. Drawing from the sociology of emotion and modern theories of social identity, the theory asserts that joint activity in the form of frequent exchange unleashes positive emotions and perceptions of relational cohesion. In turn, relational cohesion is predicted to be the primary cause of commitment behavior in a range of situations. Here we outline the theory of relational cohesion, tracing its development through the present day, and summarize the corpus of empirical evidence for the theory's claims. We conclude by looking ahead to future projects and discussing some of the more general issues informed by our work.

INTRODUCTION This paper reviews a program of research on the d e v e l o p m e n t of relational cohesion in exchange relations (Lawler, 2001, 2002; Lawler & Thye, 1999; Lawler, Thye & Yoon, 2000; Lawler & Yoon, 1993, 1996, 1998). Relational

Group Cohesion, Trust and Solidarity, Volume 19, pages 139-166. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2

139

140

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

cohesion is defined as the perception by individuals in an exchange relation that their relationship is a unifying element or force in the social situation (Lawler & Yoon, 1996). Such perceptions lead to higher levels of commitment and collectively-oriented behavior. Commitment here refers to the strength of the tie between a person and a social unit. The social unit may be a relation, group, network, organization, ethnic community, and so forth. Commitment is empirically manifest in behaviors such as staying in the social unit despite alternatives, providing unilateral benefits to others in the social unit, or collaborating with others to produce joint goods. In our theoretical research program, the focus is on how commitment behaviors develop within social exchange relations. The theory proposes an emotional-affective explanation for the development of cohesion and commitment in exchange relations (Lawler & Yoon, 1996). This explanation was adapted from a theory of affective attachments published in 1992 (see Lawler, 1992a). According to that theory, emotional-affective attachment to a collective unit develops to the extent that the unit fosters a "sense of control." Sense of control is based on structural variations in the degree of choice, and the capacity for actors to engage in "means-ends" deliberation (Elster 1986). Groups, relations, or organizations that enhance actors' sense of control become objects of affective attachment, and those that constrain or limit actors' sense of control become objects of affective detachment (Lawler, 1992a, 1997; Mueller & Lawler, 1999). The rationale is that a sense of control fosters positive emotions that are partially attributed to the social units perceived as responsible for the sense of control. The theory of relational cohesion adapts the aforementioned principles to social exchange relations. Social exchange structures and situations entail considerable uncertainty and ambiguity (Molm & Cook, 1995). As such, the formation of exchange relations should heighten actors' "sense of control" in these situations. If so, according to the theory of affective attachments, social exchange should generate positive emotions and feelings that are attributed to the exchange relation. These ideas were the backdrop and starting point for our program of theory and research on relational cohesion. Over the past decade, our work has evolved to provide a number of insights regarding the emergence of commitment in exchange relations. The research focuses on a number of questions including: (1) Under which structural conditions do purely instrumental exchange relations develop into expressive ones that are valued in their own right? (2) What kinds of exchanges are most likely to elicit relational cohesion and commitment? (3) How do emotions impact the psychological sense of coming together and the perception of a shared identity? (4) How do structural properties such as the shape of the network

The Theory of Relational Cohesion

141

affect the prospects for cohesive dyadic relations? (5) Whether and how do overarching group or common identities influence cohesion and commitment in exchange? The purpose of this paper is to pull together our published research on these questions, demonstrate the theoretical progression over time, and extract some general principles applicable beyond the program. We intend this review for those who seek a sense of the research program as a whole, and for those who are unfamiliar with the research or familiar with only select pieces of it.

BACKGROUND The theory of relational cohesion and attendant research program are anchored in social exchange theory, a tradition of micro-sociology in which interpersonal encounters are modeled as exchanges. Beginning in the late 1950s, social exchange theory included a number of theoretical statements connecting social structure, individual action, and the distribution of valued resources (Blau, 1964; Ekeh, 1974; Emerson, 1972a, b, 1976; Homans, 1961; Thibaut & Kelley, 1959). Early work in this tradition focused primarily on power and related processes in dyadic encounters (Blau, 1964; Cook & Emerson, 1978; Emerson, 1964; Homans, 1961; Molm, 1994; Thibaut & Kelley, 1959). But later, Emerson (1972a, b, 1981) proposed a network-oriented view that quickly gained prominence. Emerson's basic point was that dyadic exchanges occur in the context of larger networks and cannot be understood in isolation of them. This shift in perspective shed new light on the complexity of social exchange and posed important challenges for the field. An enduring issue has been to understand the interplay between social structures, the strategies and tactics used in negotiating an exchange, and the allocation of resources among individuals embedded in networks. For more than two decades, researchers intrigued by power dynamics and exchange processes have spawned a number of theoretical statements and empirical tests (Bacharach & Lawler, 1981; Bonacich & Bienenstock, 1993; Burke, 1997; Cook, Emerson, Gillmore & Yamagishi, 1983; Cook & Yamagishi, 1992; Ekeh, 1974; Friedkin, 1992; Markovsky, Wilier & Patton, 1988; Markovsky et al., 1993; Skvoretz & Wilier, 1993; Thye, 2000; Thye, Lovaglia & Markovsky, 1997; Walker et al., 2000; Wilier, 1999; Wilier & Anderson, 1981). And while these theories differ in their theoretical assumptions and methods of prediction, all assert that exchange transpires when two or more actors seek to jointly produce benefits they cannot produce alone. The search for benefit implies that actors are self-interested; that benefit cannot be produced

142

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

alone implies that actors are interdependent. Self-interest and interdependence are key properties of virtually all contemporary theories of exchange. This is also our theoretical point of departure. We seek to understand a phenomenon that has fallen beyond the purview of traditional exchange theory (see Cook & Emerson, 1984 for an exception) - namely, the way that social relations become salient objects toward which actors orient their exchange behaviors. Commitment is broadly defined as the strength of an attachment to another social unit such as a group, organization, or community (Kanter, 1968, 1972). In the abstract, commitment represents a person-to-group bond that is distinct from inter-personal bonds. Parsons (1951) suggested that person-to-group attachments could involve instrumental (i.e. utilitarian), affective (i.e. emotional), or normative (i.e. legitimated) bonds and saw these as an important foundation for social order. Kanter (1986, 1972) echoes these distinctions in her discussion of commitment as continuance, cohesion, and control. Important for the development of our research program is that both Parsons (1951) and Kanter (1968, 1972) recognize the importance of instrumental and affective foundations for commitment. Building on and elaborating these ideas, we explicitly show how affective attachments can develop from behavior that is initially instrumental. The traditional exchange-theory explanation for commitment focuses on instrumental conditions, in particular, uncertainty reduction (Cook & Emerson, 1984). The argument is that commitment develops because repeated exchanges foster a sense of predictability in the situation (Emerson, 1981; Kollock, 1994). Consider a car manufacturer who repeatedly buys parts from a supply dealer. Given a series of successful transactions, the two should come to learn more about one another, develop a common set of procedures or expectations for the exchange, and perhaps learn to trust one another given a history of successful encounters. These represent "benefits" in an uncertain market where the properties of alternative partners are unknown or unknowable (Kollock 1994). Recent work by Kollock (1994, 1999) confirms that uncertainty reduction is one basis for commitment in exchange relations. In contrast, the theory of relational cohesion is centered on an emotionalaffective explanation for commitment in exchange. Like other exchange theories, the actors of relational cohesion theory are driven initially by self-interest. That is, the theory presumes that actors are motivated to exchange so they can produce benefits not otherwise attainable. The theory also recognizes, however, that actors have the ability to experience, interpret, and reproduce emotional reactions to exchange outcomes. The orienting idea is that the very act of exchange represents joint social activity characterized by problems of coordination and uncertainty. As such, when exchange is successful, actors should experience positive emotional reactions; when exchange is unsuccessful,

The Theory of Relational Cohesion

143

actors should experience negative emotional reactions. It is these emotional reactions that determine the bonds of individuals to one another and to the relation itself. The notion that emotions are integral to the development of social relations is as old as sociology itself. Durkheim (1915) was perhaps the first to explicate this idea in The Elementary Forms of Religious Life. He noted that when individuals engage in joint social activity, they experience an emotional uplift which heightens their sense of group membership and collectively-oriented behavior. More recently Collins (1981, 1989) has elaborated and further developed this idea in his theory of interaction ritual chains. For Collins (1975, 1981), an interaction ritual is a basic unit of interaction that involves two or more participants who have a mutual focus on an object or action. He asserts that a sense of group membership is most likely to emerge when the participants experience a common mood or emotion that strengthens over time. Lawler and Thye (1999) suggest that most exchange relations involve a common focus and shared emotion. As such, exchange relations contain the basic seeds for incipient group membership and commitment. Our relational cohesion research program brings together these ideas to explain how joint social activity generates salient personto-group bonds in the context of exchange. In the following pages, we review the early research on negotiated exchange that set the stage for the theory of relational cohesion. We then describe the basic theory as proposed by Lawler and Yoon (1996) and the first series of empirical tests. Next we summarize new directions the theory has taken and trace its development through the present day. We conclude by extracting and discussing some general propositions and issues around which our current research is organized.

EARLY RESEARCH Early work in the relational cohesion research program established the theoretical and empirical importance of emotions in the commitment process. Lawler and Yoon (1993) posit four mechanisms through which exchange induces commitment. First, as noted above, frequent exchange reduces the level of uncertainty associated with the focal relation (Cook & Emerson, 1984; Williamson, 1975). Second, frequent social exchange often leads actors to expand the types of goods exchanged, thereby providing multiple sources of benefit from a single relation (Emerson, 1981; Foa, 1971; Tallman, Gray & Leik, 1991). Third, frequent exchange enhances the satisfaction with or attraction to the exchange partner who is providing benefit (Byrne, Clore & Smeaton, 1986; Homans, 1961). Finally, frequent social exchange should make

144

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

the relationship a salient object of awareness to which actors attribute their positive emotions (Lawler 1992b). Based upon the fourth idea, Lawler and Yoon (1993) offer a model of commitment that set the stage for the theory of relational cohesion. The overarching idea of this model can be abstracted as follows: Exchange conditions that

maximize the potential for joint benefit result in: (i) repeated agreements; (ii) positive emotions; and (iii) commitment behavior. The model postulates a process that begins at the point where structural conditions promote frequent agreement. Specifically, Lawler and Yoon (1993) assert that equal relative power and integrative bargaining are important structural factors that result in frequent exchange. Equal power is defined in terms of relative dependence, i.e. actors have equal or unequal alternatives available to them. Integrative bargaining is defined as negotiations where the contract zone is not fixed (unlike distributive bargaining), and joint-problem solving is possible (Walton & McKersie, 1965). Whereas equal power is likely to promote mutual concessions along a zerosum dimension of profit, integrative bargaining allows subjects to engage in logrolling strategies that foster jointly-beneficial outcomes.~ Both forces should promote frequent agreements between the parties to negotiated exchange. A key idea is that the very act of reaching an agreement will produce mild, everyday positive emotions. Two kinds of emotion are considered important in the relational cohesion program: pleasure/satisfaction and interest/excitement (see also Izard, 1971, 1977; Larsen & Diener, 1992; Russell et al., 1980). Pleasure/satisfaction is defined as a state of "feeling gratified" while interest/ excitement represents a state of "feeling energized." The idea that social exchange produces positive emotions coheres with Homans' (1961) account of "sentiment" in social relations. Modem theories of emotion broadly recognize these two dimensions of positive emotion as having distinct effects on attributions and behaviors (Izard, 1977; Russell, 1980; Weiner, 1985, 1986). At the heart of the commitment model is the notion that positive emotions produce commitment behavior (Lawler & Yoon, 1993). Two kinds of commitment behavior initially were studied, stay behavior and gift giving. Stay behavior refers to remaining in a focal relationship despite available altematives. Staying in a relationship, given an alternative, is the most widely recognized indicator of commitment in the organizational literature (Halaby & Weakliem, 1989; Kanter, 1972; Mueller & Price, 1990). Gift giving refers to a unilateral, noncontingent, symbolic gesture such as sending a flower or card to a friend. The theory asserts that individuals who experience positive emotion are likely to stay in the focal relationship and to give each other small token gifts. Lawler and Yoon (1993) reported the first experimental test designed to evaluate the emotional/affective core of the model. Briefly, the experiment involved

The Theory of Relational Cohesion

145

dyadic negotiations in which each person had a fixed, standing alternative partner. The negotiations transpired between two subjects in separate rooms, each of whom attempted to maximize their benefit in the focal relation. In the event two subjects could not reach an agreement~ each subject automatically earned some level of profit from their standing alternative partner. This alternative was a simulated other with whom the focal subject did not negotiate. The primary independent variables (see Lawler & Yoon, 1993 for details) were power/dependence (equal vs. unequal) and the type of bargaining (integrative vs. distributive). The primary dependent measures suggested by the theory were agreement frequency, positive emotions, and commitment behavior. The results of the experiment provide strong support for the model. Under conditions of equal relative power, subjects were more likely to make mutual concessions and these resulted in significantly more agreements. Furthermore, there was some evidence that integrative bargaining had a positive impact on the degree of logrolling, which in itself, also fostered higher mutual agreements. This was particularly the case early in the negotiation process. In turn, agreement frequency was significantly related to interest/excitement though not pleasure/satisfaction. Finally, the results of this study found positive emotion to be a good predictor of commitment behavior. Specifically, interest/excitement had large direct effects on continuance (i.e. stay behavior) and symbolic (i.e. gift giving) forms of commitment. Overall, the experiment provided support for the notion that emotions mediate the impact of exchange on commitment. Two years later, a second experimental study was conducted to investigate the role of total power in the commitment process (Lawler, Yoon, Baker & Large, 1995). Total power is defined as the sum of each actor's power (see Bacharach & Lawler, 1981; Molm, 1987). Total power is essentially "mutual dependence" in Emerson's (1972a, b) terms, which he conceived of as the structural foundation for social cohesion. In relations where total power is high, actors are more dependent on one another for valued goods compared to relations wherein total power is low. In this research, the main hypothesis was that greater total power generates more commitment in the form of gift giving through the exchange-to-emotion intervening process. This hypothesis was affirmed, and thus provided additional support for the importance of the exchange-to-emotion link. Importantly, the results from a post-experimental questionnaire measure indicated that more frequent exchange increased the "perceived closeness" of the relation. This finding raised the question of whether a sense of cohesion should be theorized as part of the commitment process. To summarize, early research in the relational cohesion research program established the importance of emotions in the commitment process. The first empirical test indicated that structural bargaining conditions - equal power and

146

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

integrative bargaining issues - produced more frequent agreement between actors in negotiated exchange. The second showed that higher total power in the relationship (i.e. mutual dependence) also generated more frequent exchange. The importance of these findings is that frequent exchange represents a joint social activity that has emotional effects on actors. Of special significance is that both empirical studies found that more frequent exchange produced more positive emotions; in one study the key emotion produced was interest/excitement (Lawler & Yoon, 1993) and in the other it was pleasure/satisfaction (Lawler, Yoon, Baker & Large, 1995). Which emotion was most important was unclear at that point. Finally, emotional reactions were shown to be the proximate cause of commitment behaviors in both studies. These initial experiments set the stage for the theory of relational cohesion, to which we now turn.

THE THEORY OF RELATIONAL COHESION The theory of relational cohesion emerged as a more general account of the commitment process (Lawler & Yoon, 1996). It included a number of important shifts or departures from earlier work. First, whereas earlier work focused on the bargaining process itself and the offer strategies actors employ (i.e. concession making and logrolling), the theory of relational cohesion was more explicitly focused on structural power conditions (i.e. both equal and total power as determined by the quality of alternatives). Although this shift was relatively subtle, it had important effects on the subsequent theory development. Second, there was more theoretical emphasis on the process whereby individuals come to see themselves as members of a common social unit, and a clear distinction between person-to-person and person-to-group bonds. Thus, the concept of "relational cohesion" - the sense of coming together, of something larger that unifies actors and actions - became the central focus of our theorizing. This change was inspired not only by some of the results discussed above, but also by theories of group formation and social identity, where the movement from individual (or interpersonal) to collective (or group) orientations is well understood (Brown, 2000; Tajfel & Turner, 1979, 1986). Finally, the behavioral indicators of commitment were expanded to include stay behavior, unilateral gift giving, and partaking (investment) in a new joint venture involving the potential for malfeasance. These behavioral indicators parallel the three dimensions of commitment identified by Kanter (1968, 1972). The main propositions of the theory can be expressed as a causal model depicted in Fig. 1. P o w e r is conceptualized as an exogenous structural potential that is distinct from p o w e r u s e (i.e. behavioral tactics or strategies) and p o w e r o u t c o m e s (i.e. the division of profit in exchange) (Bacharach &

The Theory of Relational Cohesion

147

q t

F

.o

t

q~

¢., 0

¢.. © o} Q

b©

t

/ II

f.

\ l

0 IZ

"6 1-

itu

148

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

Lawler, 1981; Emerson, 1972a, b; Lawler, 1992b; Molm, 1987, 1990). The theory conceptualizes dyadic power in terms of relative dependence. According to Emerson's (1972a, b, 1981) power dependence theory, the power of A over B (PAB) is equivalent to the dependence of B on A (DBA). The dependence of B on A, in turn, is a joint function varying: (i) directly with the value of resources controlled by A; and (ii) inversely with the availability of resources outside the A - B relationship. The theory of relational cohesion predicts that when actors A and B are equally dependent on one another for valued goods (i.e. relative power is equal), and their mutual or average dependence is high (i.e. total power is high), then A and B should find it easy to reach agreement. These dimensions of power (relative and total) are exogenous structural conditions that create the incentive for A and B to exchange. The core of the theory describes an endogenous process, whereby frequent social exchange unleashes emotions, perceptions, and cognitions that bind group members together and to the group as a whole. Emotions are defined as relatively short-lived positive or negative feeling states (Izard, 1991; Kemper, 1978). Frequent social exchange is predicted to elicit emotion; it should lead to positive emotion in much the same way that jointly accomplishing a task with another person leads to satisfaction, enjoyment, or excitement. Collins (1981, 1989) also recognizes that emotions are important in social relations, asserting that emotional energy is one type of "glue" that can bind members together when they share a common focus (such as a joint social exchange). Consistent with earlier research, the theory of relational cohesion focuses on two dimensions of positive emotion: pleasure/satisfaction and interest/ excitement (Lawler & Yoon, 1993; Watson & Tellegen, 1985). The theory asserts that emotions are generalized responses that immediately follow a positive exchange outcome, and these responses are themselves stimuli that lead to further cognitions and attributions regarding the group (Bandura, 1986; Izard, 1977; Weiner, 1985, 1986). Specifically, the theory asserts that positive emotions cause actors to seek attributions for the causes of those emotions. Actors may attribute their positive emotions to their own actions (e.g. "I contributed much to the exchange"), the actions of the other party (e.g. "my partner contributed much to the exchange") or the relation between them (e.g. "We worked well together to accomplish the exchange"). The theory presumes that since exchange is inherently a joint social activity, actors will partially attribute their emotions to their relation. This implies that the actors will come to perceive the exchange relation itself as a salient object of awareness, and the concept of relational cohesion captures this idea. The theory claims that positive emotions induce a shift in cognitive awareness, such that actors come to see their relation as being more important, stable, and valued over time.

The Theory of Relational Cohesion

149

Finally, the theory predicts that relational cohesion is the proximate cause of commitment behavior. The theory asserts that perceptions of relational cohesion lead the exchange relation to become an expressive object of attachment. In Berger and Luckmann's (1966) terms, the relation becomes "objectified" as actors become more aware of the relation as a distinct source of positive emotions. As this occurs, actors are more likely to engage in a variety of prosocial activities such as giving one another small token gifts, continuing to stay in the exchange relation, and initiating new collaborations that involve risk of opportunism or malfeasance. The effect of this is to further expand and strengthen the exchange relation. In this way, commitment behaviors emerge from and reflect the positive emotional impact of social exchange. The initial test of the theory was conducted by Lawler and Yoon (1996). They reported data from a series of three experiments, each addressing a different form of commitment behavior (i.e. gift giving, stay behavior, and contribution to a joint venture). They also directly tested the chain-process specified by the theory. Each experiment involved two subjects who negotiated exchange for twelve episodes under conditions of (high vs. low) total power and (equal vs. unequal) relative power. The subjects represented two companies, one attempting to buy a resource from the other. The setting simulated negotiations across a large number of "years" or episodes. At select points, and in accord with the theoretical model (Fig. 1), Lawler and Yoon measured each "moment" of the theorized sequence. These measures included: (i) agreement frequency; (ii) positive emotions in the form of interest/excitement and pleasure/ satisfaction; (iii) relational cohesion; and (iv) commitment behavior. The temporal sequence specified by the theory was created in the experimental context, and the research tested the set of relations predicted by the model. The results of the experiments provide strong and consistent support for the theory (Lawler & Yoon, 1996). High total power and equal relative power produced more frequent agreement in exchange. Frequent exchange, in turn, had a positive direct effect on both pleasure/satisfaction and interest/excitement, as predicted. Positive emotions had a positive direct effect on relational cohesion, also as predicted. The data revealed that both interest/excitement and pleasure/satisfaction had direct positive effects on relational cohesion, when each emotion was included as the sole predictor of relational cohesion. However, when both emotions were included simultaneously to predict relational cohesion, only pleasure/satisfaction was significant. Overall, while both emotions are important, pleasure/satisfaction played a stronger role in the commitment process. Finally, there was uniform support for the notion that relational cohesion is the proximate cause of commitment. In fact, with all variables in the model included, relational cohesion was the strongest and most significant

150

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

predictor across all three forms of commitment - stay behavior, gift giving, and contribution. The theory makes strong claims about the sequence of indirect steps through which structural power conditions promote commitment, and these were confirmed at each step by the research. To summarize, the theory and research on relational cohesion identifies an endogenous process through which structures of dependence affect the relational commitment. The first moment in the process is the frequency of exchange; the second is positive (or negative) emotions and feelings; and the third is a perception of the relation as a cohesive object. These three moments are tied together as a conceptual unit. By implication, a structural condition that changes the frequency of exchange should change correspondingly the strength of this endogenous process; a structural condition under which exchanges do not produce positive emotions should inhibit or prevent the process from operating; and if the emotions experienced are not attributed in part to the relation, they will not generate perceptions of cohesion. This endogenous process can be used to compare different relations within a network, or the same relation across time. It also helps to understand when the exchange relation, as such, takes on value. The endogenous process can be construed as "creating value" internally, within a given relation, and in the context of repeated interactions. The stronger the endogenous process, ceteris paribus, the more intrinsic or expressive value actors accord their exchange relation. Our emphasis on the endogenous process represents a departure from traditional rational choice theory, where the value (or expected utility) of a social relation is imposed exogenously through pecuniary motives (Elster, 1986; Heath, 1975). In contrast, our theory assumes that pecuniary value associated with structural power conditions is imposed exogenously, while nonpecuniary value develops through the endogenous process. This has potentially important implications for the role of emotions in contracting and trading relations among economic or political organizations, when the same individuals repeatedly negotiate for their respective organizations. EXTENSION

1: R E L A T I O N A L

COMMITMENTS

WITHIN N E T W O K S The next significant development in the relational cohesion research program occurred just two years after the initial publication of the theory. In 1998, Lawler and Yoon applied the theory to a more complex problem: How and when do dyads embedded in a larger social network become committed to one another? Whereas previous work explicitly focused on single dyadic exchange relation, the move to "network embedded" dyads broadened the scope of the theory and forged deeper connections to other branches of exchange theory (e.g.

The Theory of Relational Cohesion

151

Cook & Emerson, 1978; Cook et al., 1983; Markovsky et al., 1988) and to social identity theory (Rabbie & Horowitz, 1988; Tajfel & Turner, 1979, 1986). It enabled us to address whether "pockets of relational cohesion" develop in exchange networks, particularly for dyads that have the highest frequency of exchange. This is a direct implication of the relational cohesion theory reviewed above. This extension dealt with dyadic-level commitments in two networks: the branch and the stem (see Fig. 2). In the Fig. 2 networks each letter represents a person and each line represents an opportunity to exchange. Assuming that each position can make only one exchange per round, the branch is a strong power network because A can never be excluded while two of the more peripheral actors (B, G or D) must always be excluded (Simpson & Wilier, 1999). Strong power networks such as the branch tend to produce a "bidding war" wherein the low power actors make increasingly favorable offers to A in order to avoid exclusion. This results in large profit advantages for the central actor A. Important for our purpose is that the branch can be viewed as a network consisting of three dyadic relations (A-B, A-G, and A-D) in which A has a relative power advantage. In contrast, the stem represents a weak power network because no single individual can, without cost, consistently exclude others from exchange (Lovaglia et al., 1995; Markovsky et al., 1993; Thye et al., 1997). Weak power networks are characterized by more moderate profit differences across positions, and the tendency for some positions to be included more frequently than others. As a result of these forces, the stem network tends to "break" into two distinct exchange relations: an equal power dyad (B-G) and an unequal power dyad (A-D). Thus, the stem represents a network that contains both equal and unequal

"\ f

A

B

D

G

Strong Power - Branch

D Weak Power - Stem

Fig. 2. Two Exchange Networks.

152

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

relative power dyads embedded in the same social context. One question addressed in this research is whether and how network-based power alters the commitment process of relational cohesion theory. A second purpose is to determine whether relational cohesion, as a form of social bonding, is affected by an exogenous group identity. Research in the social identity tradition indicates two distinct mechanisms that produce group identities: social categories and interdependence. A central finding is that when social identities are activated in a group context, a variety of prosocial behaviors are likely to ensue. For instance, individuals sharing a common group identity are more likely to be cooperative, collectively-oriented, altruistic, and responsive to group goals rather than purely personal interests. Applied to relational cohesion theory, this implies that commitment is more likely to emerge when the individuals of a network share a common group identity. Lawler and Yoon (1998) tested these ideas using four experimental conditions in which subjects negotiate exchange in either the branch or stem network, and with or without a common group identity. The main predictions were that: (1) all dyadic relations in the branch will exchange with equal frequency; and (2) the G - D relation of the stem will exchange with greater frequency than A-B. The more frequent exchange along G - D should, according to the chain logic of relational cohesion theory, produce greater positive emotion, a heightened sense of relational cohesion, and more behavioral acts of commitment relative to the A-B relation. Lawler and Yoon (1998) also predicted that commitment and cohesion would be stronger when a common group identity is evoked. To implement this idea, in half of the experimental sessions the members of the network were portrayed as "departments" within a larger organization. In the other hall the participants were simply told they were competitors that had an interest in trading with others (Lawler & Yoon, 1998). The main findings were as follows. First, as predicted, there were no differences in exchange frequencies across the three dyadic relations of the branch network. However, when the members of the branch shared an exogenous group identity, profit differences were reduced. It appears that if network members have a common group identity, they take this into account such that low power actors yield less than they otherwise would, and high power actors are more conciliatory. Second, in the stem network, actors in the equal power relation (G-D) found it somewhat easier to reach agreement than those in the unequal power relation (A-B). Moreover, the commitment process was stronger in the equal than the unequal relation. That is to say, there was greater pleasure/ satisfaction, interest/excitement and relational cohesion in the equal power G - D relation compared to the unequal power A-B relation. Further, regressions within the unequal power A-B relation of the stem indicate frequent exchange

The Theory of Relational Cohesion

153

did not produce positive emotions; the endogenous process broke down at this point. Unequal power relations generated weaker effects than equal power relations along each moment of the endogenous process. Of particular note, the crucial effect of exchange frequency on positive emotion was present only under equal power. In sum, support for the theory was quite strong. An unexpected finding of this study was that the low power actor in the stem (D) seemed to respond to the fact that s/he had no alternative partners. Actor D reported negative emotional reactions to exchange, and there was a marginal negative relationship between this actors' perception of cohesion and commitment behavior (stay behavior). That is to say, higher cohesion actually reduced commitment. Our interpretation for these patterns was that for actor D in the stem, exchanges with A are essentially "forced" and involuntary. Some degree of choice or voluntarism may be important to the development of cohesion and commitment in exchange relations (Deci, 1975). In fact, this is implied by a classic idea from Homans (1961). He argued that sentiment develops in relations especially if each actor has an alternative partner, but their alternative relation is likely to generate lower benefits than the focal relation. Having the option of another partner (however poor) may be necessary for the emotional/affective process to operate. This would explain the response of the low-power actor in the stem network (Lawler & Yoon, 1998). The importance of having choices or options also is suggested by the theory of affective-attachments (Lawler, 1992a, 1997). That theory suggests that a voluntaristic exchange relation should generate a heightened sense of control, more positive emotions, and stronger affective attachment to the relevant social unit, in this case, the exchange relation. In order to examine the role of choice or voluntarism, we recently conducted another experiment to compare the relational-cohesion process in voluntary vs. involuntary exchange relations (Lawler, Thye and Yoon unpublished). The box network shown in Fig. 3 created the appropriate experimental conditions. Here: (i) each actor had two potential exchange partners; (ii) the prospective exchanges had different value so that each actor had a clear partner preference; and (iii) of the two key exchange relations, A-B was voluntary (i.e. actors A and B were each other's first choice) and G - D was involuntary (i.e. actors G and D were each other's second choice). The G - D relation essentially formed "on the rebound," because preferred partners A and B were exchanging with one another. The results of this study showed that the commitment process was present in both the voluntary A-B and the involuntary G - D exchange relations, but it was consistently stronger in the voluntary A-B relation. Thus, there is support not only for the role of relational cohesion in equal-power relations, as shown in the stem network, but also for the affect

154

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

1200

A

B 600

600

G

D 400 Fig. 3. The Box Network.

theory notion that the emotional process is stronger when actors perceive control over the choice of an exchange partner (Lawler, 1992a, 1997).

EXTENSION 2: GROUP COHESION IN PRODUCTIVE EXCHANGE The next significant development in the relational cohesion research program simultaneously expanded the theory along two fronts. First, the theory was tested in a new productive exchange context. Productive exchange is one of four basic forms of exchange identified by exchange theorists (see Emerson, 1976, 1981; Molm & Cook, 1995), the others being negotiated, reciprocal, and generalized. Productive exchange is group-oriented and indirect. That is, the exchange occurs from person-to-group. Negotiated exchange and reciprocal exchange are inherently dyadic. Negotiated exchange entails the direct exchange of offers and counter offers and explicit, binding agreements before benefits transfer. Reciprocal exchange involves sequential giving over time, without an agreement, and with unspecified obligations for reciprocity. Generalized exchange is comprised of a sequence of unilateral giving, over time, among three or more actors in which they give and receive from different others. A common form is a closed chain of giving, in which A gives to B, B gives to C, and C gives to A. This is often termed "chain generalized" in the exchange literature (Ekeh, 1974; Yamagishi & Cook, 1993). Like productive exchange, generalized exchange is a multi-actor form involving three or more individuals. These different forms of exchange are themselves underlying structural conditions, capable of generating variations in cohesion and solidarity (see Lawler, 2001 for a theoretical analysis).

