Researching Entrepreneurship

Researching Entrepreneurship

INTERNATIONAL STUDIES IN ENTREPRENEURSHIP Series Editors: Zoltan J. Acs University of Baltimore Baltimore, Maryland USA David B. Audretsch Indiana University Bloomington, Indiana USA Other books in the series: Black, G. The Geography of Small Firm Innovation Tubke, A. Success Factors of Corporate Spin-Offs Corbetta, G., Huse, M., Ravasi, D. Crossroads of Entrepreneurship Hansen, T., Solgaard, H.S. New Perspectives in Retailing and Store Patronage Behavior

Researching Entrepreneurship

by Per Davidsson Jönköping International Business School, Jönköping University, Jönköping, Sweden

Springer

eBook ISBN: Print ISBN:

0-387-23054-8 0-387-22838-1

©2004 Springer Science + Business Media, Inc. Print ©2004 Springer Science + Business Media, Inc. Boston All rights reserved No part of this eBook may be reproduced or transmitted in any form or by any means, electronic, mechanical, recording, or otherwise, without written consent from the Publisher Created in the United States of America

Visit Springer's eBookstore at: and the Springer Global Website Online at:

http://www.ebooks.kluweronline.com http://www.springeronline.com

TABLE OF CONTENTS

CHAPTER 1. WHAT IS ENTREPRENEURSHIP? 1 1 On the Variety of Definitions and Views of Entrepreneurship My Proposed View of the Entrepreneurship Phenomenon 6 New Offer as Entrepreneurship 9 New Competitor as Entrepreneurship 9 Geographical Market Expansion as Entrepreneurship 10 Organizational and Ownership Changes Are Not Entrepreneurship 11 Business as Usual and Non-Entrepreneurial Growth 12 Entrepreneurship as Micro-Level Behavior with Macro-Level Implications 12 Degrees of Entrepreneurship? 14 Summary and Conclusion 16 CHAPTER 2. ENTREPRENEURSHIP AS RESEARCH DOMAIN 17 Why Distinguish Between the Phenomenon and the Research Domain? 17 Previous Attempts at a Domain Delineation 18 My Suggested Domain Delineation 21 Uncertainty and Heterogeneity 22 Processes of Emergence; Behaviors in the Interrelated Processes of Discovery and Exploitation 23 Real or Induced, and Completed as Well as Terminated 25 Across Organizational Contexts 25 New Business Ventures; Venture Ideas and Their Contextual Fit 26 28 Antecedents and Outcomes on Different Levels of Analysis 30 Summary and Conclusion CHAPTER 3. THIS THING CALLED “THEORY” 33 33 Confessions of a Sinner 33 Theory Is No Mystery 34 The Need for Abstraction and Understanding 39 The Role(s) of Theory in the Research Process 39 Theory as Guide to Research Design Mark I: The Theory Test Theory as Guide to Research Design Mark II: Understanding the Phenomenon 44 through an Eclectic Framework Theory as Tool for Interpretation: The Theory Test 46 47 Theory as Tool for Interpretation: The Eclectic Framework Approach 48 Theory as Tool for Interpretation: Post Hoc Theorizing 50 Is It the Theory or the Data That Is Supported or Should Be Rejected? 51 Do We Need Specific Entrepreneurship Theory? 52 Summary and Conclusion CHAPTER 4. GENERAL DESIGN ISSUES 55 55 Getting Started at Last 55 “Qualitative” and “Quantitative” studies 55 The Need for “Qualitative” Entrepreneurship Research 57 “Quantitative” vs. “Qualitative”—a Confused Debate Bad Research Practice: Addressing “Quantitative” Questions with “Qualitative” 59 Research Entrepreneurship Research as the Study of Processes of Emergence of New 61 Ventures 63 Laboratory Research Methods

vi

RESEARCHING ENTREPRENEURSHIP

64 Summary and Conclusion 67 CHAPTER 5. SAMPLING ISSUES 67 A Different Look at Sampling 68 Social Science Is Not Opinion Polls 70 Sampling Individuals 72 Sampling Emerging New Ventures Identifying an Eligible Sample of On-Going Independent Venture Start-ups 73 77 Sampling On-Going Internal Venture Start-ups 80 Sampling Firms 80 Size, Size Distribution, and Heterogeneity along Other Dimensions 83 Relevance 89 Sampling Industries (or Populations) 89 Size, Size Distribution, and Heterogeneity along Other Dimensions 91 Relevance 92 Sampling Spatial Units 93 Relevance 95 Size, Size Distribution, and Heterogeneity along Other Dimensions 98 Sampling Other Units of Analysis 98 Response Rates 99 Summary and Conclusions 101 CHAPTER 6. OPERATIONALIZATION ISSUES 101 A 90-degree Turn 101 A Note on Levels of Measurement Validity and Reliability Issues 105 110 Some Balancing Exercises 113 Operationalization Issues on the Individual Level Operationalization Issues on the Level of the New, Emerging Venture 115 115 Operationalizing Resources 121 Operationalizing the Venture Idea 124 Operationalizing The External Environment 125 Operationalizing Behaviors 129 Operationalizing Outcomes 131 Operationalization Issues on the Firm Level 135 Operationalization Issues on Aggregate Levels 139 Summary and Conclusion CHAPTER 7. SPECIAL TOPIC: PREPARING A “SECONDARY” DATA SET 141 If You Don’t Have It, Don’t Try It 141 …Or Do the Job Needed to Make It Work 143 144 Use Prior Knowledge Combine Different Sources of Data 145 Check Quality and Make Corrections 153 Other Observations 155 Summary and Conclusion 156 CHAPTER 8. SPECIAL TOPIC: JOB CREATION AS THE DEPENDENT 159 VARIABLE 159 Why Care About Job Creation? 160 Data Coverage Static Comparison vs. Dynamic Analysis 163 164 Gross vs. Net Job Creation 166 The Regression-to-the-Mean Effect Organic vs. Acquisition Growth, and Job Creation vs. Economic Development 170 171 A Few More Details to Consider 173 Summary and Conclusions 175 CHAPTER 9. SPECIAL TOPIC: THE POWER OF REPLICATION 175 Sampling and Significance Testing Revisited

Contents Replicating Others Replicating One Another: Harmonized Research Collaboration Replicating Yourself Summary and Conclusion CHAPTER 10. A QUICK LOOK AT ANALYSIS METHOD Let’s Make This a Short One Heterogeneity and Analysis Method Analysis Implications of the Minority Nature of Entrepreneurship Analysis Implications of Entrepreneurship as Process Summary and Conclusion NOW THAT WE’RE DONE REFERENCES INDEX

vii 176 180 184 188 189 189 189 190 191 192 195 197 211

BEFORE WE BEGIN...

This is a methods book. Sort of. Why did I write such a book? Well, it is not going to become the most spectacular case study in successful entrepreneurship, that’s for sure! Despite the tremendous growth the field of entrepreneurship research has enjoyed since I entered it in the mid 1980s, the world market for a book by the title “Researching Entrepreneurship” is so limited that one can guarantee that it won’t make its author rich. The upside of that is that you can trust it is an honest book. I write it because I want to share my experiences and not with the intent to maximize profits; hence I do not have to compromise with my convictions in order to reach my goals. Although the market may be limited, I think there is a need for a book of this kind in that niche. There is not only growth but also progress in entrepreneurship research. However, it would be pointless to deny that there is also confusion and widespread frustration in the entrepreneurship research community, and a feeling that the field has not advanced as much as it should. For example, although entrepreneurship research has developed both conceptually and empirically, those advancements have not always gone hand in hand. One thing I try to achieve in this book is to spell out some of the implications the conceptualization of entrepreneurship has for empirical design and analysis. I thus think there may be a need for a book of this kind. One of my doctoral students also made me realize that I was one out of relatively few people who could (and fewer still who would) write such a book. Although this insight makes me feel terribly old (Hey, I’m just 46!), there are actually relatively few people who have been in the field for close to twenty years, or who have been involved in a doubledigit number of comprehensive empirical studies on entrepreneurship. I also came to realize that writing a method book—of sorts—on entrepreneurship research was within relatively easy reach for me because I have already written more than a dozen papers or manuscript (sub-)sections with method reflections and attempts to codify the method experiences gained through intense empirical work, from the early pamphlet (and keynote address) Entrepreneurship Research: How Do We Get Further (Davidsson, 1992)—which gives voice to frustrations and experiences during my years as doctoral student in this young and not so high-standing field in the late 1980s—to the recent The Domain of Entrepreneurship Research: Some Suggestions (Davidsson, 2003a), which in many ways is a direct forerunner to the present volume. Now, as the work proceeded I brought to mind more and more old manuscripts, and came to think of additional method issues for which I could (or previously had tried to) codify the tacit knowledge that I had accumulated over time. So, in the end the task wasn’t within as easy reach as I originally thought. For whom is this book intended? Doctoral programs and courses focusing on entrepreneurship and entrepreneurship research are obvious primary targets. The book is not to intended to substitute more general method textbooks, though, but supplement them by adding “entrepreneurship research meaning” to more general expositions, and add details that standard textbooks do not cover. I hope that my book also will help established researchers who visit or migrate into the field of entrepreneurship research to see research opportunities and to avoid repeating mistakes that other entrepreneurship researchers and I have made already. If I am successful, this category will find useful information herein which is otherwise hard to find or takes years of own practice to gain. Colleagues among long established entrepreneurship researchers are also a target group. Although this group collectively has much more to teach me than I have to offer them, and although they obviously already have the method knowledge and insights they need to continue

x

RESEARCHING ENTREPRENEURSHIP

their careers, I hope this book can give some colleagues advice and inspiration that help them refresh and perfect their research agendas. In some chapters and sections, I treat topics that are of relevance for social science research in general or at least well outside of the narrow topic of entrepreneurship study. I do this because I think these topics are typically less well treated in standard method textbooks, and often lack coverage in social science research education. For example, this goes for the reasoning on empirical and theoretical representativeness as well as and the related statistical inference and replication issues in chapters 5 and 9; some of the general operationalization issues in the early parts of 6, and the problems and solutions involved in working with secondary data covered in Chapter 7. Although I realize that it is unlikely to happen in a big way, I hope that these sections will find (satisfied) readers also outside of the small, albeit growing, camp of entrepreneurship researchers and research students. This book does not give anything near a complete and balanced coverage of all types of research approaches, methods, or techniques. Like I already said, it is intended to supplement rather than substitute more general method texts. The contents are admittedly and deliberately colored by my own specific experiences and expertise. My dissertation study Continued Entrepreneurship and Small Firm Growth Davidsson, 1989a, 1989b; 1991) was based on a cross-sectional phone+mail survey of small business owner managers, and the study Culture and Entrepreneurship (Davidsson, 1993; 1995a; 1995c; Davidsson & Delmar, 1992) used cross-sectional mail survey directed at the general population. There is nothing unique about that; published research in entrepreneurship is dominated by crosssectional (mail) surveys. I can claim more unique insights when in comes to longitudinal, repeated survey (panel) studies, using a combination of phone and mail data collection methods. This is through my involvement with the Panel Study of Entrepreneurial Dynamics, or PSED (Gartner, Shaver, Carter & Reynolds, 2004; Reynolds, 2000) and as one of the principal investigators in its Swedish sister project (Davidsson & Honig, 2003; Delmar & Davidsson, 2000); through the project The 1994 Start-up Cohort (Dahlqvist, Davidsson & Wiklund, 2000) and its derivative project New Internal Ventures (Chandler, Dahlqvist & Davidsson, 2003), and as regards corporate entrepreneurship also through the project Entrepreneurship in Different Organizational Contexts (Brown, Davidsson & Wiklund, 2001; Wiklund, Eliasson & Davidsson, 2002). I have also acted as supervisor for doctoral students using yet other comprehensive, longitudinal survey studies. Likewise, I can claim extensive experience from research based on comprehensive, longitudinal, secondary data sets from the projects Business Dynamics in Sweden Davidsson, Lindmark & Olofsson, 1994a, 1994b, 1995, 1996, 1998a, 1998b) and High Growth Firms 1987-1996 (Davidsson & Delmar, 2003; Davidsson & Henreksson, 2002; Davidsson, Kirchhoff, Hatemi-J & Gustavsson, 2002; Delmar, Davidsson & Gartner, 2003). I have much more limited experience from experimental or quasi-experimental research (Davidsson, 1986; Davidsson & Wahlund, 1992) and I have not used qualitative techniques for data collection and analysis since the pilot study for my dissertation project (Davidsson, 1986), although I have co-supervised several Ph.D. students applying such methods. The experiences shared in this book thus build mainly on insights from longitudinal, quantitative data from custom-designed surveys or from secondary sources. This does not reflect the view that such approaches are necessarily “better” for all purposes. In fact, some of my favorite references in entrepreneurship build on completely different approaches; often more qualitative (Bhave, 1994; McGrath, 1999; Sarasvathy, 1999a, 2001; Shane, 2000). However, it is when it comes to broadly based, quantitative approaches that I can claim some level of expertise, and it is therefore that type of approach that will dominate this book. I hope this will not make those who think quantitative methods are not their cup of tea close the book

preface

xi

and put it away at this stage already. Critique is often more interesting and useful when it comes from somebody who has deep knowledge of the critiqued, and the reader shall find that there is no shortage in this book of direct and indirect critique of some research practices that are common among researchers who use quantitative approaches. I fully agree with those who think that research is dull when it consists of taking an existing data set as it is, running a few regressions, and determining which relationships are statistically significant or not—period. Luckily, good quantitative work consists mainly of intellectual challenges that are far greater and more intriguing than that, when it comes to research design and interpretation of results. You will not find a lot of philosophy of science arguments or references in this book. Let me tell you a secret: philosophers of science often do not know much about conducting empirical research—they simply don’t have that experience and expertise. They can therefore not advise you on many of the issues dealt with in this book. In an ideal world, the author of a book like this should have both kinds of expertise—philosophies of science and deep empirical research practice. Unfortunately, I stay far short of being an expert on philosophies of science. But I’m not completely ignorant, nor do I think philosophy of science issues are unimportant. Reflection upon the foundations of knowledge production is critically important, and admittedly lacking in a lot of research of “my” kind. However, I do not believe in having a faith when it comes to philosophy of science. I can think of no more narrow-minded and unacademic attitude than thinking that “all the good guys think like us” or “our approach is the best”. So I tend to be an eclectic (or pragmatic) skeptic, accepting and refuting arguments from several camps. Early in my career, Scandinavian colleagues would equate my revealed preference for quantitative data with a positivist philosophy of science. And, for sure, there is some positivist heritage in the kind of work that I do. However reading descriptions of positivism vs. hermeneutics (popular in the late 1980s) I could easily see that the research process I had been through in my dissertation work had—with its wrestling back and forth between theory and data— much more semblance with a hermeneutic and abductive (no aliens involved, though!) research process than a positivist or hypothetico-deductive one. And reading August Comte in (translated) original made me less than impressed. If he could read my work he would be equally unimpressed, since I deviate so much from true positivist ideals. No, quantitative data don’t make you a positivist; nor do qualitative data make you a social constructionist or your approach hermeneutic. Let’s face it: data don’t know how they are going to be used! For certain purposes, like gender issues in organizations, I think social constructionism is highly useful and illuminating. But I am not a constructionist. When it becomes a nihilist faith or leads to denial of the obvious (for example, that real differences between male and female bodies exist and have real effects), I find the perspective just boring, cowardly, or outright stupid. Although I do not find it totally convincing in terms of logical coherence, scientific realism is probably the school of thought that comes closest to the practically workable middle ground that I find most useful for guiding empirical research. Method textbooks tend to be used rather than read. One reason for this is that they are often written in such a way that reading them is, should we say, less than enjoyable. Especially when the reader is an already established researcher I have no difficulty with this book being used in parts, by looking up specific issues one at a time. However, I would also like it to be a bearable experience to read it from cover to cover. Therefore, while hopefully retaining enough seriousness and credibility, I will try to refrain from dull academic jargon and unnecessarily heavy style. I will also spice up the exposition with numerous examples and actual research results, some of which have not been published before.

xii

RESEARCHING ENTREPRENEURSHIP

Many people and organizations have contributed to this book. Far too many, in fact, for it to be possible to mention them all individually. A multitude of organizations, the single most important being the Knut & Alice Wallenberg Foundation, have funded the research on which the book builds. Several employers have been kind and clever enough to allow me considerable time to do research rather than having me do things I am less apt for—although heading a department of 40+ people for the last couple of years almost made this book never happen. Many colleagues have inspired, criticized and, in other ways, contributed to this work. A collective that I have to mention here is the researchers and administrators within and affiliated with the Program on Entrepreneurship and Growth in Small and Medium-sized Firms (PEG), which I have had the pleasure of leading during the last few years. I would also like to single out and especially thank Frédéric Delmar, Leif Lindmark, Christer Olofsson David Storey, Paul Reynolds and Johan Wiklund. Their contributions have been absolutely crucial for important aspects of this book. Indepth collaboration with experts at Statistics Sweden has been essential especially for chapter 7. Finally, thousands of respondents, a somewhat smaller number of interviewers and a data manager have also been indispensable for the realization of this book. I will stick to my habit of not dedicating this book to someone I love, because I think it an odd practice to “give” to people what does not interest them. Neither will I thank my family for their assistance in my writing of this book, because I don’t think they did help that much. On the contrary, they constantly tried to drag me away from the computer and lure me into all sorts of distractions that they perceived to be better use of my time. For this, I do thank them with all my heart. Per Davidsson

CHAPTER 1

WHAT IS ENTREPRENEURSHIP?

ON THE VARIETY OF DEFINITIONS AND VIEWS OF ENTREPRENEURSHIP Researching entrepreneurship is fun, fascinating, frustrating—and important, if you ask me. One of the fascinations is the richness of the phenomenon, which leads to one of the greatest frustrations, namely the lack of a common understanding of what precisely entrepreneurship is. Let me put it this way: there is no shortage of suggestions as to what the phenomenon “entrepreneurship” really consists of. Here are a few examples: new entry (Lumpkin & Dess, 1996) the creation of new enterprise (Low & MacMillan, 1988) the creation of new organizations (Gartner, 1988) a purposeful activity to initiate, maintain and aggrandize a profit-oriented business (Cole, 1949) taking advantage of opportunity by novel combinations of resources in ways which have impact on the market (Wiklund, 1998) the process by which individuals—either on their own or inside organizations—pursue opportunities without regard to the resources they currently control (Stevenson & Jarillo, 1990) the process of creating something different with value by devoting the necessary time and effort; assuming the accompanying financial, psychological, and social risks; and receiving the resulting rewards of monetary and personal satisfaction (Hisrisch & Peters, 1989) To expand the list, we may note that without offering formal definitions, Drucker (1985) as well as Bull & Willard (1993) favor a Schumpeterian view (Schumpeter, 1934). The former associates entrepreneurship with innovative and change-oriented behavior, whereas the latter include also task-related motivation, expertise, and expectation of gain for self.

2

RESEARCHING ENTREPRENEURSHIP

Kirzner (1983) offered the following compilation of roles assigned to the entrepreneur by various economic theorists: a specific kind of labor service assuming the risk innovator arbitrageur coordinator, organizer, or gap-filler providing leadership exercising genuine will acting as a pure speculator acting as employer acting as superintendent or manager acting as a source of information being alert to opportunities as yet overlooked in the market Using an empirical approach to the question of what entrepreneurship is, Gartner (1990) found the following eight themes to emerge when professional users (academic and other) of the entrepreneurship concept were asked about its inherent meaning: the entrepreneur innovation organization creation creating value profit or non-profit growth uniqueness the owner-manager Similarly, a content analysis of journal articles and books performed by Morris, Lewis & Sexton (1994; back translated from Kufver, 1995) yielded the following most common definitional keywords: start; form; create new business innovation; new product; new market pursuit of opportunities risk taking; risk management; uncertainty pursuit of profit; personal advantages new production methods management coordination of resources value creation

What is Entrepreneurship?

3

Tired yet? At this point it should be superfluous to point out that no one can claim to have the one, true answer to the question of what the phenomenon “entrepreneurship” truly is. If anything is a social construction, language is. So far, in the social game of filling the entrepreneurship concept with meaning, none of the existing and partially overlapping constructions seems to have achieved dominance over the others. Some of the issues on which the views on entrepreneurship differ are the following: Is entrepreneurship something that is restricted to the commercial sector; is it an economic phenomenon, or something that can present itself within any area of human endeavor? Is entrepreneurship restricted to small or new or owner-managed firms, or can it be executed by or within organizations of any age, size, and governance structure? Is entrepreneurship an innate characteristic (disposition), a type of behavior, or does it involve a special type of outcome (for example, is success required)? Does something have to be purposeful in order to amount to entrepreneurship, or can processes involving luck and serendipity qualify? Is innovation required, or can imitative initiatives exemplify entrepreneurship? Is risk-taking a necessary requirement? Does entrepreneurship involve the discovery (or creation) of opportunities; the exploitation of opportunities, or both? Is it solely a micro-level phenomenon, or is entrepreneurship a meaningful concept on more aggregate levels as well? The language games we play regarding the meaning of entrepreneurship is a funny type of games where—unlike sports—it is totally conceivable that two opposing players both determine that they (according to their own rules) won the game, whereas the spectators—that is, the fellow researchers who read the arguments—find that both sides scored points, but since they did not play the same game on the same field, it wouldn’t be very meaningful to appoint a winner. Like sports, however, those language games are something some people (like researchers in this field) think are extremely interesting and important, whereas others couldn’t care less. I tend to be somewhat ambivalent on the importance of precise, inherently consistent, and agreed-upon definitions. I am pretty sure, however, that underneath the various constructions of entrepreneurship we shall find interesting and important social realities. Part of me says, “Forget definitions—let’s just go and learn and tell about those important realities!” Another part of me, however, strongly feels that in order to do just that, a researcher must have a very clear idea of what that social reality is, and be able to communicate that idea, be it shared or not by most of the readers. I will let that other part of me speak for a while now.

4

RESEARCHING ENTREPRENEURSHIP

Some of the variations in entrepreneurship definitions, I believe, are relatively minor and of little import. They largely reflect the same underlying social reality and therefore the differences in the finer nuances do not confuse communication or hinder knowledge accumulation. Other differences, however, may have such effects and therefore cannot be disregarded as easily. Over the years, I have come to the conclusion that the different entrepreneurship definitions actually address two relatively distinct social realities. The first of those is the phenomenon that some people, rather than working for somebody else under an employment contract, strike out on their own and become self-employed, or team owner-managers of an independent business. This implies a radically different risk/reward structure with a much wider span of possible financial outcomes, and a more fluid border between work and leisure. Often these new economic entities involve some element of innovation at start-up, and some degree of innovativeness may be needed in order to survive in this capacity over time. However, it is a well-known fact that the majority of independent businesses are relatively stable operations in mature and low to medium value-added industries. Some of those independent businesses will grow in size, which implies that the owners will face different types of management challenges and transitions over time. Often other family members than the original founder are involved in the business, and eventually the transfer of the ownership and management of the business either within or outside the family will become a major issue for the founders. When the concept “entrepreneurship” is used for this social reality, topics like self-employment, small business management, stages-of development models, and family business issues become aspects of entrepreneurship. In short, entrepreneurship is anything that concerns independently owned (and often small) firms and their owner-managers. The second social reality that emerges as a major underlying theme in entrepreneurship definitions is that the development and renewal of any society, economy or organization requires micro-level actors who have the initiative and persistence to make change happen. Institutions as well as market and organizational structures may facilitate or hinder change and development. However, those structures do not create any change—and they certainly do not change themselves— in the absence of human actors. In the end, it is the unique knowledge, perceptions and goals of individuals equipped with the drive to take action accordingly that initiate novelty. In order for those new initiatives to have lasting impact, however, they need to create value or save resources. When the entrepreneurship concept is used as a label for this social reality, quite a different set of topics become integral part of it: innovation along any of Schumpeter’s (1934) five dimensions; corporate venturing and organizational rejuvenation (Sharma & Chrisman, 1999), and change-agency outside of the forprofit sector. The start-up of new, independent ventures, it would seem, is the only natural candidate for inclusion under both views. A problem with many definitions of entrepreneurship, as well as many implicit views on this phenomenon, is that they cover in fact an amalgam of the two social realities described above. My personal development over time has certainly been a drift—like the overall tendency in the international research community—from

What is Entrepreneurship?

5

embracing an entrepreneurship/small business view towards being more inclined to include the creation of new business ventures within any organizational context and at the same time becoming more reluctant to include just any aspects of ownershipmanagement. I think, that in order to do entrepreneurship research good, by doing good entrepreneurship research, the researcher needs to take sides here. I’d be very pleased if I could convince my readers so far, whether or not they decide to follow the specific direction I will outline below. The choice is actually not only between the two alternatives outlined. There are also more restricted or refined alternatives. In order to discuss these we take the help of Figure 1:1.

Figure 1:1.

Possible delineations of the entrepreneurship phenomenon

One obvious alternative is to choose the entire square B as one’s view of entrepreneurship. I personally do not se the logical or linguistic reasons for doing so. We have seen above that entrepreneurship is widely connoted with quite an array of things that are definitely not necessary characteristics of independent businesses. If one wants to reserve the concept for independent firms the intersection A and B— entrepreneurial small business, if you like—would seem a more attractive alternative. This would include, for example, new firm formation; small firm innovation; internationalization and certain other aspects of growth of small, independent firms, and possibly the rejuvenation of family businesses as a result of ownership and management succession. This is not, however, my own preferred choice. Neither will I argue for the inclusion of the entire square A. The view of the entrepreneurship phenomenon that I am going to elaborate on below—and which was first developed and presented in Davidsson (2003a)—is instead a more restricted alternative illustrated by the square A', demarcated by dotted lines at the left and bottom. That is, I propose a “micro-

RESEARCHING ENTREPRENEURSHIP

6

level novel initiative” view, but for reasons detailed below, I confine it to a market context, thus excluding non-market activities such as (certain) not-for-profit endeavors and internal, organizational change per se. Activities undertaken by existing or emerging independent businesses are certainly included in this view, but only as long as they imply the introduction of novelty to the market.

MY PROPOSED VIEW OF THE ENTREPRENEURSHIP PHENOMENON Hoping that the reader remembers that I have already pointed out that no one can claim to have the right answer to the question of what entrepreneurship really is— here is what I suggest. I propose that a fruitful way to define entrepreneurship is the notion in Austrian economics that entrepreneurship consists of the competitive behaviors that drive the market process (Kirzner, 1973, pp. 19-20). This does not imply a general preference for Kirzner’s theorizing over, for example, Schumpeter’s (1934) or Baumol’s (1993). I favor this definition because it is succinct and gives a satisfactorily clear delineation of the role of entrepreneurship in society. Firstly, it is based jointly on behavior and outcomes. The behavior part is necessary for not losing track of the fact that micro-level decisions and actions are needed for any change to be introduced. As regards the outcome part, I argue that when we think of entrepreneurship as a societal phenomenon, it is a distinctive advantage to include an outcome criterion and make clear, for example, that mere contemplation over radically new ideas or vain introduction of fatally flawed ones do not amount to “entrepreneurship”. Entrepreneurship makes a difference, or else it isn’t entrepreneurship. In order to “drive the market process” the activity has to have some direct or indirect success. To those readers who get itches at this stage, I can only say I hope that the next chapter will solve the problem. So please stay tuned, because I will relax outcome as a necessary criterion when discussing entrepreneurship as a scholarly domain. When defining the entrepreneurship phenomenon, I prefer a perspective that portrays it as micro-level behavior that can have hugely important macro-level implications. Secondly, it puts entrepreneurship squarely in a market context and makes clear that it is the suppliers who exercise entrepreneurship—not customers, legislators, or natural forces that also affect outcomes in the market. When suppliers engage in entrepreneurship, they introduce new, improved or competing offerings in an emerging or already existing market. They thereby drive the market process in one or more of the following ways: 1. 2. 3.

They provide customers with new choice alternatives, potentially giving some of those customers more value for their money. They stimulate incumbent actors to improve their market offerings in their turn, which increases efficiency and/or effectiveness of those actors. If successful, they attract other new entrants to the market, thus further increasing competitive pressures towards improved efficiency and effectiveness.

What is Entrepreneurship?

7

Importantly, driving the market process does not require that the first mover makes a profit but refers to the suppliers as a collective. Even if it eventually loses out the first mover contributes to driving the market process if subsequently someone else gets it right, which leads to a lasting change in the market. Admitting that change-inducing micro-level initiatives are undertaken in nonmarket contexts, I believe it an advantage to restrict the use of the specific term “entrepreneurship” for the market or market-like contexts, that is, when the setting involves customers, suppliers and (potential) competitors or very close equivalents to those. The main reason for this restriction is that I think it is valuable for the progress of entrepreneurship research to make the concept as distinct and well defined as possible. Now, broad agreement in the research community is probably not to be hoped for, but then at least I think it important that individual researchers carrying out specific research projects base their use of the entrepreneurship concept on a notion as clear as the one suggested here. Moreover, those who want to include novelty through “new combinations” (Schumpeter, 1934) in any domain of human behavior in the concept of “entrepreneurship” have reason to contemplate the full implications of this choice. This would not only allow, for example, novelty in the arts and in the organization of humanitarian aid activities into the picture, but also novelty in crime and warfare. And it would certainly make the events of September 11, 2001, an entrepreneurship masterpiece. To conceive of a fully fueled passenger jet as a missile and to combine the idea of hi-jacking with that of kamikaze attacks is certainly innovative, and in terms of impact—economic and otherwise—it has few parallels. However, regarding these attacks as driving market processes is farfetched, and this author would therefore suggest they be not regarded an instance of entrepreneurship.