The Theory of Relational Cohesion

155

To date, almost all of the research on social exchange involves "direct exchange" in the form of negotiated or reciprocal transactions (Cook & Emerson, 1978; Markovsky, Wilier & Patton, 1988; Thye, 2000; Wilier, 1999; see Yamagishi & Cook, 1993 for an exception). Productive exchange has received very little attention from exchange theorists; yet, it can have important implications for how exchange processes lead to group formation. We first clarify the distinct properties of productive exchange, and then describe an extension of relational cohesion theory and empirical test in a productive exchange context. Lawler, Thye and Yoon (2000) conceptualize productive exchange in terms of four basic properties. First, productive exchange involves two or more individuals who combine resources to produce a single, socially produced outcome. This outcome can be an object (such as a book chapter authored by three colleagues) or another event (such as a department potluck dinner). Second, productive exchange entails higher degrees of interdependence compared to the other forms of exchange (Molm, 1994). The reason is that productive exchange relations can be disrupted by any single member, and thus, require the assent of all members to be successful. Third, productive exchange poses significant coordination problems for the actors. Given the high levels of interdependence and the multi-actor nature of the productive exchange context, the production of joint activity is simply more challenging than in dyadic exchange. Finally, in productive exchange, inputs flow from person-to-group and benefits flow from group-to-person. All other forms of exchange entail person-to-person inputs and flows of benefit. The second major contribution of this research is that we empirically compare the emotional-affective process of relational cohesion theory to an uncertainty reduction process (Lawler, Thye & Yoon, 2000). Recall that the traditional exchange theory explanation for commitment is that frequent exchanges reduce uncertainty (Cook & Emerson, 1984). That is, actors who exchange frequently should learn more about one another, come to find each others' behavior more predictable, and come to learn they are similarly oriented to the exchange (Cook & Emerson, 1984; Emerson, 1981; Kollock, 1994, 1999). Building on this idea, we expanded the relational cohesion model to test whether uncertainty reduction is a distinct, yet complementary, pathway to commitment vis-?a-vis emotion. Figure 4 shows the revised theoretical model tested in this research. In general terms, the pathway through uncertainty reduction can be construed as a boundary defining process through which exchange partners become salient, distinctive, and set off relative to other potential partners. Social identity theorists frequently use the term "distinctiveness" to describe the property of

156

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

\

/ 2

0~

g3

.=

O

o

®

d:

,4

/ m

o~

la.

\ m

Q.

o

i--e.-

tic

"r" o" kl.I

The Theory of Relational Cohesion

157

being attached to a group or larger social category (Brewer, 1991, 1993). Our boundary defining process can be construed as a process through which a sense of distinctiveness emerges. Moreover, as the focal exchange relation becomes more familiar and predictable, it should be perceived as less risky relative to others. It has long been established in cognitive psychology that, all else being equal, individuals tend to become more risk aversive when the decision context is "framed" positively (Tversky & Kahneman, 1981). In our study, the negotiations among actors are framed positively because they stand to gain more in the focal group or relation than elsewhere. As such, risk aversion processes should render the focal group or exchange relation a salient object of awareness. At the same time, the pathway through positive emotion can be construed as a social bonding process. The positive emotion from frequent exchange is, in part, generated by the other actors and the group context itself. Such emotions are "rewards" generated by the exchange and completion of joint activity. As such, actors should strive to reproduce these rewards and also think about their proximate causes. To the extent that the group is perceived as a cause of the positive emotional experience, the group itself should come to take on expressive value in its own right (Tyler, 1990, 1994). It is now generally understood that boundary defining and social bonding are distinct, complementary mechanisms that produce group cohesion (Yoon & Thye, 2002). The theory of relational cohesion reflects this logic. The revised theoretical model was tested using a new productive exchange experimental protocol. In this context, three actors faced a task in which they could produce greater joint benefits if they all collaborated than if they operated alone or worked with another group. Consistent with the concept of productive exchange defined above, the exchanges were structured such that: (i) actors in this context were deciding whether to engage in a single collaborative effort that would produce a pool of joint profit; (ii) for an exchange to be consummated, all actors had to agree to the exchange; (iii) the exchange would allocate the pool of profits across actors; and (iv) offers were made simultaneously and independently which posed significant coordination problems. Overall, joint efforts produced joint profits (actor-to-group flow of benefits) and the joint profit benefited each of the actors (group-to-actor flow of benefits). Structural power conditions were manipulated by varying the relative (equal vs. unequal) and total (high vs. low) dependence of each member on the group (see Lawler, Thye & Yoon, 2000). Consistent with the earlier tests of relational cohesion theory, dependence was operationalized as the quality (expected value) of a fixed outside offer that could be accepted in the event the focal group did not reach agreement. Under these conditions, subjects exchanged for a total of

158

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

16 episodes. At several points in time exchange frequency, positive emotion, predictability, and relational cohesion were measured. Additionally, two kinds of commitment behavior were studied. After the 13th episode, subjects could either give one another small token gifts as a symbol of their relationship (i.e. gifts of small pieces of candy) or they could invest some of their earnings in a new joint venture that involves considerable risk but could provide substantial benefits (i.e. investment in a 3-person prisoner's dilemma game). The analysis focused on whether or not behavioral commitment emerges through the sequence of steps shown in the model. Here we briefly review the main findings for each step in turn. First, as predicted, the data indicate that structural power conditions have a significant effect on exchange frequency. Under conditions of high total dependence (i.e. the expected payoff from the alternative group is smaller than the expected payoff from the focal group) and equal relative dependence (i.e. the expected payoff from the alternative group is the same for each member of the focal group), the actors found it easier to reach agreement. In turn, frequent social exchange had a significant direct effect on both positive emotion and uncertainty reduction (i.e. predictability). These findings are important because they: (i) replicate and further verify the emotional effects of frequent exchange; and (ii) support the hypothesis that exchange also generates uncertainty reduction or predictability. The latter finding is consistent with standard exchange-theoretic explanations for commitment and recent empirical tests (e.g. see Kollock, 1994). The next step in the theorized causal chain indicates that both uncertainty reduction and positive emotion increase perceptions of group cohesion. The results indicate that positive emotion has a significant effect on perceptions of group cohesion, as hypothesized, but uncertainty reduction does not. In short, the data indicate that positive emotions are more important to the development of group cohesion than uncertainty reduction. If uncertainty reduction is an important factor in the development of commitment, then it must operate through a pathway other than perceptions of cohesion. The emotional affective process at the core of relational cohesion theory receives significant support, whereas the role of the uncertainty reduction process is questioned. Finally, the theory predicts that group cohesion is the proximal cause of both gift giving and contributions to a social dilemma, our measures of commitment. The results for this prediction are mixed, but yield interesting implications. Consistent with virtually all research in the relational-cohesion program, perceived cohesion had a significant effect on gift giving. However, group cohesion did not significantly affect the propensity of actors to invest in a new venture (i.e. cooperate in the social dilemma). In previous work on dyads, relational cohesion effects have been found for this form of commitment

159

The Theory o f Relational Cohesion

behavior (Lawler & Yoon, 1996). The difference could be due to the fact that the obstacles to cooperation are known to be more difficult in a three-person prisoners' dilemma than in a two-person prisoners' dilemma. The addition of a third person heightens uncertainty and makes trust more difficult for actors under these conditions. We suspected initially that this would make it even more likely that the uncertainty-reduction pathway would be present and operate through group cohesion. Another possibility is that uncertainty reduction does indeed operate, but not through perceived group cohesion or, by implication, the assumed boundary-defining process. To investigate this, we changed the original theoretical model to include several new pathways suggested by prior theory and by our data. The results revealed a direct effect of perceived predictability on the investment-dilemma form of commitment. Thus, uncertainty reduction does operate in the productive exchange context, but not in the way we originally theorized. What makes this alternative pathway to commitment important is that it can be interpreted in terms of trust. Trust - defined as expectations of cooperation by others (Pruitt & Kimmel, 1977) - is a key concept in theory and research on social dilemmas and how they are resolved (Axelrod, 1984; Kollock, 1994, 1999; Komorita & Parks, 1996; Yamagishi, 1986). To be trusted, one must first be predictable, so in this regard, predictability can be construed as a necessary (though not sufficient) condition for the emergence of trust. If so, we should observe a direct relationship between predictability and investment, as we did. To summarize, we find considerable evidence for a dual-process conception of commitment wherein emotional-affective and uncertainty-reduction mechanisms promote different forms of commitment behavior. Of particular importance for relational cohesion theory is that the emotional/affective process operates as a separate and independent mediating process leading to commitment behavior. Other processes - such as uncertainty reduction, trust and norm formation - have been emphasized in research on exchange, contracting, and social dilemmas (e.g. Cook & Emerson, 1984; Macy & Skvoretz, 1998; Williamson, 1975, 1981; Yamagishi, 1986). Relational cohesion theory is unique in its emphasis on the emotional aspects of social exchange, and in demonstrating how relations become salient and valued in and of themselves. SUMMARY

OF IMPLICATIONS

Over the past decade our research program has evolved to include new theoretical concepts and empirical phenomena on how relational cohesion and commitment develop in exchange relations (Lawler, 1992a, 1997, 2001, 2002;

160

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

Lawler & Thye, 1999; Lawler & Yoon, 1993, 1996, 1998; Lawler, Thye & Yoon, 2000). There are both implicit and explicit implications that capture what we have done, and are a springboard for moving beyond the domains we have specifically studied. Below, we set forth a series of general principles that abstract the program's broad implications, and foreshadow our current and future directions. Proposition 1: Structures of equal power and of high total power (mutual dependence) are most prone to relational or group commitment. Power is typically seen as a dark, negative, destructive force that divides, differentiates, and stratifies human actors. Using a power-dependence perspective (Emerson, 1972a, b), we theorize a positive side to power, suggesting how and when structural power conditions promote cohesion in and commitment to social units. Virtually all of our research since 1993 indicates that, ceteris paribus, equal power relations are more likely to yield exchanges, positive emotions, perceptions of relational cohesion, and behavioral acts of commitment. The same notions apply to relations of higher total power or mutual dependence in social structures. While we have applied these notions to particular exchange networks (modified line, branch, stem, and box networks), the underlying principles should be broadly applicable to social networks. A question that needs to be addressed is whether the notion of cohesion at the dyadic level can be extended to the network level. Are there network conditions likely to generate positive emotions, perceptions of cohesion, and a sense of coming together across the entire network? We have developed a concept of network cohesion that specifies two such network conditions: (i) the proportion of relations within a network that are equal power, and (ii) the degree of relational density in the network (Thye & Lawler, 1999). The main assertion is that exchange networks containing a high degree of equal power relations and many direct ties among actors will unleash the endogenous process of relational cohesion theory at the network level. This could help us understand if, and when, exchange networks become minimal groups (Tajfel & Turner, 1986). Proposition 2: Joint social activities produce positive emotions; and positive emotions are an important mechanism producing micro social order or solidarity. This casts the endogenous process more broadly. An important underlying assumption is that social exchange is a joint activity par excellence. That is, it can only be accomplished with others. As task-oriented behavior, exchange is

The Theory of Relational Cohesion

161

likely to make people feel good or bad after the fact, and as a joint task, they may infer similar or shared emotional experiences (Collins, 1981). One can understand the conditions under which structures generate commitment behavior (via emotions) in terms of jointness and shared experience (Lawler & Thye, 1999; Lawler, Thye & Yoon, 2000; Lawler & Yoon, 1996). Equal power may generate more sense of a joint task and shared emotions than unequal power relations, and the same can be said for higher levels of total power or mutual dependence. Perhaps the most unique feature of our research program, compared to others that focus on cohesion and commitment, is that positive emotion is the primary mechanism connecting exchange to outcomes such as cohesion, commitment, and solidarity (cf. Doreian & Fararo, 1998; Markovsky & Lawler, 1994). A next step is to look more closely at the nature of exchange tasks, how they generate emotion, and the conditions under which social-unit attributions are made to relations and networks. This step is undertaken in recent theorizing (Lawler, 2001, 2002). Lawler (2001) has proposed an affect theory of social exchange that places an emoting actor at the center of exchange theorizing. He argues that different structural forms of exchange - productive, negotiated, reciprocal, and generalized (Emerson, 1981; Molm & Cook, 1995) - entail tasks with different degrees of jointness; and, this effects the solidarity of resulting relations or groups. Tasks that involve more jointness ostensibly promote a stronger sense of shared responsibility for the results of exchange. Shared responsibility, in turn, promotes attributions of emotion to relevant social units, because these are the context for actors' common focus and activity. This theory can help explain when and how networks become groups on a cognitive or behavioral level (see Lawler, 2001). For example, the conditions for high network cohesion - dense ties of equal power (Thye & Lawler, 1999) - may also generate a greater sense of shared or collective responsibility at the network level. Future research will test these and other predictions of the affect theory of social exchange.

Proposition 3: Social identity processes undergird and strengthen cohesion and commitment. Over the past decade, the theory of relational cohesion has forged closer connections to the principles of social identity (Brewer, 1993; Brown, 2000; Tajfel & Turner, 1986). Emphasizing the interdependence and common fate underlying exchange relations, and the distinctiveness of focal vs. alternative relations, it is not difficult to imagine that relational cohesion involves "incipient group formation" (Lawler & Thye, 1999; Lawler & Yoon, 1996). Whereas early research established the importance of emotions (Lawler & Yoon, 1993), later

162

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

research added "relational cohesion" to the endogenous process, and suggested that the emotions are a source of incipient group formation. Relational cohesion might be construed as a form of "group identity" driven by common emotional experiences of actors (Collins, 1981, 1989). An important next step is for us to investigate the conditions under which larger and more complex exchange networks develop into groups. This is a fundamental question posed by Emerson (1972a, b, 1981) and never fully answered by social exchange theorists. At this stage we now have the theoretical apparatus to undertake this task, given: (i) strong and cumulative support for relational-cohesion theory (Lawler, Thye & Yoon, 2000; Lawler & Yoon, 1993, 1996); (ii) the theoretical extension to network-level cohesion (Thye & Lawler, 1999); and (iii) principles of the affect theory of social exchange (Lawler, 2001). With this framework already in place, there is little doubt our research can move toward answering Emerson's (1972a, b, 1981) question: When do networks of self-interested actors develop into groups of collectivelyoriented actors? The theory of relational-cohesion is unique in its emphasis on the role of emotions in social exchange and its focus on the mediating processes through which social structures strengthen or weaken the cohesiveness of exchange relations. There are a wide variety of ways that emotions can play a role in social exchange (see Lawler & Thye, 1999 for an analysis of these). In the theory of relational cohesion, the primary focus is on how emotions induce "social formations." We argue that when emotions mediate social formations, the relevant social units take on value in and of themselves. For this reason, emotion-based cohesion renders social units - be they relations, networks, or groups - expressive and symbolic of shared activity, common experience, and collective affiliation. Emotions offer a way to understand how social exchange processes generate social formations with considerable resilience, for good or ill. NOTE 1. Logrolling is a bargaining strategy that occurs when subjects trade off concessions across multiple commodities. For instance, logrolling occurs if subject A concedes more of commodity x, while subject B concedes more of commodity y. ACKNOWLEDGMENTS The order of authorship does not reflect differential contributions. This research was supported by National Science Foundation grants SBR-9022192, SES-

The Theory o f Relational Cohesion

163

9222668, SBR-9514860, SBR-9817706, and SBR-9816259. The authors thank Marty Kuhn for his programming expertise, and Christine Witkowski for comments on a previous version. REFERENCES Axelrod, R. (1984). The Evolution of Cooperation. New York: Basic Books Inc. Bacharach, S. B., & Lawler, E. J. (1981). Bargaining: Power, Tactics, and Outcomes. San Francisco, CA: Jossey-Bass. Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory. Englewood Cliffs, NJ: Prentice-Hall. Berger, P. L., & Luckmann, T. (1966). The Social Construction of Reality. New York: Doubleday. Blau, P. (1964). Exchange and Power in Social Life. New York: John Wiley and Sons, Inc. Bonacich, P., & Bienenstock, E. J. (1993). Assignment Games, Chromatic Number and Exchange Theory. Journal of Mathematical Sociology, 17, 243-259. Brewer, M. B. (1991). The Social Self: On Being the Same and Different at the Same Time. Journal of Personality and Social Psychology Bulletin, 17, 475-482. Brewer, M. B. (1993). Social Identity, Distinctiveness, and In-group Homogeneity. Social Cognition, 11, 150-164. Brown, R. (2000). Group Processes (2nd ed.). Oxford, U.K.: Blackwell Publishers. Burke, P. (1997). An Identity Model for Network Exchange. American Sociological Review, 62, 134-150. Byrne, D., Clore, G. L., & Smeaton, G. (1986). The Attraction Hypothesis: Do Similar Attitudes Affect Anything? Journal of Personality and Social Psychology, 51, 1167-1170. Collins, R. (1975). Conflict Sociology. New York: Academic. Collins, R. (1981). On the Microfoundations of Macrosociology. American Journal of Sociology, 86, 984-1014. Collins, R. (1989). Toward a Neo-Meadian Sociology of Mind. Symbolic Interaction, 12, 1-32. Cook, K. S., & Emerson, R. M. (1978). Power, Equity, and Commitment in Exchange Networks. American Sociological Review, 43, 721-739. Cook, K. S., & Emerson, R. M. (1984). Exchange Networks and the Analysis of Complex Organizations. In: S. B. Bacharach & E. J. Lawler (Eds), Research on the Sociology of Organizations (Vol. 3, pp. 1-30). Greenwich, Conn.: JAI Press. Cook, K. S., Emerson, R. M., Gillmore, M. R., & Yamagishi, T. (1983). The Distribution of Power in Exchange Networks: Theory and Experimental Evidence. American Journal of Sociology, 89, 275-305. Cook, K. S., & Yamagishi, T. (1992). Power in Exchange Networks: A Power Dependence Formulation. Social Networks, 14, 245-265. Deci, E. L. (1975). Intrinsic Motivation. New York: Plenum Press. Doreian, P., & Fararo, T. (1998). The Problem of Solidarity: Theories and Models. Amsterdam: Gordon and Breach Publishers. Durkheim, E. (1915). The Elementary Forms of Religious Life. New York: Free Press. Ekeh, P. (1974). Social Exchange Theory. Cambridge, MA: Harvard University Press. Elster, J. (1986). Introduction. In: J. Elster (Ed.), Rational Choice (pp. 1-33). New York: New York University Press. Emerson, R. M. (1964). Power-Dependence Relations: Two Experiments. Sociometry, 27, 282-298.

164

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

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). Boston: 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). Boston: Houghton-Mifflin. Emerson, R. M. (1976). Social Exchange Theory. Annual Review of Sociology, 2, 335-362. Emerson, R. M. (1981). Social Exchange Theory. In: M. Rosenberg & R. H. Turner (Eds), Social Psychology: Sociological Perspectives (pp. 30~5). New York: Basic Books, Inc. Foa, U. G. (1971). Interpersonal and Economic Resources. Science, 171, 345-350. Friedkin, N. (1992). An Expected Value Model of Social Power: Predictions for Selected Exchange Networks. Social Networks, 14, 213-229. Halaby, C. N., & Weakliem, D. (1989). Worker Control and Attachment to the Firm. American Journal of Sociology, 95, 549-591. Heath, A. (1975). Rational Choice and Social Exchange. Cambridge, England: Cambridge University Press. Homans, G. L. (1961). Social Behavior: Its Elementary Forms. New York: Harcourt Brace Jovanovich. Izard, C. E. (1971). The Face of Emotion. New York: Appleton-Century-Crofts. Izard, C. E. (1977). Human Emotions. New York: Plenum Press. Izard, C. E. (1991). The Psychology of Emotion. New York: Plenum Press. Kanter, R. M. (1968). Commitment and Social Organization: A Study of Commitment Mechanisms in Utopian Communities. American Sociological Review, 33, 499-517. Kanter, R. M. (1972). Commitment and Community: Communes and Utopias in Sociological Perspective. Cambridge, MA.: Harvard University Press. Kemper, T. D. (1978). A Social lnteractional Theory of Emotions. New York: Wiley. Kollock, P. (1994). The Emergence of Exchange Structures: An Experimental Study of Uncertainty, Commitment, and Trust. American Journal of Sociology, 100, 315-345. Kollock, P. (1999). The Production of Trust in Online Markets. In: S. R. Thye, E. J. Lawler, M. W. Macy & H. A. Walker (Eds), Advances in Group Process (Vol. 16, pp. 99-123). Stanford, Conn: JAI Press. Komorita, S. S., & Parks, C. D. (1996). Social Dilemmas. Boulder, CO: Westview Press. Larsen, R. J., & Diener, E. (1992). Promises and Problems With the Circumplex Model of Emotion. In: M. S. Clark (Ed.), Emotion (pp. 25-59). Newbury Park, CA: Sage. Lawler, E. J. (1992a). Choice Processes and Affective Attachments to Nested Groups: A Theoretical Analysis. American Sociological Review, 57, 327-339. Lawler, E. J. (1992b). Power Processes in Bargaining. The Sociological Quarterly, 33, 17-34. LaMer, E. J. (1997). Affective Attachments to Nested Groups: The Role of Rational Choice Processes. In: J. Skvoretz, J. Szmatka & J. Berger (Eds), Status, Networks, and Structures: Theory' Development in Group Processes. Stanford: Stanford University Press. Lawler, E. J. (2001). An Affect Theory of Social Exchange. American Journal of Sociology, 107, 321-352. Lawler, E. J. (2002). Micro Social Orders. Social Psychology Quarterly, 65(1), 4-17. Lawler, E. J., & Thye, S. R. (1999). Bringing Emotions into Social Exchange Theory. Annual Review of Sociology, 25, 217-244. Lawler, E. J., Thye, S. R., & Yoon, J. (2000). Emotion and Group Cohesion in Productive Exchange. American Journal of Sociology, 106(3), 61~657. Lawler, E. J. (unpublished). Voluntarism and Commitment in Exchange Networks. Manuscript available from the authors upon request.

The Theory of Relational Cohesion

165

Lawler, E. J., & Yoon, J. (1993). Power and the Emergence of Commitment Behavior in Negotiated Exchange. American Sociological Review, 58, 465-481. Lawler, E. J. (1996). Commitment in Exchange Relations: Test of a Theory of Relational Cohesion. American Sociological Review, 61, 89-108. Lawler, E. J. (1998). Network Structure and Emotion in Exchange Relations. American Sociological Review, 63, 871-894. Lawler, E. J., Yoon, J., Baker, M. R., & Large, M. D. (1995). Mutual Dependence and Gift Giving in Exchange Relations. In: B. Markovsky, J. O'Brien & K. Heimer (Eds), Advances in Group Process (Vol. 15). Greenwich, Conn: JAI Press. Lovaglia, M. J., Skvoretz, J., Wilier, D., & Markovsky, B. (1995). Negotiated Outcomes in Social Exchange Networks. Social Forces, 74(1), 123-155. Macy, M. W., & Skvoretz, J. (1998). The Evolution of Trust and Cooperation Between Strangers: A Computational Model. American Sociological Review, 63, 638~560. Markovsky, B., & Lawler, E. J. (1994). A New Theory of Group Solidarity. In: B. Markovsky, J. O'Brien & K. Heimer (Eds), Advances in Group Processes (Vol. 11, pp. 113-137). Grenwich, Conn: JAI Press. Markovsky, B., Wilier, D., & Patton, T. (1988). Power Relations in Exchange Networks. American Sociological Review, 53, 220-236. Markovsky, B., Skvoretz, J., Wilier, D., Lovaglia, M. J., & Erger, J. (1993). The Seeds of Weak Power: An Extension of Network Exchange Theory. American Sociological Review, 58, 197-209. Molm, L. (1987). Power Dependence Theory: Power Processes and Negative Outcomes. In: E. J. Lawler & B. Markovsky (Eds), Advances in Group Processes (Vol. 4, pp. 171-198). Grenwich, Conn: JAI Press. Molm, L. (1990). Structure, Action, and Outcomes: The Dynamics of Power in Social Exchange. American Sociological Review, 55, 427-447. Molm, L. (1994). Dependence and Risk: Transforming the Structure of Social Exchange. Social Psychology Quarterly, 57, 163-89. Molm, L., & Cook, K. (1995). Social Exchange and Exchange Networks. In: K. Cook, G. Fine & J. House (Eds), Sociological Perspectives on Social Psychology (pp. 209-235). Boston: Allyn and Bacon. Mueller, C. W., & Lawler, E. J. (1999). Conurutment to Nested Organizational Units: Some Basic Principles and Preliminary Findings. Social Psychology Quarterly, 62, 325-346. Mueller, C. W., & Price, J. L. (1990). Economic, Psychological and Sociological Determinants of Voluntary Turnover. Journal of Behavioral Economics, 9, 321-335. Parsons, T. (1951). The Social System. New York: Free Press. Pruitt, D. J., & Kimmel, M. J. (1977). Twenty Years of Experimental Gaming: Critique, Synthesis and Suggestions for the Future. Annual Review of Psychology, 28, 363-392. Rabbie, J. M., & Horowitz, M. (1988). Categories vs. Groups as Explanatory Concepts in Intergroup Relations. European Journal of Social Psychology, 19, 172-202. Russell, J. A. (1980). A Circumplex Model of Affect. Journal of Personali~ and Social Psychology, 39(6), 1161-1178. Simpson, B., & Wilier, D. (1999). A New Method For Finding Power Structures. In: D. Wilier (Ed.), Network Exchange Theo~ (pp. 270-284). Westport, CT: Praeger. Skvoretz, J., & Wilier, D. (1993). Exclusion and Power: A Test of Four Theories of Power in Exchange Networks. American Sociological Review, 58, 801-818. Tajfel, H., & Turner, J. C. (1979). An Integrative Theory of Intergroup Conflict. In: W. G. Austin & S. Worchel (Eds), The Social Psychology of Intergroup Relations. Monterey, CA: Brooks/Cole.

166

SHANE R. THYE, JEONGKOO YOON AND EDWARD J. LAWLER

Tajfel, H. (1986). The Social Identity Theory of Intergroup Behavior. In: S. Worchel & W. G. Austin (Eds), Psychology of Intergroup Relations (pp. 7-24). Chicago, IL: NelsonHall. Tallman, I., Gray, L., & Leik, R. (1991). Decisions, Dependency, and Commitment: An Exchange Based Theory of Group Development. In: E. J. Lawler, B. Markovsky, C. Ridgeway & H. H. Walker (Eds), Advances in Group Processes (Vol. 9). Greenwich, Conn: JAI Press. Thibaut, J. W., & Kelley, H. H. (1959). The Social Psychology of Groups. New York: John Wiley and Sons, Inc. Thye, S. R. (2000). A Status Value Theory of Power in Exchange Relations. American Sociological Review, 65, 407-432. Thye, S. R., & Lawler, E. J. (1999). Collaborative Research on Social Exchange and Network Cohesion. Proposal to the National Science Foundation (funded May 15, 1999). Thye, S. R., Lovaglia, M. J., & Markovsky, B. (1997). Responses to Social Exchange and Social Exclusion in Networks. Social Forces, 75(3), 1031-1047. Tversky, A., & Kahneman, D. (1981). The Framing of Decisions and The Psychology of Choice. Science, 211,453-458. Tyler, T. R. (1990). Why People Obey the Law: Procedural Justice, Legitimacy, and Compliance. New Haven, CT: Yale University Press. Tyler, T. R. (1994). Psychological Models of the Justice Motive: Antecedents of Distributive and Procedural Justice. Journal of Personality and Social Psychology, 67, 850-863. Walker, H. A., Thye, S. R., Simpson, B., Lovaglia, M. J., Wilier, D., & Markovsky, B. (2000). Network Exchange Theory: Recent Developments and New Directions. Social Psychology Quarterly, 63(4), 324-337. Walton, R. E., & McKersie, R. B. (1965). A Behavioral Theory of Labor Negotiations: An Analysis of a Social Interaction System. New York: ILR Press. Watson, D., & Tellegen, A. (1985). Toward a Consensual Structure of Mood. Psychological Bulletin, 98, 219-235. Weiner, B. (1985). An Attributional Theory of Achievement Motivation and Emotion. Psychological Review, 92, 548-573. Weiner, B. (1986). An Attributional Theory of Motivation and Emotion. New York: Springer-Verlag. Willer, D. (1999). Network Exchange Theory. Westport CT: Praeger. Willer, D., & Anderson, B. (Eds) (1981). Networks, Exchange, and Coercion. New York: Elsevier. Williamson, O. E. (1975). Markets and Hierarchies: Analysis and Antitrust Implications: A Study in the Economics of Internal Organization. New York: Free Press. Yamagishi, T. (1986). The Provision of a Sanctioning System as a Public Good. Journal of Personality and Social Psychology, 51, 110-116. Yamagishi, T., & Cook, K. (1993). Generalized Exchange and Social Dilemmas. Social Psychology Quarterly, 56, 253-248. Yoon, J., & Thye, S. R. (2002). A Dual Process Model of Organizational Commitment. Work and Occupations, 29, 97-124.

EMBEDDED TRUST: CONTROL AND LEARNING Vincent Buskens and Werner Raub

ABSTRACT This paper discusses two mechanisms through which social embeddedness can affect trust among actors in cooperative relations. Trust can be based on past experiences with a partner or trust can be built on possibilities for sanctioning an untrustworthy trustee through own or third-party sanctions. These two mechanisms are labeled learning and control. The mechanisms are often left implicit or discussed in isolation in earlier research. Learning and control can operate at different levels: at the dyadic level and at the network level. We argue that for understanding trust the two mechanisms should be studied simultaneously, theoretically as well as empirically. We show that this is more easily said than done by addressing some of the theoretical as well as empirical issues. We offer preliminary evidence of the simultaneous working of the learning and control mechanisms at the dyadic level and the network level.

1. INTRODUCTION People encounter every day trust situations with persons they have never seen before or with others they already know for a considerable time, with complete strangers or with friends of their friends. For example, professors lend books

to colleagues or students. Most people are more reluctant to lend their books

Group Cohesion, Trust and Solidarity, Volume 19, pages 167-202. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2

167

168

VINCENT BUSKENS AND WERNER RAUB

to a complete stranger than to somebody they know or, at least, their friends know. Likewise, if somebody has ever returned your book with coffee spots on every other page, you might be more reluctant to lend him a book again. If nobody would be willing to lend books, everybody has to buy all books himself, which is more expensive compared to the situation that books can be shared among a group of people. Therefore, trust increases efficiency in human interactions (Arrow, 1974). Lending books to colleagues or students resembles social exchange (Blau, 1964 [1996]). Trust also plays a crucial role in economic exchange, for example, in inter-film relations. Buyers trust that suppliers will deliver good products without writing contracts that cover all possible contingencies (Macaulay, 1963). Negotiating and writing such contracts is costly and reduces the profits related to the transaction for both parties. Therefore, trust can make transactions more efficient but the problem remains that the supplier often has an incentive to abuse trust and obtain an extra profit in a transaction, for example, by selling an inferior product for the price of a high-quality one. If the buyer considers this risk to be too high, she will not trust the supplier. Before we continue, we need to make explicit what we mean by trust. The definitions used in the literature vary considerably. We assume that actors are rational in the sense that trust is only possible if, for the trustor, the expected outcome of placing trust is preferred over the expected outcome of not placing trust. This is similar to Williamson's (1993) notion of "calculative trust." Trust can be problematic because the trustor is uncertain about the abilities and competencies of the partner. In the literature (e.g. Barber, 1983; see also Snijders, 1996, Chap. 1 for a discussion), this is sometimes labeled as a problem of "confidence." Here, we address trust problems due to incentives of the trustee rather than his abilities: The trustee has an incentive to abuse trust and the trusting actor has something to lose if trust is abused. Coleman (1990, pp. 97-99) distinguishes four elements that define a trust situation between a trnstor and a trustee. (1) Placing trust by the trustor allows the trustee to honor or abuse trust. This action would not have been possible without placing trust by the trustor. (2) The trustor regrets placing trust if trust is abused, but benefits from honored trust. (3) The trustor voluntarily places resources in the hands of the trustee without formal safeguards. (4) There is a time-lag between placement of trust and the action of the trustee. The Trust Game as shown in Fig. 1 is a game-theoretic representation of a transaction that closely resembles Coleman's definition of a trust situation

Embedded Trust: Control and Learning

169

(see also: Camerer & Weigelt, 1988; Dasgupta, 1988; Kreps, 1990). The Trust Game starts with a move by the trustor, who chooses between placing trust and not placing trust. If the trustor does not place trust, the game is over and the trustor obtains a payoff PI, while the trustee receives P2" If the trustor places trust, the trustee chooses between honoring trust and abusing trust. If the trustee honors trust, the trustor and trustee receive, respectively, R i > Pi, i = 1, 2. If the trustee abuses trust, the trustor receives Sj < P~ and the trustee T2 > R~. We assume that these payoffs represent utilities for the actors. Assume that the Trust Game with all payoffs is known to both actors in the game. The trustee will abuse trust if trust is placed, because the payoff for abusing trust is larger than the payoff for honoring trust (T2 > R2). The trustor realizes that the trustee will abuse trust if trust is placed. Thus, because the trustor is better off not placing trust than when she does place trust and trust is abused (S I < P0, a rational trustor will not place trust. Bold lines indicate these moves of the trustee and the trustor in Fig. 1. Note that "abuse trust" is not observed if trust is not placed. The outcome constitutes a social dilemma because both actors are worse off in the situation that trust is not placed than in the case where trust is placed and honored (Ri > P). The Trust Game is a one-sided version of the Prisoner's Dilemma, since in the Trust Game there is only one actor who can profit from being "nasty."

~stor no tru7

/I

~ust

a b u s e / ( ~ k r u tilnor

/

t7

Pl

S~

P2

~Qustor

T2

~st

R~ R2

no t r u /

~ast

/

~rustee

abuse/ ~n~r

/

t r T / ¢

1 1

0 5

3 3

Fig. 1. Extensive Form of a Trust Game. R~ > P~, R2 > P2, P~ > Sv and T2 > R:. Note: The right-hand Trust Game is a numerical example. Bold lines indicate the equilibrium path of play.

170

VINCENT BUSKENS AND WERNER RAUB

Definition o f a trust situation. A trust situation is a transaction between two actors that resembles a Trust Game.

The discussion of the Trust Game highlights two issues. First, the trustfulness of the trustor, i.e. the extent to which the trustor places trust in the trustee. Second, the trustworthiness of the trustee, i.e. the extent to which the trustee honors trust if trust is placed. The analysis of the Trust Game suggests that the trustfulness of the trustor largely depends on her expectation of the trustworthiness of the trustee. If the trustor expects the trustee to abuse trust, she will not place trust. We will focus here on explaining the trustfulness of the trustor (who will be our "focal actor"), while the trustworthiness of the trustee is considered as one of the important predictors for trustfulness. We use the term trust to refer to the trustfulness of the trustor. It is important to note that trust is not something that is intrinsic to a particular person. Trust depends on the characteristics of both actors in a trust situation, on the relation between these two actors, and on the specific properties of the trust situation (Snijders, 1996). Finally, we define the size of the trust problem in a trust situation. We define T2-R 2 as the (size of the) trust problem, which equals the incentive for the trustee to abuse trust in a given trust situation. 1 This gives us the opportunity to study the extent to which a trustor trusts a trustee. The extent to which a trustor trusts a trustee in a certain situation equals the maximum size of the trust problem for which the trustor still will trust the trustee. As can be seen from the discussion of the Trust Game, trust is difficult in isolated encounters. Most trust situations, however, occur in a social context, where actors meet each other and each other's acquaintances more often over time. This is often referred to as "embeddedness" in the social science literature (e.g. Granovetter, 1985). In this paper, we discuss two mechanisms through which embeddedness in a social context may affect trust, which we label learning and control. 2 Learning refers to the possibility for actors to improve their choices in given interactions using experiences from past interactions. In contrast with assumptions in simple formal models of trust, actors are often uncertain about many aspects of an interaction with another actor, for example, the incentives of other actors or the properties of the social context. Both actors have certain beliefs about such parameters, for example, the trustor has some idea about the value of T2, but might not be completely certain about this value. Moreover, the trustee has a belief about the trustor's belief about the value of T2. Actors obtain information from previous trust situations about the incentives, beliefs, and other characteristics of other actors. This information can be obtained from

Embedded Trust: Control and Learning

171

the focal actor's past interactions with the trustee or from third parties who had interactions with the trustee. If a trustee has been trustworthy in past interactions, a trustor might become more convinced that the trustee will be trustworthy again in similar interactions in the future. Learning effects have been discussed frequently in the sociological literature (Granovetter, 1985, p. 400; Coleman, 1990; Burt & Knez, 1995a). Often, only the positive side of the emergence of trust as a result of positive experiences is emphasized, but Coleman as well as Butt and Knez show that trust can also decrease, or distrust can increase, if negative information about a trustee is revealed. Past interactions can affect present interactions not only via information and learning. Past interactions can have implications in terms of relation specific investments that are made with respect to a given partner. For example, if one starts using a certain type of software, it becomes often costly to switch to another type of software. Moreover, as far as learning is concerned, we concentrate on learning about characteristics of a partner. Actors might also learn about suitable mechanisms for coping with trust problems. For example, after some experience, writing a suitable contract becomes easier compared to the first time one writes a contract. Such aspects of learning are largely neglected in this paper. Control refers to the fact that trustors realize that trustees have short-term incentives for abusing trust, but that some long-term incentives for the trustee are under control of the trustor. The trustee has to take into account the threat that the trustor will not trust him in future transactions after he abuses trust and that the trustor might search for another partner. Moreover, a trustor can inform other trustors about untrustworthy behavior by the trustee, which might induce other potential partners of the trustee to refrain from trusting him. These sanction threats imply long-term losses for the trustee if they are executed. Therefore, trustors are expected to place trust if the trustee's long-term incentives of honoring trust are larger than the short-term incentives of abusing trust and if they do not have information that the trustee acted untrustworthy in the past. The mechanisms underlying control are also referred to as reciprocity (Gouldner, 1960; Blau, 1964 [1996]; Voss, 1982) or conditional cooperation (Taylor, 1987). Control through social networks can take rather extreme forms such as social ostracism by social groups if untrustworthy behavior is detected (see Wechsberg's, 1966, pp. 81-86 famous example of the diamond merchants). Control through social networks might be problematic due to conflicting perceptions and motives of the potential actors who have to execute the sanctions. It is possible that a buyer has positive experiences with a dealer, but receives negative signals from others about this dealer. In such a situation, it

172

VINCENT BUSKENS AND WERNER RAUB

seems less probable that the buyer neglects her own experiences and implements sanctions against the dealer. The farther information travels through a network, the more actors will doubt the content of the information (for example, see: Lorenz, 1988; Williamson, 199 pp. 153-155; Blumberg, 1997, pp. 208-210; Buskens, 2002, pp. 18-20). This implies that third-party effects of control are smaller than one would expect if such issues are neglected. Consequently, a careful research design in which third-party effects of control and learning can be disentangled is crucial for a convincing argument that control through third parties actually occurs. Two examples illustrate why learning as well as control is essential for understanding trust. In the first example learning is crucial. One prefers to hire a nanny with good references from a friend over a nanny without such references. Moreover, if one hires a nanny for a second child, one will first consider the nanny one had before. If she was trustworthy and seemed to treat your child well, she will probably be hired again. Clearly, in this example in which making the wrong choice might have dramatic consequences, learning about positive past behavior of the trustee is important. As a second example, consider long-term buyer-supplier relations. In such relations, control can be expected to play a key role in providing trust. Buyers often realize that the businesses of suppliers depend to a considerable extent on their reputation as trustworthy suppliers. If buyers can switch easily to other suppliers or if the supplier has many other customers in the neighborhood of the buyer, sanctions are relatively easy to implement for the buyer and, likewise, rather costly for the supplier. Therefore, if expected losses from abused trust are limited for the trustor and potential future losses for the trustee are obvious and can be expected to be sizeable, trustors will be able to base trust primarily on control. Table 1 summarizes the four embeddedness effects on trust studied in this paper. Although learning and control matter, it is difficult to figure out how they are interrelated and how bilateral effects relate to network effects. To what extent do control opportunities matter if you do not know anything about a

Table 1.

Summary of Four Effects on Trust Considered in this Paper. Two Levels Network

Two Mechanisms

Dyad

Learning

Information about the trustee from own past experiences. Possibilities to sanction a trustee oneself after abuse of trust.

Control

Information from third parties about their past experiences with a trustee. Possibilities to sanction a trustee through third parties after abuse of trust.

Embedded Trust: Control and Learning

173

certain partner? Perhaps the partner is very short-sighted and does not care about any future sanctions. And, if someone has own experiences with a partner, to what extent are experiences of network contacts taken into account? Purely intuitive reasoning will not easily lead to clear conclusions about the combined effects of learning and control. Consequently, development of formal theory is inevitable to derive hypotheses on the combined effects of these two mechanisms. However, in this paper we will concentrate on arguing that both effects matter, that there is theory for both mechanisms separately, and that there is empirical evidence for both mechanisms. Theory in which both mechanisms co-occur is almost non-existent and available theory is not very satisfying in various respects. In addition, we think that up to now there has been hardly any empirical evidence in which both mechanisms are clearly disentangled. Therefore, this paper offers some evidence for the co-occurrence of learning and control and how both mechanisms interact. We hope that the empirical evidence presented in this paper proves useful in directing further theoretical research. In the following section, we summarize theory concerning learning and control mechanisms related to trust. Section 3 offers evidence for learning and control from existing literature. Section 4 discusses in more detail two experiments in which evidence is found for the combination of learning and control. Finally, Section 5 presents ideas about how existing research on this topic can be extended and improved.

2. THEORIES 2.1. Learning Models "Pure" learning models are models in which actors do not consider the payoffs of their opponents when making own decisions) These models assume only a minimal level of rationality for the actors. Actors adapt their behavior based on past experiences using a limited set of rules. Two main types of learning models can be distinguished: belief-based learning models and choicereinforcement learning models. 4 In belief-based learning models, an actor plays best-responses against his beliefs about the probabilities for certain moves of other actors. He does not take into account that other actors adapt behavior as a result of his behavioral changes. Updating of beliefs is based on experiences with and information about past behavior of the other actors. Simultaneously, actors change their behavior to play best-responses given the updated beliefs. In choice-reinforcement learning models, actors adapt their propensities for playing certain moves on the basis of the payoffs they received themselves in

174

VINCENT BUSKENS AND WERNER RAUB

the past. If they are satisfied with current play they continue their behavior, otherwise they change their behavior. In economics, the development of learning models has been growing rapidly. Many studies are concerned with behavior "in the long run" and investigate the extent to which predictions of learning models are in the long run similar to predictions of more "rational" models in which the players take into account the incentives of other players (see, for example, Fudenberg & Levine, 1998). Another use of learning models in economics is to explain subjects' behavior in experiments (Roth & Erev, 1995; Camerer & Ho, 1999; Erev & Roth, 1999). Since the actors in learning models act rather mechanically, learning models can straightforwardly be represented in computer simulations. Sociologists have used this opportunity to study what might happen in large groups of actors that have to cope with social dilemma situations (for some recent examples: Messick & Liebrand, 1995; Flache, 1996; Heckathorn, 1996; Lomborg, 1996; Buskens & Snijders, 1997; Macy & Skvoretz, 1998). Most of these studies provide conditions under which cooperation problems or collective good problems can be solved in societies of actors who adapt their behavior if they are not satisfied with the outcome of present behavior) For example, Flache (1996) demonstrates that, in a pure learning model, dense social networks can lead to very high as well as very low production of collective goods depending on the bilateral relations of the actors. The effect largely disappears if actors take into account that they can be controlled by others in future periods of play. This is just one example that learning models do not need to lead to the same conclusion as control models. Formal learning models have not considered trust situations as defined here. However, there are some more informal studies on learning in trust situations. For example, Gulati (1995a) argues that strategic inter-firm alliances are less equity based if the alliance partners were partners before. Equity creates a joint ownership that "effectively deters opportunistic behavior" (p. 88). Gulati interprets this as an indication that trust grows with (positive) past experiences. Lawler and Yoon (1996) as well as Lawler et al. (2000), in their theory of relational cohesion, likewise argue that positive experiences in dyadic exchange relations as well as in triadic exchange induce more contributions of the actors in a joint venture which they conceptualize as a social dilemma. Their theory focuses on disentangling learning from earlier exchanges in the sense of uncertainty reduction and the emotional/affective consequences of earlier exchanges. Burt and Knez (1995a) show that dense networks amplify trust as well as distrust. Their argument is that due to the homogeneity of opinions in a dense network, actors become very convinced about some information because they receive the information disproportionately often. Clearly, these informal

Embedded Trust: Control and Learning

175

analyses correspond with what we would expect from formal learning models about trust. Given that a trustor is not concerned about the payoffs of the trustee, the trustor's only consideration is whether the probability that she receives R t if she places trust is large enough (see also Coleman, 1990, Chap. 5). In a belief-based learning model, this implies that the percentage p of positive experiences in the past should be larger than (Pt-S1)/(RI-St), since this is the condition for which pR~ + (1-p)S~ > P~ so that the expected utility of placing trust exceeds the expected utility from no trust. In a choice-reinforcement learning model, the trustor should have obtained R~ often enough after she placed trust to let her decide that placing trust leads to the most favorable outcome. And, although we will not elaborate on formal analysis of learning models for the Trust Game, it can generally be concluded that such models predict that if a trustor has more positive past experiences, the probability that she will trust a trustee is larger, while if she has more negative information, she will trust the trustee less. Buskens (2002, Chap. 4) shows how we can use this argument to derive hypotheses about the effects of structural properties of a trustor's network on the extent to which the trustor will trust a trustee under the assumption that information is either predominantly positive or negative. If information is mixed or ambiguous, this model does not provide predictions about effects on trust of learning through network embeddedness. Learning models do not provide hypotheses about control effects because they assume that trustors do not consider the payoffs of the trustee but condition their behavior purely on past experiences. Therefore, by definition, we cannot derive effects of control, which would be based on arguments about future payoffs for the trustee, from learning models. "Control" elements might be included in strategies that are used in learning models. For example, a pre-programmed strategy might prescribe that the trustor does not place trust in an encounter after trust has been abused, and this might be a well-performing strategy. TIT-FOR-TAT would be an obvious example (see Axelrod, 1984). Still, this is something else than placing trust because of the anticipation that the trustee honors trust due to the fact that he would be worse off in the long run by abusing trust. Based on the arguments and models mentioned above, we can derive five testable hypotheses, including three hypotheses about learning: (1) Trust increases with the potential gains for the trustor from honored trust (R1-St) and (2) decreases with the potential loss from abused trust (Pt-St). Under the assumption that information about the trustee is positive, (3) trust increases with the number of past experiences the trustor has had with the trustee; (4) trust increases with the extent to which the trustor receives information about the

176

VINCENT BUSKENS AND WERNER RAUB

trustee from other trustors (indegree); and (5) trust increases with the density of the network of trustors.

2.2. Control Models Models about control in trust situations are predominantly based on gametheoretic arguments. This is not surprising since trust based on control involves primarily considerations of the trustor about the trustee's incentives to abuse trust. The interdependence of two (or more) actors involved in trust situations is what drives the solutions of these models. First, it has to be noted that if the Trust Game is played only once between the same trustor and trustee, and there is no relation to any games with third parties that might be informed about the outcome of the Trust Game, there cannot be any control. Control is based on the fact that if the trustee abuses trust, a trustor who knows about this will not place trust, or at least will not place trust for some time after the abuse of trust. As a consequence, the trustee experiences a considerable loss because he will not receive more than P2 if he is not trusted. Hence, the trustee is trustworthy and the trustor can be trustful. We consider here "pure" control models. In these models it is assumed that players play rational and that all elements of the game are common knowledge (Rasmusen, 1994, p. 44). Then, by definition, hypotheses about learning cannot be derived from these models because all players are perfectly informed about the incentives and, implicitly, about the characteristics of the other players. 6 The infinitely repeated Trust Game analyzed by Kreps (1990) is an example of such a control model. In this model, the Trust Game is repeated after each period with a continuation probability c. Kreps shows that always placing trust and always honoring trust is an equilibrium outcome if players value future payoffs high enough and the continuation probability is high enough. More precisely, assume players value the payoff X, in a Trust Game that is n periods away as wn X, 0 < w < 1. Then, c × w > (T2-R2)/(T2-P 2) is a necessary and sufficient condition for always placing trust and always honoring trust to be an equilibrium outcome. Under this condition, the long-term losses for the trustee are larger than the short-term gains from abusing trust if the trustor does never place trust after any abuse of trust. In subsequent research, models for the repeated Trust Game have been extended to allow for a situation in which there is not one trustor but a network of trustors who can communicate about the behavior of the trustee (Raub & Weesie, 1993; Buskens, 1998; Weesie et al., 1998; Buskens & Weesie, 2000a; Buskens, 2002). 7 Weesie et al. (1998) find that the denser the network of the trustors, the more these trustors can trust the trustee, but this model allows only for a homogeneous network of

Embedded Trust: Control and Learning

177

trustors. Heterogeneous networks are introduced by Buskens (1998, 2002). The conclusion from these models is that not only network density has a positive effect on trust, but also the extent to which a trustor transmits information to other trustors (outdegree). The larger the outdegree of a trustor, the faster she transmits information about the trustee to other trustors in the network. Since these other trustors will not trust the trustee anymore, or at least will trust him less, if they receive negative information about the trustee, a trustor with a large outdegree has a large sanction potential against the trustee. This allows her to trust the trustee even if the temptation for the trustee to abuse trust is fairly large. In pure control models, we do not find effects of past transactions of the trustors with the trustee or effects of the indegree of the trustors. In control models, trust is purely based on the potential sanctions of the trustor after an eventual abuse of trust. However, since the trustors perfectly know the incentives of the trustee, they will never place trust if the trustee would abuse trust. Therefore, trust will never be abused in equilibrium. Consequently, there is no relevant information to be communicated about the outcomes of past interactions under the condition that there are no deviations from equilibrium play. This is also the reason why, for example, the incentives of the trustors do not affect the equilibria either. As we have seen before, the threshold value for c × w is related only to payoffs of the trustee. This is a striking contrast with the learning models in which the trustor's payoffs determine trust. We derive six testable hypotheses from the control models. Three of these hypotheses focus explicitly on control mechanisms: (1) Trust decreases with the trust problem, or the temptation for the trustee to abuse trust (T2-R2); (2) trust increases with the loss for the trustee from not placing trust (T~-P2) and (3) with the extent to which trustees value payoffs in the future (w); (4) trust increases with the probability that the same trustor and trustee have more interactions in the future (c). Moreover, trust increases with (5) the outdegree of the trustor and (6) the density of the trustor's network.

2.3. Combining Learning and Control There are a few models integrating effects of learning and control. In these models, trustors are uncertain about the incentives of the trustee, but trustors know the distribution of incentives to abuse trust in the population of trustees. It is assumed that there exist some trustees who never have an incentive to abuse trust. Dasgupta (1988) and Camerer and Weigelt (1988) discuss such a (repeated) Trust Game with incomplete information. Also, Camerer and Weigelt test the model in an experiment, which we will discuss in the following section.

178

VINCENT BUSKENS AND WERNER RAUB

Technical details of this model have been provided by Bower et al. (1997). One conclusion is that the payoffs of the trustor are the main predictors for trust. This indicates that even in a setting in which there is only a limited amount of uncertainty for the trustors, the predictions of the "pure" control models about the payoffs of the players seem to lose much of their predictive power. Buskens (2000) extends the model with an exit option for the trustor and with a second trustor who communicates with the first about the behavior of the trustee. It is shown that an exit option does not have an effect on trust in this setting. Furthermore, communication between two trustors affects trust only if each trustor informs the other with a high probability about the trustee's behavior, but not if only one of the trustors informs the other trustor with a high probability. 8 Some of the results of these models are quite counterintuitive. In particular, note that in certain periods in the game, the probability that trustors trust increases (!) with the temptation for the trustee to abuse trust and decreases (!) with the probability that two trustors communicate about the trustee's behavior. We will not elaborate here on the details behind these results. Explaining the model implications would require the introduction of too many technicalities. The important point is that the implications are not only counterintuitive but also inconsistent with some experimental findings. The complexity of the equilibria implies that there is hardly any reason to assume that actors actually (behave as if they) calculate the equilibria in such complicated situations. Still, Camerer and Weigelt (1988) claim that the equilibria have predictive power and can explain some features of the behavior of real actors. However, Neral and Ochs (1992) show that the actors do not act in accordance with the less intuitive prediction that trustfulness increases with temptation for the trustee in some periods of the game. Consequently, we consider models in which learning and control are combined as very preliminary. The assumption that trustors know the distribution of incentives for the trustees might be too strong. Finding realistic information assumptions seems to be a more important problem than finding at least some assumptions for which game-theoretic solutions can be found, since outcomes of game-theoretic models tend to vary considerably depending on the information assumptions made (as we have seen from the different analyses related to the repeated Trust Game). We will not solve this problem in this paper (see Buskens, 2001, for some further discussion). We only want to show that to understand trust, we need a model in which learning and control are addressed simultaneously. Since the information assumptions seem to matter considerably for the predictions about behavior of the actors, it is important to obtain more knowledge about how trustors take into account their information in embedded trust situations with incomplete information. 9

179

Embedded Trust: Control and Learning

3. EMPIRICAL EVIDENCE FOR LEARNING AND CONTROL

IN TRUST

SITUATIONS

A typical feature of many empirical studies on trust in economic exchange as well as experimental settings is that they offer evidence for effects of dyadic embeddedness or network embeddedness, but it is hardly ever the case that we can determine whether the effects are due to learning, control, or a combination of the two mechanisms. It is often acknowledged, as we will see, that embeddedness is important for trust (Granovetter, 1985). Still, analysis of the mechanisms through which embeddedness works is often lacking. For example, in a field study of four enterprises, Larson (1992) shows that relations between firms start with small transactions. Transactions become larger over time as trust emerges if the transactions progress smoothly. This is an example illustrating that trust evolves gradually and in a stepwise fashion (see: Blau, 1964 [1996]; Dasgupta, 1988; Coleman, 1990, p. 104). After the "trial period," cooperative efforts can become more extensive, while the transactions are controlled through social relations only: " 'they trust us to pay and we trust them to give us a reasonable price' " (Larson, 1992). The trial period could be used to investigate the incentives of a partner, but also to obtain a better idea about the probability that more transactions will follow in the future. In an ethnographic study, Uzzi (1996, 1997) shows that failure rates of dress apparel firms are lower the stronger the relationship between partners. Uzzi explains this as a result of the comparative advantage of trust in strong relationships. The "black box" with unexplained mechanisms seems even larger here. It remains not only unexplained how strong relations affect trust, but also why trust increases performance. ~° Uzzi (1996, p. 681) seems to prefer the learning explanation since he provides circumstantial evidence that trustors trust a trustee even in final transactions in embedded ties. Gulati (1995a, b) finds that the probability that firms form alliances is larger if they have been involved in alliances with the same partner before. Gulati interprets this finding as an indication that previous, probably positive, experiences enlarge trust among partners. Moreover, the probability that partners in alliances use equity as a formal governance mechanism decreases with the number of previous alliances between the partners. Complementary experimental findings are due to Lawler and Yoon (1996) who present evidence showing that positive experiences in repetitive exchanges positively affect cooperation in the Prisoner's Dilemma (see Lawler et al., 2000, for similar evidence for N-person Prisoner's Dilemmas). Similarly, Baker et al. (1998) find that interorganizational ties between advertising agencies and their clients have a smaller probability of being dissolved if they have already existed for a longer

180

VINCENT BUSKENS AND WERNER RAUB

period. Although these findings are clearly explained in terms of learning positive past experience increases trust, while trust enlarges the probability to stay together - a control interpretation seems likewise plausible. The increased probability to stay together improves control opportunities, which is also a reason to trust each other better. Some studies on trust in economic exchange are noteworthy for addressing the effects of embeddedness at the dyadic level on the investment in formal arrangements such as investments in contracting. We consider the lack of investment in formal arrangements as an indication for trust among partners since such arrangements provide, for example, compensation for the trustor in case of untrustworthy behavior by the trustee. In a study on 72 subcontracting relationships, Lyons (1994) finds that the probability for arranging the relationship with a formal contract decreases with the number of years subcontractors have been trading with their most important customers. Blumberg (1997, Section 4.2) uses a complex measure for the investment in formal arrangements (comprising both time and money invested in formal arrangements and negotiations, and the extensiveness of the contract). He finds that both measures decrease and, thus, that trust increases with the extent to which the partners had transactions in the past. These results support the learning hypothesis that positive experiences increase trust. Blumberg actually distinguishes between the effect of past transactions and transactions expected with the partner in the future, but he does not find an effect of the transactions partners expect in the future. Batenburg et al. (2002) study relations between buyers and suppliers of IT products. Their dependent variable is a combination of time and money spent in searching for a partner, negotiating with the partner, and the extensiveness of the contract. Such a dependent variable represents investments in the ex ante planning of transactions. They find that these investments decrease if the partners had transactions in the past. Furthermore, they find that the investments decrease even more if the partners already had past transactions and expect more transactions in the future. They do not find an effect of expected future transactions if the partners had no previous transactions. Their explanation employs two arguments. First, costly investments in ex ante planning are less necessary if more future transactions are expected because of the sanction opportunities from subsequent transactions. This is a control effect based on the expectation of future transactions. Second, however, it is worthwhile to invest more in formal arrangements if more future transactions are expected, because these investments can be used again in subsequent transactions. This is an investment effect due to the expectation of future transactions. The driving force of this effect is that relation specific investments associated with a focal

Embedded Trust: Control and Learning

181

transaction affect the incentive structure of future transactions. Such effects have been ignored above in our theoretical discussion of learning and control. Combining the arguments on control and on the investment effect, it is unclear what the total effect of future transactions will be. However, a negative interaction effect between past and future on ex ante planning is indeed expected, since the investment effect will be larger in initial transactions compared to later transactions (see: Raub, 1996; Raub & Snijders, 2001). Another explanation for such an interaction effect could be that control only plays a role if the partners have sufficient information about each other and that uncertainties about an unknown partner are simply too large to allow for reliance on the effectivity of future sanctions already in the first transaction. Buskens et al. (2000) use the same data, but their dependent variable is the number of issues addressed in the contract that was written for a transaction. They find that less items are included in the contract after positive transactions in the past, but no effect of transactions expected in the future is found for this dependent variable. Buskens et al. (2000) also address the effects of network embeddedness on trust. They find that there are less issues addressed in the contract if the buyer and supplier are located closer to each other. Their explanation is that buyers and suppliers who are located closer to each other are probably embedded in a denser network. Although alternative explanations might be possible, this is an indication that network embeddedness increases trust. Obviously, being located close to one another improves learning as well as sanction opportunities, so it is unclear whether this effect is due to learning or control. In experimental research, cooperative relations are extensively studied using repeated Prisoner's Dilemmas (see Colman, 1982, Chap. 7 for an overview). Findings suggest that subjects often end up in cooperative play, but also might run into mutually defective choices. One explanation for this is that subjects try to learn and test whether another subject is trustworthy. This can lead to mutually cooperative choices. However, defective choices of untrustworthy partners probably result in implementing sanctions associated with control strategies, implying defective choices with partners who are untrustworthy. More directly related to the issues discussed here are the experiments by Camerer and Weigelt (1988) and Neral and Ochs (1992). In these experiments, games are played that consist of eight or six times repeated Trust Games with incomplete information. Incomplete information in these games implies that trustors are not sure whether the trustee has an incentive to abuse trust. Thus, there is a positive probability that a trustor meets a trustee for whom T2 < R2, which implies that such a trustee will never abuse trust. All other elements of the Trust Game are the same. These experiments show empirically that trustees

182

VINCENT BUSKENS AND WERNER RAUB

care more about their reputation in the initial periods of play than at the end of the eight or six periods to be played. Moreover, trustors test whether the trustee is trustworthy, especially in the earlier periods. The more positive experiences a trustor has, the more she is convinced that she is playing with a trustee who does not have any incentive to abuse trust. However, the observed probability of placing trust by the trustor does not increase as the end of the game comes nearer. This is consistent with the theory, since the trustor has to realize that also the trustee has an incentive to make her believe that he does not have an incentive to abuse trust, but he will abuse trust toward the end of the game. Of course, in real-life situations, actors hardly ever know in advance which transaction with the partner will be the last one. The outcomes in the experiment clearly show learning in early periods, but simultaneously a concern about decreasing control opportunities if the end of the game approaches. In the absence of learning, a backward induction argument would cause that no trust can be placed in any of the rounds in a finitely repeated Trust Game.ll In the absence of control, trust can never decrease after positive experiences. Kollock (1994) studies trust experimentally in situations in which buyers are uncertain about the quality of a seller's products. Two groups of four subjects ("buyers" and "sellers") can trade with each other for twenty rounds. In each round, buyers and sellers are given five minutes to make offers about prices, while the sellers advertise the quality of their goods. The sellers can choose from three levels of quality at which they want to produce. The advertisement about quality does not need to be trustworthy. After each round the buyers are informed about the quality of the product they bought. Profit margins are largest for low-quality products sold for high prices. But selling high-quality products for high prices is preferred by all parties over selling low-quality products for low prices. In these experiments, most sellers seek to obtain a reputation for selling high-quality products and in the meantime obtaining high prices for their products. Sellers who start to sell low-quality products for high prices eventually end up with mutually sub-optimal outcomes, namely, selling low-quality products for low prices. A major difference between Kollock's experiments and the repeated Trust Games experiments of Camerer and Weigelt (1988) is that in Kollock's experiment the game with twenty rounds was played only once, while Camerer and Weigelt let the subjects play the eight times repeated Trust Games 75-100 times with changing partners. Camerer and Weigelt show that subjects need some time to experience what the better strategies are. In the initial games in their experiments, a considerable number of subjects end up in sub-optimal outcomes, but this number decreases with experience in playing the repeated games. Summarizing, both experiments show that trustors test the trustees and try to learn about them in the course of the game. Moreover, trustors

Embedded Trust: Control and Learning

183

execute control strategies if it is necessary to do so. Finally, it takes trustors and trustees some experience before they are able to figure out how to combine learning and control opportunities in an optimal way. Most evidence discussed above is related to embeddedness at the level of the dyad. It is more difficult to find clear empirical evidence unambiguously supporting a causal relationship between social networks and trust. In the sociological literature, cohesive social structures and trust are often seen as being almost equivalent. For example, Gellner (1988, p. 143) states this explicitly and repeatedly. In these studies, trust is treated as being either present or absent in a society. Actors know each other and have similar norms leading to trustworthy behavior. Diamond merchants (Wechsberg, 1966) and the stock exchange market (Weber, 1921, pp. 191-193) form cohesive societies where trust reaches high levels. Ellickson (1991) describes an example of "high trust" among farmers in Shasta County, California. Disputes between the farmers arise when trespassing cattle damage neighboring fields or when there are collisions between cattle and vehicles. However, as long as the disputes are not too serious, neighbors are able to settle them informally, thereby saving the cost of law-suits. Formal law is used to settle conflicts only in exceptional cases. Anthropological studies likewise provide examples where reciprocal behavior without formal arrangements is the rule rather than the exception in close communities (see Sahlins, 1972, Chap. 5). Although these studies provide evidence that trust can be extensive in cohesive networks, they do not explain why trust emerged and continues to exist. Learning could be part of the explanation, because newcomers will not receive negative information about actors in the network. Moreover, imitation will lead to trustful behavior after some positive experiences. However, if trust is exclusively based on learning, the situation does not seem to be very stable. One untrustworthy actor could easily invade a society where trust is omnipresent and exploit the trustful actors. Therefore, control is important in maintaining a stable level of trust. Actors who are untrustworthy must be expelled or punished severely enough for other actors to see that untrustworthy behavior is only worthwhile in the very short term. ~2 Some evidence suggests that control alone is not sufficient for the emergence and maintenance of trust. Burt and Knez (1995a, b) show that not only trust emerges in dense social networks, but trust and distrust may occur simultaneously in the same group of actors. Moreover, strong trust as well as strong distrust are especially prevalent in dense parts of the network. Control effects, in principle, explain strong trust rather than strong distrust. The reason is that in a dense network control is effective and, therefore, no untrustworthy behavior will occur. However, the fact that distrust might increase with the