8

RESEARCHING ENTREPRENEURSHIP

Figure 1:2

Firm and market newness of economic activities

Put in slightly different words, entrepreneurship according to the suggested perspective consists of the introduction of new economic activity that leads to change in the marketplace (cf. Herbert Simon in Sarasvathy, 1999b, pp. 2, 11). This illustrated in Figure 1:2. Note that “new” along the market axis means either that an entirely new market emerges, or that an activity is new to an existing market. In the latter case, “new” could mean the launch of an innovation, but merely entering as a new competitor could also qualify. Likewise along the firm axis, “new” means that either the new activity is an independent start-up, implying that a new firm emerges as a result, or it is an internal new venture, which means that the firm has previously not been making this particular market offering. Under the suggested definition the left hand side of the figure—quadrants I and IV—exemplify entrepreneurship, whereas quadrants II and III do not. This conjures also with the argument developed at some length by Baumol (1993) in that imitative entry and internationalization are included in the concept, whereas, for example, take-over is excluded.

What is Entrepreneurship?

9

New Offer as Entrepreneurship The first entry in quadrant I reads “New offer”. This is when something so new is introduced that a new market is created (Bhave, 1994, p. 231; Sarasvathy, 1999a) or at least no supplier has previously made the same offer in the same market. There is hardly any disagreement among scholars that this should be included in the concept of entrepreneurship, although some might want to restrict the inclusion to situations where a new and/or independent firm is behind the new offer. The first category, new product or service, corresponds to Schumpeter’s (1934) “new product” and Bhave’s (1994) notion of “product novelty”, respectively, and requires no further explanation. The second category, new bundle, refers to any combination of product and service components that—as a package deal—is unique relative to what has previously been offered on the market, although no individual component may be strictly new. This is what Bhave (1994) calls “new business concept”, and what Amit & Zott (2001) have in mind when they talk about “new business model”—as long as the concept or model includes newness as perceived by buyers and competitors. In some cases it amounts to Schumpeter’s (1934) category “re-organization of an entire industry”. An illustrative case is IKEA. The newness IKEA brought to the market was not so much the piece of furniture in use, but in the division of labor among different actors—including the consumer—in the production and distribution of the end product. IKEA would also qualify under the third category included in “new offer”, new price/value relation. This does not create a new market but drives the market process because it changes consumer choices and give other competitors reason to change their offerings. Consequently, Kirzner (1973, pp. 23-24) explicitly discusses offering the same product at a lower price as one form of entrepreneurship. A process innovation or organizational change (quadrant II) may often be the underlying cause of a new price/value relation, but this is not necessarily the case. It may also represent a strategic change that relies on expected economies of scale or experience, or a switch from low volume/high margin to high volume/low margin strategy.

New Competitor as Entrepreneurship The second main entry in quadrant I is “New competitor”. That is, I suggest that not only innovative but also imitative entry be included in the entrepreneurship concept (cf. Aldrich, 1999; Aldrich & Martinez, 2001). This is when a new, start-up firm enters the market, or an existing firm launches a new product line in a situation where other firms already supply the market with essentially the same product. Now, the reader may wonder whether the author is incapable of seeing the difference between the entry of yet another hairdresser or mom-and-pop store on the one hand, and the venture-capital backed launch of a new, high-potential biotech firm on the other. Well, let’s look back and see how we have defined entrepreneurship. Does the new hairdresser provide customers with a new, potentially better alternative? If nothing else is special about the new actor, it will at least have a different location, which may be more convenient for some customers. And the closest competitors

10

RESEARCHING ENTREPRENEURSHIP

may well find reason to reduce their price or shape up their service level in order to limit the damage caused by the new competitor. Hence, the reason for imitative entry to be included in the entrepreneurship concept is that such entry drives the market process in the sense that consumers get additional choices and incumbent firms get reason to change their behavior to meet this new competition. Another vantage point for this argument is that studies have found that entry with complete lack of novelty tends not to appear empirically (Bhave, 1994, p. 230; Davidsson, 1986). No entrant is a perfect clone of an existing actor. Therefore, trying to include an innovativeness criterion in the definition of entrepreneurship would create problems. Rather than drawing the line at zero innovation (which would exclude no cases) one would be forced to define an arbitrary minimum limit of innovativeness across different industries and types of novelty. This problem is aggravated by the fact that what appears new in one market may be a blueprint copy of what already runs successfully in a different market. For example, in his intriguing historical reconstruction of the automated restaurant industry in Sweden, (Gratzer, 1996) shows that when this concept revolutionized the restaurant market in Sweden it was already a maturing niche in the US restaurant market. All in all, then, there are several good reasons to include imitative market entry in the entrepreneurship phenomenon. However, if the new entrant is inferior along all dimensions it will neither succeed nor influence other actors’ behavior, and then it does not constitute an instance of entrepreneurship. Moreover, there are important differences between more and less innovative start-ups that call for a notion of “degrees of entrepreneurship”. This is an issue we will return to later on in this text.

Geographical Market Expansion as Entrepreneurship Defining entrepreneurship the way we have done makes it logical to include also quadrant IV—geographical market expansion—in the concept of entrepreneurship. Again, some readers may find this to be over-extending the entrepreneurship concept. What makes the “simple” repetition of old success recipes in new contexts entrepreneurial? Again, the answer lies in the fact that we have defined entrepreneurship from a market perspective. Although by now, the activities are (largely) no longer new from the firm’s perspective, their introduction in new markets—if not totally unsuccessful—drives the market process in these new places. When business model innovators like McDonalds, IKEA, Dell or the free newspaper Metro enter their nth country market, it may well be as revolutionary for the consumers and competitors in that market as it was for consumers and competitors in the markets where these businesses originated. If the entry is successful, it reflects Schumpeter’s (1934) “new market” category of economic development. The alternative to require newness to the firm as a criterion would lead to less desirable consequences. For example, had Southwest Airlines successfully copied their own concept in the European market it would not constitute entrepreneurship. If instead a new actor (Ryan Air) copied the concept and took it to the European market it would count as entrepreneurship. This is less than satisfactory from any perspective, and from a market perspective it makes no sense at all.

What is Entrepreneurship?

11

Organizational and Ownership Changes Are Not Entrepreneurship By contrast, according to our conceptualization the organizational and ownership changes listed in quadrant II do not by themselves constitute entrepreneurship. At this point, after the generosity awarded to imitative start-ups and geographical expansions, some readers may be outright annoyed or disturbed to find internal reorganizing, no matter how dramatic and creative, to be excluded from the entrepreneurship concept. But please bear with me a few more lines or perhaps a few more pages. Perhaps you can appreciate the internal logic of the argument, regardless of whether you are inclined or not to fully accept the definition of entrepreneurship that I develop here. It is certainly conceivable (and likely) that reorganization facilitates the creation of new economic activity by the organization. However, it is not necessarily the case that organizational and ownership changes lead to such effects. It is also conceivable that organizational units that are transferred to new ownership and/or undergo internal reorganization experience changes in job satisfaction and/or financial performance without at all changing the consumers’ choice options or influencing the behavior of competitors. Actually, there are at least four cases: a) an organizational or ownership change is intended to lead to more new market offerings by the firm, and does so, b) same as (a) but the intended increase in new market offerings does not happen, c) the change is undertaken for other reasons and has no effect on the firm’s market offerings, and d) the change is undertaken for other reasons but has the unintended effect of also making the firm more entrepreneurial in terms of introducing novelty in the marketplace. I think it is valuable not to lump together all those cases and include them in the notion of entrepreneurship. Instead, I see it as valuable to conceptually separate the organizational or ownership change from its effects. With our market-based definition of entrepreneurship it is the (successful or influential) launching of new business activities that might follow from it, and not the organizational change itself, that constitutes entrepreneurship. Whether increased launch of novelty to the market was an intentional outcome or not is not of significance. The argument is perhaps easier to accept if we move to the level of societal organization. Politicians can decide on changes in how society is organized and introduce de-regulation or other institutional changes which create opportunity in market x and therefore an increase in competitive behaviors that drive the market process in that market. In other words, the result is more entrepreneurship in market x. According to my argument, it is not the politician who exercises entrepreneurship in market x, but the micro-level actors in that market. The political decision facilitates entrepreneurship. In the same way, a manager may facilitate entrepreneurship through organizational change, but it is the market related activities that may result, and not the organizational change per se, that constitute entrepreneurship. This conceptual distinction is also the reason why I refrain from including Schumpeter’s (1934) “new production method” and “new source of supply”, as well as Bhave’s (1994) “novelty in production technology”, in the definition of the entrepreneurship phenomenon (cf. Davidsson, 2003a; Kirzner, 1983, p. 288).

12

RESEARCHING ENTREPRENEURSHIP

According to my argument, it is only when these events are translated into new offers or a new price/value relation in the market that we see entrepreneurship. As we shall see in the next chapter, the study of how organizational change relates to discovery and exploitation of new venture ideas remain an important question for entrepreneurship as a scholarly domain. Business as Usual and Non-Entrepreneurial Growth Turning now to quadrant IV, “Business as usual” here is, at first glance, as easy to exclude from the notion of entrepreneurship, as “New offer” in quadrant I was easy to include. But there does not seem to exist full agreement even here. First, we have von Mises’ denial of the existence of such a thing as “business as usual” when saying that “In any real and living economy every actor is always an entrepreneur” (Mises, 1949, p. 253). One can argue that no market action is completely void of novelty. For example, when a daily newspaper carries out the totally expected and routine actions of producing a new issue and distributing it to its subscribers and usual sales outlets, it is a new issue, and not yesterday’s paper, that is being distributed. Competitors will equally routinely read it, and it cannot be ruled out that some part of the contents may have a twist that inspires the competitor to do something in a future issue, which it otherwise would not have done. In other words, we find an element of “competitive behaviors that drive the market process” in these routine actions. Although this seems to lead to a delimitation problem similar to the arbitrary innovation criterion discussed above, my conclusion in this case goes in the other direction. That is, there is a lot of “known products for known buyers” activity going on that is so clearly predominantly of a “business as usual” character that it is not very difficult to classify it as such, both conceptually and empirically, and thus exclude it from our definition of entrepreneurship. The issue of non-entrepreneurial growth is tricky for slightly different reasons (see Davidsson, Delmar & Wiklund, 2002, for an elaborate discussion). When an economic actor exploits a venture idea, there will be no well-defined moment at which “entry” ends and “continued, routine exploitation” begins. Schumpeter (1934) held that mere volume expansion was not entrepreneurial, whereas he included the opening of new markets. It is a similar distinction I have in mind here. By “nonentrepreneurial growth” I mean passively or re-actively letting existing activities grow with the market. This would not provide much cause for alert among competitors nor give customers new choices.

Entrepreneurship as Micro-Level Behavior with Macro-Level Implications I pointed out in the early parts of this chapter that one important feature of the entrepreneurship definition I have chosen is that it portrays entrepreneurship as micro-level behavior with hugely important implications for more aggregate levels of analysis. Simplistic conceptions of venture outcomes typically classify them as successes or failures. However, when you consider multiple levels of analysis a more complex picture emerges. It is, for example, possible that a new venture that fails miserably has important positive effects on the economy-at-large, because both

What is Entrepreneurship?

13

those involved and others learn for the future and can come up with smarter solutions that would not have been within reach without the initial “failure.” This is what Figure 1:3 is getting at. “Venture” could here mean the sole activity of a new firm or a new, additional activity by an established firm. Thus, “venture” should not be interpreted (necessarily) as new firm or company, but as a new-to-the-market activity as discussed above. If we turn first to quadrant I, we find ventures that are successful in themselves and which produce net utility to society as well. These ventures are analytically unproblematic. Their successful entries into the market no doubt “drive the market process” and hence they exercise entrepreneurship under the definition I have suggested. Likewise, the failed ventures in quadrant III pose no trouble. They represent launching efforts that do not take off financially, and neither do they inspire followers or incumbent firms so that the eventual net effect becomes positive on the societal level. They are, so to speak, completely vain efforts.

Figure 1:3

Outcomes on different levels for new ventures (new economic activities)

The catalyst ventures in quadrant IV are a more interesting category. And they are probably much more ubiquitous than we might first think. Although not successful on the micro level—perhaps because they are outsmarted by followers or retaliating incumbents—they do “drive the market process” precisely because they bring forth such behavior on the part of other actors. An unsuccessful venture that inspires more profitable successors does not complete the entrepreneurial process but it no doubt contributes to the entrepreneurship phenomenon. The total effect on the economy is not necessarily smaller for catalysts than for “success ventures”. Catalyst ventures may therefore be a very important category from a societal point

14

RESEARCHING ENTREPRENEURSHIP

of view (cf. Low & MacMillan, 1988; McGrath, 1999). This should serve as a warning against too simplistic a view on micro-level failure. The ventures in quadrants I and IV, then, represent entrepreneurship whereas the failed ventures in quadrant III do not. What about the “Re-distributive” ventures in quadrant II? These are ventures that yield a surplus on the micro-level while at the same time the societal outcome is negative. Examples could be trafficking with heavy drugs or—as in an actual case in Sweden—a graffiti removal operation whose owners used nighttime to generate demand for their business. Thus, in these cases those involved in the venture enrich themselves at the expense of collective wealth. Does this represent entrepreneurship? I would argue that the theoretical status of “redistributive” ventures is determined by the answer to “towards what?” entrepreneurship drives the market process. Schumpeter (1934) and Kirzner (1973, p. 73) give seemingly contradictory answers to that question. On closer look, however, the movement from Schumpeter’s (local) equilibrium and the movement towards Kirzner’s (global) equilibrium are in full agreement insofar as that entrepreneurship drives the market process towards more effective and/or efficient use of resources. Therefore I hold that—admitting a sense of comfort and relief— there is theoretical ground to suggest that “re-distributive” ventures do not represent entrepreneurship. Entrepreneurship leads to improved use of resources in the economic system as a whole, and the re-distributive ventures in Figure 1:3 do not fulfill that criterion. Portraying the possible outcomes the way I have done it in Figure 1:3 is, of course, still a radical simplification. Outcomes are described as dichotomous and no explicit time horizon was introduced. Further, outcomes on only two out of many possible levels (for example, venture, firm, industry, region, nation, world) were discussed. In practice, assessing exactly where individual ventures fit into this framework would in many cases be very difficult, and contingent on the time perspective. For example, if one enjoyed the luxury of dealing with an economy characterized by a perfect and just institutional framework it would be easy to argue that re-distributive ventures equal illegal ventures. Regrettably, we will have to live with the fact that in real economies “legal, yet re-distributive” and “illegal, yet, socially beneficial” ventures are both possible. Nonetheless, I think it is useful to highlight the distinctions made here and to note that as theoretical categories not only “success ventures” but also “catalyst ventures” carry out the entrepreneurial function in the economy, whereas neither “failed ventures” nor “re-distributive ventures” fulfill this role.