184

VINCENT BUSKENS AND WERNER RAUB

density of the network is a strong indication that trustors adjust their beliefs about the trustworthiness of a trustor on the basis of information they obtain from network contacts and that untrustworthy behavior occurs in spite of the existence of a dense network. In recent years, several studies into the relations between organizations with respect to network governance have provided evidence for learning and control (Nohria & Eccles, 1992; Swedberg, 1993; Smelser & Swedberg, 1994; Kramer & Tyler, 1996). "Network govemance" is used as a label for a group of firms that engage in exchange relationships and the exact meaning of the label varies considerably between the studies (see Jones et al., 1997). Most of the research has a descriptive focus describing which kinds of networks emerge and which firms are more likely to be in the center of a network (see Barley et al., 1992). Kogut et al. (1992) show that biotechnology firms tend to search for partners within a cluster of firms with whom they can work cooperatively. However, the causes and consequences of the choices made by these firms are mostly ignored. Gerlach (1992) goes one step further, showing that Japanese firms are more tightly connected in organizational networks than American firms. Therefore, Japanese firms can rely more on inter-firm networks of supply and distribution than American firms. As a result, Japanese firms are smaller than American firms because the latter incorporate more frequently production processes within the firms. Finally, Gulati (1995b) finds that social networks help firms to obtain information about facilities and the abilities of potential partners. The result is that alliances occur more often among partners who have more common ties with third parties. Gulati and Gargiulo (1999) is one of the recent studies in which more specific network properties such as centrality are used to explain, in this case, alliance formation. Stuart and Robinson (Robinson & Stuart, 2001; Stuart & Robinson, 2001) also distinguish between different network mechanisms expressing the importance of having a good reputation based on past transactions and having sanctioning opportunities. Their core independent variables are the alliance experience of the trustor ("client") and trustee ("agent" or "target"), the centrality of the trustor and trustee in the network, and the status of previous partners of the trustee. They also include a variable representing past alliances among the two partners. These studies provide strong evidence for effects of dyadic embeddedness and network embeddedness on trust by explaining the use of informal network management mechanisms rather than more formal and contractual management. Again, however, it is impossible to distinguish clearly between leaming and control because Stuart and Robinson use centrality measures such that the ties considered are symmetric and can be used for sending as well as receiving information. Moreover, the evidence in these network studies is based on the

185

Embedded Trust: Control and Learning

assumption that the network structure for alliance formation corresponds largely with the network structure of communication among the relevant firms. If this assumption does not hold, leaming and control can be the result of ties other than the alliance ties. The evidence for leaming and control through social networks discussed above is still far from conclusive. While there is evidence that trust can emerge in dense social networks, it remains unclear what drives the emergence of trust. Is it learning or is control through the threat of sanctions more important? Probably, the empirical evidence is limited because theoretical explanations guiding the search for empirical evidence are scarce. Researchers have primarily focused on establishing the relationship between embeddedness and trust considering at most one mechanism that drives this relationship. The proposed interdependence between learning and control, however, asks for an integrated approach that allows for disentangling these two mechanisms. We will return to this issue later in the conclusions. Before, we discuss some more specific evidence for the co-occurrence of learning and control in trust situations. 4. T W O

VIGNETTE

STUDIES

Available survey data do not offer more than circumstantial evidence to support that control and learning mechanisms play a role in durable relations. Most research has not been designed to distinguish learning and control effects explicitly and, consequently, although effects of social embeddedness are found, the precise nature of the mechanisms producing the effects remains unclear. Even if one would aim beforehand at distinguishing learning and control, it is a complex task to develop an appropriate research design, since convincing theory about the two mechanisms is still in the early stages of development. Experiments are useful to provide more control over the variation of key variables and experiments can be designed explicitly to resemble more closely assumptions made in theoretical models. In this way, experimental research cannot only provide us with evidence on hypotheses derived from theoretical models, but they can also shed light on the extent to which certain assumptions in existing models actually are problematic. Thus, experimental evidence can often guide further development of theoretical models more accurately than survey research. We discuss two experiments in which subjects are presented with hypothetical transactions. The subjects are asked to imagine that these "vignettes" are actual transactions and they answer questions about their behavior related to these transactions (see, for example: Rossi, 1979; Rossi & Nock, 1982 on vignette experiments). The two experiments are rather different as far as the

186

VINCENT BUSKENS AND WERNER RAUB

design is concerned, but they are both designed specifically to test effects of different dimensions of dyadic embeddedness and network embeddedness on trust. In the first experiment, purchase managers of Dutch companies are asked to answer questions about hypothetical transactions. The description of the transactions incorporates information about characteristics such as price and importance, but also about the relationship of the buyer with the supplier. Questions that had to be answered by the purchase managers included questions about the extent to which buyers would negotiate about the terms of the transactions and questions about arrangements they want to include in a formal contract for the given transaction. In the second experiment, students are asked to compare pairs of situations for buying a used car, while properties of the relation between the buyer and the car dealer are varied. There are a number of important differences between these two experiments. The type of transaction is clearly different. In the first experiment, subjects rate vignettes by indicating the level of investments in negotiations and contracting, while in the second experiment subjects compare different vignettes rather than providing ratings. In the first experiment, subjects are purchase managers with considerable experience in the type of transactions discussed in the experiment, while in the second experiment the subjects are students. In the first experiment, transaction characteristics are varied while they are kept constant in the second experiment. In the first experiment, learning and control through third parties are not explicitly distinguished, while the second experiment provides a first attempt to do so. An overview of the dependent and independent variables as well as other core features of both experiments is presented in Table 2. The table also indicates how the variables that represent social embeddedness are related to learning and control. In this paper, we cannot go into all details about the design of the experiments. Additional information about the experiment on transactions in buyer-supplier relations can be found in Rooks et al. (2000). Further details on the experiment on buying a used car are provided in Buskens and Weesie (2000b). 4.1. Vignettes on Transactions in Buyer-Supplier Relations

In this experiment, respondents were requested to consider transactions from their daily business practice as purchase managers of medium-sized and large Dutch firms. A vignette consists of a description of a transaction for which eight variables are varied. The formulations for the different values of the variables are given in Table 3. The first three variables are related to transaction

187

Embedded Trust: Control and Learning Table 2.

Summary of the Empirical Evidence on Trust from Two Vignette Studies Considered in this Paper. Experiment 1

Experiment 2

Transactions in buyer-supplier relations Purchase managers 40 348 Effort invested in management

Buyinga used car

Volume of transaction Specific investments Uncertainty

Volume of transaction

Dyadic embeddedness

Past (learning) Future (control)

Past (learning) Future (control)

Network embeddedness

Degree (learning and control)

Density (learning and control) Third-partyinformation (learning) Outdegree (control)

Type of transaction Subjects Number of subjects Number of transactions Dependent variable Independent variables Transaction characteristics

Institutional embeddedness

Students 125 1249 Choicebetween two dealers

Exit opportunities (control) Country

characteristics, namely, the volume of the transactions, the extent to which the buyer needs to make specific investments for implementation of the transaction, and the extent to which the buyer is able to judge the quality of the product at the time of the transaction (uncertainty). Transaction cost arguments suggest that all three variables should have positive effects on the extent to which the buyer is concerned about obtaining a low-quality product, because of the supplier's opportunities and incentives for abusing trust as well as the potential damage for the buyer (see, for example, Williamson, 1985). Hence, these variables are expected to affect trust negatively. The fourth and fifth variable describe the bilateral relation between the buyer and the supplier. The past variable indicates whether the buyer and supplier have done business before with one another. Thus, the past variable represents learning. The future variable indicates whether buyer and supplier expect to do more business in the future. Clearly, this variable represents control. The sixth and the seventh variable are related to the network of the buyer. The degree variable is an indicator for the extent to which the buyer knows

188

VINCENT BUSKENS AND WERNER RAUB

Table 3.

D e s c r i p t i o n o f the V a r i a b l e s in the First V i g n e t t e E x p e r i m e n t .

Variable

Value

Volume

0 1 2

Specific investments

0 1 2

Uncertainty

Past

0 1 2 0 1

Future

2 0 1

Degree

0 1

2 Exit 0 opportunities 1 2 Country 0 1

2 3 4

Text A transaction with a small volume, namely less than US$ 5000 on a yearly basis. A transaction with a reasonable volume, namely, about 5% of the total purchase volume of your finn. A transaction with a very large volume, namely, about 18% of the total purchase volume of your finn. A transaction for which extra investments by your finn are not necessary. A transaction for which your firm has to make small investments, such as investments in specific machines and equipment. A transaction for which your finn has to make considerable investments, such as investments in specific machines and equipment. A known product about which your firm has the required expertise. A known product but one about which your firm has only limited expertise. A brand new product about which your finn has only very limited expertise. A supplier with whom your finn has never done business before. A supplier with whom your firm has a relationship in which minor problems occurred. A supplier with whom your finn has a long and successful relationship. It is uncertain how long your finn will continue with the supplier. Your finn expects to do business for a long period with the supplier. You do not know any business partners of the supplier. You know some business partners of your supplier, but your firm does not do business with them. You do business with other business partners of the supplier. A product for which there are no alternative suppliers on the market. A product for which there are some alternative suppliers on the market. A product for which there are many alternative suppliers on the market. A supplier from Eastern Europe. A supplier from Japan. A supplier from the United States. A supplier from Germany. A supplier from the Netherlands.

o t h e r b u s i n e s s p a r t n e r s o f the supplier. T h i s variable r e p r e s e n t s l e a r n i n g as well as control. T h e exit variable d e s c r i b e s the e x t e n t to w h i c h the b u y e r has a l t e m a t i v e s for b u y i n g a similar product. T h e r e f o r e , this variable r e p r e s e n t s control. A n additional v a r i a b l e i n d i c a t e s in w h i c h c o u n t r y the s u p p l i e r is located. This is an i n d i c a t o r for "institutional e m b e d d e d n e s s . " T h e i d e a is that trust is m o r e difficult i f the s u p p l i e r c o m e s f r o m a d i f f e r e n t institutional setting. W e d o not d i s c u s s this v a r i a b l e any f u r t h e r in this paper. A m o r e d e t a i l e d

Embedded Trust: Control and Learning

189

discussion can be found, for example, in Weesie and Raub (1996). The precise formulation of the variables is given in Table 3. Trust can be measured in different ways in this experiment. The subjects were asked how much time they would invest in negotiations and contracting for the transaction, and which departments would be involved in these negotiations. We assume that purchase managers want to invest more time in negotiations if they trust the supplier less. The subjects also answered whether or not they wanted to include a legal item (from a list of fourteen items) in a contract for the given transaction. Again, the assumption is that trust is smaller if a purchase manager wants to incorporate more safeguards in the contract. We discuss in detail the analyses for a dependent variable that is the product of the time invested in negotiations and the number of departments that would participate in these negotiations. Regression analysis is used with Huber standard errors modified for clustering. This is necessary since each purchase manager answered all questions for up to ten vignettes (Huber, 1967; Rogers, 1993). The analyses in Table 4 demonstrate the importance of the transaction characteristics to explain trust in a given transaction. Purchase managers take more precautions for larger transactions for which they have to invest more and about which they have less information. These variables define in fact the trust problem in terms of the payoffs in a Trust Game. The trust problem can be

Table 4.

Regression Analysis of Effort Invested in Management.

Independent variable Volume Specific investments Uncertainty Past Future Degree Exit opportunities Institutionalembeddedness Past x future Constant Explained variance Number of subjects Number of observations

Hypothesis

Model 1

Model 2

+ + + ? -

0.63** 0.16"* 0.16* ~0.14* -0.02 ~).08~ -0.04 -0.10*~

0.64** 0.14" 0.17** -0.13 * 4.02 ~3.09" ~).04 -0.10"* ~).32"* -1.26 0.51 40 348

-1.27 0.49 40 348

**, *, and ~ represent two-sided significanceat respectivelyp < 0.01, p < 0.05, and p < 0.10 based on Huber standard errors modified for clustering.

190

VINCENT BUSKENS AND WERNER RAUB

alleviated by learning and control based on embeddedness as we see through the main effects of past and degree, and the interaction effect of past and future. Trust increases with transactions in the past and with the extent to which the buyer knows business partners of the supplier. Moreover, given that the buyer and supplier have done business in the past, the trust problem decreases considerably if the buyer and supplier expect to do more business in the future. The effect of exit opportunities is in the expected direction, but not significant. The effects for degree and exit opportunities are both somewhat unstable depending on the exact specifications of the model. If we estimate randomeffect models with two random components, one on the level of the vignettes and one on the level of the subjects, both effects are close to significance but not very strong. The effects of the different countries turned out to be nicely ordered and, therefore, we added a one-dimensional variable for institutional embeddedness. Germany was clearly closest to the Netherlands on this scale followed by the United States and Japan. Purchase managers wanted to invest the most time in negotiating with suppliers from Eastern Europe. We did similar analyses with the extent to which the purchase managers wanted each of fourteen items in a contract for a given vignette as the dependent variable. It turns out that volume, specific investments, and uncertainty are the predominant predictors, although the standard errors of the effects vary considerably with the exact specification of the models. Remarkably, however, is a strong negative effect of exit opportunities on the extensiveness of the contract. Moreover, though purchase managers tend to write more extensive contracts with suppliers from Germany, the United States, and Japan, they do not do so the in case of Eastern Europe. This analysis does not show any other effects of embeddedness. It seems that the two different management mechanisms "contracting" and "negotiating" are, at least to some extent, complementary mechanisms to overcome a trust problem. Probably, contracting seems of limited value in Eastern Europe due to the problem of enforceability, but this is compensated with more negotiating. On the other hand, exit opportunities can be used in negotiations to obtain a better deal, which might lengthen negotiations, but the supplier should stick to the verbal agreements since the exit threats remain available for the buyer. We conclude that this experiment confirms the existence of a strong effect of embeddedness on trust due to learning within a dyadic relation. Moreover, the interaction effect of past and future indicates not only the existence of a control mechanism, but also that the occurrence of control is contingent on some previous learning opportunities or the importance of the investment effect of contracting that has been discussed in Section 3 above. This finding is in correspondence with the results of the survey on IT transactions of Batenburg

Embedded Trust: Control and Learning

191

et al. (2002). Concerning third-party effects, we find that knowing other business partners of the supplier increases trust. It is unclear whether this effect is due to learning or control, since these third parties can be used to obtain information, but they also can be informed in case of untrustworthy behavior of the supplier to extend control opportunities. The effect of exit opportunities on contracting indicates that the purchase managers realize that this provides them with alternative sanction opportunities implying that the supplier is less likely to act untrustworthy if he has more competitors. In the light of our previous discussion, two aspects of this first experiment are problematic. First, there is considerable variation in transaction characteristics and although we reach a considerable level of explained variance in the statistical models, we think that this variation creates likewise a considerable amount of noise. This noise might cause that the network effects, which are considerably smaller than the effects of the transaction characteristics, are partly obscured by the unexplained variance in the trust problem due to these transaction characteristics. Second, the experiment does not distinguish between learning and control due to the exchange of information with other buyers of the supplier. Both problems are addressed in the following experiment. 4.2. Vignettes on Buying a Used Car

In the second experiment, we asked students to compare situations for buying a used car. We presented students with pairs of vignettes describing such a transaction and asked them which one they preferred. The experiment was held in the U.S. (Chicago) and the Netherlands (Utrecht and Tilburg). Six variables are varied at the vignettes. The first variable indicates the price of the car (U.S.$1000 or U.S.$4000). Within each pair of vignettes between which subjects had to choose, the price of the car is held constant. Consequently, the price does not have a direct effect on the choices made by the subjects, but it might be the case that some embeddedness variables are more important for cheap cars than for expensive cars. In other words, the volume variable is added only to test for interaction effects of the size of the trust problem and embeddedness variables. Five other variables represent embeddedness: past, future, density, third-party information, and outdegree. The precise formulations are given in Table 5. Past, indicating learning at the dyadic level, is operationalized as whether the buyer has bought a car from the dealer before and was satisfied, or did never buy a car from the dealer. We only study the difference between no information and positive information from past experiences with the dealer, disregarding negative or mixed information. We expect that a vignette in which the buyer

192

Table 5. Variable Volume Past Future Density

VINCENT BUSKENS AND WERNER RAUB Description of the Variables in the Second Vignette Experiment. Value Text 0 1 0 1 0 1 0 1

Third-party 0 information 1 Outdegree

0 1

You can buy a car for about US$1000. You can buy a car for about US$4000. You never bought a car from The Autoshop before. You bought a car from The Autoshop before and you were satisfied. You will move to the other side of the country in a few weeks. You do not expect to move out of town soon. The Autoshop is an unknown garage in your neighborhood. The Autoshop is a well-knowngarage and has many customers in your neighborhood. As far as you know, none of your friends have bought a car from The Autoshop You have friends who bought a car from The Autoshop and they were satisfied. You do not have a close link with the owner of The Autoshop. The owner of the garage and you are members of the same football team.

had positive experiences with the dealer is more attractive than a vignette in which the buyer had no positive experiences with the dealer. Future, indicating control at the dyadic level, is operationalized as whether or not the buyer expects to move to the other side of the country soon. Control is more difficult for a buyer if she moves to the other side of the country. Moreover, the probability that the buyer has future transactions with the dealer is smaller if the buyer moves. Therefore, it is expected that a vignette is more attractive if the buyer does not expect to move than if the buyer expects to move. One problem with the operationalization is that, theoretically, the effect of future is based on the sanctions of the buyer anticipated by the dealer. Therefore, strictly speaking, future can be expected to affect the behavior of the buyer only if the dealer is informed about the buyer's plans to move. Another problem is that if a buyer moves, the possibilities of control through the network also become smaller. We will come back to this issue in the discussion of outdegree. Density differentiates between a dealer whose garage is or is not well-known in the neighborhood of the buyer. If more potential customers in the neighborhood know the dealer, the buyer probably knows more other customers and it is likely that there are more ties among these other customers. Therefore, learning as well as control of a well-known garage through the network of customers can be more effective than learning about or control of a garage that is not well-known. Although we cannot distinguish learning and control for this

Embedded Trust: Control and Learning

193

variable, density is added because it is a global network characteristics that is not directly linked to the specific individual network position of the buyer and the dealer. Third-party information is operationalized as whether or not the buyer has information from friends about transactions of these friends with the garage. Again, only positive information is included, because we focus on the difference between no and positive information rather than the effect of negative or mixed information. Therefore, third-party information is expected to have a positive learning effect on the attractiveness of a vignette. Outdegree is operationalized as whether or not both the buyer and the dealer are members of the same sports team. This is a measure for outdegree, because the number of acquaintances the buyer and dealer have in common is expected to be larger if the buyer and dealer are members of the same sports team. Common membership provides the buyer with possibilities of controlling the dealer through reputational sanctions both in his business and as a team member. These sanctions can include discouraging others to buy from the dealer, but also social sanctions during activities at the sports team. A rational dealer should be concerned about these sanction opportunities of the buyer. Therefore, outdegree is expected to have a positive control effect on the attractiveness of a vignette. Note that this effect of common membership may be smaller if the buyer expects to move to the other side of the country (future). In the formulation for outdegree, we did not mention anything about past behavior of the dealer to prevent the effect being interpreted as learning. Of course, the buyer has also a better potential for learning about the dealer's behavior, but given that the social setting allows for very extensive social sanctions and the fact that there is no indication whether information would be positive or negative, the control interpretation seems predominant for this variable. An advantage of this formulation of outdegree is that the theoretical assumption of "common knowledge about the network" is unlikely to be violated because the buyer and the dealer both know that they are members of the sports team. DiMaggio and Louch (1998) demonstrate that many buyers prefer a relative as a dealer for a used car rather than a dealer with whom they have no social relationship. A problem with this operationalization is that there may also be a group of buyers who are reluctant to buy a car from a team member. The relationship between the team members may be spoiled if the car happens to have a defect. Moreover, if it is not very obvious that an acquaintance has acted untrustworthy, it is questionable whether a buyer will actually execute sanctions against this acquaintance because the costs of sanctions for the buyer herself are probably relatively high. Obviously, this alternative interpretation of outdegree would

194

VINCENT BUSKENS AND WERNER RAUB Table 6.

Independent variable Past Future Density Third-party information Outdegree Number of subjects Number of observations

Probit Analyses of the Choice of Vignettes. a Hypothesis + + + + +

All

Chicago

1.09"* 0.99** 0.57** 0.61"* 0.71"* 0.67** 0.83** 0.77** 0.26** 0.18 125 40 1249 400

Utrecht

Tilburg

1.19"* 0.61"* 0.73** 0.89** 0.28* 72 720

1.39"* 0.30 0.73** 0.86** 0.51' 13 129

a The effect sizes are proportional to the parameter estimates and, hence, not reported. ** and * represent two-sided significance at respectively p < 0.01 and p < 0.05 based on Huber standard error modified for clustering.

make a dealer less attractive if he is a team member, and would therefore reduce the theoretically predicted outdegree effect. The choices in this experiment can be analyzed using a random utility model for multiple choice situations (McFadden, 1973) if we assume that subjects assign to each vignette a certain utility that depends in a linear manner on the attributes of the vignette. This statistical model implies that we estimate a probit model in which the differences in the values for the variables within a pair of vignettes are the independent variables, and whether a certain vignette is chosen is the dependent variable. The coefficients resulting from this model are interpretable straightforwardly as regression coefficients such that a positive coefficient implies that a certain embeddedness variable contributes positively to the utility of a vignette. Standard errors in the analyses are modified for clustering of observations within subjects. The outcomes for the analyses are presented in Table 6 for all sessions together and for the sessions in the different cities separately. Note the remarkable resemblance of the results for the different sessions. No statistically significant differences occur between the sessions. Therefore, we focus now on the aggregated results. All embeddedness variables have positive effects on the utility of a vignette, which we interpret as having a positive effect on the probability that the subject expects to obtain a car that is worth the money paid for it. Thus, all types of embeddedness have a positive effect on trust. The largest and most significant effects are those of the learning variables: past and third-party information, j3 Positive information clearly enhances trust. Future, density, and outdegree have likewise positive effects on trust, which implies that control is important at the dyadic as well as at the network level. Remember that this evidence for a control mechanism is somewhat problematic since these

Embedded Trust: Control and Learning

195

variables are subject to alternative explanations. There might be other disadvantages for buying a car just before you move, for example, because you need to find another garage if there is any problem with the car in the future. The alternative interpretation of the outdegree variable would imply that the effect should be in the other direction. This is not the case and, consequently, the result is in correspondence with the arguments and results of DiMaggio and Louch (1998). However, we cannot exclude that actors prefer to trust well-known others over unknown others for other reasons than the control reasons advocated in this paper. From the theory in Buskens and Weesie (2000a), we expected that the effects of embeddedness at the level of the dyad would become smaller compared to the effects of network embeddedness for larger transactions. Unfortunately, we did not find any interaction effects in this experiment. A reason for this might be due to the design of the experiments in which the corresponding variables were row-wise presented next to each other. This might cause that subjects concentrate on differences to simplify their decision task and disregard variables having the same values within a pair of vignettes. As a consequence, interaction effects can only be found by comparing pairs of vignettes that differ in both interacting variables with pairs of vignettes that do not differ in both variables. This decreases the number of cases considerably that can be used for testing the interaction effects. 5. C O N C L U S I O N In this paper we have outlined two mechanisms that foster trust in cooperative relations. One of these mechanisms is learning from previous behavior of a trustee, the other being control of the trustee through possibilities for future sanctioning of untrustworthy behavior. Both mechanisms work at a dyadic as well as at a network level. Two vignette experiments have shown evidence for learning as well as control. The first experiment on transactions in buyer-supplier relations demonstrates that learning through past transactions with the same partner and control through expected future transactions with the partner facilitate trust between business partners. In addition, we find evidence that these two effects of dyadic embeddedness are interrelated. Network effects exist in addition to the dyadic embeddedness effects, although we cannot disentangle learning and control in the degree variable that represented network embeddedness in the first experiment. In the second experiment, we are able to distinguish between learning and control through social networks. Again, learning and control affect trust through embeddedness at the dyadic level and at the level of the network. Learning at the dyadic level is represented through

196

VINCENT BUSKENS AND WERNER RAUB

previous experiences of the buyer with the dealer, while control at the dyadic level is represented through a variable specifying whether or not the buyer is about to move. In addition to these embeddedness effects at the dyadic level, trust is facilitate in this experiment through learning at the network level due to third-party information from common friends as well as through control at the network level via common membership of a sports club. We find additional support for the effect of network embeddedness from a variable for density, where network density represents learning as well as control. Consequently, we find consistent support for the positive effects of learning and control on trust at the dyadic as well as the network level in this experiment. The empirical evidence accumulated in this paper has not only shown that trust prevails in embedded settings, but also that explaining trust requires to consider different mechanisms. Four effects have been distinguished, resulting from two mechanisms that operate at two levels. Trust can be based on bilateral learning and control mechanisms as well as on network learning and control mechanisms. While all four effects are found in the literature, there is hardly any knowledge about the relative importance of the four effects and the interplay between the two mechanisms and the two levels at which they occur. Existing theoretical research provides underpinning for all four effects, but a satisfactory model that incorporates all four simultaneously does not yet exist. New modeling efforts face the difficulty of finding a fruitful combination of analytic tractability and realistic assumptions about what information actors have and how they use relevant sources of information. On the one hand, models with more realistic informational assumption are often difficult to analyze. On the other hand, knowledge about what realistic informational assumptions would be is limited, because most empirical research has not succeeded in clearly disentangling the effects of learning and control mechanisms. Disentangling the effects could provide some evidence on the relative importance of these mechanisms and evidence about changes in the importance of different effects related to different circumstances. To overcome these limitations, we propose a two-step empirical and theoretical approach. More experimental research is necessary to obtain better insights in the relative importance of the different mechanisms. Experiments should be designed such that subjects are involved in abstract trust situations embedded in a social context that allows for communication among trnstors between transactions with trustees. The experiments should explicitly provide insights in how subjects use information they obtain from other subjects in the networks and whether or not they try to sanction untrustworthy behavior by informing other subjects in the network. In this way, the experiments enable the development of new models built on assumptions, for example, about

Embedded Trust: Control and Learning

197

information exchange, that have an empirical basis rather than on assumptions chosen exclusively on the basis of introspection of researchers and mathematical tractability. Moreover, the experiments can be used to obtain initial insights in circumstances that affect the importance of learning vs. control in trust situations. Some of the arguments in this paper support that learning becomes more important compared to control if the trust situations presents more uncertainties. Specific uncertainties, for instance, about the incentives of the trustee, can be varied in order to study variations in the use of learning or control mechanisms depending on these kinds of uncertainties. Results of such experiments can inspire new theoretical models on the relative effects of learning vs. control, depending on circumstances that affect trust situations. Based on these new theories, survey designs can be developed that allow for variations in learning and control variables such that the predicted effects can be distinguished.

NOTES 1. Alternatively, one could define the temptation (T2-R2)](T2-P2) as the size of the trust problem if one wants to incorporate the potential loss for the trustee if the trustor refuses to place trust (see Snijders, 1996). 2. The distinction we make here resembles one made by Lewicki and Bunker (1995) between calculus-based trust and knowledge-based trust. The major difference, however, is that we do not consider different types of trust, but argue that the extent of trust depends on learning and control simultaneously. Moreover, we do not believe that earlier on in a relation trust is more based on control (calculus-based) and only later becomes knowledge-based. On the contrary, there is evidence that control becomes more important if actors already have information about each other (Batenburg et al., 2002). Note that trust based on learning is possible before actors actually meet, because social networks can provide learning opportunities as will be explained below. Note also that Burt (e.g. 1992, 2000) discusses a related distinction between "information benefits" and "control benefits" associated with social capital and "structural holes." A major difference with our analysis here is that Burt focuses on information and control benefits for individual performance in competitive situations, while we address how learning and control can be beneficial for both trustor and trustee in mitigating a social dilemma such as the Trust Game. Trustor and trustee are better off if trust is placed and honored due to learning and control. 3. Game-theoretic models with incomplete information, which combine learning and control and assume that actors take incentives of their opponents into account are discussed later in this paper. 4. See Camerer and Ho (1999) for a comparison and an attempt for unification of these models. 5. An alternative interpretation of these models is that actors using strategies that lead to lower payoffs obtain less off-spring, which causes that they become less numerous in subsequent periods.

198

VINCENT BUSKENS AND WERNER RAUB

6. One might argue that learning is still possible in these models, since there are often many equilibria and it is not clear why players should choose the same equilibrium to start with. We disregard this issue, assuming that players coordinate instantly on the same equilibrium (see, for example, Fudenberg & Levine, 1998, p. 20). 7. See Raub and Weesie (1990) for an earlier analysis of a similar scenario for the Prisoner's Dilemma. 8. The model indicates that whether or not a probability can be considered high depends on the temptation of the trustee to abuse trust. The higher the temptation, the higher the communication probability should be. 9. An interesting question that we owe to the editors is whether trust should be based on both mechanisms, control and learning. Or, alternatively, is it possible that trust is produced and stabilized only as a consequence of one of the two mechanisms? It follows from the theory outlined here that if the learning mechanism is strong enough in the sense that the probability p of meeting a trustworthy trustee has become high enough, the trustor places trust even without opportunities for control. However, in this case trust will be occasionally abused, namely, when the trustee is in fact untrustworthy. Conversely, the theory implies that given sufficient control opportunities, trust will be placed and honored even without learning. 10. From the description of the Trust Game, it is clear why trust increases performance since trusting partners earn more than partners who do not trust each other. 11. The backward induction argument states that since trust will be abused in the last period, trust cannot be placed in the last period. Consequently, trust will be abused in the last but one period, so no trust will be placed in this period either. This argument continues up to the first period (see, for example, Rasmusen, 1994, pp. 122-123, for the same argument about the Prisoner's Dilemma). 12. In game-theoretic terms, a combination of strategies in which abuse of trust is not sanctioned cannot be a subgame-perfect Nash equilibrium. A similar situation can be found in Axelrod's (1984) computer experiments, in which naive cooperators in Prisoner's Dilemmas are exploited by strategies that initiate defection. 13. We do not have any strong evidence for which effects are larger than others since the scales of the variables cannot be compared and the strength of effects depends largely on the specifics of the formulations of the variables.

ACKNOWLEDGMENTS Stimulating comments of and discussions with Jeroen Weesie and Chris Snijders are gratefully acknowledged. We also acknowledge useful suggestions of the editors of this volume. The vignette studies are a result of joint work with Gerrit Rooks, Robert Selten, Frits Tazelaar, and Jeroen Weesie. Financial support was provided by the Netherlands Organization for Scientific Research (NWO) under grants S 96-168 and PGS 50-370. Financial support for one of the vignette studies reported in this paper was provided by the NEVI Research Foundation (NRS) of the Dutch Association for Purchase Management (NEVI). Buskens worked on this paper during a research visit at the Department of Sociology at the University of Chicago, while Raub worked on the paper during a

Embedded Trust: Control and Learning

199

fellowship at the Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS), Wassenaar.

REFERENCES Arrow, K. J. (1974). The Limits of Organization. New York: W. W. Norton & Company. 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. Barber, B. (1983). The Logic and Limits of Trust. New Brunswick: Rutgers University Press. Barley, S. R., Freeman, J., &Hybels, R. C. (1992). Strategic Alliances in Commercial Biotechnology. In: N. Nohria & R. G. Eccles (Eds), Networks and Organizations: Structure, Form, and Action (pp. 311-347). Boston: Harvard Business School Press. Batenburg, R. S., Raub, W., & Snijders, C. (2002). Contacts and Contracts: Temporal Embeddedness and the Contractual Behavior of Firms. Research in the Sociology of Organizations (forthcoming). Blau, P. M. (1964 [1996]). Exchange and Power in Social LiJe. New Brunswick, NJ: Transaction Publishers. Blumberg, B. F. (1997). Das Management von Technologiekooperationen: Partnersuche und Verhandlungen mit dem Partner aus empirisch-theoretischer Perspektive. Amsterdam: Thesis Publishers. Bower, A. G., Garber, S., & Watson, J. C. (1997). Learning about a Population of Agents and the Evolution of Trust and Cooperation. International Journal of Industrial Organization, 15, 165-190. 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. Research in Organizational Behavior, 22, 345423. Burt, R. S., & Knez, M. (1995a). Kinds of Third-Party Effects on Trust. Rationality and Society, 7, 255-292. Burt, R. S., & Knez, M. (1995b). Trust and Third Parties. Working paper, Graduate School of Business, University of Chicago. Buskens, V. (1998). The Social Structure of Trust. Social Networks, 20, 265-289. Buskens, V. (2000). Trust in Triads: Effect of Exit, Control, and Learning. Games and Economic Behavior (forthcoming). Buskens, V. (2001). Third-Party Effects in Cooperation Problems: An Integrated Approach. ISCORE Paper 195. Utrecht University. Buskens, V. (2002). Social Networks and Trust. Dordrecht, Boston: Kluwer Academic Publishers. Buskens, V., Raub, W., & Weesie, J. (2000). Networks and Contracting in Information Technology Transactions. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations: Theoretical and Empirical Models for Organizations and Households (pp. 77-81). Amsterdam: Thela Thesis. Buskens, V., & Snijders, C. (1997). 'Individual Heuristics and the Dynamics of Cooperation in Large Groups': Additional Results Using Analytical Methods. Psychological Review, 104, 792-800. Buskens, V., & Weesie, J. (2000a). Cooperation via Networks. Analyse und Kritik, 22, 44-74.