Degrees of Entrepreneurship? I said earlier that the inclusion of imitative entry called for a discussion of “degrees” of entrepreneurship (cf. Davidsson, 1989a; Schafer, 1990; Tay, 1998). Realizing the variations in scope and impact of “competitive behaviors that drive the market process” it seems natural to treat entrepreneurship not as a dichotomous variable, but to say that some ventures show more entrepreneurship than others. But what should be the criterion by which we judge the degree of entrepreneurship? There are at least three possibilities:

What is Entrepreneurship?

15

The degree of (direct and indirect) impact on the economic system. If we choose this criterion we stay true to our definition of entrepreneurship as the competitive behaviors that drive the market process. In a theoretical discussion of entrepreneurship, then, this should be the preferred criterion, simply because it is the most correct one. For empirical research practice, the criterion has severe shortcomings because impact can only be assessed after the fact and not in real time, and because even then it can be very difficult to obtain even roughly correct estimates of total direct and indirect effects on a complex economic system. This, however, are problems that should bother us in the next chapter rather than this one. A variation (or an indicator) of the “degree of impact” criterion is the criterion “amount of wealth created”. Needless to say, this suffers from the same kind of assessment problems. The degree of novelty to the market. This is intuitively appealing in the sense that what is more creative is seen as a higher degree of entrepreneurship. Although the above-discussed problem of comparing very different kinds of novelty pertains to this criterion it has the advantage that it is not totally impossible to assess in real time. The main downside is the following. Innovative new activities that are successful are likely to have larger market impact on average. However, it may actually be more difficult for an innovative venture to be successful at all. There is no guarantee that a high degree of novelty ascertains market effect—history is full of weirdo inventions that nobody wanted! Some seemingly marginal innovations revolutionize markets and create great private and societal wealth whereas some radical innovations have marginal impact or fail altogether. Therefore, when market effect is part of the definition of entrepreneurship the degree of novelty is at best a rough proxy for degree of entrepreneurship. The degree of novelty to the actor. Sometimes you here expressions like “That was very entrepreneurial of you” or “That was a very entrepreneurial move for that firm”. Presumably, this means that the action was radically different from what that actor has done before. The problem is that the same action was not necessarily very novel or valuable as the market sees it. Relating the degree of entrepreneurship to the history of the actor rather than to the market in this way has highly undesirable consequences. For one thing, this type of criterion actually makes previous inactivity or conservatism increase an actor’s potential for showing a high degree of entrepreneurship! Moreover, it is a criterion that regards it more entrepreneurial to do something totally unrelated to one’s prior experience. Theories as well as empirical findings suggest this may not be a wise move (Barney, 1991; Sarasvathy, 2001; Shane & Venkataraman, 2000). I would therefore discourage the use of this kind of criterion for “degree of entrepreneurship”. In all, although there is a conceptual need for discussing “degrees of entrepreneurship”, there is no easy or straightforward way to actually assess such variation. But if theorizing and researching were easy tasks, they wouldn’t be much fun! Of the less-than-perfect but available alternatives, the degree of impact on the economic system is the criterion that best matches the definition of entrepreneurship that I have proposed. Degree of novelty (either to the market or to the actor) is better regarded as a possible cause of variations in the degree of entrepreneurship (or impact of entrepreneurship) than being a direct measure of such variation.

16

RESEARCHING ENTREPRENEURSHIP

SUMMARY AND CONCLUSION There are almost innumerable suggestions in the literature concerning what entrepreneurship really is. Noting that no one can claim to have the correct answer, I have proposed that defining entrepreneurship as the competitive behaviors that drive the market process has much to commend it. I think so for the following reasons. This definition emphasizes behavior rather than assuming a dispositional stance that has proven largely unfruitful (Gartner, 1988). It also includes an outcome that is successful or influential. Jointly this implies that the processes of discovery and exploitation are included and that mere contemplation over radical ideas, or the introduction of completely vain innovation, are not examples of entrepreneurship. Further, the definition restricts entrepreneurship to a market context, which gives a more precise and coherent characterization of this phenomenon. At the same time, the definition is permissive in that it does not take a restrictive stand on purposefulness, innovation, organizational context, or ownership and personal risktaking. Most importantly, it links micro to macro by portraying entrepreneurship as micro-level phenomenon with important effects on more aggregate levels. Finally, relative to many other alternatives, I would argue that the suggested definition has advantages in terms of being clearly delimited, logically coherent, and easy to communicate. Of course, my arguments will not convince everybody. To those who want entrepreneurship to mean “anything that concerns independent businesses” I can only say “I’m sorry!” Our interests have a certain degree of overlap, but our views on the entrepreneurship phenomenon are fundamentally different. Therefore, much of the remainder of this book may be of less value for such readers. Other aspects of the perspective I have outlined, which may be hard to swallow for some, could be the inclusion of an outcome criterion and the exclusion of organizational change per se as entrepreneurship. Here I am more optimistic, and ask doubtful readers to please try to make it through the next chapter. As I see it, there is good reason to be hopeful that our differences will be sorted out there. Stay tuned!

CHAPTER 2

ENTREPRENEURSHIP AS RESEARCH DOMAIN

WHY DISTINGUISH BETWEEN THE PHENOMENON AND THE RESEARCH DOMAIN? Now that we have devoted an entire chapter to discussing what entrepreneurship is, there shouldn’t be much need for a chapter delineating the research domain “entrepreneurship”, should there? Entrepreneurship as a research domain aims at better understanding of the phenomenon we call “entrepreneurship”, so now that we “know” what it is, why not just go out and study it? Paradoxically, the research domain cannot be equated to the study of empirical cases known to qualify under the definition of entrepreneurship that we discussed in the previous chapter. How can that be? Most importantly, although including an outcome criterion is desirable when we discuss entrepreneurship as a societal phenomenon, it becomes a burden when we think of entrepreneurship as a research domain. This is because we have to be able to study entrepreneurship as it happens, before the outcome is known. It would be awkward indeed not to know until afterwards whether one was doing “entrepreneurship research” or not. It would also be a bit hard on the researcher to require that every empirical study of “entrepreneurship” should await and assess the outcome on every relevant level. Researchers must be allowed to go deeply into aspects of the process without following up on the outcomes—and still be acknowledged for doing “entrepreneurship research”. That is, attempts to offer buyers new choices should suffice. A second very important reason for making a distinction between the phenomenon and the research domain is that previous and current entrepreneurship practice does not necessarily have all the answers needed to develop normative theory about entrepreneurship. That is, there may be better ways to learn meaningful things about entrepreneurship than finding real cases of “best practice”. To study what successful entrepreneurs have done is important, but an even more important and interesting question is what could be done. As entrepreneurship scholars we should be able to answer such questions, too, if we are the experts at abstracted

18

RESEARCHING ENTREPRENEURSHIP

sense making that we claim to be (Davidsson, 2002). This implies that the research domain should include purely theoretical development as well, and that empirical entrepreneurship research may be well advised to study also induced entrepreneurial situations, such as experiments or simulations (cf. Baron & Brush, 1999; Fiet & Migliore, 2001; Sarasvathy, 1999a). Yet other reasons for distinguishing between the phenomenon and the research domain also deserve mentioning. The behavior-plus-outcome definition lures one into a retrospective view that compresses time and de-emphasizes the process aspects of entrepreneurship. It may therefore be advisable to have a domain delineation that explicitly highlights the process nature of entrepreneurship. To study the processes as they happen is important in order to avoid selection and hindsight biases; topics we will revert to in chapters to come. Further, although the inclusion of a socially beneficial outcome clarifies the role of entrepreneurship in the economy, it may have detrimental effects on the long term credibility of entrepreneurship research in political and fellow academic circles if we portrayed the micro-processes that we study as “always good”. When the creation of new economic activity is studied real time or the outcomes for other reasons have not been carefully assessed, it is advisable for entrepreneurship researchers to have an open attitude to the possibility of different types of outcomes on different levels.

PREVIOUS ATTEMPTS AT A DOMAIN DELINEATION I am certainly not the first to suggest that we need a delineation of the research domain, and not just a definition of the phenomenon. So maybe we can get some help here? The (American) Academy of Management Entrepreneurship Division Domain Statement (cited from Gartner, 2001) reads as follows: Specific Domain: the creation and management of new businesses, small businesses and family businesses, and the characteristics and special problems of entrepreneurs. Major topics include: new venture ideas and strategies; ecological influences on venture creation and demise; the acquisition and management of venture capital and venture teams; self-employment; the ownermanager; management succession; corporate venturing and the relationship between entrepreneurship and economic development.

Well, on second thought...maybe not? As I see it, there is absolutely nothing wrong with any of the areas of interest listed above. Neither is there anything wrong with a researcher embracing all of them. I tend myself to have a keen interest in most of what is included in the statement, as evidenced by the research I have conducted and published over the years. But over those same years I have become increasingly concerned about including them all under the same entrepreneurship label. That is, I share the fear that it is precisely this kind of all-inclusive delineation that gives the entrepreneurship domain a “hodgepodge” or “potpourri” appearance, which hinders theory development and academic legitimacy (Gartner, 2001; Low, 2001; Shane & Venkataraman, 2000). Referring back to Chapter 1, the Entrepreneurship Division Domain Statement is a disharmonic mix of the “independent business” view (“management of...small businesses and family businesses”; “self-employment; the owner-manager; management succession”) and the “micro-level novel initiative” view (“creation of...new businesses”; “new

Entrepreneurship as Research Domain

19

venture ideas and strategies; ecological influences on venture creation”; “corporate venturing”). Despite probably being a fair description of the research interests members of that association represent—and, again, there is nothing wrong with those interests—the Entrepreneurship Division domain statement is not a very effective starting point for optimal knowledge accumulation concerning the phenomenon we were dealing with in Chapter 1. Many readers may have been surprised—and more than so—that I did not include Shane & Venkataraman’s (2000) entrepreneurship definition in the opening of the previous chapter. This would seem a peculiar omission as theirs has arguably been the most influential conceptual contribution to entrepreneurship research in recent years (possibly equaled by Sarasvathy, 2001). The reason is that Shane & Venkataraman (2000) wisely suggested not just another attempt at defining the entrepreneurship phenomenon, but precisely the scholarly domain. So here is the more proper place to discuss their definition of the field of entrepreneurship, which reads: [T]he scholarly examination of how, by whom, and with what effects opportunities to create future goods and services are discovered, evaluated, and exploited (Venkataraman, 1997). Consequently the field involves the study of sources of opportunities; the processes of discovery, evaluation, and exploitation of opportunities; and the set of individuals who discover, evaluate, and exploit them (p. 218).

They further point out the following three sets of research questions as especially central: 1) why, when and how opportunities for the creation of goods and services come into existence; 2) why, when and how some people and not others discover and exploit these opportunities; and 3) why, when and how different modes of action are used to exploit entrepreneurial opportunities. In the subsequent dialogue they agreed with Zahra & Dess (2001) that the outcomes of the exploitation process represent a fourth important set of research questions, adding that outcomes on the level of industry and society should be considered as well (cf. Venkataraman, 1996, 1997; Zahra & Dess, 2001). As regards antecedents of the process and its outcomes they emphasize the characteristics of individuals and opportunities as the first-order forces explaining entrepreneurship and hold that environmental forces are second order (Shane & Venkataraman, 2001). They describe their approach as a disequilibrium approach (cf. Shane & Eckhardt, 2003). They highlight variations in the nature of opportunities as well as variations across individuals. In short, they depict the economy as fundamentally characterized by heterogeneity. Further, they point out that entrepreneurship does not require, but can include, the creation of new organizations (cf. Simon in Sarasvathy, 1999b, pp. 11; 41-42; Van de Ven, 1996). I have detailed elsewhere (Davidsson, 2003a) the many merits I think this domain delineation has, and will not repeat all of that here. Suffice it to say that it is largely in line with the entrepreneurship definition we discussed in Chapter 1, and that the many positives arguably made Shane & Venkataraman’s framework the best effort so far to delineate entrepreneurship as a distinct research domain. One of the few debatable points is the general primacy given to the individual and the “opportunity”. This does not seem go give much room for entrepreneurship research on more aggregate levels of analysis (cf. Zahra & Dess, 2001). A more important question mark, perhaps, is their adopting Casson’s (1982) definition of opportunity

20

RESEARCHING ENTREPRENEURSHIP

as “those situations in which goods, services raw materials and organizing methods can be introduced and sold at greater than their cost of production” (cf. Singh, 2001). I would argue that this makes their definition get stuck only halfway towards clearly distinguishing between the phenomenon and the domain. Shane & Venkataraman (2000) hold that, among other things, we should study with what effects “opportunities” are exploited. But with Casson’s definition of opportunity, entrepreneurship becomes characterized by certainty rather than uncertainty regarding one important aspect of the effects of the pursuit of opportunity: it is profitable. As I see it, Casson’s definition is compatible with the definition of the entrepreneurship phenomenon that we developed in the previous chapter, but largely unhelpful for entrepreneurship as a research domain. This is because defining “opportunity” this way is inconsistent with having the outcomes of entrepreneurship as an open research question. This apparent weakness of Shane & Venkataraman’s exposition points at a more general problem in the entrepreneurship literature, namely that “opportunity” is becoming a central concept but one which often is illconceptualized or applied in an inconsistent manner. We will have reason to come back to this problem later on. Several of Bill Gartner’s many writings on entrepreneurship (for example, Gartner, 1988, 1990, 1993, 2001; Gartner & Brush, 1999) can also be regarded more as attempts to delineate the field of research than defining and describing the phenomenon. Gartner’s view—which he is careful to present as a suggestion for redirection rather than a formal “definition” (Gartner, 1988)—is that entrepreneurship is the creation (or emergence; cf. Gartner, 1993) of new organizations. This choice of focus has two origins. One was a perceived lack of treatment of organizational emergence in organization theory. Somehow organizations were assumed to exist; theories started with existing organizations (cf. Katz & Gartner, 1988). The other was a frustration with the pre-occupation that early entrepreneurship research had with personal characteristics of entrepreneurs. For these reasons, Gartner (1988) suggested that entrepreneurship research ought to focus on the behaviors in the process of organizational emergence. This view certainly has a lot to commend it. For one thing, it has a clearly defined focus, addressing terrain that economics as well as management studies have treated in a step-motherly fashion. This clear focus gives promise of giving unique contribution and avoiding over-extending the field of entrepreneurship research. Further, Gartner’s view has a strong process orientation. The main problem I have with Gartner’s (1988) approach is that whereas organizing is an important aspect of the exploitation process, he does not emphasize the discovery process (cf. Shane & Venkataraman’s domain delineation above). Further, his approach directs no or only cursory attention to the possibility of alternative modes of exploitation for given “opportunities” (Shane & Venkataraman, 2000; Van de Ven, Angle & Poole, 1989). If interpreted as a delineation of the (entire) research domain his take on entrepreneurship appears overly narrow in these regards. In short, I see Gartner’s focus as the natural task for an organization theorist to take on within a somewhat broader domain. Below I will try to create precisely that: a somewhat broader, yet sufficiently precise, domain delineation. What an incredibly pretentious thing to do! Well, the

Entrepreneurship as Research Domain

21

reason that I dare try is that I can stand on the shoulders of Gartner and Shane & Venkataraman, as well as their predecessors. The little trick I will attempt below is the sewing together of their respective perspectives while ironing out the little wrinkles I think I’ve found, in order to arrive at a coherent domain delineation, tailor-made for entrepreneurship research.

MY SUGGESTED DOMAIN DELINEATION First, I take from Gartner (1988) the idea that entrepreneurship research should study behavior in the process of emergence. That’s three very important components right there: behavior, process and emergence. Based on Shane & Venkataraman (2000) I take the point that we should distinguish between two sub-processes: discovery and exploitation (I include “evaluation” in the discovery process). Further, in line with the view of entrepreneurship that we developed in Chapter 1, I agree with their notion that entrepreneurship research should not study only or primarily the emergence of new (independent) organizations, but the emergence of new market offerings (they say “new goods and services”) through different modes of exploitation. From Venkataraman (1997), Shane & Venkataraman (2001) and Zahra & Dess (2001) I also adopt the idea that entrepreneurship research should study a variety of outcomes on different levels. The final element I take from Shane & Venkataraman (2000) is the idea that entrepreneurship research should adopt as a fundamental assumption that the economy is characterized by heterogeneity (they discuss this under the “disequilibrium” label). To this I only need to add two little pieces, which I have touched upon already. The first is to adopt the additional fundamental assumption that the economy is also characterized by uncertainty. The second is that empirical entrepreneurship research need not and should not be restricted to the study of empirical cases known to qualify as “entrepreneurship” à la our definition of that phenomenon in the previous chapter. Entrepreneurship research should also study failure and induced processes of emergence. Piecing it all together, I arrive at the following (cf. Davidsson, 2003a): Starting from assumptions of uncertainty and heterogeneity, the domain of entrepreneurship research encompasses the study of processes of (real or induced, and completed as well as terminated) emergence of new business ventures, across organizational contexts. This entails the study of the origin and characteristics of venture ideas as well as their contextual fit; of behaviors in the interrelated processes of discovery and exploitation of such ideas, and of how the ideas and behaviors link to different types of direct and indirect antecedents and outcomes on different levels of analysis.

Now, I can assure that there is no shortage of information hidden in those few lines, so it would be really nice if at this point the reader could stop, reflect, reread—and perhaps start counter-arguing or asking follow-up questions. After playing that game for a couple of rounds, I’d be delighted if the reader imbibed my own elaborations below.

22

RESEARCHING ENTREPRENEURSHIP

Uncertainty and Heterogeneity It could be debated whether one should really let this type of assumptions restrict a research domain. My rationale for including them is that I firmly believe that a theory or a research design that assumes that economic aggregates (such as an industry, or demand) are made up of the sum of identical micro-level entities, is unlikely to be a fruitful starting point for understanding or researching the entrepreneurship phenomenon. For example, individuals are heterogeneous with respect to experience, skills and cognitive capacity (Cohen & Levinthal, 1990; Conner & Prahalad, 1996; Shane & Venkataraman, 2000; Van de Ven et al, 1989) and also have heterogeneous motivations (Birley & Westhead, 1994). Two important aspects of organizational heterogeneity are governance structure (Coase, 1937; Foss, 1993; Williamson, 1999) and resources (Barney, 1991; Cohen & Levinthal, 1990; Collins & Montgomery, 1995; Foss, 1993; Galunic & Rodan, 1998; Greene, Brush & Hart, 1999; Penrose, 1959; Teece, Pisano & Shuen, 1997). Whether or not a new venture evolves within an existing organization the external environment in a broader sense will also be heterogeneous (Baumol, 1990; Chandler & Hanks, 1994) and the characteristics of the external environment may have profound effects on what venture ideas are attractive and likely to succeed (Zahra & Dess, 2001). Heterogeneity also occurs over time. Individuals and organizations learn and change over time and whether or not they choose to remain in the “same” environment, the characteristics of the environment are not stable, either (Aldrich, 1999; Aldrich & Martinez, 2001; Miner & Mezias, 1996). It follows from all this heterogeneity that the universe of perceptible and profitable opportunity is not the same for all individuals or organizations, and that therefore they will come up with different venture ideas and different exploitation strategies. Importantly, they will also have different views on what constitutes a successful or acceptable outcome (Gimeno, Folta, Cooper & Woo, 1997; Venkataraman, 1997). Neither do I think it is illuminating for the understanding of this phenomenon to start from a view of reality as characterized by certainty and calculable risk alone. I’d be the last to argue that all decisions for all actors are non-calculable. However, the situations in which behaviors aimed at creating new economic activity are undertaken often have this characteristic. That is, information collection and processing, careful planning and calculation cannot give a conclusive and reliable answer as to whether something will be successful or not. Only (trial) implementation will tell. In short, such situations have a substantial element of genuine, Knightian uncertainty (Knight, 1921). That is, the future is not only unknown, but also unknowable (Sarasvathy, Dew, Velamuri & Venkataraman, 2003). Here I disagree with the same Kirzner (1973) that I leant so heavily on in the first chapter. Very rarely are entrepreneurial situations certain in the way Kirzner portrays them. In one passage Kirzner likens entrepreneurial opportunity with realizing that a free ten-dollar bill is resting in one’s hand, ready to be grasped. If we should use the ten-dollar bill metaphor at all, I would suggest the true situation is more like spotting the bill from your balcony. From that distance one would face the (calculable) risk that the bill was for anything from one to a hundred dollars. But moreover, while you dash down the stairs it may blow away, or someone else may

Entrepreneurship as Research Domain

23

get it before you, or it may turn out upon closer look that it was not a real money note, after all. There is no way the finder can tell before she takes the decision to run down the stairs. Or maybe one should change the perspective and view opportunity as the uncertain prospect of perhaps being able to make one’s own ten-dollar bills— and get away with it? Either way, in order to understand behaviors in such situations it is important to start from a theoretical perspective that acknowledges or even emphasizes uncertainty.

Processes of Emergence; Behaviors in the Interrelated Processes of Discovery and Exploitation Shane & Venkataraman state as their point of departure (2000, p. 217) that “For a field of social science to have usefulness it must have a conceptual framework that explains and predicts a set of empirical phenomena not explained or predicted by conceptual frameworks already in existence in other fields.” One of Gartner’s (1988; 1993; 2001) great strengths is that he has pointed out one such empirical phenomenon: the process of emergence. Other fields of research simply haven’t done a very good job here, and therefore entrepreneurship research can make a real contribution if it takes on this challenge. I agree with Shane & Venkataraman (2000) that both discovery and exploitation are required for entrepreneurship to happen, and that both should be studied in entrepreneurship research. So again, I disagree with Kirzner’s (1973, p. 47) claim that “Entrepreneurship does not consist of grasping a free ten-dollar bill which one has already discovered to be resting in one’s hand; it consists of realizing that it is in one’s hand and that it is available for the grasping.” That is, he holds that entrepreneurship consists solely of discovery; exploitation is presumably “something else”. But returning to the balcony, nothing much happens if we just note that a tendollar bill seems to be lying down there, does it? How Kirzner makes restricting entrepreneurship to (instantaneous) discovery match his notion that entrepreneurship consists of the “competitive behaviors that drive the market process” beats me. There seems to be an underlying assumption in his reasoning that every actor who perceives an opportunity not only knows with certainty that it really is an opportunity, but also necessarily acts upon it. Entrepreneurship researchers know that such is not the case. Many of us just have to exercise a little introspection to realize that. Our emphasis on the interrelated processes of discovery and exploitation as new economic activities emerge implies that a very central set of research questions for entrepreneurship research concerns what individuals and other economic entities actually do when they initiate, refine, and realize ideas for new business ventures. With a slight rewrite of Shane & Venkataraman’s third central research question we get: Why, when and how are different modes of action used to discover and exploit venture ideas?

This area needs much more investigation over and above the tentative steps that have been taken so far (e.g., Bhave, 1994; Carter, Gartner & Reynolds, 1996;

RESEARCHING ENTREPRENEURSHIP

24

Chandler, Dahlqvist & Davidsson, 2002; 2003; Delmar & Shane, 2002; Fiet & Migliore, 2001; McGrath, 1996; Samuelsson, 2001; Sarasvathy, 1999a) The term discovery may be suspected to reflect an objectivist view on venture ideas. That is, the term seems to suggest that they somehow exist “out there”, ready to be discovered. This is not a perspective I purport. Rather, like Shane & Eckhardt (2003), I use the term “discovery” to maintain consistency with prior literature, despite its potentially misleading connotations. Discovery refers to the conceptual side of venture development, from an initial idea to a fully developed business concept where many specific aspects of the operation are worked out in great detail, especially as regards how value is created for the customer and how the business will appropriate some of the value (Amit & Zott, 2001; de Koning, 1999b, p. 121). Importantly, discovery is a process—the venture idea is not formed as a complete and unchangeable entity at a sudden flash of insight. Thus, it includes not only what is elsewhere called “idea generation”, “opportunity identification” and “opportunity detection”, but also “opportunity formation” and “opportunity refinement” (Bhave, 1994; de Koning, 1999a, 1999b; Gaglio, 1997). Exploitation is a negatively loaded word is some contexts, and may therefore evoke negative associations. In the present context, I would suggest it is a netral term referring to the decision to act upon a perceived opportunity, and the behaviors that are undertaken to achieve its realization. The exploitation process deals primarily with resource acquisition and co-ordination, as well as market making (see Shane & Eckhardt, 2003; cf. also Sarasvathy, 1999a; Van de Ven, 1996). This includes all research questions pertaining to the organizing of new ventures, that is, the research agenda that Gartner (1988; 2001) emphasizes. Exploitation thus simply means the attempted realization of ideas. Like discovery, exploitation is a process that may or may not lead to the attainment of profit or other goals. The emphasis on the interrelatedness of the two processes is based on empirical insights (Bhave, 1994; Sarasvathy, 1999a, 2001). I think discovery and exploitation are best conceived of as overlapping processes. This is what Figure 2:1 tries to portray.

Figure 2:1

The interrelation between discovery and exploitation

For example, an entrepreneurial process may start with an individual perceiving what she thinks is an opportunity for a profitable business [discovery]. In the efforts to make this business happen, contacts with resource providers and prospective customers [exploitation] make it clear that the business as initially conceived will

Entrepreneurship as Research Domain

25

not be viable [feedback to discovery]. The individual changes the business concept accordingly [discovery] and continues her efforts to marshal and coordinate the resources needed for the realization of the revised business concept [exploitation]. Although the above process starts with an element of discovery, this is not necessarily always the case. Empirical research suggests that venture creation processes can follow almost any sequence (Carter et al, 1996; Gartner & Carter, 2003), and Bhave’s (1994) study indicates that the insight (or discovery) that a problem solution one has developed for one’s own needs may present a business opportunity often comes rather late in a process that initially did not have the creation of a new venture as a goal.

Real or Induced, and Completed as Well as Terminated These are issues that we dealt with already in the beginning of the present chapter. The practicing entrepreneurs the world has seen so far do not necessarily have all the answers. That is, pure theory development and laboratory research methods may sometimes prove better avenues to arrive at normatively valid results and theories (Fiet, 2002). As a case in point, one of the most interesting and influential developments in recent years, namely Sarasvathy’s reasoning on effectuation vs. causation processes, emanates from research on induced (or hypothetical) entrepreneurial processes (Sarasvathy, 1999a, 2001). Further, if we were to study successfully completed cases only, there is no telling whether terminated cases shared the same characteristics. This is especially important with regard to risk-taking and its correlates. Risk-taking should increase the span of possible outcomes. That is, the entrepreneur who takes risks should be rewarded with a greater likelihood of great success. At the same time, however, that entrepreneur incurs an increased risk of making a big splash. If our research design censors the terminated cases, we will systematically misinterpret the effects of risky strategies and actions.

Across Organizational Contexts Again, this has been thoroughly dealt with already. In Chapter 1 we parted from the “independent business” perspective on entrepreneurship. Shane & Venkataraman (2000) make a major point of this issue, emphasizing different modes of exploitation (such as internal venturing vs. the setting up of a new firm) as a core set of research questions for entrepreneurship research. This is in apparent conflict with Gartner’s perspective. It is important to note, however, that Gartner’s “creation of new organization” should not necessarily be read as “creation of new, owner-managed firms”. Gartner (1988, p. 28) explicitly discusses internal venturing. Although he— arguably with good reason—regards the emerging new firm as a particularly promising arena for studying it, his interest is in “organizing” in the Weickian sense (Gartner, 2001, p. 30, cf. Gartner & Carter, 2003), not necessarily the creation of formal and legally defined organizations. Across organizational contexts has additional meaning beyond opening up for the study of discovery and exploitation both in emerging and existing firms, small

26

RESEARCHING ENTREPRENEURSHIP

and large, owner-managed or else. This is also where we can start inviting back to the party those organizational changes in quadrant II of Figure 1:2, which in the previous chapter were defined as not being instances of entrepreneurship. Change in organizational context as explicitly related to the creation of new, market-related activity is clearly within the entrepreneurship research domain. Studies referred to by Ucbasaran, Westhead & Wright (2001. p. 64) showing that management buy-outs are followed by increased development of new products, are therefore examples of entrepreneurship research. Likewise, longitudinal empirical tests of Stevenson’s argument that certain organizational changes would facilitate entrepreneurship in established organizations (Stevenson, 1984; Stevenson & Jarillo, 1986) would clearly be instances of entrepreneurship research (Eliasson & Davidsson, 2003). Those latter examples presume a shift of “organizational context” within the same organizational entity. The emerging venture may also lead a life that cuts across several different organizations. What originates as an idea by an independent inventor may be acquired into an existing small firm, which is later acquired by a large organization, which decides to spin out this particular part of their business operations. This points to the need for studies of processes of emergence that use the venture idea itself as the unit of analysis (Davidsson & Wiklund, 2001). Such studies would follow samples neither of individuals nor of organizations, but precisely new, emerging activities—i.e., venture ideas and what evolves around them—from their conception and through whatever changes in human champions and organizational contexts might occur along the way. In some cases what originated as a de novo start-up is transferred to an existing firms; in other cases what originates within a firm may be spun out at an early stage. This is something we will also return to in later chapters.