200

VINCENT BUSKENS AND WERNER RAUB

Buskens, V., & Weesie, J. (2000b). An Experiment on the Effects of Embeddedness in Trust Situations: Buying a Used Car. Rationality and Society, 12, 227-253. Camerer, C., & Ho, T.-H. (1999). Experience-Weighted Attraction Learning in Games: Estimates from Weak-Link Games. In: D. V. Budescu, I. Erev & R. Zwick (Eds), Games and Human Behavior (pp. 31-51). Mahwah, NJ: Lawrence Erlbaum. Camerer, C., & Weigelt, K. (1988). Experimental Tests of a Sequential Equilibrium Reputation Model. Econometrica, 56, 1-36. Coleman, J. S. (1990). Foundations of Social Theory. Cambridge, MA: The Belknap Press of Harvard University Press. Colman, A. (1982). Game Theory and Experimental Games: The Study of Strategic Interactions. Oxford: Pergamon Press. Dasgupta, P. (1988). Trust as a Commodity. In D. Gambetta (Ed.), Trust: Making and Breaking Cooperative Relations (pp. 49-72). Oxford: Blackwell. DiMaggio, P., & Louch, H. (1998). Socially Embedded Consumer Transactions: For What Kinds of Purchases Do People Most Often Use Networks? American Sociological Review, 63, 619--637. Ellickson, R. C. (1991). Order Without Law: How Neighbors Settle Disputes. Cambridge, MA: Harvard University Press. Erev, I., & Roth, A. E. (1999). On the Role of Reinforcement Learning in Experimental Games: The Cognitive Game-Theoretic Approach. In: D. V. Budescu, I. Erev & R. Zwick (Eds), Games and Human Behavior (pp. 53-77). Mahwah, NJ: Lawrence Erlbaum. Flache, A. (1996). The Double Edge of Networks: An Analysis of the Effect of Informal Networks on Cooperation in Social Dilemmas. Amsterdam: Thesis Publishers. Fudenberg, D., & Levine, D. K. (1998). The Theory of Learning in Games. Cambridge, MA: MIT Press. Gellner, E. (1988). Trust, Cohesion, and Social Order. In: D. Gambetta (Ed.), Trust: Making and Breaking Cooperative Relations (pp. 142-157). Oxford: Blackwell. Gerlach, M. L. (1992). The Japanese Corporate Network: A Blockmodel Analysis. Annual Review of Sociology, 16, 435-463. Gouldner, A. W. (1960). The Norm of Reciprocity: A Preliminary Statement. American Sociological Review, 25, 161-178. Granovetter, M. S. (1985). Economic Action and Social Structure: The Problem of Embeddedness. American Journal of Sociology, 91,481-510. Gulati, R. (1995a). Does Familiarity Breed Trust? The Implications of Repeated Ties for Contractual Choice in Alliances. Academy of Management Journal, 38, 85-112. Gulati, R. (1995b). Social Structure and Alliance Formation Patterns: A Longitudinal Study. Administrative Science Quarterly, 40, 619~52. Gulati, R., & Gargiulo, M. (1999). Where Do Interorganizational Networks Come From? American Journal of Sociology, 104, 1439-1493. Heckathorn, D. D. (I 996). The Dynamics and Dilemmas of Collective Action. American Sociological Review, 61, 250-277. 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, 1, 221-233. Jones, C., Hesterly, W. S., & Borgatti, S. P. (1997). A General Theory of Network Governance: Exchange Conditions and Social Mechanisms. Academy of Management Review, 22, 911-945. Kogut, B., Sham W., & Walker, G. (1992). The Make-or-Cooperate Decision in the Context of an Industry Network. In: N. Nohria & R. G. Eccles (Eds), Networks and Organizations: Structure, Form, and Action (pp. 348-365). Boston: Harvard Business School Press.

Embedded Trust: Control and Learning

201

Kollock, P. (1994). The Emergence of Exchange Structures: An Experimental Study of Uncertainty, Commitment, and Trust. American Journal of Sociology, 100, 313-345. Krackhardt, D. (1992). The Strength of Strong Ties: The Importance of Phitos in Organizations. In: N. Nohria & R. G. Eccles (Eds), Networks and Organizations: Structure, Form, and Action (pp. 216-239). Boston: Harvard Business School Press. Kramer, R. M., & Tyler, T. R. (Eds) (1996). Trust in Organizations: Frontiers of Theory and Research. Thousand Oaks, CA: Sage. Kreps, D. M. (1990). Corporate Culture and Economic Theory. In: J. E. Alt & K. A. Shepsle (Eds), Perspectives on Positive Political Economy (pp. 90-143). Cambridge: Cambridge University Press. Larson, A. (1992). Network Dyads in Entrepreneurial Settings: A Study of the Governance of Exchange Relationships. Administrative Science Quarterly, 37, 76-114. Lawler, E. J., Thye, S. R., & Yoon, J. (2000). Emotion and Group Cohesion in Productive Exchange. American Journal of Sociology, 106, 616-657. Lawler, E. J., & Yoon, J. (1996). Commitment in Exchange Relations: Test of a Theory of Relational Cohesion. American Sociological Review, 61, 89-108. Lewicki, R. J., & Bunker, B. B. (1995). Trust in Relationships: A Model of Development and Decline. In: B. B. Bunker & J. Z. Rubin (Eds), Conflict, Cooperation, and Justice: Essays Inspired by the Work of Morton Deutsch (pp. 133-173). San Francisco: Jossey Bass Publishers. Lomborg, B. (1996). The Evolution of Social Structure in the Iterated Prisoner's Dilemma. American Sociological Review, 61, 278-307. Lorenz, E. H. (1988). Neither Friends nor Strangers: Informal Networks of Subcontracting in French Industry. In: D. Gambetta (Ed.), Trust: Making and Breaking Cooperative Relations (pp. 94-107). Oxford: Blackwell. Lyons, B. R. (1994). Contracts and Specific Investments: An Empirical Test of Transaction Cost Theory. Journal of Economics and Management Strategy, 3, 257-278. Macaulay, S. (1963). Non-Contractual Relations in Business: A Preliminary Study. American Sociological Review, 28, 55-67. Macy, M. W., & Skvoretz, J. (1998). The Evolution of Trust and Cooperation between Strangers: A Computational Model. American Sociological Review, 63, 638-660. McFadden, D. (1973). Conditional Logit Analysis of Qualitative Choice Behavior. In: P. Zarembka (Ed.), Frontiers in Econometrics (pp. 105-142). New York: Academic Press. Messick, D. D., &Liebrand, W. B. G. (1995). Individual Heuristics and the Dynamics of Cooperation in Large Groups. Psychological Review, 102, 131-145. Neral, J., & Ochs, J. (1992). The Sequential Equilibrium Theory of Reputation Building: A Further Test. Econometrica, 60, 1151-1169. Nohria, N. & Eccles, R. G. (Eds) (1992). Networks and Organizations: Structure, Form, and Action. Boston: Harvard Business School Press. Rasmusen, E. (1994). Games and lnformation: An Introduction to Game Theory. Oxford: Blackwell. Raub, W. (1996). Effects of Temporal Embeddedness on Ex Ante Planning under Incomplete Information. Raub, W., & Snijders, C. (2001). A Reluctant Match: Models for the Analysis of Trust in Durable Two Party Relations. In: Y.-H. Tan & C. Castelfranchi (Eds), Trust and Deception in Virtual Societies (pp. 195-200). Dordrecht, Boston: Kluwer Academic Publishers. Raub, W., & Weesie, J. (1990). Reputation and Efficiency in Social Interactions: An Example of Network Effects. American Journal of Sociology, 96, 626~554.

202

VINCENT BUSKENS AND WERNER RAUB

Raub, W., & Weesie, J. (1993). Symbiotic Arrangements: A Sociological Perspective. Journal of Institutional and Theoretical Economics, 149, 716-724. Robinson, D. T., & Stuart, T. E. (2001). Just How Incomplete Are Incomplete Contracts: Evidence from Biotech Strategic Alliances. Working paper, Graduate School of Business, University of Chicago. Rogers, W. M. (1993). Regression Standard Errors in Clustered Samples. Stata Technical Bulletin, 13, 19-23. Rooks, G., Raub, W., Selten, R., & Tazelaar, F. (2000). Cooperation between Buyer and Supplier: Effects of Social Embeddedness on Negotiation Effort. Acta Sociologica, 43, 123-137. Rossi, P. H. (1979). Vignette Analysis: Uncovering the Normative Structure of Complexity. In: R. K. Merton, J. S. Coleman & P. H. Rossi (Eds), Qualitative and Quantitative Social Research: Papers in Honor of Paul F. Lazarsfeld (pp. 176-186). New York: The Free Press. Rossi, P. H., & Nock, S. L. (Eds) (1982). Measuring Social Judgments: The Factorial Survey Approach. Beverly Hills, CA: Sage. Roth, A. E., & Erev, I. (1995). Learning in Extensive-Form Games: Experimental Data and Simple Dynamic Models in the Intermediate Term. Games and Economic Behavior, 8, 164-212. Sahlins, M. (1972). Stone Age Economics. Chicago: Aldine. Smelser, N. J., & Swedberg, R. (Eds) (1994). The Handbook of Economic Sociology. New York: Russell Sage. Snijders, C. (1996). Trust and Commitments. Amsterdam: Thesis Publishers. Stuart, T. E., & Robinson, D. T. (2001). The Origins oflnterorganizational Networks: Probation Until Reputation. Working paper, Graduate School of Business, University of Chicago. Swedberg, R. (Ed.) (1993). Explorations in Economic Sociology. New York: Russell Sage. Taylor, M. (1987). The Possibility of Cooperation. Cambridge: Cambridge University Press (Revised edition of Anarchy and Cooperation, London: Wiley 1976). 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. Voss, T. (1982). Rational Actors and Social Institutions: The Case of the Organic Emergence of Norms. In: W. Raub (Ed.), Theoretical Models and Empirical Analyses: Contributions to the Explanation of Individual Actions and Collective Phenomena (pp. 101-128). Utrecht: Explanatory Sociology Publications. Weber, M. (1921). Wirtschaft und Gesellschafi. Ttibingen: Molar 1976. Wechsberg, J. (1966). The Merchant Bankers. New York: Bedminster Press. Weesie, J., Buskens, V., & Raub, W. (1998). The Management of Trust Relations via Institutional and Structural Embeddedness. In: P. Doreian & T. Fararo (Eds), The Problem of Solidarity: Theories and Models (pp. 113-138). Amsterdam: Gordon and Breach. Weesie, J., & Raub, W. (1996). Private Ordering: A Comparative Institutional Analysis of Hostage Games. Journal of Mathematical Sociology, 21, 201-240. Williamson, O. E. (1985). The Economic Institutions of Capitalism. New York: The Free Press. Williamson, O. E. (1993). Calculativeness, Trust, and Economic Organization. Journal of Law and Economics, 36, 453-486. Williamson, O. E. (1996). The Mechanisms of Governance. Oxford: Oxford University Press.

GROUP REACTIONS TO LOYALTY AND DISLOYALTY John M. Levine and Richard L. Moreland

ABSTRACT The causes and consequences of loyalty and disloyalty in groups have received little attention from social psychologists. This chapter analyzes how groups respond to loyalty (defined as staying in a group, even though one could obtain a better outcome by leaving, because staying benefits the group) and disloyalty (defined as leaving a group, because one can obtain a better outcome by doing so, even though leaving harms the group). Attention is given to factors that influence the valence and intensity of group responses to loyalty and disloyalty on the part of both ingroup and outgroup members.

INTRODUCTION 'Your sentence,' the court's dreaded formula ran, 'is that you be led back to prison; laid on a hurdle, and so drawn to the place of execution; there to be hanged, to be cut down alive, your privy members cut off and cast into the fire, your bowels burnt before your eyes, your head smitten off, your body quartered and divided at the King's will. God have mercy on your soul. Amen!' (Hibbert, 1971, p. 50). [The recipient of this 16th century treason sentence, John Fisher, Bishop of Rochester, got off lightly. He was merely beheaded, after which his head was parboiled and displayed on a pole on London Bridge.]

Group Cohesion, Trust and Solidarity, Volume 19, pages 203-228. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2 203

204

JOHN M. LEVINE AND RICHARD L. MORELAND

Some 40 years ago, Solomon Asch noted both the importance of group loyalty and the neglect of this phenomenon in social psychology. Asch wrote, "Loyalty to the group. A worker may be convinced that a call to a strike is unwise, but will lay down his tools because he believes that the welfare of his union will be best served by his acquiescence. This quite human and powerful attitude seems not to have found credence in our psychology" (1959, p. 382). With few exceptions, social psychologists have devoted little attention to loyalty in subsequent years. But what exactly is loyalty? Several definitions have been offered. According to Scott (1965), loyalty is " b e i n g . . . a devoted member of the group, never criticizing it to outsiders, and working hard to get it ahead of other groups" (p. 24). Brewer and Brown (1998), following Sumner (1906), stressed "adherence to ingroup norms and trustworthiness in dealings with fellow ingroup members" (p. 560), whereas Brewer and Silver (2000) focused on "willingness of group members to exert effort, pay costs, or sacrifice personal benefits on behalf of the group as a whole" (p. 162). Finally, Zdaniuk and Levine (2001) stated that "loyalty implies more than actual or intended progroup behavior. In addition, lack of concern for personal welfare is necessary" (p. 502). Three important characteristics of loyalty can be derived from these definitions. First, loyalty involves behavior that enhances (or is meant to enhance) group welfare. 1 Second, loyalty involves personal sacrifice - the willingness to forgo individual rewards in the service of group goals. Finally, loyalty occurs in intergroup contexts and often involves effort to benefit the ingroup at the expense of outgroups.

ORIGINS OF LOYALTY: THE ROLE OF GROUP IDENTIFICATION Although loyalty is sometimes viewed as a personality disposition (e.g. James & Cropanzano, 1994), we agree with Brewer and Silver's (2000) emphasis on the causal role of group identification. Consistent with social identity theory (Tajfel & Turner, 1986) and self-categorization theory (Turner, Hogg, Oakes, Reicher & Wetherell, 1987), Brewer and Silver argue that group identification occurs to the extent that an individual adopts a depersonalized self-representation, that is, a view of the self as a group member rather than a unique individual. Further, they contend that "Shared distinctiveness is the group property that is most likely to engage high levels of attachment and identification with a given social group or category" (p. 155). Shared distinctiveness in turn depends on a group's ability to satisfy members' needs for both inclusion and differentiation (Brewer, 1991).

Group Reactions to Loyalty and Disloyalty

205

There are several interrelated reasons why group identification might produce loyalty. For example, identification increases feelings of similarity to other members and liking for them (Brewer & Brown, 1998; Wilder, 1986), which in turn increases the motivation to help them (Dovidio, 1984; Stem, 1995). Identification also triggers collectivist motivation, "whereby an individualistic self-interested orientation is matched or superceded by the motivation to maximize joint or collective interests" (Brewer & Silver, 2000, p. 161; see also Batson, 1998; Dawes, van de Kragt & Orbell, 1988; Tumer et al., 1987). The hypothesis that group identification produces loyalty (defined as progroup behavior accompanied by personal sacrifice) is certainly plausible. However, there is surprisingly little direct evidence for this effect. Although there is strong support for the notion that group identification increases progroup behavior, only a handful of studies provide evidence for the role of personal sacrifice. Research on ingroup bias indicates that people often assign higher rewards to ingroup than to outgroup members, even when the allocation decision is private and allocators do not directly benefit from their decision (Brewer & Brown, 1998; Diehl, 1990). In addition, ingroup bias is more likely when loyalty, as opposed to equality, is a salient norm (Hertel & Kerr, 2001). However, ingroup bias does not entail personal sacrifice (i.e. allocators do not lose anything by favoring ingroup members). On the contrary, recent evidence suggests that such bias occurs mainly when allocators expect reciprocity from ingroup members (Gaertner & Insko, 2000). Research on individual mobility shows that people who strongly identify with a low-status group are less motivated to move to a high-status group than are those who weakly identify (Ellemers, Spears & Doosje, 1997, Experiment 1; see also Doosje, Ellemers & Spears, 1999). However, there is little evidence that people who remain in a low-status group do so because of selfless concern about what will happen to the group if they leave. It is just as likely that their behavior is motivated by selfish concems about possible sanctions from ingroup members and/or lack of acceptance by outgroup members (cf. Branscombe & Ellemers, 1998). Similar ambiguity arises in studies of organizational tumover, which typically do not focus on movement from one organization to another (e.g. Abrams, Ando & Hinkle, 1998; Mael & Ashforth, 1995; Mathieu & Zajac, 1990; see also Ellemers, Kortekaas & Ouwerkerk, 1999). Although these studies indicate that people who are strongly identified with or highly committed to an organization are more likely to stay in it than are people whose identification or commitment is lower, in many cases staying may be motivated by a concem for personal, rather than organizational, welfare. Consistent with this argument, social exchange analyses of the decision to stay in vs. leave an organization or

206

JOHN M. LEVINE AND RICHARD L. MORELAND

group emphasize personal, rather than collective, rewards and costs (e.g. Moreland & Levine, 1982; Moreland, Levine & Cini, 1993; Rusbult & Farrell, 1983; Rusbult, Farrrell, Rogers & Mainous, 1988; see also Lawler, Thye & Yoon, 2000). The complex motivations underlying staying in or leaving a group were revealed in a recent study by Zdaniuk and Levine (2001), who assessed the impact of group identification on two forms of loyalty: (1) staying in a group when doing so benefits the group but harms oneself; and (2) leaving a group when doing so benefits the group but harms oneself. Findings indicated that identification with the group increased the likelihood that people would stay in it, even though doing so helped the group in the former case and harmed it in the latter case. This pattern of results was interpreted in terms of adherence to a non-abandonment norm, which stipulates that people who identify with a group should remain in it and receive the same outcomes as everyone else. Interestingly, participants' apparent failure to consider group welfare in deciding whether to stay or leave did not mean they were insensitive to this issue. Those who stayed and helped the group felt more moral than did those who left and harmed the group. In contrast, those who stayed and harmed the group felt less moral than did those who left and helped the group. Identification with or commitment to an organization or group can affect other behaviors besides staying and leaving. These include job performance, tardiness, absenteeism, and extra-role (citizenship) behavior, such as helping co-workers and volunteering for difficult assignments (Mathieu & Zajac, 1990; Meyer & Allen, 1997; Morrow, 1993; Randall, Fedor & Longenecker, 1990). However, as in the case of turnover, it is not clear that these behaviors are always (or even usually) motivated by concern for collective, as opposed to personal, welfare. In analyzing organizational commitment from a social identity perspective, Ouwerkerk, Ellemers and De Gilder (1999) argued that affective commitment (Meyer & Allen, 1991), which is closely related to social identification, is a stronger predictor of employee behaviors than are other kinds of commitment. Moreover, they suggested that some consequences of affective commitment are better indicators of willingness to exert effort on behalf of the organization than are others. Job performance, tardiness, and absenteeism, for example, are often heavily influenced by organizational control mechanisms that constrain employees to behave in ways that benefit the organization. For this reason, these behaviors cannot be confidently attributed to selfless concern for the organization (sacrifice). In contrast, extra-role (citizenship) behavior seems to provide clearer evidence of group-oriented motivation on the part of employees (see also Organ & Ryan, 1995; Tyler & Blader, 2001). Even in these cases, however, it is difficult to rule out the possibility that employees are at

Group Reactions to Loyalty and Disloyalty

207

least partially motivated by the belief that "voluntary" efforts to help the organization will yield personal benefits down the road (Bolino, 1999). The desire to achieve personal goals, including positive social identity, may also be important in studies showing that identification with or commitment to work groups produces increased effort on their behalf (e.g. Karau & Williams, 2001; Ouwerkerk, de Gilder & de Vries, 2000; Worchel, Rothgerber, Day, Hart & Butemeyer, 1998). In contrast to the research discussed above, other lines of work provide stronger evidence that group identification can increase progroup behavior that involves personal sacrifice. For example, in a study pitting personal welfare against group welfare, Barreto and Ellemers (2000) forced participants to choose between working on tasks that would contribute to their own performance or to their group's performance. When group norms favored progroup behavior, high identifiers chose to help the group whether or not they were accountable to fellow members (i.e. would have to justify their decisions). In contrast, low identifiers only made progroup responses when they were accountable (see also Ellemers, Barreto & Spears, 1999). Additional evidence linking group identification, progroup behavior, and personal sacrifice comes from research on social dilemmas. In non-iterated dilemmas, cooperation necessarily involves personal sacrifice, because each group member obtains a higher payoff for self-oriented behavior (defection) than for group-oriented behavior (cooperation), no matter what others do. 2 A number of studies indicate that group identification enhances cooperative responding in social dilemmas (e.g. Brewer & Kramer, 1986; Dawes et al., 1988; De Cremer & van Dijk, in press; De Cremer & van Vugt, 1999; Kramer & Brewer, 1984; Kramer & Goldman, 1995). GROUP

REACTIONS

TO LOYALTY/DISLOYALTY

Having defined loyalty and discussed the role that group identification plays in producing it, we now address our primary goal in this chapter, namely analyzing how groups respond to loyalty and disloyalty. Although these behaviors can occur in contexts where intergroup relations are not salient, we believe that loyalty and disloyalty matter most when they influence the outcome of intergroup competition. Therefore, we focus on situations in which: (a) two groups (an ingroup and an outgroup) are in a competitive relationship; (b) people who leave one group join the other group; and (c) people increase the competitive advantage of whichever group they belong to. In these situations, loyalty is defined as staying in the group, even though one could obtain a better outcome by leaving, because staying benefits the group. In contrast, disloyalty

208

JOHN M. LEVINE AND RICHARD L. MORELAND

is defined as leaving the group, because one can obtain a better outcome by doing so, even though leaving harms the group. 3 It is useful to compare group responses to loyalty with group responses to personally rewarding progroup behavior, defined as staying in the group when doing so benefits both oneself and the group. Although parallel logic might suggest comparing group responses to disloyalty with group responses to personally punishing anti-group behavior, defined as leaving the group when doing so harms both oneself and the group, it is difficult to imagine the conditions that would produce such behavior.

Reactions to Loyal and Disloyal Ingroup Members There are several reasons why the ingroup's ability to compete with the outgroup is strengthened by loyal ingroup members and weakened by disloyal ingroup members. For example, loyal members contribute their task-relevant knowledge and skills to the ingroup and deny these to the outgroup. In addition, by demonstrating their commitment to the ingroup in the face of temptation to defect to the outgroup, these members increase the perceived value of the ingroup and decrease the perceived value of the outgroup in the eyes of other members. This in turn should increase other members' motivation to work for the ingroup and reduce their motivation to join the outgroup. Conversely, disloyal members deny their skills and knowledge to the ingroup and provide these to the outgroup, along with "secret" information about how the ingroup operates. Moreover, by demonstrating their lack of commitment to the ingroup in the face of temptation to defect, these members decrease the perceived value of the ingroup and increase the perceived value of the outgroup. This in turn should reduce other members' motivation to work for the ingroup and increase their motivation to join the outgroup (cf. Levine, Moreland & Ryan, 1998). All of this suggests that loyal ingroup members will elicit more favorable reactions than will disloyal ingroup members. Finally, positive reactions to loyal members and negative reactions to disloyal members should be stronger in groups that elicit high rather than low identification from their members. Relevant research. Several lines of research, though not explicitly focusing on loyalty and disloyalty as defined above, are consistent with these hypotheses. There is substantial evidence, for example, that ingroup members who dissent from group consensus on opinion issues are liked less than those who conform (see Levine, 1989; Levine & Thompson, 1996). Both consistent deviants and conformers who become deviants are liked less than deviants who become conformers and consistent conformers (e.g. Levine, Sroka & Snyder, 1977). A number of factors influence how much rejection ingroup deviants receive. For

Group Reactions to Loyalty and Disloyalty

209

example, deviants are rejected more when they advocate extreme positions (e.g. Hensley & Duval, 1976) or interfere with group goal attainment (e.g. Kruglanski & Webster, 1991) and when the group is cohesive (e.g. Schachter, 1951) or possesses norms prohibiting deviance (e.g. Janis, 1982). Festinger (1950) identified two motives that underlie pressure toward group uniformity and thereby rejection of members who undermine consensus. The first motive, group locomotion, involves the desire to move toward valued group goals. The second motive, social reality, involves the desire to create and maintain a subjectively valid view of the world (Levine & Higgins, 2001). Recent research on the "black sheep effect" has identified another important reason why ingroup deviants are rejected - namely the threat they pose to other members' social identity (Marques & Paez, 1994; Marques, Abrams, Paez & Hogg, 2001; see also Iwao, 1963). Evidence from several studies demonstrates that unlikeable ingroup members are evaluated more negatively than similar outgroup members, whereas likeable ingroup members are evaluated more positively than similar outgroup members. Analyses of the black sheep effect suggest that groups derogate ingroup deviants because doing so protects their positive social identity, which is most vulnerable when deviants violate prescriptive norms that distinguish the ingroup from the outgroup (Marques, Abrams & Serodio, 2001). Consistent with this analysis, individuals with higher group identification are more likely to reject ingroup deviants than are individuals with lower identification (Coull, Yzerbyt, Castano, Paladino & Leemans, 2001; Marques, Abrams, Paez & Martinez-Taboada, 1998, Experiment 4). Moreover, there is evidence that derogation of ingroup deviants increases the group identification of those who engage in this activity (Marques et al., 1998; see also Marques et al., 2001). In a related line of work, Branscombe, Wann, Noel and Coleman (1993) assessed reactions to loyal vs. disloyal in-group and out-group members, where loyalty was defined as unwavering support for one's group (a college basketball team) regardless of its performance and disloyalty was defined as contingent support (high when the group was successful and low when it was unsuccessful). Branscombe et al. found that in-group members who were loyal when their group was threatened received the highest evaluations, whereas those who were disloyal when their group was threatened received the lowest evaluations. Moreover, loyal in-group members were evaluated most positively and disloyal in-group members were evaluated most negatively by participants who identified strongly with their group. Branscombe et al. also found that loyalty elicited more favorable reactions than what we earlier called "personally rewarding progroup behavior." They noted that, "When the in-group member behaves loyally following a win, then the individual's loyalty may not have

210

JOHN M. LEVINE AND RICHARD L. MORELAND

been genuinely tested, and so some element of doubt about the target might temper the otherwise positive evaluation" (p. 382). Loyalty and disloyalty can also be defined in terms of how former members of a group evaluate remaining members, with positive evaluations constituting loyalty and negative evaluations constituting disloyalty. To investigate reactions to this kind of loyalty/disloyalty, Moreland and McMinn (1999) created six-person groups, had them work on a task, and then split them into two three-person groups. These smaller groups then worked on a new task (creation of a university recruiting poster), after which they evaluated their own poster and one produced either by ingroup members (i.e. former members of their original six-person group) or outgroup members (i.e. former members of a different six-person group). More positive evaluations were given to posters produced by ingroup than by outgroup members, and this effect was stronger when participants' identification with their original group was high rather than low. Next, participants received a negative evaluation of their own poster that ostensibly came from either ingroup or outgroup members. This evaluation produced more distress when it reflected the views of ingroup members (and hence reflected disloyalty) than when it reflected the views of outgroup members, and again this effect was stronger when participants were highly identified with their original group. Given the focus of this chapter, we are particularly interested in how group members respond to loyalty defined as remaining in the group and disloyalty defined as leaving the group to join another (defection). According to Tajfel and Turner (1986), people who defect are heartily disliked, particularly when there is an intense conflict of interest between the two groups. They suggest that, "This is precisely the situation...in which it is extremely difficult for an individual to conceive of the possibility of "betraying" his or her own group by moving to the opposing group. Although this does happen on occasion, sanctions for such a move are, on the whole, powerful, and the value systems (at least in our cultures) are in flagrant opposition to i t . . . The intensity of explicit intergroup conflicts of interests is closely related in our cultures to the degree of opprobrium attached to the notion of "renegade" or "traitor" (p. 10). A similar point was made by Branscombe and Ellemers (1998) in their discussion of the risks associated with individualistic strategies for dealing with membership in a low-status group. We could locate only two experiments that explicitly investigated group responses to defection. In the first experiment, Singer, Radloff, and Wark (1963) placed participants into competing teams to discuss human relations problems. A confederate in each team, after initially agreeing with group consensus, either disrupted team performance (heretic) or defected to join the competing team

Group Reactions to Loyalty and Disloyalty

211

(renegade). Over the course of the experiment, participants became more negative toward the renegade than toward the heretic, other members of the ingroup, or other members of the outgroup. In the second experiment, Charlton and Bettencourt (2001) allowed members of a low-power group to overhear a fellow member either succeed or fail to gain entry into a high-power group. As predicted, the successful member was liked more than the failing member, particularly when group boundaries were relatively impermeable. Charlton and Bettencourt suggested that social mobility elicited a positive, rather than a negative, reaction in their experiment because the defector did not actively devalue his or her original group (cf. Branscombe et al., 1993) and because the two groups were in a cooperative, rather than a competitive, relationship. They indicated that they "might expect rejection of an in-group member who attempted to join a higher power adversarial group, especially if the move were detrimental to the low power group" (p. 89). They also noted that participants in their study may have believed the defector was motivated by concern for collective rather than personal gain, although if this were true, the successful defector should have been viewed positively in the permeable as well as the impermeable boundary condition. In summary, a number of studies are potentially relevant to the question of how groups respond to loyalty and disloyalty on the part of ingroup members, where: (a) loyalty is defined as staying in a group, even though one could obtain a better outcome by leaving, because staying benefits the group; and (b) disloyalty is defined as leaving a group, because one can obtain a better outcome by doing so, even though leaving harms the group. However, with the partial exception of the Singer et al. (1963) experiment, none of these studies directly examined loyalty/disloyalty as we have defined them here. This is surprising, given the strong responses that loyalty and disloyalty evoke in many natural groups. As the quote at the beginning of the chapter suggests, the punishments associated with disloyalty are often quite negative, including public condemnation, imprisonment, and death. In contrast, the rewards associated with loyalty are often quite positive, including public acclaim, medals for heroism, and election to high office. Given the lack of theoretical and empirical work on this important topic, it seems worthwhile to consider some of the factors that may affect the levels of approbation/opprobrium that loyal and disloyal ingroup members elicit. Characteristics of the person. One potentially important characteristic of a loyal or disloyal person is his or her perceived motive for remaining in the group or defecting to a competing group (cf. Levine, 1989). When someone stays, the amount of personal loss (sacrifice) incurred by staying affects the