New Business Ventures; Venture Ideas and Their Contextual Fit “Business ventures” here should be interpreted broadly. It includes independent start-ups as well as new internal ventures, and also new market offers that are so limited that the actors involved do not necessarily conceive of them as entire “ventures”. However, in line with our putting entrepreneurship in a market context in the previous chapter the suggested domain delineation is restricted to new business ventures. The reader may have noted that I have put “opportunities” within quotation marks in several places above, and started to sneak in the concept Venture Idea in its stead. This is very, very intentional, of course. The term “opportunity” refers to something not yet realized. The increased use of the this term in entrepreneurship research therefore signals the sound development that the field is really turning towards a focus on emergence, rather than starting from existing firms and established business founders. However, there is a huge linguistic problem with adopting “opportunity” as a central concept in entrepreneurship research. By almost any definition, an opportunity is something known to be favorable. But we just said we did not like that, by making uncertainty a fundamental assumption in our domain delineation. That is, the use of the term “opportunity” for an unproven venture idea is fundamentally opposed to acknowledging uncertainty as an inescapable aspect of

Entrepreneurship as Research Domain

27

the environment of the emerging activity and/or organization that the entrepreneurship scholar tries to understand. At the time, the actors cannot know whether or not they pursue an “opportunity”. Neither can the researcher. “Opportunity” also seems to be a notion that arouses a lot of controversy among entrepreneurship researchers. Some seem to regard opportunity as objectively existing in the environment, whereas others hold that opportunity is created by the entrepreneur (cf. Davidsson, 2003a). To a considerable extent this is a semantic battle, where a lot of the confusion arises from the fact that the term “opportunity” is used in at least four different ways in the entrepreneurship literature: i) for a set of external conditions known in retrospect to be favorable (to some people) for the successful discovery and exploitation of new business activities, ii) for a set of external conditions thought (by some people) but not proven to be existing and favorable for the successful discovery and exploitation of new business activities, iii) for specific new venture initiatives known in retrospect to be viable, and iv) for specific new venture initiatives that are currently being pursued but whose viability is not yet proven. As I see it, “opportunity” should really only be used for the i) and possibly the iii) category. I suggest Venture Idea as a more appropriate and less controversial term for category iv) [which over time will also include category iii) as a subset]. A venture idea may start as a very rudimentary idea of a technically possible product, or the perception of an unsolved problem that a market segment would be willing to pay to get solved, if one could find a solution to the problem. Over time it may be changed, honed, and elaborated to qualify as what others would called a business concept or a fully developed (conception of a) business model. The venture idea is, so to speak, the focal object of the discovery and exploitation processes. Referring back to Figure 1:2, venture ideas are ideas for new products or services or bundles thereof; introducing a new price/value relation; imitative entry, and new markets. Relating also to the heterogeneity issue, this shows that venture ideas come in different flavors. A seriously under-research area, I would argue, concerns the characteristics of new venture ideas and how these characteristics relate to antecedents, behaviors and outcomes. Samuelsson (2001; 2004) represents one of the few entrepreneurship studies that have explored the nature and effects of characteristics of venture ideas, and followers are needed. Although an abundance of studies have tried to assess the characteristics of entrepreneurs, very few have focused on the characteristics of the venture ideas they pursue. Interestingly, this disproportionate interest in the individual is shared by diffusion research, where only about one percent of the close to 4,000 studies have focused on the characteristics of the innovation, whereas more than half of them focus on the individuals who adopted them (Rogers, 1995). An explanation for this might be the general human tendency that psychologists have dubbed “the fundamental attribution error”. This is to seek explanations to events in terms of the characteristics of the individuals involved, also when structural or situational factors are the true determinants (Ross, 1977). Researchers beware! The above said does not mean that entrepreneurship research should forget about “real opportunities”. For the first central research question that Shane &

28

RESEARCHING ENTREPRENEURSHIP

Venkataraman (2000) pointed out, we may be well advised to maintain this objectivist stance. With a slight rewrite, this central research question reads: Why, when, where, how and for whom does opportunity for the creation of new goods and services come into existence?

I have here added “where” and “for whom”. I have also changed “opportunities” to “opportunity” as an uncountable to veer away from the view that a finite number of well specified, ready-to-use opportunities exist out there, waiting to be discovered (such as lost ten-dollar bills, for example). Questions concerning “real” opportunity can be asked for different types of entities or levels of analysis, for example for nations, regions or other spatial units over time or across space, as well as for organizations, industries or population sub-groups. Asking such questions is a prerequisite for building strong theory about where opportunity will emerge in the future. Building such theory is a challenging but important aspect of scholarship in entrepreneurship, which feeds directly into entrepreneurship education (cf. Davidsson, 2002) where learning where to look for opportunity should be one of the most central features (cf. Drucker, 1985; Vesper, 1991). However, proven opportunity can only be studied in retrospect. And it gets worse. When you think of it, it is impossible to know the universe of not-yet-developed, but potentially viable, venture ideas. In one context there may be abundant opportunity but little actual venturing, because of cultural inhibitions or because non-entrepreneurial opportunity is also abundant. In another context there may be high levels of desperate venturing due to lack of other alternatives. Therefore, not even the number of venture ideas that are both acted upon and proven successful is a direct measure of objective opportunity density. It is inescapable that whatever measure is used for opportunity density, it will be a proxy measure. Finally, as regards contextual fit, questions of this kind also arise from the heterogeneity issue. Seriously under-researched questions concern fit between individuals’ prior knowledge and (information about) the new venture idea (e.g. Cooper, Folta & Woo, 1995; Fiet, 1999; Shane, 2000); relatedness between organizations’ prior knowledge, resources or capabilities and (information about) the new potential venture (e.g., Cohen and Levinthal 1990; Teece et al, 1997; Van de Ven 1996); fit between existing resources and what strategies can lead to the venture’s success (e.g., Chandler & Hanks, 1994) and fit between characteristics of the new potential venture and current user practices (e.g., Raffa, Zollo & Caponi, 1996). Empirically based knowledge on these issues is limited, which means abundant opportunity for research contributions.

Antecedents and Outcomes on Different Levels of Analysis This should be easy enough. It is standard research practice to ask questions about antecedents and outcomes. But let’s see. First, generalizing Shane & Venkataraman’s (2000) second research question to several levels of analysis, and substituting “venture ideas” for “opportunities”, we arrive at the following:

Entrepreneurship as Research Domain

29

Why, when and how do individuals, organizations, regions, industries, cultures, nations (or other units of analysis) differ in their propensity for discovery and exploitation of new venture ideas?

This is straightforward enough. One implication is that entrepreneurship research can be conducted on any level of analysis as long as antecedents on that level are explicitly related to discovery and exploitation of new venture ideas. Thus, again, we can re-invite the organizational issues in quadrant II of Figure 1:2. The relationships between organizational characteristics and change on the one hand, and discovery and exploitation of new venture ideas on the other, are important questions for entrepreneurship research. However, those who think narrowly of entrepreneurship as dealing with the firm level of analysis should reflect on the fact that there are many other levels of analysis that are of equal relevance on the entrepreneurship research agenda. The opportunities and challenges involved in researching entrepreneurship on those different levels of analysis will be the central theme in chapters to come. The “propensity” in the question should not be limited to quantitative but also to qualitative differences. For example, due to the distinction between “opportunitybased” and “necessity-based” entrepreneurship, nations and regions may have similar firm start-up rates for very different reasons, and representing very different levels of real, profitable opportunity (Davidsson, 1995a; Reynolds, Camp, Bygrave, Autio & Hay, 2001). The same problem is likely to occur on the organizational level. A firm desperately struggling for its survival may take more new initiatives than a firm that is doing well, even if better objective opportunity is available for the latter (March & Sevón, 1988). To Shane & Venkataraman’s (2000) three original questions we can add a fourth: What are the outcomes on different levels (e.g., individual, organization, industry, society) of efforts to exploit venture ideas?

The first implication here is that the interest of entrepreneurship research is very, very far from restricted to the question of the financial performance of firms. At this stage of reading, migrants and visitors from strategic management should start to understand why the notion of entrepreneurship as a sub-field of strategy is, should we say, somewhat incomplete. As we have delineated entrepreneurship research here, the strategic management questions that are also entrepreneurship questions constitute a corner of the totality of the entrepreneurship domain. It should be clear by now that many disciplines and sub-disciplines cover different aspects of the research domain we have delineated. It should be equally clear that no one other existing discipline or sub-discipline covers the entirety of what we here see as entrepreneurship research. I would suggest that in showing a genuine interest in outcomes on different levels, and in providing a more refined and empirically informed view on “failure”, entrepreneurship can distinguish itself from other fields and make strong contributions to social science at large (cf. Low, 2001; Venkataraman, 1997). The question of when successful venture level outcomes are and are not associated with successful outcomes on the societal level, and vice versa, is highly relevant but seldom asked. It is conceivable that under certain circumstances the successful

30

RESEARCHING ENTREPRENEURSHIP

pursuit of ideas for new ventures does not benefit society (cf. Baumol, 1990). It is also possible to conceive of a situation where entrepreneurial efforts on the whole benefit society while at the same time the most likely outcome on the micro-level is a loss—and that therefore the rational decision is to refrain from entrepreneurship (cf. Olson in Sarasvathy, 1999b, p. 35). Both of these situations represent important problems that entrepreneurship research can help societies to solve or avoid. The question of differential outcomes on different levels can also be asked from the perspective of the corporate manager: when and why does and does not new venturing—successful or not on the venture level—contribute to company performance? Again, because of potential learning and cannibalization the answer is not a simple one to one relationship between venture- and organizational level outcomes. Referring back to Figure 1:3, the issue of catalyst ventures, then, is of particular interest. Too narrow or simplistic a view on “failure” may lead to gross misrepresentation of the benefits of attempts to create new business activity, on micro- as well as aggregate levels. What in a narrow perspective appears to be a “failure” may instead be a beneficial “catalyst” either because those directly involved in the “failure” learn for the future or because others imitate. A possible outcome of deeper and more refined research into apparent “failure” is that pure failure as defined in Figure 1:3 is far less usual than previously thought (cf. Gimeno et al, 1997, pp. 69, 72). I think one of the first things entrepreneurship scholars should try to get rid of is the bias against failure. In addition to the “catalyst” potential, both theory and empirical evidence actually suggest that experimentation that may end in failure as well as the demise of less effective actors are necessary parts of a well-functioning market economy (Davidsson et al, 1995; Eliasson, 1991; Reynolds, 1999; Schumpeter, 1934). We should not forget that there are qualitatively different types of outcomes, too. Entrepreneurial processes do not only have financial outcomes, and affect not only those directly involved in the project. Supplementary outcome assessment may concern, e.g., satisfaction, learning, imitation and retaliation. For researchers who have the creativity and guts to be unconventional there are plenty of opportunities— or should I say “research ideas”?—that await your discovery and exploitation.

SUMMARY AND CONCLUSION In this chapter I have argued that even though the object of entrepreneurship research is to understand the phenomenon we call entrepreneurship, our research cannot be delimited to the study of proven empirical instances of entrepreneurship. Instead, I suggested the following domain delineation for entrepreneurship research: Starting from assumptions of uncertainty and heterogeneity, the domain of entrepreneurship research encompasses the study of processes of (real or induced, and completed as well as terminated) emergence of new business ventures, across organizational contexts. This entails the study of the origin and characteristics of venture ideas as well as their contextual fit; of behaviors in the interrelated processes of discovery and exploitation of such ideas, and of how the ideas and behaviors link to different types of direct and indirect antecedents and outcomes on different levels of analysis.

Entrepreneurship as Research Domain

31

Building on a combination and extension of earlier contributions by Gartner (1988; 1993; 2001) and Shane & Venkataraman (2000; 2001) the domain I suggest for entrepreneurship research is broader than either of these predecessors. This inclusive attitude may lead to a less distinctive domain. If one looks closely at the above domain statement, it is clear that most core research questions in entrepreneurship would fit in some existing discipline or sub-discipline. It is equally clear, however, that entrepreneurship is not in its entirety a sub-division of any one established discipline or field of research. More importantly, if left to the disciplines, there is no guarantee that a lot of research would be conducted on the most central questions of entrepreneurship, as we have here outlined that domain. Many of these questions may be peripheral to every discipline (cf. Acs & Audretsch, 2003b). Therefore, a failure to collectively cover the entrepreneurship research agenda is neither a problem nor a shortcoming on the part of the existing disciplines. When maximum knowledge development about entrepreneurship is the vantage point, however, this is a very real and important problem. This is the most important raison d’être for entrepreneurship research as a distinctive domain and research community. Now, after this long warming up it’s about time we get to the real stuff: methodrelated opportunities and challenges of entrepreneurship research. So that’s what we’ll throw ourselves over next, and stick to for the remainder of this book: empirical design and analysis issues. Oh, well, perhaps not...there was this little thing called “theory” that we have to deal with first...

CHAPTER 3

THIS THING CALLED “THEORY”

CONFESSIONS OF A SINNER I confess! I am a sinner! I haven’t always practiced what I preach as far as theory goes. Some of the projects I have been involved in, and where I have enjoyed access to excellent empirical data, haven’t been as theory-driven or theory-developing as they should. Pressed for time and in the face of intriguing empirical relationships, I have sometimes neglected the conceptual side of research. But that is really my loss. No matter how intriguing an empirical result may seem here and now, it is the sense making of theory that makes it travel through space and stand the test of time. It is theoretical interpretations that uncover the implications of empirical results, so that they can properly guide practitioner behavior and the design of continued research efforts. In short, theory is crucially important. In the absence of theory, empirical research will be poorly designed, and the results will have little meaning. Besides, I’m not the worst of sinners. As will be boasted below, one of the main contributions of my very first attempts in the field of entrepreneurship research—i.e., my dissertation project—was to increase the level of abstraction (Davidsson, 1991)1. So despite sometimes having sinned, and knowing that there are colleagues who would raise an eyebrow at my writing this particular chapter, I am a great fan of theory. Below I will try to explain why. I will start with an attempt to de-mystify theory—something which seems to be needed at least when addressing students. I will then dwell on the advantages of abstractions for some time, before turning to the different roles of theory in the research process. After that, we should be ready for discussing whether entrepreneurship needs its own theory development. Perhaps theories from existing disciplines and fields of research suffice?