212

JOHN M. LEVINE AND RICHARD L. MORELAND

person's perceived motives, which in turn affect how loyal he or she is perceived to be. Because personal loss is an inhibitory cause for staying, it increases the likelihood that staying will be attributed to facilitative causes, such as wanting to help the ingroup and harm the outgroup (Kelley, 1971). The more staying is attributed to these motives, the more loyal the person will seem to be. Finally, perceived loyalty should affect liking, with more loyalty producing more liking. Parallel predictions can be made regarding reactions toward someone who leaves the ingroup. When this occurs, the amount of personal gain obtained by leaving affect the person's perceived motives, which in turn affects how disloyal he or she is perceived to be. Because personal gain is a facilitative cause for leaving, it decreases the likelihood that leaving will be attributed to alternative (and more negative) facilitative causes, such as wanting to harm the ingroup and help the outgroup. The less leaving is attributed to these motives, the less disloyal the person will seem to be. Finally, perceived disloyalty should affect liking, with less disloyalty producing more liking. In sum, an ingroup member who stays will be liked more when staying involves a large rather than a small personal loss, whereas an ingroup member who leaves will be liked more when leaving involves a large rather than a small personal gain. These predictions are consistent with evidence indicating that responses to deviants are affected by their attributed motives (e.g. Burnstein & Worchel, 1962; Levine & Ruback, 1980; Levine et al., 1977). The s t a t u s of loyal and disloyal people is also likely to affect the reactions they elicit from other group members. In general, loyal high status people will be liked more than loyal low status people, whereas disloyal high status people will be liked less than disloyal low status people. There are at least three reasons for this prediction. First, because high status people have more task-relevant skills than do low status people and thus are more instrumental to group goal attainment, their decision to remain in the group (loyalty) is more rewarding and their decision to leave (disloyalty) is more punishing. Second, because high status people are closer to the group prototype than are low status people (Hogg, 2001), their behavior has more impact on other members' social identity. Compared to low status people who remain in the group, high status people who remain do more to reaffirm other members' identity and hence are more rewarding. Compared to low status people who leave the group, high status people who leave do more to undermine other members' identity and hence are more punishing. Third, because high status people elicit more respect from other group members than do low status people, they are more likely to be imitated. Compared to low status people, then, high status people who stay will inhibit defection by other members, and high status people who leave will stimulate defection. Given that groups want to retain their members, high status people

Group Reactions to Loyalty and Disloyalty

213

who remain are a great asset, which should cause them to be strongly liked, whereas high status people who leave are a great liability, which should cause them to be strongly disliked. The tenure of loyal and disloyal people in the group may also be important. Though tenure is often positively related to status (Insko, Thibaut, Moehle, Wilson, Diamond, Gilmore, Solomon & Lipsitz, 1980), this is not necessarily the case, and the amount of time people have spent in the group before exhibiting loyalty or disloyalty may have an independent effect on other members' reactions to them. The longer people have been in the group, the more their decision to stay or leave may seem to reflect: (a) accurate knowledge of what the group is like; and (b) thoughtful consideration of the future benefits and costs of membership. The more knowledge and consideration that are assumed to underlie the decision, the more informed the decision will appear to be, which in turn should affect other members' reactions to it. In addition, the longer people have been in the group, the greater the investment other members will have made in them, which in turn should affect reactions to their decision to stay or leave. For both of these reasons, people who stay in the group should be liked more when their (pre-loyalty) tenure was long rather than short, whereas people who leave the group should be liked less when their (pre-disloyalty) tenure was long rather than short. Finally, the personal relationships that loyal and disloyal people have with other group members may influence reactions to their decision to stay or leave, in part because these relationships can serve as levers for social influence. People who stay in the group and have close friends there can use their relationships to convince these friends to remain. Similarly, people who leave the group and have close friends there can use their relationships to convince these friends to defect. Because of their power to help or harm the group by stabilizing or destabilizing its membership, loyal people with close ties to other members are probably liked more than loyal people without such ties, whereas disloyal people with close ties to other members are probably liked tess than disloyal people without such ties. There is another reason why disloyal people with close ties to other members may be disliked, even if they are not successful in producing defection - their friends may view the disloyalty as a personal betrayal, which in turn will anger and demoralize them. And to the extent this demoralization interferes with group performance, other members who were not friends of those who left will be angry. Characteristics of the ingroup. Characteristics of the ingroup can also affect reactions to loyalty and disloyalty, and these reactions also depend on the rewards and costs of these two kinds of behavior for the group. More specifically, groups under high threat should respond more positively to loyalty

214

JOHN M. LEVINE AND RICHARD L. MORELAND

and more negatively to disloyalty than should groups under low threat (cf. Branscombe et al., 1993). The group's role in mediating rewards is likely to affect members' perceptions of threat and thereby their reactions to people who stay or leave. The more important these rewards, the more sensitive members will be to behavior that facilitates and inhibits their attainment. So, as the importance of group-mediated rewards increases, positive responses to people who stay and negative responses to people who leave should both increase. Relevant rewards can be either tangible, such as obtaining a monetary payoff, or intangible, such as maintaining positive social identity. Consistent with this argument, research indicates that people who inhibit the attainment of important rewards elicit strong negative reactions from other members of their group (e.g. Berkowitz & Howard, 1959; Kruglanski & Webster, 1991; Marques & Paez, 1994; Marques et al., 2001). Another factor likely to influence members' reactions to people who stay or leave is group performance. Members of failing groups should feel high threat, which should cause them to respond very positively to people who stay and very negatively to people who leave. In contrast, members of successful groups should feel low threat, which should reduce the intensity of their reactions to loyal and disloyal people. In assessing group performance, members are likely to consider how well the group did in the past, how well it is doing in the present, and how well it will do in the future. Although there is probably a negative relationship between group performance and the extremity of reactions to loyalty and disloyalty, this relationship may not be linear. For example, when group performance has been, is, and seems likely to remain extremely low, members may lose commitment to the group and feel empathy for those who decide to leave. In such cases, "rats leaving a sinking ship" may not earn the enmity of fellow group members who remain behind (cf. Charlton & Bettencourt, 2001). An important factor in group performance is staffing level, defined as the relationship between the number of people who belong to the group and the number needed to perform the group task (Barker, 1968). Understaffed groups have too few members to perform their tasks; adequately staffed groups have exactly as many members as they need; and overstaffed groups have too many members. By definition, understaffed groups are likely to perform worse than adequately or overstaffed groups. For this reason, members of understaffed groups should feel high threat, causing them to respond very positively to people who stay and very negatively to people who leave (cf. Cini, Moreland & Levine, 1993). Similar, though less extreme, responses are likely in adequately staffed groups. Though these groups are currently in good shape, their performance could deteriorate if even a few members defected. Therefore, they too will respond positively to people who stay and negatively to people who leave, though with

Group Reactions to Loyalt)) and Disloyalty

215

lower intensity than in adequately staffed groups. What about overstaffed groups? One might assume that they will respond in just the opposite fashion, that is, negatively to people who stay and positively to people who leave. Recall, however, that we are focusing on situations in which two groups are in a competitive relationship, people who leave one group join the other group, and people increase the competitive advantage of whichever group they belong to. In this situation, even overstaffed groups cannot afford to punish loyalty and reward disloyalty. Therefore, we would also expect them to respond positively to people who stay and negatively to people who leave, though with even lower intensity than in adequately staffed groups. Finally, it is likely that all three types of groups will respond more strongly to both loyalty and disloyalty when: (a) the group has recently lost members; and (b) the pool of prospective members is small. Characteristics of the intergroup relationship. As noted earlier, we are concerned with loyalty and disloyalty in competitive intergroup situations, where a person's decision to stay in a group increases its competitive advantage and a decision to leave (joining the competing group) has the opposite effect. In order to understand how people respond to ingroup members who are loyal and disloyal, it is thus necessary to consider the impact of intergroup competition on intragroup processes (cf. Levine et al., 1998). Evidence indicates "there is usually more ingroup bias, less intergroup liking, and greater intergroup discrimination when groups are objectively in competition than when they are independent or must cooperate to achieve a common goal" (Brewer & Brown, 1998, p. 565). Therefore, the more intergroup competition there is (because important rewards are at stake, the groups have a history of hostile relations, and so on), the more motivated members should be to enhance the competitive advantage of their ingroup and diminish the competitive advantage of the outgroup. This in turn should produce more positive reactions toward people who stay and more negative reactions toward people who leave. In addition to the level of intergroup competition, the anticipated outcome of that competition is likely to affect reactions to loyalty and disloyalty on the part of ingroup members, because outcomes influence perceived threat. In our discussion of ingroup characteristics, we argued that failing groups feel more threatened than successful groups, which leads them to respond more positively to people who stay and more negatively to people who leave. Although, as this argument suggests, a group's absolute performance level is likely to influence its reactions to loyal and disloyal members, its relative performance vis-a-vis an outgroup may be even more important. To the extent the outgroup has performed better in the past and is performing better in the present, ingroup members will fear defeat in the future, which in turn should increase the intensity of their reactions to loyalty and disloyalty (cf. Branscombe et al., 1993).

216

JOHN M. LEVINE AND RICHARD L. MORELAND

Other factors besides the overall performance levels of the two groups may affect ingroup members' predictions about the likely outcome of intergroup competition (and hence the amount of threat they feel). One such factor is temporal change in the performance of the groups. For example, ingroup members are likely to be more worried if the outgroup's performance has increased over time, while their own has decreased, than by the opposite trend. The history of member movement between the two groups may also be important. Ingroup members are likely to be more worried, for example, if they have recently lost more members to the outgroup than they have gained from this group. Finally, anticipated changes in the nature of the competitive task may influence perceived threat. For example, ingroup members are likely to be more worried if the task is changing in a way that favors the outgroup than if it is changing in a way that favors the ingroup. The similarity between the ingroup and outgroup on such dimensions as goals, member characteristics, and structural features is yet another factor that may influence reactions to loyalty and disloyalty on the part of ingroup members. Similarity might be important for two reasons. First, because group members' social identity is based on the (positive) distinctiveness of their group vis-a-vis an outgroup (Tajfel & Turner, 1986), social identity theory suggests that a similar outgroup poses a greater threat than does a dissimilar outgroup, which in turn leads to greater effort to positively differentiate the ingroup from the outgroup. Evidence confirms this prediction for people who identify highly with their ingroup (Jetten, Spears & Manstead, 1999, 2001). Therefore, when members identify highly with their ingroup, the more similar the ingroup and outgroup, the more ingroup members who stay should be liked and the more those who leave should be disliked. Second, the similarity of the ingroup and outgroup may affect the attributions that group members make for loyalty and disloyalty, which in turn may affect the perceived rewards and costs of these behaviors and thereby the intensity of members' reactions to them. It seems likely that, the greater the similarity between the ingroup and outgroup, the more staying will be attributed to attraction to the ingroup and the more leaving will be attributed to rejection of the ingroup. If so, greater intergroup similarity should increase positive responses to people who stay and negative responses to people who leave.

Reactions to Loyal and Disloyal Outgroup Members Just as a group's ability to compete with an outgroup is strengthened by loyal ingroup members and weakened by disloyal ingroup members, so it is weakened by loyal outgroup members and strengthened by disloyal outgroup

Group Reactions to Loyalty and Disloyalty

217

members. For example, loyal outgroup members contribute their task-relevant knowledge and skills to the outgroup and deny these to the ingroup. In addition, by demonstrating their commitment to the outgroup in the face of temptation to defect to the ingroup, these members increase the perceived value of the outgroup and decrease the perceived value of the ingroup in the eyes of other outgroup members. This in turn should increase other members' motivation to work for the outgroup and reduce their motivation to join the ingroup. Conversely, disloyal outgroup members deny their skills and knowledge to the outgroup and provide these to the ingroup, along with "secret" information about how the outgroup operates. Moreover, by demonstrating their lack of commitment to the outgroup in the face of temptation to defect, these members decrease the perceived value of the outgroup and increase the perceived value of the ingroup. This in turn should reduce other members' motivation to work for the outgroup and increase their motivation to join the ingroup (cf. Levine et al., 1998). All of this suggests that loyal outgroup members will elicit less favorable reactions than will disloyal outgroup members (cf. Abrams, Marques, Bown & Henson, 2000). Moreover, these reactions should be stronger when ingroup members feel high rather than low identification with their own group (cf. Abrams et al., 2000). In general, we expect characteristics of the person, the ingroup, and the intergroup relationship that affect reactions to loyal and disloyal ingroup members to have opposite effects on reactions to loyal and disloyal outgroup members. This is because factors that increase the value of loyalty by ingroup members decrease the value of loyalty by outgroup members, whereas factors that decrease the value of disloyalty by ingroup members increase the value of disloyalty by outgroup members. Regarding characteristics of the person, we predict that loyal outgroup members (who refuse to defect to the ingroup) will be liked less when they incur high rather than low personal loss, because willingness to incur high loss for staying indicates strong desire to help the outgroup and harm the ingroup. Here, greater perceived loyalty elicits lower liking. In addition, we predict that disloyal outgroup members (who defect to the ingroup) will be liked more when they obtain low rather than high personal gain, because willingness to defect for low gain indicates strong desire to harm the outgroup and help the ingroup. Here, greater perceived disloyalty elicits higher liking. Parallel arguments lead to predictions that: (a) loyal outgroup members with high status will be liked less than those with low status, whereas disloyal outgroup members with high status will be liked more than those with low status; (b) loyal outgroup members will be liked less when their (pre-loyalty) tenure was long rather than short, whereas disloyal outgroup members wilt be liked more when their

218

JOHN M. LEVINE AND RICHARD L. MORELAND

(pre-disloyalty) tenure was long rather than short; and (c) loyal outgroup members with close ties to other members will be liked less than those without such ties, whereas disloyal outgroup members with close ties to other members will be liked more than those without such ties. Our remaining predictions regarding reactions to loyalty and disloyalty are also reversed when outgroup, rather than ingroup, members are considered. Regarding characteristics of the ingroup, we expect that loyal outgroup members will be liked less and disloyal outgroup members will be liked more as: (a) the importance of ingroup-mediated rewards increases; (b) ingroup performance decreases; and (c) ingroup staffing level decreases. Regarding characteristics of the intergroup relationship, we predict that loyal outgroup members will be liked less and disloyal outgroup members will be liked more as: (a) the level of intergroup competition increases; (b) the relative performance of the ingroup decreases; and (c) the similarity between the ingroup and outgroup increases. This analysis suggests that loyal outgroup members (who remain in the outgroup) receive the same negative reaction as disloyal ingroup members (who join the outgroup), because both kinds of people decrease the ingroup's competitive advantage. In contrast, disloyal outgroup members (who join the ingroup) receive the same positive reaction as loyal ingroup members (who remain in the ingroup), because both kinds of people increase the ingroup's competitive advantage. There is reason to believe, however, that reactions to loyal and disloyal outgroup members may sometimes differ from reactions to disloyal and loyal ingroup members, respectively. On the one hand, reactions to loyal outgroup members may be more positive than our analysis suggests. Loyal outgroup members, though disliked because they reduce the ingroup's competitive advantage, may nonetheless earn ingroup members' grudging respect for remaining in the outgroup when it is not in their personal interest to do so. This reaction is sometimes observed in battlefield situations, where soldiers on the winning side show respect to soldiers on the losing side who fought valiantly rather than surrendering (e.g. Ulysses S. Grant's generous treatment of Confederate troops at Appomatox). A related point was made by Moscovici (1976) in discussing how numerical minorities are perceived and evaluated by majorities. He argued that although minorities tend to be disliked, they are often admired for maintaining an unpopular position in the face of majority opposition. On the other hand, reactions to disloyal outgroup members (outgroup defectors) may be more negative than our analysis suggests. This may occur for two related reasons. First, outgroup defectors may elicit antipathy, because they violate a widely held norm that people should not switch sides during intergroup conflict (cf. Zdaniuk & Levine, 2001). Second, outgroup defectors

Group Reactions to Loyalty and Disloyalty

219

may not be trusted, because of fear that they will betray the ingroup in the future (cf. Moreland & Levine, in press). In assessing the likelihood that outgroup defectors will betray the ingroup and hence do not deserve trust, ingroup members are likely to consider: (1) the defector's current orientation toward the ingroup (positive, negative); and (2) the defector's probable future orientation toward the ingroup (positive, negative). Crossing these two factors yields four kinds of defectors (see Table 1). Reliable defectors are people whose current and future orientations toward the ingroup are believed to be positive. They are perceived as motivated to help the ingroup in the present (e.g. because they have recently engaged in freely chosen and costly actions that help the ingroup) and as likely to remain committed to this goal in the future (e.g. because they have cut all personal ties with members of the outgroup). Such people will elicit long-term trust from ingroup members. Unreliable defectors are people whose current orientation toward the ingroup is believed to be positive, but whose future orientation is likely to be negative. They are seen as motivated to help the ingroup in the present (e.g. because they are being paid to do so), but as likely to lose commitment to this goal in the future (e.g. because the outgroup will pay them more to switch sides). Such people may elicit some trust in the short term, but will not be given access to information that could be used against the ingroup in the future. Unreliable spies are people whose current orientation toward the ingroup is believed to be negative, but whose future orientation is likely to be positive. They are seen as motivated to harm the ingroup in the present (e.g. because they are employed by the outgroup), but as likely to lose commitment to this goal in the future (e.g. because the ingroup can force them to change sides). Such people will not elicit trust in the short term, but may be granted a modicum of trust after they change sides. Finally, reliable spies are people whose current and future orientations toward the ingroup are believed to be negative. They are seen as motivated to harm the ingroup in the present and as likely to remain committed to this goal in the future (e.g. because they are ideologically committed to the outgroup). Such people will not elicit either short-term or long term-term trust. Research by Prislin and her colleagues on reactions to losing vs. gaining a majority position in a group is consistent with the argument that defection by outgroup members is not always rewarding. Prislin, Limbert, and Bauer (2000) found that although participants who lost support for their position (and thereby shifted from the majority to the minority faction) had intense negative reactions to the experience, those who gained support (and thereby shifted from the minority to the majority faction) did not respond positively to the experience. These effects were replicated in a subsequent study when the social

220

JOHN M. LEVINE AND RICHARD L. MORELAND Table 1.

A typology of disloyal outgroup members (defectors). Defector's presumed current orientation to ingroup Positive Negative

Defectors' probable future orientation to ingroup

Positive Negative

Reliable defector Unreliablespy Unreliable defector Reliable spy

situation was construed as a group, but not when it was construed as an aggregate (Prislin, Brewer & Wilson, in press). An interesting aspect of Prislin's findings is that participants disliked a group in which they gained support as much as a group that consistently opposed them. These findings seem inconsistent with other evidence showing that group members like an individual who moves from disagreement to agreement with their position much more than one who consistently disagrees (e.g. Levine, Saxe & Harris, 1976; Levine et al., 1977; Schachter, 1951). As Prislin suggested, the difference between the two sets of findings may reflect the belief that individual change is more genuine and durable than group change.

FUTURE DIRECTIONS We have offered hypotheses about several potentially important determinants of the valence and intensity of group members' responses to loyalty and disloyalty on the part of ingroup and outgroup members. These determinants are listed in Table 2. Although we touched on many issues, space considerations precluded a complete analysis of this rich and intriguing topic. Three issues in particular deserve future theoretical and empirical attention. The first issue concerns the number of people who are loyal or disloyal. Although we focused on single individuals who either stay in or leave their group, subgroups can behave in the same ways. It is likely that reactions to subgroups exhibiting a particular behavior will be of the same valence, but of greater intensity, than reactions to individuals exhibiting that behavior. This is because, holding other factors (e.g. status) constant, a subgroup will have more impact than an individual on the ingroup's ability to compete with the outgroup. Moreover, this is likely even if the subgroup consists of only two members (cf. Moscovici & Lage, 1976; Tindale, Davis, Vollrath, Nagao & Hinsz, 1990). Thus, subgroups that show loyalty to the ingroup should elicit more positive reactions than individuals who do so, whereas subgroups that show disloyalty to the ingroup should elicit more negative reactions than individuals who do

Group Reactions to Loyalty and Disloyalty Table 2.

221

Determinants of group members' responses to loyalty/disloyalty by ingroup/outgroup members.

Characteristics of the person

Perceived motive (based on personal gain or loss) Status in the group Tenure in the group Personal relationships with other group members Characteristics of the ingroup

Group's role in mediating rewards Group's performance Group's staffing level Characteristics of the intergroup relationship

Level of intergroup competition Anticipated outcome of competition (based on ingroup's relative performance) Similarity of ingroup and outgroup

so. Conversely, subgroups that show loyalty to the outgroup should elicit more negative reactions than individuals who do so, whereas subgroups that show disloyalty to the outgroup should elicit more positive reactions than individuals who do so. It is also likely that reactions to all four kinds of subgroups will vary positively with their size, though this relationship may not be linear (cf. Clark, 1999). Although social psychologists have shown little interest in loyal and disloyal subgroups, recent efforts have been made to understand the causes of schisms, defined as "the division of a group into subgroups, and the ultimate secession of at least one subgroup from the parent group" (Sani & Reicher, 1998, p. 624). Based on their analyses of schisms in the Italian Communist Party and the Church of England, Sani and Reicher (1998, 1999, 2000) concluded that schisms arise when subgroups: (a) view themselves as defending the group's essence from competing subgroups that seek to subvert it; and (b) do not feel they have a voice in arguing their position within the group. A second issue that deserves consideration is how group members behaviorally express their positive and negative reactions to loyal and disloyal behavior on the part of ingroup and outgroup members. It has been suggested, for example, that negative reactions to ingroup deviants sometimes involve punishments designed to change the deviant's opinion and other times involve efforts to exclude the deviant from the group (Israel, 1956; Orcutt, 1973).

222

JOHN M. LEVINE AND RICHARD L. MORELAND

Among the exclusionary responses that groups use in dealing with ingroup deviants is ostracism, defined as "any act or acts of ignoring and excluding of an individual or groups by an individual or group" (Williams, 2001, p. ix). Group members have multiple options for responding to loyalty and disloyalty by ingroup and outgroup members, and these options may serve different goals. In the case of ingroup loyalty, group members are motivated to reward the person for his or her behavior, to signal other ingroup members that loyalty is rewarding, and to ensure that the person remains loyal in the future. To achieve these goals, group members are likely to deliver various kinds of rewards, such as verbal praise, money, and medals, to the loyal person. In the case of ingroup disloyalty, group members are motivated to punish the person for his or her behavior, to signal other ingroup members that disloyalty is dangerous, and perhaps to rehabilitate the person. To achieve the first and second goals, group members are likely to deliver various kinds of punishments, including verbal abuse, monetary fines, ostracism, imprisonment, torture, and even death, to the disloyal person. But to achieve the third goal, group members must reduce the severity of their punishment and hold out the possibility of reward for future loyalty. In the case of outgroup loyalty, group members are motivated to punish the person for his or her behavior, to signal other outgroup members that loyalty is dangerous, and perhaps to seduce the person or other outgroup members into becoming disloyal. To achieve the first and second goals, group members are likely to deliver various kinds of punishments, including imprisonment or death, to the loyal person. But to achieve the third goal, group members must again reduce the severity of their punishment and hold out the possibility of reward for future disloyalty. Finally, in the case of outgroup disloyalty, group members are motivated to reward the person for his or her behavior, to signal other outgroup members that disloyalty is rewarding, and to ensure that the person remains disloyal in the future. To achieve these goals, group members are likely to deliver various kinds of rewards, such as verbal praise, money, and medals, to the disloyal person. A third issue worthy of future attention involves the situation in which different group members have different perceptions of what constitutes loyal and disloyal behavior. For example, cases may arise in which an individual feels that the group would benefit by his or her leaving (Zdaniuk & Levine, 2001), but believes that other members would not perceive this benefit and instead would view leaving as treason. Suppose, for example, that a member feels that he or she could help the ingroup by working as a double agent (ostensibly defecting to the outgroup as a way to gain information useful to the

Group Reactions to Loyalty and Disloyalty

223

ingroup), but, in order for this to work, the defection would have to appear genuine to all or most ingroup members. In such cases, individuals are likely to experience conflict over what to do - should they help the ingroup by defecting, but incur the wrath of other members by doing so, or should they hurt the ingroup by staying, but retain the affection of other members by doing so? Given the power of groups to constrain the behavior of their members, it is not surprising that few people have the courage to risk being labeled traitors, even though by (nominally) defecting to the outgroup they would help, rather than harm, their ingroup (cf. Staub, 1997; Winslow, 1998).

NOTES 1. In contrast, Brewer and Silver (2000) view loyalty as an action potential that mediates the relationship between group identification and progroup behavior. 2. In iterated dilemmas, the situation is not so clear, because others may retaliate for selfish behavior by defecting in subsequent rounds (Pruitt, 1998). Moreover, as Batson and his colleagues have shown (Batson, Ahmad, Yin, Bedell, Johnson, Templin & Whiteside, 1999; Batson, Batson, Todd, Brummett, Shaw & Aldeguer, 1995), self-interest and collective interest are not the only motives that can operate in social dilemmas. They found that empathy for particular group members reduces allocations that benefit the group as a whole. 3. If we omit the constraint that people increase the competitive advantage of whichever group they belong to, additional forms of loyalty and disloyalty are possible (Zdaniuk & Levine, 2001). Thus, if a person's presence is harmful to a group but staying is personally beneficial, then loyalty is demonstrated by leaving the group and disloyalty is demonstrated by staying.

ACKNOWLEDGMENTS Preparation of this chapter was supported by Army Research Institute Contract D A S W 0 1 - 0 0 - K - 0 0 1 8 . The views, opinions, and/or findings contained in this paper are those of the authors and should not be construed as an official Department of the A r m y position, policy, or decision. Thanks are extended to Leslie Hausmann for comments on an earlier draft.

REFERENCES Abrams, D., Ando, K., & Hinkle, S. (1998). Psychological attachment to the group: Cross-cultural differences in organizational identification and subjective norms as predictors of workers' turnover intentions. Personality and Social Psychology Bulletin, 24, 1027-1039. Abrams, D., Marques, J. M., Bown, N., & Henson, M. (2000). Pro-norm and anti-norm deviance within and between groups. Journal of Personality and Social Psychology, 78, 906-912.

224

J O H N M. L E V I N E A N D R I C H A R D L. M O R E L A N D

Asch, S. E. (1959). A perspective on social psychology. In: S. Koch (Ed.), Psychology: A Study of a Science (Vol. 3, pp. 363-383). New York: McGraw-Hill. Barker, R. G. (1968). Ecological psychology. Stanford, CA: Stanford University Press. Barreto, M., & Ellemers, N. (2000). You can't always do what you want: Social identity and selfpresentational determinants of the choice to work for a low-status group. Personality and Social Psychology Bulletin, 26, 891-906. Batson, C. D. (1998). Altruism and prosocial behavior. In: D. T. Gilbert, S. T. Fiske & G. Lindzey (Eds), The Handbook of Social Psychology (Vol. 2, pp. 282-316). Boston, MA: McGrawHill. Batson, C. D., Ahmad, N., Yin, J., Bedell, S. J., Johnson, J. W., Templin, C. M., & Whiteside, A. (1999). Two threats to the common good: Self-interested egoism and empathy-induced altruism. Personality and Social Psychology Bulletin, 25, 3-16. Batson, C. D., Batson, J. G., Todd, R. M., Brummett, B. H., Shaw, L. L., & Aldeguer, C. M. R. (1995). Empathy and the collective good: Caring for one of the others in a social dilemma. Journal of Personality and Social Psychology, 68, 619-631. Berkowitz, L., & Howard, R. C. (1959). Reactions to opinion deviates as affected by affiliation need (n) and group member interdependence. Sociometry, 22, 81-91. Bolino, M. C. (1999). Citizenship and impression management: Good soldiers or good actors? Academy of Management Review, 24, 82-98. Branscombe, N. R., & Ellemers, N. (1998). Coping with group-based discrimination: Individualistic vs. group-level strategies. In: J. K. Swim & C. Stangor (Eds), Prejudice: The Target's Perspective (pp. 243-266). San Diego, CA: Academic Press. Branscombe, N. R., Wann, D. L., Noel, J. G., & Coleman, J. (1993). In-group or out-group extremity: Importance of the threatened social identity. Personality and Social Psychology Bulletin, 19, 381-388. Brewer, M. B. (1991). The social self: On being the same and different at the same time. Personality and Social Psychology Bulletin, 17, 475--482. Brewer, M. B., & Brown, R. J. (1998). Intergroup relations. In: D. T. Gilbert, S. T. Fiske & G. Lindzey (Eds), The Handbook of Social Psychology (Vol. 2, pp. 554-594). Boston, MA: McGraw-Hill. Brewer, M. B., & Kramer, R. M. (1986). Choice behavior in social dilemmas: Effects of social identity, group size, and decision framing. Journal of Personality and Social Psychology, 50, 543-549. Brewer, M. B., & Silver, M. D. (2000). Group distinctiveness, social identification, and collective mobilization. In: S. Stryker, T. J. Owens & R. W. White (Eds), Self, Identity, and Social Movements (pp. 153-171). Minneapolis, MN: University of Minnesota. Burnstein, E., & Worchel, P. (1962). Arbitrariness of frustration and its consequences for aggression in a social situation. Journal of Personality, 30, 528-541. Charlton, K., & Bettencourt, B. A. (2001). In-group members' perceptions of a social mobility attempt: Effects of boundary permeability and outcome. Group Dynamics: Theory, Research, and Practice, 5, 81-91. Cini, M. Moreland, R. L., & Levine, J. M. (1993). Group staffing levels and responses to prospective and new members. Journal of Personality and Social Psychology, 65, 723-734. Clark, R. D. III (1999). Effect of number of majority defectors on minority influence. Group Dynamics: Theory, Research, and Practice, 3, 303-312. Coull, A., Yzerbyt, V. Y., Castano, E., Paladino, M.-P., & Leemans, V. (2001). Protecting the ingroup: Motivated allocation of cognitive resources in the presence of threatening ingroup members. Group Processes and Intergroup Relations, 4, 327-339.

Group Reactions to Loyalty and Disloyalty

225

Dawes, R. M., van de Kragt, A. J. C., & Orbell, J. M. (1988). Not me or thee but we: The importance of group identity in eliciting cooperation in dilemma situations: Experimental manipulations. Acta Psychologica, 68, 83-97. De Cremer, D., & van Dijk, E. (in press). Reactions to group success and failure as a function of identification level: A test of the goal-transformation hypothesis in social dilemmas. Journal

of Experimental Social Psychology. De Cremer, D., & van Vugt, M. (1999). Social identification effects in social dilemmas: A transformation of motives. European Journal of Social Psychology, 29, 871-893. Diehl, M. (1990). The minimal group paradigm: Theoretical explanations and empirical findings. In: W. Stroebe & M. Hewstone (Eds), European Review of Social Psychology (Vol. 1, pp. 263-292). Chichester, England: Wiley. Doosje, B., Ellemers, N., & Spears, R. (1999). Commitment and intergroup behaviour. In: N. Ellemers, R. Spears & B. Doosje (Eds), Social Identity: Context, Commitment, Content (pp. 84-106). Oxford, England: Blackwell. Dovidio, J. F. (1984). Helping behavior and altruism: An empirical and conceptual overview. In: L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 17, pp. 361-427). New York: Academic Press. Ellemers, N., Barreto, M., & Spears, R. (1999). Commitment and strategic responses to social context. In: N. Ellemers, R. Spears & B. Doosje (Eds), Social Identity: Context, Commitment, Content (pp. 127-146). Oxford, England: Blackwell. Ellemers, N., Kortekaas, P., & Ouwerkerk, J. W. (1999). Self-categorisation, commitment to the group and group self-esteem as related but distinct aspects of social identity. European Journal of Social Psychology, 29, 371-389. Ellemers, N., Spears, R., & Doosje, B. (1997). Sticking together or falling apart: In-group identification as a psychological determinant of group commitment vs. individual mobility. Journal of Personality and Social Psychology, 72, 617-626. Festinger, L. (1950). Informal social communication. Psychological Review, 57, 271-282. Gaertner, L., & Insko, C. A. (2000). Intergroup discrimination in the minimal group paradigm: Categorization, reciprocation, or fear? Journal of Personality and Social Psychology, 79, 77-94. Hensley, V., & Duval, S. (1976). Some perceptual determinants of perceived similarity, liking, and correctness. Journal of Personality and Social Psychology, 34, 159-168. Hertel, G., & Kerr, N. L. (2001). Priming in-group favoritism: The impact of normative scripts in the minimal group paradigm. Journal of Experimental Social Psychology, 37, 316-324. Hibbert, C. (1971). Tower of London. New York: Newsweek. Hogg, M. A. (2001). Social categorization, depersonalization, and group behavior. In: M. A. Hogg & R. S. Tindale (Eds), Blackwell Handbook of Social Psychology: Group Processes (pp. 56-85). Malden, MA: Blackwell. Insko, C. A., Thibaut, J. W., Moehle, D., Wilson, M., Diamond, W. D., Gilmore, R., Solomon, M. R., & Lipsitz, A. (1980). Social evolution and the emergence of leadership. Journal of Personality and Social Psychology, 39, 431-448. Israel, J. (1956). Self-evaluation and Rejection in Groups: Three Experimental Studies and a Conceptual Outline. Uppsala: Almqvist & Wiksell. Iwao, S. (1963). Internal vs. external criticism of group standards. Sociometry, 26, 410-421. James, K., & Cropanzano, R. (1994). Dispositional group loyalty and individual action for the benefit of an ingroup: Experimental and correlational evidence. Organizational Behavior and Human Decision Processes, 60, 179-205. Janis, I. L. (1982). Groupthink (2nd ed.). Boston, MA: Houghton Mifflin.

226

J O H N M. L E V I N E A N D R / C H A R D L. M O R E L A N D

Jetten, J., Spears, R., & Manstead, A. S. R. (1999). Group distinctiveness and intergroup discrimination. In: N. EUemers, R. Spears & B. Doosje (Eds), Social Identity: Context, Commitment, Content (pp. 107-126). Oxford, England: Blackwell. Jetten, J., Spears, R., & Manstead, A. S. R. (2001). Similarity as a source of differentiation: The role of group identification. European Journal of Social Psychology, 31, 621-640. Karau, S. J., & Williams, K. D. (2001). Understanding individual motivation in groups: The Collective Effort Model. In: M. Turner (Ed.), Groups at Work: Theory and Research (pp. 113-141). Mahwah, NJ: Erlbaum. Kelley, H. H. (1971). Attribution in Social Interaction. Morristown, NJ: General Learning Press. Kramer, R. M., & Brewer, M. B. (1984). Effects of group identity on resource use in a simulated commons dilemma. Journal of Personality and Social Psychology, 46, 1044-1057. Kramer, R. M., & Goldman, L. (1995). Helping the group or helping yourself?. Social motives and group identity in resource dilemmas. In: D. A. Schroeder (Ed.), Social Dilemmas: Perspectives on Individuals and Groups (pp. 49-67). Westport, CT: Praeger. Kruglanski, A. W., & Webster, D. M. (1991). Group members' reactions to opinion deviates and conformists at varying degrees of proximity to decision deadline and of environmental noise. Journal of Personality and Social Psychology, 61, 212-225. Lawler, E. J., Thye, S. R., & Yoon, J. (2000). Emotion and group cohesion in productive exchange. American Journal of Sociology, 106, 616-657. Levine, J. M. (1989). Reaction to opinion deviance in small groups. In: P. Paulus (Ed.), Psychology of Group Influence (2nd ed., pp. 187-231). Hillsdale, NJ: Erlbaum. Levine, J. M., & Higgins, E. T. (2001). Shared reality and social influence in groups and organizations. In: F. Butera & G. Mugny (Eds), Social Influence in Social Reality: Promoting Individual and Social Change (pp. 33-52). Gottingen: Hogrefe and Huber. Levine, J. M., Moreland, R. L., & Ryan, C. S. (1998). Group socialization and intergroup relations. In: C. Sedikides, J. Schopler, & C. A. lnsko (Eds), Intergroup Cognition and Intergroup Behavior (pp. 283-308). Mahwah, NJ: Erlbaum. Levine, J. M., & Ruback, R. B. (1980). Reaction to opinion deviance: Impact of a fence straddler's rationale on majority evaluation. Social Psychology Quarterly, 43, 73-81. Levine, J. M., Saxe, L., & Harris, H. J. (1976). Reaction to attitudinal deviance: Impact of deviate's direction and distance of movement. Sociometry, 39, 97-107. Levine, J. M., Sroka, K. R., & Snyder, H. N. (1977). Group support and reaction to stable and shifting agreement/disagreement. Sociometry, 40, 214-224. Levine, J. M., & Thompson, L. (1996). Conflict in groups. In: E. T Higgins & A. W. Kruglanski (Eds), Social Psychology: Handbook of Basic Principles (pp. 745-776). New York: Guilford. Mael, F. A., & Ashforth, B. E. (1995). Loyal from day one: Biodata, organizational identification, and turnover among newcomers. Personnel Psychology, 48, 309-333. Marques, J. M., & Paez, D. (1994). The "black sheep effect": Social categorization, rejection of ingroup deviates, and perception of group variability. In: W. Stroebe & M. Hewstone (Eds), European Review of Social Psychology (Vol. 5, pp. 37-68). Chichester, England: Wiley. Marques, J. M., Abrams, D., Paez, D., & Martinez-Taboada, C. (1998). The role of categorization and in-group norms in judgments of groups and their members. Journal of Personality and Social Psychology, 75, 976--988. Marques, J. M., Abrams, D., Paez, D., & Hogg, M. A. (2001). Social categorization, social identification, and rejection of deviant group members. In: M. A. Hogg & R. S. Tindale (Eds), Handbook of Social Psychology: Group Processes (pp. 400-424). Malden, MA: Blackwell.

Group Reactions to Loyalty and Disloyalty

227

Marques, J. M., Abrams, D., & Serodio, R. G. (2001). Being better by being right: Subjective group dynamics and derogation of in-group deviants when generic norms are undermined.

Journal of Personality and Social Psychology, 81,436-447. Mathieu, J. E., & Zajac, D. M. (1990). A review and meta-analysis of the antecedents, correlates, and consequences of organizational commitment. Psychological Bulletin, 108, 171-194. Meyer, J. P., & Allen, N. J. (1991). A three-component conceptualization of organizational commitment. Human Resources Management Review, 1, 61-89. Meyer, J. P., & Allen, N. J. (1997). Commitment in the Workplace: Theory, Research, and Application. Thousand Oaks, CA: Sage. Moreland, R. L., & Levine, J. M. (1982). Socialization in small groups: Temporal changes in individual-group relations. In: L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 15, pp. 137-192). New York: Academic Press. Moreland, R. L., & Levine, J. M. (in press). Socialization and trust in work group. Group Processes

and Intergroup Relations. Moreland, R. L., Levine, J. M., & Cini, M. A. (1993). Group socialization: The role of commitment. In: M. Hogg & D. Abrams (Eds), Group Motivation: Social Psychological Perspectives (pp. 105-129). London: Harvester-Wheatsheaf. Moreland, R. L., & McMinn, J. G. (1999). Gone but not forgotten: Loyalty and betrayal among ex-members of small groups. Personality and Social Psychology Bulletin, 25, 1476-1486. Morrow, P. C. (1993). The theory and measurement of work commitment. Greenwich, CT: JAI Press. Moscovici, S. (1976). Social influence and social change. New York: Academic Press. Moscovici, S., & Lage, E. (1976). Studies in social influence III: Majority vs. minority influence in a group. European Journal of Social Psychology, 6, 149-174. Orcutt, J. D. (1973). Societal reaction and the response to deviation in small groups. Social Forces, 52, 259-267. Organ, D. W., & Ryan, K. (1995). A meta-analytic review of attitudinal and dispositional predictors of organizational citizenship behavior. Personnel Psychology, 48, 775-802. Ouwerkerk, J. W., Ellemers, N., & de Gilder, D. (1999). Group commitment and individual effort in experimental and organizational contexts. In: N. Ellemers, R. Spears & B. Doosje (Eds), Social Identity: Context, Commitment, Content (pp. 185-204). Oxford, England: Blackwell. Ouwerkerk, J. W., de Gilder, D., & de Vries, N. K. (2000). When the going gets tough, the tough get going: Social identification and individual effort in intergroup competition. Personality and Social Psychology Bulletin, 26, 1550-1559. Prislin, R., Brewer, M. B., & Wilson, D. J. (in press). Changing majority and minority positions within a group vs. an aggregate. Personality and Social Psychology Bulletin. Prislin, R., Limbert, W. M., & Bauer, E. (2000). From majority to minority and vice versa: The asymmetrical effects of losing and gaining majority position within a group. Journal of Personality and Social Psychology, 79, 385-397. Pruitt, D. G. (1998). Social conflict. In: D. T. Gilbert, S. T. Fiske & G. Lindzey (Eds), The Handbook of Social Psychology (4th ed., Vol. 2, pp. 470-503). Boston: McGraw-Hill. Randall, D. M., Fedor, D. B., & Longnecker, C. O. (1990). The behavioral expression of organizational commitment. Journal of Vocational Behavior, 36, 210-224. Rusbult, C. E., & Farrell, D. (1983). A longitudinal test of the investment model: The impact on job satisfaction, job commitment, and turnover of variations in rewards, costs, alternatives, and investments. Journal of Applied Psychology, 68, 429438. Rusbult, C. E., Farrell, D., Rogers, G., & Mainous, A. G. (1988). Impact of exchange variables on exit, voice, loyalty, and neglect: An integrative model of responses to declining job satisfaction. Academy of Management Journal, 31, 599~527.

228

J O H N M. L E V I N E A N D R I C H A R D L. M O R E L A N D

Sani, F., & Reicher, S. (1998). When consensus fails: An analysis of the schism within the Italian Communist Party (1991). European Journal of Social Psychology, 28, 6234545. Sani, F., & Reicher, S. (1999). Identity, argument, and schism: Two longitudinal studies of the split in the Church of England over the ordination of women to the priesthood. Group Processes and Intergroup Relations, 2, 279-300. Sani, F., & Reicher, S. (2000). Contested identities and schisms in groups: Opposing the ordination of women as priests in the Church of England. British Journal of Social Psychology, 39, 95-112. Schachter, S. (1951). Deviation, rejection, and communication. Journal of Abnormal and Social Psychology, 46, 190-207. Scott, W. (1965). Values and Organizations: A Study of Fraternities and Sororities. Chicago: RandMcNally. Singer, J. E., Radloff, L. S., & Wark, D. M. (1963). Renegades, heretics, and changes in sentiment. Sociometry, 26, 178-189. Staub, E. (1997). Blind vs. constructive patriotism: Moving from embeddedness in the group to critical loyalty and action. In: D. Bar-Tal & E. Staub (Eds), Patriotism: In the Lives of Individuals and Nations (pp. 213-228). Chicago: Nelson-Hall. Stem, P. C. (1995). Why do people sacrifice for their nations? Political Psychology, 16, 217-235. Sumner, W. (1906). Folkways. New York: Ginn. Tajfel, H., & Turner, J. C. (1986). The Social Identity Theory of intergroup behavior. In S. Worchel & W. G. Austin (Eds), Psychology of Intergroup Relations (2nd ed., pp. 7-24). Chicago: Nelson-Hall. Tindale, R. S., Davis, J. H., Vollrath, D. A., Nagao, D. H., & Hinsz, V. B. (1990). Asymmetrical social influence in freely interacting groups: A test of three models. Journal of Personality and Social Psychology, 58, 438-449. Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. S. (1987). Rediscovering the Social Group: A Self-categorization Theory. Oxford, England: Blackwall. Tyler, T. R., & Blader, S. L. (2001). Identity and cooperative behavior in groups. Group Processes and Intergroup Relations, 4, 207-226. Wilder, D. A. (t986). Social categorization: Implications for creation and reduction of intergroup bias. In: L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 19, pp. 291-355). New York: Academic Press. Williams, K. D. (2001). Ostracism: The Power of Silence. New York: Guilford Press. Winslow, D. (1998). Misplaced loyalties: The role of military culture in the breakdown of discipline in peace operations. Canadian Review of Sociology and Anthropology, 35, 345-367. Worchel, S., Rothgerber, H., Day, E. A., Hart, D., & Butemeyer, J. (1998). Social identity and individual productivity within groups. British Journal of Social Psychology, 37, 389-413. Zdaniuk, B., & Levine, J. M. (2001). Group loyalty: Impact of members' identification and contributions. Journal of Experimental Social Psychology, 37, 502-509.

SEQUENCE AND SOLIDARITY Steven E. Clayman

ABSTRACT This paper develops a conversation analytic perspective on social solidarity, focusing on the organized practices through which solidary relations are maintained within interaction. Previous research on preference organization is reviewed and synthesized, and it is demonstrated that this robust mode of organization tends to suppress discordant actions while promoting solidary actions. The suppression of discordant actions involves practices that: (1) mitigate such actions, as well as; (2) minimize the likelihood of their occurrence. Conversely, solidary actions tend to be: (1) not mitigated; and (2) delivered in ways that maximize the likelihood of their occurrence.

INTRODUCTION This paper illuminates the phenomenon of social solidarity from a perspective that is not usually represented in theoretical discussions of this concept, namely that of conversation analysis (henceforth CA). It may at first seem that C A would have little to contribute to this area, given its concern with the fundamental organization of talk-in-interaction and its preference for naturalistic observation over formal theorizing. However, as Heritage (1984, pp. 265-280) has observed, a range of CA findings have clear implications for the problem of solidarity from the standpoint of how solidary relations are achieved and maintained within actual social situations in which persons are interactionally engaged (see also Brown & Levinson, 1987, pp. 38--41; Holtgraves, 1992). This Group Cohesion, Trust and Solidarity, Volume 19, pages 229-253. Copyright © 2002 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0898-2

229

230

STEVEN E. CLAYMAN

represents a distinctive contribution to the understanding of this core sociological phenomenon. Most approaches to the problem of solidarity are fundamentally motivational in emphasis. Those in the Durkheimian/Parsonian tradition focus on shared values and internalized norms of conduct, while rational choice theorists emphasize calculations of expected utility, but both are primarily concerned with the question of what motivates or drives individuals to act in a minimally conflictual/maximally cooperative manner. CA, in contrast, offers what might be termed a procedural approach to the problem of social solidarity. Here motivational theorizing is subsidiary to examining what interactional participants actually do - identifying and analyzing the practices they engage in to avoid conflict and promote solidary relations. A similar emphasis on the procedural foundations of solidarity can also be found in Brown and Levinson's (1987) study of the language of politeness. However, the verbal strategies identified by Brown and Levinson operate exclusively at the level of the design of individual actions or turns at talk, whereas CA studies incorporate a concern with the level of sequential organization. Here the focus is on the design of sequences of action and of turns-within-sequences as they bear on the maintenance of solidarity. In a variety of ways, interactants exploit these sequential arrangements as resources for suppressing actions that are uncooperative, disaffiliative, or otherwise discordant, while systematically promoting actions that are relatively harmonious. This paper will review a range of sequential phenomena that figure in this process. These phenomena have been analyzed under the rubric of preference organization, which is one of the most thoroughly researched areas within conversation analysis (Heritage, 1984, pp. 265-280; Schegloff, forthcoming) and refers to a conventionalized set of practices through which certain interactional outcomes are promoted or favored visa vis other outcomes. While not all manifestations of preference organization are relevant to the accomplishment of solidarity, a substantial subset does have this import and forms the focus of the present analysis. Extending the analysis developed by Heritage (1984, pp. 265-280), the objective here to provide an up-to-date synthesis of the findings in this area, draw out the implications for solidarity, and explore the scope and generality of this organization of human interaction. SEQUENTIAL

OUTCOMES

AND SOLIDARITY

While most actions produced within interaction are in some way consequential for what others may do in response, some actions are especially confining in that they establish the relevance of a narrow range of responses (Schegloff, forthcoming). Consider those sequence-initiating actions that call for a response

Sequence and Solidarity

231

in either of two diametrically opposed forms. Assessments (e.g., "The dinner was wonderful)" may be followed by either agreement or disagreement; invitations (e.g., "Let's have lunch on Thursday)" may be either accepted or declined; requests (e.g., "Pass the salt please)" may be either granted or refused; and so may offers, proposals, and the like. These altemative responses differ in their implications for relations between the interactants, with the first alternative (agreement or acceptance) generally supportive of solidary relations, while the second (disagreement or rejection) is discordant and threatening to solidarity (see Table 1). This point may seem obvious, but it is worth dwelling for a moment on how it is so. The response alternatives differ, first, in terms of their cooperativeness. By accepting a request or agreeing with an assessment (that is, by producing one of the responses in the middle column of the table), one advances the course of action that has been initiated and thus collaborates with the agenda being pursued by the prior speaker; rejections and disagreements (grouped in the right column of the table) are by contrast distinctly uncooperative. The responses also differ in terms of their affiliativeness. While acceptances and agreements may be taken to indicate that the respondent is aligned with and has some positive regard for the prior speaker, rejections and disagreements are experienced as disaligning and hostile. In short, both on dimensions of cooperativeness and affiliativeness, the response alternatives have substantially different ramifications for relations between the interactants. These ramifications can be understood in terms of different aspects of "face" identified by Erving Goffman (1967) and elaborated by Brown and Levinson (1987) as basic human desires characteristic of all competent adults. Within that framework, negative face refers to the desire to be free from imposition and to have one's autonomy and prerogatives honored and respected. Positive face refers to the desire to have an favorable self-image that is validated by others. Against this backdrop, the cooperativeness and affiliativeness dimensions discussed above bear directly on negative and positive face, respectively. Insofar as acceptances and agreements are cooperative, they respect the prior speaker's

Table 1. Initial Action

Solidary Response

Discordant Response

assessment invitation request offer proposal

agreement acceptance granting acceptance acceptance

disagreement declination refusal rejection rejection

232

STEVEN E. CLAYMAN

prerogatives and thus affirm that speaker's negative face. Insofar as acceptances and agreements are affilliative, they display positive regard for the prior speaker and thus affirm that speaker's positive face. Conversely, rejections and disagreements are threatening to both aspects of face. Given the diametrically opposed ramifications of different sequential outcomes, these sequences of action constitute an ideal environment in which to investigate the achievement of solidarity in interaction. Such sequences facilitate a comparative analysis of how, within a common environment of action (e.g., following an assessment, or a request, etc.), subsequent face-affirming and face-threatening actions are managed and dealt with. As it turns out, interactants do not treat these alternative response types equally; they tend to act in ways that systematically favor or promote the first type of response while disfavoring the second type of response. That is, they treat the first response type as preferred vis ?t vis the second. The term "preference" here has a technical meaning that differs from its vemacular usage. It refers not to the subjective feelings of the interactants, but to public forms of conduct that are recurrent and institutionalized, and that systematically favor certain interactional outcomes over others. The entire array of practices that differentiate among interactional outcomes is what is known as preference organization, and we shall examine these practices in turn. RESPONSIVE

ACTIONS:

ASYMMETRICAL

DESIGN

The altemative responses outlined above tend to be designed in systematically different ways. Agreements and acceptances ordinarily are produced promptly following the initiation and in a straighOeorwardmanner, whereas disagreements and rejections tend to be delayed and more elaborate or complex (Davidson, 1984, 1990; Pomerantz, 1984; Sacks, 1987; Wootton, 1981). To illustrate these differences, compare the following request sequences, the first involving acceptance and the second rejection. (A guide to the transcription symbols appears in the Appendix.) (1) [Davidson, 1990, p. 150] A: We:ll, will you help me [ou:t. B: [I certainly wi:ll. (2) [Wootton, 1981, p. 63] Ch: I want a'apple .hh two p.hh apples each (1.6) M: Half an apple each = I don't think we've got so many apples.

Sequence and Solidarity

233

In excerpt 1, B accepts A's request for help in a brief and unqualified manner, and the acceptance begins a bit early, overlapping the last word of the request. In contrast, the rejection in excerpt 2 is not only delayed by a 1.6 second silence, but it is built as a partial rejection, and is followed by an account that explains and justifies the rejection. A strikingly similar asymmetry may be seen in the following invitation sequences, involving an acceptance in excerpt 3 and a rejection in excerpt 4. (3) [Heritage, 1984, pp. 265-266] A: Why don't you come and se_~eme some[times B: [I would like to (4) [Heritage, 1984, p. 266] 1 A: Uh if you'd care to come over and visit 2 a little while this morning I'll give 3 you a cup of c___Qffee. 4 B: hhh Well that's awfully sweet of you, 5 I don't think I can make it this morning 6 .hh uhm I ' m running an ad in the paper 7 and- and uh I have to stay near the phone. Both sequences begin with an invitation to get together ("come and see me" in 3 and "come over and visit" in 4). Although the initial action is quite different from the first two examples - involving an invitation rather than a request the responses exhibit a similar asymmetrical pattern. The acceptance in 3 is prompt and straightforward, while the rejection in 4 is delayed by an audible outbreath and an initial expression of appreciation (line 4), such that the rejection proper does not begin until line 5. When the rejection is finally delivered, it is qualified in various ways - marked with pro forma uncertainty ("I don't think)" and as conditional on the time of day ("this morning)". Finally, the rejection is followed by an account (lines 6-7) that provides an explicit rationale for the rejection. In general, following the framework elaborated by Heritage (1984, pp. 266-267) and Levinson (1983, pp. 334-335), dispreferred responses such as disagreements and rejections tend to include the following features: • Delays: silences preceding the delivery of the response; prefaces of various

kinds; and insertion sequences which displace the response over a series of turns.

234

STEVEN E. CLAYMAN

• Prefaces: discourse markers such as "uh" or "well"; token agreements, appreciations, and apologies; and other forms of hesitation. • Accounts: explanations for the disagreement or rejection. • Disagreement/rejection is itself mitigated: marked as uncertain, conditional, or indirect.

Asymmetries of this kind are by no means limited to distinctions between sequential responses, nor do they appear only for acts that are clearly affiliative/ disaffiliative (cf., Sacks & Schegloff, 1979; Schegloff, 1979; Schegloff, Jefferson & Sacks, 1977). Nevertheless, the relevance of preference organization for relational solidarity is highlighted by the fact that, when affiliation/disaffiliation is involved, the same asymmetrical design features recur across a wide variety of a sequence types (e.g. responses to assessments, requests, invitations, offers, proposals, and so on). Furthermore, in particular environments where the polarity of affiliation/disaffiliation is reversed, the pattern of asymmetry may be reversed as well. Although agreements are generally prompt and straightforward while disagreements tend to be delayed, mitigated, and accountable, agreeing/ disagreeing responses to accusations (Atkinson & Drew, 1979) and selfdeprecations (Pomerantz, 1984) have precisely the opposite design features. Here, where an agreement would be face-threatening, it is treated as the dispreferred alternative. The association between the design of the action and its and relational function is thus quite robust. But how do these asymmetries actually promote solidarity? Mitigating Discordant Responses

Each of the design features associated with discordant responses work to soften or mitigate the force of that action. Consider first the phenomenon of delay. The generic practice of delaying a disagreement or rejection may be taken to indicate a modicum of "reluctance" to engage in that action. Of course, such delays may in fact be generated by any number of factors, but in this environment - just before some form of rejection - they are hearable as indicating a certain lack of enthusiasm for the action in progress (cf. Lerner, 2001). This plainly softens the threat to face embodied in the rejection. Moreover, when the response is delayed by prefatory talk, the preface frequently involves a face-affirming display of "positive politeness" (Brown & Levinson, 1987). For example, when invitations are rejected, the rejection is often prefaced by an expression of appreciation to the prior speaker for having extended the invitation (see Example 4 above). Correspondingly, disagreements are often prefaced by token agreements (e.g. "Yes that's true, b u t . . . " ) . The

Sequence and Solidarity

235

import of such prefaces is that, insofar as they are face-affirming, they counterbalance the ensuing face-threatening action and by being produced first are given temporal priority over it. When accounts are offered to explain why a request, invitation, etc., is being rejected, such accounts overwhelmingly involve reference to circumstantial factors that prevent the speaker from accepting. Speakers thus present themselves as unable, rather than unwilling, to accept (Heritage, 1984, pp. 269-273). In Example 4 above, an invitation to get together is rejected on the grounds that the respondent has a newspaper ad running and so she "has to stay near the phone" (see also Example 2). Thus, out of all the possible reasons that could in principle explain a rejection, speakers select "no fault" accounts which propose that the rejection should not be taken as a sign of disregard toward the prior speaker. Finally, insofar as the disagreement/rejection is itself mitigated, it is plainly reduced in force. In Example 2 above, the mother only partially rejects her child's request for apples, and in Example 4 the invitation to get together is marked as slightly uncertain ("I don't t h i n k . . . " ) and conditional on the time of day ("this morning)". These various practices combine to mitigate the threat to face and solidary relations embodied in disagreements and rejections. Conversely, the absence of such features in agreements and acceptances means that these solidary actions are produced with greater forcefulness and impact. Avoiding Discordant Responses One particular design feature - delayed positioning - not only blunts the force of an uncooperative/disaffiliative response, but also facilitates the avoidance of such responses altogether. To see how this is so, consider that the asymmetries of preference organization are commonplace and institutionalized, so that interactants, having been exposed to these patterns since birth, have at least a tacit grasp of them. On this basis, interactants can anticipate the type of response that is forthcoming purely on the basis of its initial form. Thus, following the completion of an assessment, request, etc., any delay in responding - even mere silence - may be interpreted as the first move toward some form of disagreement/rejection. This pattern of inference is illustrated in the following excerpt from a telephone conversation, in which C's invitation (line 1) is followed by a 0.4 second silence (line 2). (5) [Davidson, 1984, p. 105 (simplified)] 1 C: Well you can both sta:y.

236 2 3

STEVEN E. CLAYMAN

C:

(0.4) Got plenty a' roo:m,

C deals with the silence (line 3) in a way that reveals her understanding of what it portends. She counters a possible reason for rejecting the invitation (i.e. insufficient room to accommodate houseguests) by reassuring her recipient that this is not a problem. In so doing, she shows that she hears the antecedent silence as indeed rejection-implicative, as foreshadowing a trouble or problem with the invitation that she then seeks to address. A similar pattern may be observed in the following - an initial invitation (line 1) receives only silence (line 2), which the recipient treats as rejection-implicative (line 3). (6) [Davidson, 1984, p. 105] A: C ' m o n down he:re, = it's oka:y, (0.2) A: I got lotta stuff, = I got bee:r en stuff 'n In short, delay in this sequential environment can be an early harbinger of interactional discord. Furthermore, such delay provides the initiator of the sequence with what is, in effect, an opportunity space that can be used to forestall the incipient rejection/disagreement. The first speaker can, upon hearing a rejection-implicative delay, withdraw the initial action or revise it so as to make it more acceptable. In Examples 5 and 6 above, the party that extended the invitation subsequently elaborates it in such a way as to entice a reluctant recipient and thus push for acceptance. Similarly, the author of an assessment may, upon hearing a disagreement-implicative silence, back down from the assessment in a way that facilitates agreement: (7) 1 2 3 4

[Pomerantz, 1984, p. 77] B: . . . an' that's not an awful lotta fruitcake (1.0) B: Course it is. A little piece goes a long way. C: Well that's right.

Here a pessimistic assessment of the quantity of available fruitcake (line 1) is followed by a full one-second silence (line 2). This silence both foreshadows disagreement and provides the first speaker with an opportunity to revise her assessment so as to render it more optimistic (line 3). Consequently, what might have led to an expression of disagreement results instead in agreement (line 4).

Sequence and Solidarity

237

In a variation on this theme of delay as providing an opportunity space, the initiator of the sequence may also respond to delay by "guessing" or collaboratively completing the rejection, in effect articulating much of the rejection before the recipient has a chance to do so (Lerner, 1996). Such a maneuver is disarming in its import. For instance, consider this telephone call in which Jennifer calls her friend's mother - Molly - and asks to rent the latter's garage (lines 1-5). Molly initially displays some confusion as to what is being requested, asking if the purpose is to actually live in her garage (line 9). When Jennifer confirms this (line 10), Molly hesitates before responding to the now-clarified request. She first pauses for 0.3 seconds (line 11), takes an audible inbreath, and produces just the beginning of a response ("It's j u s t . . . " at line 12), but then trails off. After allowing a full 1.0 second silence to elapse, Jennifer interjects (line 14) in a way that collaboratively completes the core component of the rejection (adding a tag question to mark it as a guess): "not possible huh." (8) 1 2 3 4 5 6 7 8 9 10

[Jennifer & Molly, p. 2] Jennifer: I was just wondering y'know.hhh (0.3) could (.) d'you think you might (.) wanna rent (.) you know like the bottom part a yer: (.) g'ra::ge like to me: fer a whi:le, a sump'm like that. (0.3) Molly: Wul[1Jennifer: [I think [( Molly: [Oh- you mean for] living in: Jennifer? Jennifer: Ye:ah

11 12 Molly: 13 14 Jennifer: 15 Molly: 16

(0.3) .hh It's just (1.0) no:t possible.--h[uh [Ye:ah we- Tina tri:ed that one time...

Here the sequence does contain a rejection of sorts, but that action has been fundamentally transformed and in effect detoxified. The rejection is articulated, not by the recipient, but by the initiator of the sequence herself. This in turn enables the recipient to deliver a subsequent response that is laminated as confirmatory and affiliative (beginning with "yeah" at line 15).

238

STEVEN E. CLAYMAN

Thus, beyond softening the blow of a discordant response, delay also provides a systematic opportunity for its avoidance. Conversely, because solidary actions are done promptly, they are not easily forestalled. This structural asymmetry has clear implications for the aggregate realization of response types over time. The promptness of solidary responses systematically m a x i m i z e s the likelihood of their occurrence, while the tendency to delay discordant responses m i n i m i z e s the likelihood of their occurrence (Pomerantz, 1984). What might first have seemed like a small and insignificant detail concerning the precise timing of alternative responses turns out to have far-reaching ramifications for the tenor of interactants' relations with one another.

INITIATING ACTIONS: PRE-SEQUENCES AND OTHER PRELIMINARIES Just as responses are geared to promote solidary relations, initiating actions are as well. Central to this process, once again, is the constellation of practices that lead up to a focal action. One such practice is the p r e - s e q u e n c e , ~ which contingently precedes and foreshadows actions (e.g. requests, invitations, etc.) that initiate a c o r e o r b a s e s e q u e n c e . Consider the following pre-invitation sequence. (5) [Terasaki, 1976, p. 35] A: Whatcha doin' B: Nothin'. A: Wanna drink? Here the first speaker asks a question that is obviously not seeking information for its own sake, but is preliminary to some as-yet-unrealized action. Moreover, this preliminary character is recognized by the second speaker, who does not provide a literal answer to the question, but a "go ahead" response that prompts the first speaker to proceed to the main business at hand, which turns out to be an invitation. Pre-sequences do not seem to be as commonplace as the asymmetrical response patterns discussed above - speakers sometimes choose to launch into an invitation or request "out of the blue," without such preparatory work. Nevertheless, when pre-sequences are employed, they are similarly geared to the minimization of discordant reponses in the base sequence. Moreover, as we shall see, pre-sequences also provide for the avoidance of the base sequence altogether. This is significant because the entire sequence can strain the interactants' relationship - not only does the initiator face possible rejection,

Sequence and Solidarity

239

but the respondent is to some degree imposed upon (Brown & Levinson, 1987). For simplicity, the ensuing discussion will focus mainly on preliminaries to base sequences involving requests and invitations, although other pre-sequences and other preliminary practices will also be considered.