THEORY IS NO MYSTERY Theory is not the opposite of reality, and not the opposite of practice. These are the first things I tell students about theory. Theory is not some mystical, unworldly exercise of ivory-tower academics. On the contrary, we all use theory all the time. There’s no escape! If I walk to the street hawker in order to buy a newspaper, I do what I do because I have an experienced-based “lay theory” which says that if I walk this and that street and make turns at specific corners, I will eventually get to the street hawker, and if I give him the right amount of money and utter the magic words, I will indeed get the evening paper I was after. But this sequence of events is not a fact until I have really gone there, paid, and received my paper. In principle, delivery could have been late, or it might have been a public holiday that I had forgotten, or the hawker might have stopped trading, or something else came up. So my actions are guided by theory. In fact, when viewed this way, all of our goal1 Reprints of this article can also be found in Krueger (2002), Storey (2000) and Westhead & Wright, (2000)

RESEARCHING ENTREPRENEURSHIP

34

directed behavior is governed by some kind of theory about the workings of the world. If no form of theory should guide action, random behavior would be the only alternative. So the issue in design and interpretation of research is not whether it is guided by theory, but how articulated and suitable for the purpose that theory is. As I see it, theories are best regarded as tools. “Science is tooled knowledge”, as Schumpeter (1954, p. 7) puts it. Scholarly theories have usually been developed by clever, hardworking people trying to make their very best. This is reason to be respectful towards such theories, and not disregard them until one is certain one has a better alternative. However, the toolmakers cannot possibly know exactly what tools are needed to solve your particular problem. It is not the toolmaker’s fault if you make a mess trying to open a can with a hammer. The researcher is well advised to rummage the toolbox a little more before trying such questionable solutions. However. sometimes the toolbox does not contain the perfect theory. This is when a sound amount of creative disrespect may be needed vis-à-vis theory. Theories are not untouchables; one might need to adapt and combine them in order to get the tools needed to solve the problem. Using theory is not a reason to stop thinking for oneself! Sometimes entirely new tools may need to be developed. However, if you ask me it is more often researchers’ ignorance or creative itches than real lack of available tools that make them try to develop genuinely new theory. Mind you, there are—and should be—more skilled craftsmen than toolmakers. Theories are not the opposite of reality, but they must be simplifications of reality. Some people get annoyed when some aspects of a theory do not hold true with respect to a specific case, or that the theory does not account for the full richness of empirical cases. This could be a sign of deficit on the part of the theory, but more likely it shows that those people simply haven’t got it. That is, they have not understood what theory is and what it is good for (and regrettably this problem is not confined to students). When a theory fits perfectly with a specific case, it is no longer a theory but an idiosyncratic description of little use for understanding other cases. A useful theory must abstract and generalize, and thus neglect many of the fine details. Theorizing involves abstraction. This is what we will turn to in the next sub-section.

THE NEED FOR ABSTRACTION AND UNDERSTANDING In all honesty, the evening paper buying theory above wasn’t much of a theory, after all. The example made the point that the behavior was guided by experiencebased guesses, not facts. It may be best described as an empirical generalization, which is a forerunner to—or a very primitive form of—theory. It has very local applicability. In order to be of use outside of the very local context, one would at least have to generalize the applicability of the money-for-paper part, and add some more general understanding of in what types of locations outlets for evening papers tend to be found. There exist different definitions and descriptions of what is needed for something to qualify as a “theory”. However, two of the elements that are usually required are: 1. 2.

A set of well-defined, abstracted concepts A set of well-specified relationships among those concepts

This is something that can also be expressed as a formula, or graphically in the form of a boxes and arrows diagram. In other words, it can be expressed as a model. Many would say that a theory is more than that. For example:

This Thing Called “Theory”

3.

35

A deeper understanding of why those relationships exist, and what they imply

As regards concepts, consider the following example. A few years ago, Husqvarna AB, which happens to be located in my current home town, introduced a radically new type of lawn mower. This machine, called the Solar Mower, is like a modern sheep, which walks the garden at random and on its own, cutting the grass little by little. Solar cells generate the power needed, and random-walk software in the small internal computer makes sure the device cuts the entire lawn, within magnetic cord demarcations. For safety and security, the device is equipped with an anti-theft alarm and cuts the grass with a piece of plastic string rather than a metal knife. Got it? Seen one? The product still exists in modified form, but it has not taken off in a big way. There are many reasons for this. However, one problem in North Sweden was that this lawn mower was not “macho” enough. Up there, tractors like machines—preferably bigger than your neighbors’—were the name of the game. In the UK, market acceptance was slow for seemingly an entirely different reason: it doesn’t make stripes! Apparently, UK homeowners want stripes, just like on the soccer fields you see on TV—and preferably straighter than your neighbours’. What do we learn from this? Don’t introduce sissy lawn mowers up north in Sweden? If you are a lawn mower manufacturer considering the UK market, don’t forget the stripes? As yet, all we have are a couple of cute little marketing anecdotes. We could try to incorporate these two events in a “lawn mower launch theory”, but that theory would just be a (long) list of historical particularities that have somehow, somewhere hampered or facilitated the market acceptance of a new lawn mower. But if we from instances like these are able to distill the abstracted concept compatibility we can see that the “macho” and “stripes” issues are in a sense aspects of the same type of problem: lacking compatibility with prevailing norms. And now we have a concept that is useful not only with respect to historical cases or to lawn mowers, but one that applies to the diffusion of any innovation. Practitioners and researchers involved in other innovations in the future now have a meaningful issue to consider, along with the issues of relative advantage, complexity, trialability and observability—the other generic attributes of innovations that theorists have abstracted from empirical instances that first might seem unique and unrelated (Rogers, 1995). That’s the power of theory. The specific manifestations of those concepts will differ, but these types of problem will always remain a potential threat to innovators’ success. As another example of the need for abstraction and understanding, consider the following example. When I first entered the field of small business research it was with an interest in the growth (and non-growth) of small firms. I soon got to view growth as an instance of entrepreneurship—a view I have subsequently revised and refined (see Davidsson et al, 2002 and Chapter 1, above). By reading a large number of empirically based studies on entrepreneurship and small firm growth, the picture I got of their determinants was something like the horrors of Figure 3:1. Now, how does one deal with this? One approach would be to cover everything and estimate (or conceptually try to tease out) all relationships. This is not feasible; we would soon lose track of the important over-riding structure and arrive at the result that “the world is complex.” This is something we probably knew from the very beginning. So this is the consequence of lack of theory: even if we measure all relevant variables and estimate all relevant relationships—which is highly unlikely to happen in the first place if we do not have theoretical insights—we will not really understand much. Another possibility is to make negligibility assumptions (cf. Musgrave, 1981). That is, the theory or the empirical study covers just a few aspects; those that we believe are the most important or those that we find most interesting

36

RESEARCHING ENTREPRENEURSHIP

for other reasons. However, if some of the left-out aspects are in fact important the normal result is only weak effects of the studied factors, or worse: misinterpretation of the true nature of the relationships. Either way, using the concepts in Figure 3:1 one would at best arrive at empirical generalization, not theoretical understanding. Assume, for example, that we find a reliable negative relationship between Firm Age and Entrepreneurship. What does that mean? The relationship, no matter how strong and statistically significant, is empty and pretty meaningless without interpretation. What is it about older firms that make them less entrepreneurial? Is this a problem for older firms? If so, what can they do about it? Perhaps older firms are happily non-entrepreneurial, but this is a problem for society? If so, what can policy-makers do about that? A theoretical understanding of the relationship would answer at least some of those questions; the empirical generalization itself just leaves us wondering. A third and better way of attacking the problem of the complexity in Figure 3:1 is to move up the ladder to a higher level of abstraction. This means to include a lot of the specifics but to view them as aspects of more general concepts. A good question to ask oneself in order to raise the level of abstraction is: This is a special case of what? While reading and re-reading a large number of empirical studies when I was working on my doctoral dissertation I asked myself that question a number of times. After several attempts at summarizing the findings of previous studies in terms of more abstracted concepts I came up with three rather simple ones. All of the specifics in the “messy” picture could be regarded as aspects of Ability, Need, or Opportunity. More specifically, I let the model depicted in Figure 3:2 guide the analyses.

This Thing Called “Theory”

37

38

RESEARCHING ENTREPRENEURSHIP

This Thing Called “Theory”

39

In what sense does this model represent an improvement in terms of abstraction and understanding? Firstly, because it suggests meaningful explanations for empirical generalizations. Why would older and larger firms grow less? Because they have less Need to grow! The older and larger you are, the more likely it is that you have already attained the minimum efficient size. Why would affiliation with a growing industry and location in a major city be associated with higher (entrepreneurial) growth? Because there is more Opportunity in such environments! And do we really believe that education causes entrepreneurship? Do we believe that experience does? I would say no, but we may believe that some Ability is needed for entrepreneurship to come about and that measures of experience and education are two out of many possible indicators of Ability. When facing a specific opportunity, those with more education and experience may perceive themselves as relatively more able to exploit that opportunity (Ability -> Perceived Ability) and therefore become more motivated to seize it (Perceived Ability -> Entrepreneurial Motivation). We may also believe that those with more education and experience are better at making plans materialize (Ability -> Continued Entrepreneurship). Secondly, applying this level of abstraction makes it possible to consider many specific sources of influence without getting lost. Using Partial Least Squares Analysis—a LISREL-like structural equations modeling technique (Fornell, 1987; Fornell & Larcker, 1981)—I analyzed versions of this model with up to 72 manifest (low-level) variables in the same analysis (Davidsson, 1991). Analyzing the pairwise co-variation among such a large number of variables would lead nowhere but to bewilderment. Again, abstraction is a blessing, and theoretical understanding is what makes research more fun than is producing a largely meaningless list of empirical generalizations.

THE ROLE(S) OF THEORY IN THE RESEARCH PROCESS There are two major roles for theory in the research process, namely to guide the design and/or analysis of empirical studies and to interpret the results of empirical research or other empirical observations. The theory or theories that guide the design should logically also be used for interpretation of that same research. The converse does not necessarily hold true, although it often does. It is conceivable that theory that was not considered at the design stage may still be useful for interpreting and understanding the results. Within those two roles—design and interpretation— there is a variety of cases, some of which will be discussed below.

Theory as Guide to Research Design Mark I: The Theory Test The most obvious case that comes to mind is the pure theory test. This type of research starts either from an interest in the theory and its applicability or from an interest in the entrepreneurship phenomenon to be investigated. Either way, the researcher realizes that an existing theory has implications for an entrepreneurship phenomenon. The researcher therefore designs an empirical test of the propositions made by—or hypotheses derived from—the theory. A well-specified theory is likely to guide the researcher on issues like:

RESEARCHING ENTREPRENEURSHIP

40

What kind of units/level of analysis do I need data on (e.g., individuals or firms or networks; general population or business founders or habitual entrepreneurs)? Will cross-sectional data suffice, or do I need more than one wave of data collection? What core concepts do I need to operationalize? What relationships should be tested, and with what analysis technique? As an example, consider a test of Ajzen’s (1991) Theory of Planned Behavior (TBP) in an entrepreneurship context. The core concepts and relationships in this theory are depicted in Figure 3:3.

Figure 3:3

A graphical representation of the Theory of Planned Behavior

Behavior would in the current context mean having started a business venture, or concrete actions towards realizing a business start-up. Intentions would reflect willingness or a plan to do so. Attitude and subjective norm reflect the extent to which the individual and his/her relevant environment (peers; family; society-atlarge), respectively, regard starting a venture as a good or bad thing to do, both as judged by the individual. Perceived behavioral control would here be high for individuals who feel they have the knowledge, contacts and means needed to get a business going, and lower for those who feel they lack one or more of those requirements. A test of this theory is a piece of research that could be initiated by a researcher specializing in TBP and looking for other arenas for testing the theory’s applicability after already having established its relevance for behaviors like weight loss, use of

This Thing Called “Theory”

41

contraceptives and quitting smoking. Alternatively, we may have to do with a researcher interested specifically in why people go into business for themselves, and who in TBP has found a theory that seems promising for (partially) explaining this phenomenon. Either way, the theory would be helpful in answering the above questions. What subjects? Clearly, psychological theory gears one to the individual level of analysis, but not to practicing entrepreneurs in this case. The general population seems to be a prime suspect, possibly deducting those who already run their own businesses and people permanently out of the workforce. Could we use students? Yes, as a second best and at the cost of making the theory test much more restricted. Preferably one would then use MBAs or last year undergraduates rather than freshmen, who are for the moment rather far from making real career choices. Cross-sectional data? You would probably get away with it for a test of the left part of the model, but not for the important intention-behavior link (and you still wonder why you have seen more papers on the rather unimportant question of predicting intentions than those taking on the important task of explaining actual entrepreneurial behavior?). What concepts need to be operationalized? This is obvious from the model; just don’t think that any measure is a valid measure. Psychological concepts like these are likely to need multiple indicators whose internal consistency needs to be tested in factor and/or reliability analysis (Bryman & Cramer, 1999). What relationships should be investigated? This is also obvious from the model, but don’t forget that the non-included arrows should be empirically ruled out and not just assumed not to exist; otherwise we haven’t really tested alternatives to the theory. As to techniques, the co-occurrence of indirectly measured constructs and direct as well as indirect relationships points towards some kind of structural equations modeling technique such as LISREL or PLS (Fornell, 1987), whereas the (perhaps) dichotomous nature of the ultimate dependent variable restricts the set of available techniques in other ways. Interactions and non-linear relationships should also be tested (possibly in separate analyses), as finding those to be insignificant strengthens the support for the theory. Remember also that if we do not have temporal division between the three sets of variables in the model, we have not ruled out the possibility of reversed causality. The design suggested by the theory, then, would be a longitudinal study of a representative sample from the working age population, using validated operationalizations of the core concepts and a set of analyses, performed with adequate techniques, of the relationships predicted by the theory as well as other possible alternatives. If supportive, a well-designed test of this kind would lead us to conclude that TBP is valid for entrepreneurial behavior—and probably for a range of other behaviors as well. From the perspective of entrepreneurship research a supportive test would make our beliefs about why and how business start-ups come to be somewhat less speculative. If the outcome were negative, we would conclude that the explanations offered by TBP do not apply to this type of entrepreneurial behavior. It may seem self-evident that one should let theory guide the design. Sometimes it is not as easy as it might first seem, however, to find satisfactory solution to this matching need. I have elaborated elsewhere on the subtleties of this matter as regards the firm level of analysis (Davidsson & Wiklund, 2000). Because there are