Testing the Waters Pre-sequences often begin with a question that makes reference to some matter that might affect the outcome of the base sequence, namely acceptance or rejection. Most often, these matters concern the same sort of circumstantial factors that are typically used in the accounts that accompany rejections (Heritage, 1984, pp. 278-279; Levinson, 1983, pp. 357-358). Thus, preinvitations (like extract (5) above) often ask whether the respondent is busy or occupied; pre-requests often ask if the respondent has or is using the requested item; and so on. 2 These preliminary questions can thus be seen as a way of "testing the waters" to determine in advance if the request or invitation will be accepted. If the preliminary receives a favorable response, the speaker ordinarily proceeds to the invitation or request, as in extract (5) above. Moreover, the speaker can do so with some confidence that, having ruled out the most common grounds for rejection, an acceptance will probably be forthcoming. On the other hand, if the preliminary receives an unfavorable response, the invitation or request is generally abandoned, along with the base sequence that would otherwise follow. For example: (6) [Sacks, 1992, p. 686] Jack: How ya doin'. Say, what're ya doin'. Judy: Well, we're going out. Why. Jack: Oh, I was just gonna say come out an' come over here and talk to the people. But if you're going out you can't very well do that. In response to Jack's pre-invitation query, Judy indicates that she will be unavailable. Jack then abandons the invitation, mentioning it only as a prior intention which has now been relinquished. This type of outcome - abandonment of the base sequence altogether - is advantageous for both parties and relations between them. Not only does the respondent avoid having to reject the initiator outright, but the initiator avoids imposing upon the respondent with an unwanted invitation, these goals having been accomplished under the guise of merely (on the surface at least) "seeking

240

STEVEN E. CLAYMAN

and providing information." Even respondents who are in fact available for the social event in question, but who simply don't wish to be involved, may be drawn to provide a forestalling response. Confronted with a pre-invitation query, such a respondent may well choose to fabricate a face-saving account (e.g. "I already have plans)" rather than tell the truth and then be forced to resort to a less benign account when the actual invitation is extended (Heritage, 1984, pp. 278-279). Thus, there are clear incentives for both parties to make use of this sequence as a way of forestalling an exchange that could strain their relationship. Pre-sequences and related practices can be exploited in more subtle ways to facilitate a solidary outcome. An initially forestalling response can be probed (e.g. "How long will you be out?)" in pursuit of material that is more conducive to subsequent acceptance (Maynard, 1989, 1991). The design of the base sequence initiation may be subtlely adjusted in light of the particulars of the pre-sequence response so as to make it more acceptable (Maynard, 1992). Finally, the initial utterance may be designedly opaque as to its preliminary character, so that any exploratory "testing of the waters" occurs off-the-record and is not accountable as such (Drew, 1984).

Reversing the Initiative We have seen that pre-sequences are exploited to increase the likelihood that the base sequence will have an solidary outcome. Beyond this, pre-sequence initiations can engender another kind of interactional trajectory that also enhances solidarity. The recipient of a pre- may choose to take the initiative and do, or offer to do, what the pre- foreshadows. For instance, following a pre-request, a recipient may offer the requested item without waiting for the actual request to be delivered: (7) [Merritt, 1976, pp. 324-325] C: Do you have the pecan Danish today? S: Yes we do. Would you like one of those? C: Yes, please. Here, what could have required four turns is reduced to three, there is a reversal of interactional initiative for the core action, and this action is transformed from a projected request to an offer. This is a recurrent pattern in pre-request sequences, and it amounts to another order of preference operating at the level of sequence-initial actions - namely, a systematic preference for offers over requests (Sacks, 1992, pp. 685-692).

Sequence and Solidarity

241

The realization of this preference plainly enhances solidarity. Because offers involve the transfer of a good or service from speaker to recipient rather than the reverse, they are generally less imposing on the recipient than requests. Moreover, an offer in this environment - after a pre-request rather than "out of the blue" - plainly has an increased likelihood of being accepted. An even stronger reversal of initiative, resulting an even more compact trajectory, occurs when the recipient of a pre-request not only offers the requested item, but actually provides it then and there. (8) [Merritt, 1976, p. 339] C: Do you have coffee to go? S: Cream and sugar? ((starts to pour)) In this case, the pre- is treated as standing on behalf of the request itself, and becomes functionally equivalent to a conventionally indirect - and hence less imposing - request (Brown & Levinson, 1987). 3 Reversals of initiative also occur in invitation sequences, with preliminaries to invitations prompting self-invitations and proposals from recipients (Drew, 1984). Once again, an affiliative outcome is extremely likely, and is achieved without the first speaker having to impose on the recipient with a formal invitation. Finally, preliminaries to bad news announcements can enable the recipient to guess what the news is (Schegloff, 1988; Maynard, 1996). Here the relational implications are somewhat different. In guessing correctly, the second speaker shows him- or herself to be "on the same wavelength" as the first speaker, and partially relieves that speaker of what is often a difficult and burdensome task. It should now be clear that pre-sequences and related practices systematically favor solidary outcomes, and they do so in at least two basic ways. They enable anyone contemplating a request, invitation, etc. to determine in advance whether it is likely to be accepted, and then proceed with the projected course of action only if the prospects for acceptance are good. Pre-sequences also facilitate reversals of initiative by creating an environment where that becomes a safe and ultimately less imposing course of action.

LEVELS OF PREFERENCE: THE CASE OF QUESTION-ANSWER SEQUENCES Having reviewed similar forms of preference organization operating across a wide range of action sequences, we turn now to consider one ubiquitous type of sequence in greater detail: namely, the question-answer sequence. Here,

242

STEVEN E. CLAYMAN

although the ramifications for solidarity are less prominent (the questioner's positive face often is not at issue) such ramifications remain in terms of the questioner's negative face (i.e. the desire to have one's prerogatives respected). Thus, responses to questions may cooperate fully with the agenda established by the question and thus honor the questioner's prerogatives in launching the query; or responses may in varying degrees resist the questioner's agenda. These response alternatives are ordered by structures of preference operating at varying levels of scale.

Preference for Answers Over Non-Answers At the most macro level is a preference that operates on the gross alternatives of answering vs. not answering. This preference favors answers over non-answers, and it is manifest in ways that parallel some of the practices outlined above. While answers are non-accountable - respondents rarely explain why they are answering a question - non-answers typically are explained and justified (Heritage, 1984, pp. 249-251; Clayman, 2001). For example,while M does not provide the information requested by J's query, she furnishes a warrant for her failure to do so: (9) [Heritage, 1984, p. 250] J: But the trai:n goes. Does th'train go o:n th'boa:t? M: .h.h Ooh I've no idea:. She ha:sn't sai:d. Moreover, the accounts offered in this context tend once again to be no-fault accounts that are minimally face-threatening to the questioner. While many accounts could in principle be offered for a refusal to answer, speakers tend to select accounts that cast themselves as unable rather than unwilling to answer, thereby implicitly validating the appropriateness of the question. In the previous example, the query concerns a third party's travel arrangements. M first declines to answer J's query on the grounds that she doesn't know the answer ("I've no idea)", and this lack of knowledge is in tum accounted for by reference to the fact that the traveler "hasn't said." She thus presents herself as willing in principle to cooperate with the agenda of the question - thereby validating the question as not inappropriate or overly intrusive - but unable to do so due to circumstances beyond her control. Even in institutional environments characterized by pervasively adversarial relations between participants, a similar response pattern is maintained. In news interviews involving journalists and public figures, refusals to answer are

Sequence and Solidarity

243

generally warranted by reference to exogenous circumstances (e.g. insufficient time, the delicacy of ongoing negotiations, etc.) rather than by attacking the legitimacy of the question or the motives of the questioner (Clayman, 2001). Such accounting practices, when used in the context of what might otherwise be a face-threatening refusal, are plainly disarming in their import.

Preference for a Particular Type of Answer One step below the preference for answers over non-answers is a preference that operates on the distinction between different types or categories of answer. While there is no generic preference for answers of a given type, questions can be designed in such a way as to favor one type of answer vis a vis its alternatives. This is most easily seen for the case of yes/no questions, which can be polarized or "tilted" so as to invite either an affirmative or negative response (Heritage, forthcoming a, forthcoming b). Consider the following question, which was put to President Reagan during a press conference. After some prefatory remarks (lines 1-9), the journalist asks (at lines 10-11) whether it's time for Reagan to take "strong a c t i o n . . , to get interest rates down." (10) [Reagan, 19 Oct 1983, p. 28] 1 JRN: Mr. President, new figures out today show that 2 housing starts were down pretty sharply last month, 3 and the number of building permits went down for 4 the second month in a row. Analysts are saying 5 this could mean the economic recovery is going to 6 level off, maybe kind of peter out next year. 7 And more people are becoming concerned about 8 high interest rates. And given the big deficits 9 being projected by your own administration, 10 isn't it time for some strong action by you to 11 get interest rates down? This question strongly invites an affirmative answer, and it illustrates two distinct resources through which a preference of this sort can be encoded within a yes/no question. Consider, first, the prefatory remarks. The journalist leads up to the question by reporting a recent drop in housing starts and building permits (lines 1-4), noting that this may be a drag on the economy more generally (lines 4-6), and that interest rates are becoming a matter of widespread concern in light of projected budget deficits (lines 7-9). This prefatory material, taken

244

STEVEN E. CLAYMAN

together, implies that there is indeed a compelling need to reduce interest rates and thus favors an affirmative answer to the question. A similar preference is encoded in the linguistic form of the question itself (line 10). Notice that the question is negatively formulated - instead of asking "is it time for some strong action," the journalist asks "isn't it time . . . . "Heritage (forthcoming b) has demonstrated that negative interrogatives strongly invite an affirmative answer. Other linguistic forms for designing questions also embody a pronounced tilt (eg., tag questions and b-event statements; see Heritage forthcoming a), but negative interrogatives do so in the strongest way. Indeed, such questions are often treated as if they were asserting an opinion rather than merely asking a question (Heritage, forthcoming b). For instance, President Clinton's initial response ("I disagree with that)" to a negative interrogative (line 3) plainly treats it as an assertion to be disagreed with, rather than a question to be answered. (11) [Clinton, 7 March 1997, Simplified] Well Mister President in your ze___aa:l(.) for 1 IR: 2 funds during the last campaign.hh 3 didn't you put the Vice President (.) and Maggie and all the others in your (0.4) administration 4 top side in a very vulnerable position, hh 5 6 (0.5) I disagree with that. hh u- How are we vulnerable 7 BC: 8 because... As Heritage (forthcoming a) has observed, for interviewees who choose to resist the tilt of the question and answer in some other way, strongly polarized questions establish a higher threshold of accountability. Interviewees thus "find themselves responding in a more defensive or self-justifying way than might otherwise be the case."

Preference for Particular Lexical Items At a still finer level of detail is a preference that bears on the specific lexical items contained in the response. Yes/no interrogatives embody a systematic preference for answers that actually contain either yes or no, or a roughly equivalent lexical token, within them (Raymond, 2000). This preference for what Raymond has termed "conforming answers" over "non-conforming answers" is apparent in a variety of ways, including the fact that responses

Sequence and Solidarity

245

lacking either yes or no tend to be infrequent and are specifically warranted by the speaker in the course of responding. Consider the following pair of exchanges, both taken from interactions between health visitors - British medical professionals who routinely visit homes with newborn children - and new mothers. (12) [HV, 5A1] HV: How about your breast(s) have they settled do:wn [no:w. M: [Yeah they 'ave no:w yeah. (13) [HV, 1Cll HV: Are your breasts alright. (0.7) M: They're fi:ne no:w I ' v e stopped leaking (.) so: Both questions concern the condition of the new mother's breasts, and both are designed as yes/no questions, but they are responded to rather differently. While the first question receives a conforming response in the affirmative (note the tum-initial "yeah)", the second question receives a broadly affirmative but nonetheless non-conforming response. The differential uptake appears to be a consequence of the somewhat different manner in which the prior questions are formulated and the presuppositions that they embody as a consequence. While the first question inquires into the changing condition of the mother's breasts ("have they settled down now)" and thus presupposes some prior difficulties, the second references only the current state of the mother's breasts ("are your breasts alright)" and thus does not presuppose any prior difficulties. In the latter case, a simple yes-type conforming response could foster the mistaken impression that the mother has had no problem whatsoever with her breasts. To counteract the presuppositional loading of the latter question, the mother offers a non-conforming response that portrays a trajectory of improvement and thus alludes to earlier difficulties. As Raymond (2000) has demonstrated, non-conforming responses embody a subtle but very real challenge to the "definition of the situation" that is engaged or presupposed by the question. Responses of this sort are not entered into casually or haphazardly; they are done only "for cause," and they are designed in such a way as to specifically warrant or justify the departure from conformity. In summary, questions exert pressure for a response that takes a specific and identifiable form, and this pressure can occur at varying levels of scale - not only for answers over non-answers, but also for answers that perform a

246

STEVEN E. CLAYMAN

particular action (i.e. affirmation) over alternative actions, and even for answers that contain particular lexical items (i.e. yes or no) over those that do not. While the first order of preference is engaged by the production of a question per se, the second and third orders of preference are engaged by the specific manner in which the question is formulated and in particular the lexical and grammatical resources employed within it. Questions thus initiate a course of action with multifaceted constraints on what should be done in response. Correspondingly, responses that resist one or more of these constraints - and are thus uncooperative and threatening to solidarity - are treated as accountable and tend to be explicitly or implicitly justified, sometimes in ways that validate the legitimacy of the question and the manner in which it was framed. Uncooperative responses thus tend to be managed in such a way as to preserve solidary relations between the interactants.

GENERALITY The various practices underlying preference organization are recurrent and deeply institutionalized features of interactional conduct. A wide range of evidence supports this conclusion. Consider, first, the fact that strikingly similar practices are found across a large number of diverse sequence types, including assessments (Pomerantz, 1984), invitations (Drew, 1984), requests, offers, and proposals (Davidson, 1984, 1990). Moeover, such patterns do not appear to be particularly sensitive to status differences among interactants. Thus, when parents reject the requests of their two year old children, they tend to employ the same mitigating features as adults do when rejecting one another (although childrens' subsequent persistence tends to elicit more forceful rejections from parents; see Wootton, 1981). Preference structures are not only robust; they are also robustly linked to solidary outcomes. As noted earlier, in particular environments where the implications for solidarity are reversed, the organization of preference may be reversed as well. Thus, while assessments generally prefer agreement over disagreement, when the assessment involves a self-deprecation, disagreement becomes the face-affirming response and it is treated as preferred over agreement (Pomerantz, 1984). Similarly, following a complaint or accusation, denial of wrongdoing - a form of disagreement - becomes the solidary response and is generally preferred over an admission (Atkinson & Drew, 1979; Garcia, 1991), although there appear to be additional complexities in accusatory contexts (Dersley & Wootton, 2000). 4 Of course, disagreeing and rejecting responses become more blunt and unmitigated in the context of an ongoing argument (Goodwin, 1983, 1990; Kotthoff,

Sequence and Solidarity

247

1993). Some have suggested that preference structures are actually reversed in argumentative contexts (Kotthoff, 1993). However, given that arguments tend to be bounded and relatively short-lived episodes within interaction, it may be more accurate to say that an episode is constituted as an argumentative departure from "normal" interaction by breaching the institutionalized asymmetries of preference organization (cf. Dersley & Wootton, 2000). Similar patterns have been found beyond the domain of ordinary conversation, although with some variations. Preference structures are sensitive to specialized contingencies found in some institutional contexts. Thus, broadcast news interviewers cannot agree with their guests without departing from the turn taking system for news interviews (Heritage & Greatbatch, 1991) and undermining the neutralistic posture that they are professionally obligated to maintain (Clayman, 1988, 1992), and interviewer silence is not taken as a harbinger of disagreement (see also Atkinson, 1992 on arbitrators in small claims court). Similarly, in various gatekeeping environments - such as intake interviews at people-processing organizations - expressions of agreement may implicate a gatekeeping decision and may thus be suppressed (Lazarton, 1997). Granting these variations, preference structures strikingly similar to those in ordinary conversation have been documented in contexts ranging from courts of law (Atkinson & Drew, 1979; Garcia, 1991) to scientific laboratories and conferences (Jacoby, 1998; Lynch, 1985). Such patterns can also be found in public speaking contexts in which a single speaker interacts with a mass audience. Collective episodes of applause and booing have asymmetrical design features that parallel those of agreement/ disagreement, acceptance/rejection, and so on (Atkinson, 1984; Clayman, 1993; Heritage & Greatbatch, 1986). Farther afield, preference structures have even been found in written discourse - in the letters written by scientists to one another, agreements and disagreements are similarly asymmetrical (Mulkay, 1985). The cultural boundaries of preference organization are somewhat more difficult to assess. Research has thus far been limited mainly to data drawn from the United States and Britain. However, no substantial differences have been detected between these two cultures. Moreover, the few studies that probe beyond the Anglo-American context find substantial similarities. Thus, in a study of Swedish conversations, Lindstrom (1997) demonstrates that a particular discourse marker - a distinctively-intoned ja - functions as a common preface to rejection, hearably projects rejection, and facilitates avoidance of that action. All of this closely parallels the organization of rejecting responses among English speakers. More cross-language and cross-cultural research is needed to ascertain the true scope of preference organization and to identify possible adaptations to varying linguistic and cultural contexts.

248

STEVEN E. CLAYMAN

The form of organization described in this paper does not appear to be restricted to interactions involving human beings. Behavior patterns formally similar to at least some aspects of preference organization have long been described in studies of animal communication. Ethologists and behavioral ecologists have noted the tendency, across a wide range of species, for the behavioral antecedents of aggressive acts to become ritualized and/or conventionalized in ways that render them more conspicuous and hence effective at conveying in advance - whether truthfully or not - an intention to attack (eg., Daanje, 1950; Moynihan, 1955; Smith, 1977, Chap. 11). Because animals are attentive to even quite subtle "intention movements" in others and adjust their behavior accordingly (Davis, 1975), such movements have clear implications for the avoidance of conflict (Maynard Smith, 1974; Maynard Smith & Price, 1973). Since physical aggression is intrinsically risky to both aggressors and their opponents - injury is always a possibility - ritualized or conventionalized pre-aggressive behaviors reduce this risk by first intimidating the opponent and thus increasing the likelihood of an appeasing response that forestalls further aggression. Displays of threat and conciliation thus stand on behalf of outright conflict and provide for its avoidance. When potenial combatants are conspecifics and members of the same social group, there are further ramifications for group cohesion. Such behavior patterns represent a generic and effective solution to a fundamental problem confronted by any social group: how to maintain the cohesion and persistence of the group and its constituent social relationships in the face of potentially conflictual encounters between group members. Since even mildly competitive exchanges have the potential to escalate into relationally damaging and at times deadly conflict (Luckenbill, 1977), the integrity and persistence of the group is always at risk. Given that the problem is so widespread, it is perhaps not surprising to find a correspondingly widespread solution in parallel organizations of interaction across social species) CONCLUSION Preference organization does not offer a theory of the motivations that drive interactants to act in a solidary manner. Nevertheless, it does embody, in its own way, a theory of solidarity, albeit one whose frame of reference lies squarely within interaction itself. It offers an empirically based account of the organizational features of interaction that systematically promote solidary actions while suppressing discordant ones. It describes the asymmetrical properties of cooperative/affiliative actions vs. uncoooperative/disaffiliative actions, demonstrates that these asymmetries are produced and oriented to by

Sequence and Solidarity

249

interactants themselves, and shows how interactants exploit these asymmetries in a way that maximizes the occurrence of the former type of action relative to the latter. In short, it is a mode of social organization that operates within interaction and is implicated in the real-time process by which solidary relations are achieved and maintained. For those seeking a further explanation of the existence of preference organization itself, a rational choice framework may prove useful, insofar as it leads to consideration of the expected utility of engaging in the practices described in this paper. Indeed, the initial origins of preference organization may be explicable in terms of varying incentives associated with relationally consequential courses of action. However, a purely instrumentalist account will ultimately be insufficient, for it overlooks the extent to which the practices associated with preference organization are thoroughly institutionalized and in some respects normative in character. Indeed, the effective implementation of these practices would seem to require such institutionalization. The practices in question are finely coordinated, with interactants inferring one another's intentions on the basis of extremely minimal fragments of behavior - often just the initial trace of an action in progress - and adjusting their own subsequent behaviors accordingly. None of this would be possible without a well-established framework of interactional conventions on which to base such inferences. Accordingly, preference organization is not properly understood as the aggregate result of an array of individual calculated decisions; its asymmetries are institutionalized methods of interacting that are oriented to by interactants themselves and are centrally implicated in the process by which interactants produce these selfsame asymmetries and deploy them in pursuit of solidarity.

NOTES 1. Studies of pre-sequences and allied practices include Atkinson and Drew (1979), Sacks (1974, 1992, pp. 685-692), Schegloff (1980, 1988), and Terasaki (1976). Pre-sequences have also been analyzed in relation to conversational openings (Schegloff, 1968) and closings (Schegloff & Sacks, 1973). For other sequence-preliminary practices, see Drew (1984) and Maynard (1989, 1991, 1992, 1996). 2. In a somewhat different vein, pre-announcements often seek to determine whether the recipient has heard some items of news. 3. Several analysts have noted the close relationship between pre-sequences and conventional indirectness (Heringer, 1997; Levinson, 1983, pp. 356-364; Schegloff, 1988). 4. Other departures from the usual pettern for agreement/disagreement have been documented for responses to compliments (Pomerantz, 1978), responses attempts to huess bad news (Schegloff, 1988), and responses in argumantative environments (Goodwin, 1983, 1990, Chap. 7; Kotthoff, 1993).

250

STEVEN E. CLAYMAN

5. Such parallels do not necessarily mean that preference organization among humans is neurologically based or that it necessarily eveolved from antecedents among evolutionary ancestors (although this is a possibility). My agnosticism is signalled by my use of both the biological concept of "ritualization" and the sociological concept of "conventionalization" in the previous paragraph. The claim here is merely that preference organization, whatever its origins, is an effective and recurrent solution to a very general problem of organized social life.

ACKNOWLEDGMENTS T h e author w o u l d like to thank John Heritage and M a n n y S c h e g l o f f for reading and c o m m e n t i n g on an earlier draft o f this paper.

REFERENCES Atkinson, J. M. (1984). Public speaking and audience responses: some techniques for inviting applause. In: J. M. Atkinson & J. Heritage (Eds), Structures of Social Action: Studies in Conversation Analysis (pp. 370--409). Cambridge: Cambridge University Press. Atkinson, J. M. (1992). Displaying Neutrality: Formal Aspects of Informal Court Proceedings. In: P. Drew & J. Heritage (Eds), Talk at Work (pp. 199-211). Cambridge: Cambridge University Press. Atkinson, J. M., & Drew, P. (1979). Order in Court: The Organisation of Verbal Interaction in Judicial Settings. London: Macmillan. Atkinson, J. M., & Heritage, J. (1984). Structures of Social Action. Cambridge: Cambridge University Press. Brown, P., & Levinson, S. (1987). Politeness: Some Universals in Language Usage. Cambridge: Cambridge University Press. Clayman, S. E. (1988). Displaying Neutrality in Television News Interviews. Social Problems, 35(4), 474-492. Clayman, S. E. (1992). Footing in the Achievement of Neutrality: The Case of News Interview Discourse. In: P. Drew & J. Heritage (Eds), Talk at Work (pp. 163-198). Cambridge: Cambridge University Press. Clayman, S. E. (1993). Booing: The Anatomy of a Disaffiliative Response. American Sociological Review, 58, 110-130. Clayman, S. E. (2001). Answers and Evasions. Language in Society, 30, 403--442. Daanje, A. (1950). On the Locomotory Movements in Birds and the Intention Movements Derived From Them. Behavior, 3, 49-98. Davidson, J. (1984). Subsequent Versions of Invitations, Offers, Requests, and Proposals Dealing with Potential or Actual Rejection. In: J. M. Atlonson & J. Heritage (Eds), Structures of Social Action (pp. 102-128). Cambridge: Cambridge University Press. Davidson, J. (1990). Modifications of Invitations, Offers and Rejections. In: G. Psathas (Ed.), Interaction Competence (pp. 149-180). Washington: International Institute for Ethnomethodology and Conversation Analysis and University Press of America. Davis, J. M. (1975). Socially Induced Flight Reactions in Pigeons. Animal Behavior, 23, 597-601. Dersley, I., & Wootton, A. (2000). Complaint Sequences Within Antagonistic Argument. Research on Language and Social Interaction, 33, 375--406.

Sequence and Solidarity

251

Drew, P. (1984). Speakers' 'Reportings' in Invitation Sequences. In: J. M. Atkinson & J. Heritage (Eds), Structures of Social Action: Studies in Converstaion Analysis (pp. 129-151). Cambridge: Cambridge University Press. Garcia, A. (1991). Dispute resolution without disputing: how the interactional organization of mediation hearings minimizes argumentative talk. American Sociological Review, 56, 818-835. Goffman, E. (1976). Interaction Ritual: Essays in Face to Face Behavior. Garden City, New York: Doubleday. Goodwin, M. (1983). Aggravated Correction and Disagreement in Children's Conversations. Journal of Pragmatics, 7, 657-677. Goodwin, M. (1990). He-SaM-She-Said: Talk as Social Organization among Black Children. Bloomington: Indiana University Press. Heringer, J. T. (1977). Pre-Sequences and Indirect Speech Acts. In: E. O. Keenan & T. Bennett (Eds), Discourse Structure Across Time and Space (pp. 169-180). SCOPIL 5. University of California Linguistics Department. Heritage, J. (1984). Garfinkel and Ethnomethodology. Cambridge: Polity Press. Heritage, J. (Forthcoming a). Designing Questions and Setting Agendas in the News interview. In: P. Glenn, C. LeBaron & J. Mandelbaum (Eds), Excavating the Taken-for-Granted: Studies in Language and Social Interaction. Heritage, J. (Forthcoming b). The Limits of Questioning: Negative Interrogatives and Hostile Question Content. Journal of Pragmatics. Heritage, J., & Greatbatch, D. (1986). Generating Applause: A Study of Rhetoric and Response at Party Political Conferences. American Journal of Sociology, 92, 110-157. Heritage, J., & Greatbatch, D. (1991). On the Institutional Character of Institutional Talk: The Case of News Interview Interaction. Pp. 93-137 In: D. Boden & D. H. Zimmerman (Eds), Talk and Social Structure (pp. 93-137). Berkeley: University of California Press. Holtgraves, T. (1992). Linguistic realization of face management: implications for language production and comprehension, person perception and cross-cultural communication. Social Psychology Quarterly, 55(2), 141-159. Jacoby, S. (1998). Science as Performance: Socializing Scientific Discourse Through the Conference Talk Rehearsal (Ph.D. Dissertation, Applied Linguistics). UCLA. Kotthoff, H. (1993). Disagreement and Concession in Disputes: On the Context-Sensitivity of Preference Structures. Language in Socie~,, 22, 193-216. Lazarton, A. (1997). Preference Organization in Oral Proficiency Interviews: The Case of Language Ability Assessments. Research on Language and Social Interaction, 30, 53-72. Lerner, G. (1996). Finding "Face" in the Preference Structures of Talk-in-Interaction. Social Psychology Quarterly, 59, 303-321. Lerner, G. (2001). The Search for Delicate Formulations. Paper presented at the ASA annual meetings (August). Anaheim. Levinson, S. (1983). Pragmatics. Cambridge: Cambridge University Press. Lindstrom, A. (1997). Designing Social Actions: Grammar, Prosody, and Interaction in Swedish Conversation (Ph.D. Dissertation). Department of Sociology, UCLA. Luckenbill, D. (1977). Criminal Homicide as a Situated Transaction. Social Problems, 25(2), 176-186. Lynch, M. (1985). Art and Artefact in Laboratory Science. London: Routledge and Kegan Paul. Maynard, D. (1989). Perspective-display sequences in conversation. Western Journal of Speech Communication, 53, 91-113. Maynard, D. (1991). On the interactional and institutional bases of assymetry in clinical discourse. American Journal of Sociology, 92, 448-495.

252

S T E V E N E. C L A Y M A N

Maynard, D. (1992). On clinicians co-implicating recipients' perspective in the delivery of diagnostic news. In: P. Drew & J. Heritage (Eds), Talk at Work: Social Interaction in Institutional Settings (pp. 331-358). Cambridge: Cambridge University Press. Maynard, D. (1996). On "Realization" in Everyday Life: The Forecasting of Bad News as a Social Relation. American Sociological Review, 61, 109-131. Maynard Smith, J. (1974). The Theory of Games and the Evolution of Animal Conflicts. Journal of Theoretical Biology, 47, 209-221. Maynard Smith, J., & Price, G. R. (1973). The Logic of Animal Conflict. Nature, 246, 15-18. Merritt, M. (1976). On Questions Following Questions in Service Encounters. Language in Society, 5, 325-357. Moynihan, M. (1955). Types of Hostile Display. Auk, 72, 247-259. Mulkay, M. (1985). Agreement and disagreement in conversations and letters. Text, 5, 201-227. Raymond, G. (2000). The Structure of Responding: Type-Conforming and Nonconforming Responses to Yes/No Type Interrogatives (Ph.D. dissertation). Department of Sociology, UCLA. Pomerantz, A. (1978). Compliment Responses: Notes on the Co-operation of Multiple Constraints. In: J. Schenkein (Ed.), Studies in the Organization of Conversational Interaction (pp. 79-112). New York: Academic Press. Pomerantz, A. (1984). Agreeing and Disagreeing with Assessments: Some Features of Preferred/Dispreferred Turn Shapes. In: J. M. Atldnson & J. Heritage (Eds), Structures of Social Action: Studies in Conversation Analysis (pp. 57-101). Cambridge: Cambridge University Press. Sacks, H. (1987). On the Preferences for Agreement and Contiguity in Sequences in Conversation. In: G. Button & J. R. E. Lee (Eds), Talk and Social Organisation (pp. 54-459). Clevedon, England: Multilingual Matters. Sacks, H. (1992). Lectures on Conversation. Oxford: Blackwell. Sacks, H., & Schegloff, E. (1979). Two Preferences in the Organization of Reference to Persons and Their Interaction. In: G. Psathas (Ed.), Everyday Language: Studies in Ethnomethodology (pp. 15-21). New York: Irvington Publishers. Schegloff, E. A. (1979). Identification and Recognition in Telephone Conversation Openings. In: G. Psathas (Ed.), Everyday Language: Studies in Ethnomethodology (pp. 23-78). New York: Irvington. Schegloff, E. A. (1980). Preliminaries to Preliminaries: 'Can I Ask You a Question'. Sociological Inquiry, 50, 104-152. Schegloff, E. A. (1987). Between Macro and Micro: Contexts and Other Connections. In: J. Alexander, B. Giesen, R. Munch & N. Smelser (Eds), The Micro-Macro Link (pp. 207-234). Berkeley: University of California Press. Schegloff, E. A. (1988). On an Actual Virtual Servo-Mechanism for Guessing Bad News: A Single Case Conjecture. Social Problems, 35(4), 442-457. Schegloff, E. A. (1988). Presequences and Indirection: Applying Speech Act Theory to Ordinary Conversation. Journal of Pragmatics, 12, 55~52. Schegloff, E. A. (forthcoming). Sequence Organization. Cambridge: Cambridge University Press. Schegloff, E., Jefferson, G., & Sacks, H. (1977). The Preference for Self-Correction in the Organization of Repair in Conversation. Language, 53, 361-382. Schegloff, E., & Sacks, H. (1973). Opening Up Closings. Semiotica, 8, 289-327. Smith, W. J. (1977). The Behavior of Communicating. Cambridge, MA: Harvard University Press. Terasaki, A. (1976). Pre-Announcement Sequences in Conversation. Social Science Working Paper 99. School of Social Sciences, Irvine California. Wootton, A. (1981). The Management of Grantings and Rejections by Parents in Request Sequences. Semiotica, 37, 59-89.

Sequence and Solidari~

253

APPENDIX: TRANSCRIPT NOTATIONAL CONVENTIONS Interaction excerpts were transcribed with notational conventions adapted from those used in conversation analysis. The transcripts capture the details of speech and audience behavior, although the excerpts in this paper have been slightly simplified to enhance readability. Below is a guide to the transcription symbols used here; for a more detailed exposition, see Atkinson and Heritage (1984, pp. ix-xvi). A: A: A: A: A:

A: B: A: B: A: B: A:

Underlined items were markedly stessed. Colon(s) indicate the prior sound was prolonged. Capital letters indicate increased volume. A hyphen denotes a glottal stop or "cut-off' of sound. .hhh That's my view. Strings of "h" mark audible breathing. The hhhh At least for now. longer the string, the longer the breath. A period preceding denotes inbreath; no period denotes outbreath. That's (.) my view. Numbers in parentheses denote elapsed silence (1.3) in tenths of seconds; a period denotes a But should it be? micropause of less than 0.2 seconds. That's my view.= E•ual sians indicate that one event followed =But should it be? the other with no intervening silence. That's [my view.] Brackets mark the onset and termination [But shou]ld it of simultaneous activities. That's my ( ) Open parentheses indicate transcriber's At (least for now). uncertainty as to what was said. Words in parentheses represent a best guess as to what was said. That's my view. That's my:: view. THAT'S my view. That's my- my view.