42

RESEARCHING ENTREPRENEURSHIP

different conceptualizations of “the firm” there may be a mismatch between the theoretical and empirical notion of “firm” even though the level of analysis seems to be right. We will elaborate on this issue in Chapter 5. It may also be the case that no data can be obtained that perfectly matches the theoretical level of analysis. For example, ecological and evolutionary theories deal with “species” and “populations” of organizations that share certain competence and/or other characteristics. Empirically, however, membership of a population is often equated with having the same industry classification code, and “populations” then become equal to what industrial economists call “industries” (Aldrich, 1999, p. 224). Obtaining data that more closely mirrored the theoretical definition of organizational “species” would require costly and cumbersome collection of primary data (Gratzer, 1996; 1999). Sometimes data do not exist for the chosen level of analysis but can be aggregated from lower levels. For example, in the project Business Dynamics in Sweden (Davidsson et al, 1994a, 1994b; 1995, 1996, 1998a, 1998b) we started from data on business establishments (plants; places-of-work). Because our intended level of analysis was the region (theory so suggesting or not; this was my big “sinner” project) we aggregated the establishment level data so that in our data set the variables were, for example, the numbers of started, closed, growing and shrinking establishments per firm and industry category per region per year. Hence, a variable by the neither exotic nor poetic name EXPSIM88 means “the number or expanding single-establishment firms in this region in 1988.” In the related study Culture and Entrepreneurship (Davidsson, 1993, 1995a, 1995c; Davidsson & Wiklund, 1997) I needed data on cultural (values and beliefs) differences across regions. As no regional data on such variables were available I had to collect them from representative samples of individuals in different regions. The averages for individuals in regions were then used as regions scores (and as a bonus I got data that were excellent also for analyses on the individual level, Davidsson, 1995b). A special case of the theory testing approach is when theory is not used to design the study, but only to guide the analysis of data that are already available. Two such examples from my own research immediately come to mind. The first was during my dissertation project, when well after the data collection was over I came across Smith’s classical study—not available in any Swedish library at the time— where he coined the (at least back then) oft-cited concepts “craftsman-entrepreneur” and “opportunistic-entrepreneur” (Smith, 1967). A close reading revealed that I had data that could be used for testing no less than 22 specific claims Smith made about these two types of entrepreneur and the firms they created. So what I did was to hypothesize that if I grouped my cases into the two most homogeneous groups I could create, the resulting groups would conform to Smith’s types. And lo and behold! Judging from a two-group cluster analysis of the 22 variables the taxonomy Smith developed on the basis of an all-male and all-manufacturing sample in Michigan in the 1960s held up very well in my mixed-sex, mixed industries sample in Sweden 20 years later (Davidsson, 1988, 1989a). The important lesson for the present context, however, is not specifically about the validity of Smith’s (1967) theory, but about realizing the value of doing theory testing rather than exploratory work when both are possible. As I said in a footnote of my doctoral thesis already: “Typologies [sic] arriving at 2 to 11 groups, on the basis of different approaches and

This Thing Called “Theory”

43

with more or less of systematic empirical backing, may be found in: [nine references]. Conclusion: adding another one would be superfluous” (Davidsson, 1989a, p. 158). I take the other personal example of theory-driven analysis of already available data from my pile of work-in-very-slow-progress (know the feeling?). Firm growth sometimes is and sometimes is not an entrepreneurship issue, and it can be argued that organic growth is more likely to reflect entrepreneurship than is acquisition growth (cf. Chapter 1 and Davidsson et al, 2002). A close reading of the classic “The Theory of the Growth of the Firm” (Penrose, 1959) leads to the following Penrosian propositions about growing firms: The proportion of organic growth (to acquisition growth) will be higher for younger firms. The proportion of organic growth (to acquisition growth) will be higher for firms in young and growing industries. The proportion of organic growth (to acquisition growth) will be higher in times of economic upturn. The proportion of acquisition growth (to organic growth) will be higher for large firms. The proportion of acquisition growth (to organic growth) will be higher for incorporated firms. The growth of firms with a high proportion of acquisition growth will be more irregular over time, relative to the growth of firms with a high proportion of organic growth. A supportive test of these propositions would no doubt further strengthen Penrose’s theory, which is much more often cited than tested. It would also further our empirically based knowledge about the phenomenon of firm growth. Importantly, the value of finding empirical relationships in line with the above propositions is much greater when they emerge from a theory test than from random exploration of any possible empirical relationship in the data. This is so because Penrose (1959) provides a coherent theoretical explanation for why these relationships should be expected. If the exact same relationships emerged from data mining an array of causal mechanisms that we have not yet thought of—or methodological artifacts—could with equal justification be suspected to have yielded the results. What data would we need to test these propositions (or hypotheses)? Well, first we need data about a sufficient number of firms that actually do grow. Penrose (1959) carefully points out that her theory concerns growing firms; it does not purport to explain why some firms choose not to grow. The propositions also make clear that we need a sample of firms that vary in terms of age, size, governance and industrial affiliation, as well as data on these variables. We also need supplementary data on the relative maturity and development of industries. Further, growth is a process that occurs over time. Hence, we need longitudinal data. The last proposition makes clear that just two data points from different times will not suffice, and the third proposition also requires that the studied period include

44

RESEARCHING ENTREPRENEURSHIP

different business cycle stages. Most importantly, the data must make it possible to distinguish between organic (internal) growth and growth through acquisition. I happen to sit on data that has all those characteristics (Davidsson & Delmar, 1998, 2003; Delmar et al, 2003). So I’d better get down to having those propositions tested. I promise I will. One day. I only have to finish this book first. And that paper. See? That’s how you become a sinner...

Theory as Guide to Research Design Mark II: Understanding the Phenomenon through an Eclectic Framework A theory-testing approach as described above is often regarded the hallmark of good research. And often it is. When I entered the field of entrepreneurship research there was certainly a lack of theory-testing (cf. Low & MacMillan, 1988), and most likely there still is. Nonetheless, in the future there is a non-negligible risk that we reach the other extreme. That is, more or less all published entrepreneurship research will follow the formula take an existing theory – design a study to test it on a research question relevant to entrepreneurship research – analyze – report. This formula more or less guarantees a manageable research task and relatively short time to market (i.e., quick and relatively certain publication)—as well as a meaningful (albeit small) contribution. The problem is that application of the theory-testing formula, when over-used or mindlessly applied, may get mechanical and lead to relatively little understanding of the studied phenomenon. When research starts from a true interest in the phenomenon one will soon detect that a single theory can only offer very limited and partial insights. Theory-testing as described above deserves the position as sovereign ruler only in situations when a) all the theories that are needed to understand and explain the phenomenon already exist, and b) all the theories are part of the same puzzle, so that the insights gained from one theory test fits neatly alongside with previous knowledge development. I hold that these conditions are currently not met in entrepreneurship research. Therefore, if sinning means deviating from the ideal of straight theory-testing, one should allow oneself a little sinful escapade at times. There are some variations to the theme of “understanding the phenomenon through an eclectic framework”. One, which is the closest to the theory-testing approach just described, is to design the study to test more than one theory in parallel (but separately) in order to be able to determine which of them best explains the phenomenon (cf. Krueger, Reilly & Carsrud, 2000). That is not very sinful at all, and can be highly useful. The more representative case is when the researcher shows a little more of creative disrespect for the original theories. Theories are tools, remember? Your wish to modify the tool and combine it with other tools does not imply a shortcoming on the part of the toolmaker—s/he could not possibly know about your specific needs. In order to illustrate the merits of an eclectic framework building on elements from several theories, let us return to the Theory of Planned Behavior. What is not explained by this theory? Well, it is a theory of planned behavior, right (and it does not pretend to be more than so)? It is well known from behavior in other domains that some of those who plan to do something do not do it, whereas some of those

This Thing Called “Theory”

45

who had no intention go ahead and do it all the same (Pickering, 1981). Starting a business is no different. Bhave (1994) distilled two main routes to starting ones own business, apparently about equally common. The first does indeed start with a wish or intention to strike out on ones own, followed by search, selection and refinement of specific business ideas. The second, however, is when individuals gradually drift into independent entrepreneurship through solving a job-, hobby- or consumptionrelated problem for themselves, only to detect that others have the same problem and are willing to pay to get it fixed. This latter group would be unlikely to report an intention to start a business far in advance of actually doing so. Therefore, for those with a genuine interest in understanding the phenomenon of business start-ups, Ajzen’s (1991) theory is incomplete. This would show, for example, in the form of modest explanatory value. Although it (perhaps) gives statistically significant partial explanation of the phenomenon, it is far from giving full explanation. In order to capture also not-so-planned business start-ups we would have to add, at least, one more main variable to the model. We can call this variable entrepreneurial potential, which could be indicated by membership of vocational groups with a high instance of independent business, such as professionals and craftsmen; mastery of some unique and marketable skill; deep interest and knowledge in a hobby, and the like. Based on other research we may also want to include a variable called triggering event (such as unemployment, divorce, turning forty, winning or inheriting money) as moderator between intention and entrepreneurial potential on the one hand, and behavior on the other (cf. Shapero & Sokol, 1982). Triggering event could also be an explanatory variable in its own right. Moreover, TBP as depicted in Figure 3:3 is incomplete in that it explains behavior and psychological constructs with other psychological constructs. When they find Ajzen’s theory supported in an entrepreneurship context those with a genuine interest in understanding the phenomenon immediately ask: “Where do these attitudes, norms and control perceptions come from?” So we may want to include some more tangible, personal background and environment variables as well, and estimate their direct and indirect influence on entrepreneurial intentions and behavior (cf. Davidsson, 1995b). The model in Figure 3:2 (above), which I developed in my dissertation study, can be regarded an example of an eclectic framework approach. I have argued that such an approach can lead to a more complete understanding of the studied phenomenon. However, every framework has to be incomplete—we cannot include the whole world in our models. And—as every experienced empirical researcher knows—even with the most comprehensive model we are unlikely to explain more than half of the variance. There is just too much idiosyncratic variation and unavoidable measurement error, so it is actually sound practice to be suspicious about research that purports to reach farther than that. Moreover, using a comprehensive, eclectic framework for designing one’s study makes for a more demanding research task. Therefore, if you are an American doctoral student and approach your committee with a research proposal of this kind you can be sure that the question “Will they reject it?” belongs in the “Is the Pope Catholic?” category. I was as blessed by European freedom when conducting my thesis work as have many others been cursed by the exact same freedom to take on research tasks of an

46

RESEARCHING ENTREPRENEURSHIP

enormous magnitude. Generally speaking, “narrow down!” is good advice to doctoral candidates. All the same, I maintain that more comprehensive research built on eclectic theoretical frameworks is also needed. Entrepreneurship is a complex phenomenon, and that complexity does not go away just because we make our designs more manageable. Piling up the many small pieces from separate theory tests is important. However, that alone will not lay the puzzle for us without more comprehensive research that help us see what pieces do and do not belong in our puzzle, and make the fitting pieces fall into their proper places.

Theory as Tool for Interpretation: the Theory Test I have pointed out already that the theory that is used for designing the study should also be used for interpreting its results. Does this really need to be pointed out? I think it does, not because researchers frequently make a sudden turn to other theories when they interpret their results, but because there is this kind of really dull, quantitative research where there is no interpretation at all. Early entrepreneurship research was full of this type of studies. That is, the concluding section of the paper merely re-stated that these hypotheses were supported whereas those were rejected. Period. Yawn. For God’s sake—what does it mean? Our vantage point was our curiosity about an interesting theory or a really important societal phenomenon, right? Then maybe we should return to discussing this at the far end of the manuscript? Seems a good idea. Entrepreneurship research has become much better in this regard over the last couple of decades, but I think even more could be done. In my opinion, it is often the quality of the concluding discussion that distinguishes an excellent and highly interesting piece of research from merely publishable, standard research. So please, finish the job and devote considerable time and energy to telling us what the results (might) imply for practice as well as for future research and theory development. As regards further theory development, the interpretation of a pure theory test is pretty straightforward. The possible outcomes can be sketched as four cases. The first is that the theory holds up; all its proposed relationships are found in the data, and the theory provides strong explanation of the phenomenon under study. This shows that the theory is applicable to the investigated domain, and strengthens the general validity of the theory. The second case is limited support. That is, the hypothesized relationships hold up, but they are weak and much of the variance remains unexplained. This is likely to be the case in our TBP example. The interpretation would be that the theory is valid (also) in this domain, but that we need to add more variables and relationships to it in order to get a more complete explanation of the studied phenomenon. The third case is partial support, i.e., that some of the relationships proposed by the theory hold up empirically whereas others do not. This points at a need to modify the theory rather than—or in addition to— supplementing it. This could reflect a shortcoming of the theory that has also emerged in other empirical contexts. If so, the theory would likely benefit from a general revision, whereas a domain specific deviation from the theory’s predictions only call for domain specific modifications. Finally, the fourth case is

This Thing Called “Theory”

47

straightforward. This is when the results of the empirical test do not at all support the theory. Assuming that the empirical test is performed in the context of a welldesigned study we would then reject the theory, at least with respect to the specific phenomenon under study, and start look for other alternatives. Importantly, rejection of a theory is not a failure. From a Popperian perspective falsification is the main route to knowledge development (Popper, 1992). With respect to understanding the phenomenon, and implications for practice, the interpretation work is firstly a matter of going back to the original theory and the detailed arguments as to why the proposed—and now confirmed—relationships should manifest themselves. If we forget this important step, we are back to relatively empty empirical generalizations. To illustrate this point, let’s return for a moment to the issue of organic vs. acquisition growth. Assume we can confirm that young firms actually do grow organically whereas old ones grow through acquisition. This could mean that young firms are more creative and entrepreneurial, and therefore have the ability to come up with new products and services that make organic growth possible. Old, lethargic firms have lost this capability, and in order to grow at all their only option is to acquire other firms. Alternatively, acquisition growth is the less risky and more profitable growth strategy, but this option is not open to young firms with limited resources and legitimacy. The interpretation means a world of difference as regards implications for practitioners. Penrose’s view (1959, p. 210) is simply that young firms are less likely to have exhausted the potential inherent in their original product(s) or service(s), so she does not hypothesize a favorable “youth” characteristics that firms should try to maintain as they grow older.

Theory as Tool for Interpretation: the Eclectic Framework Approach Most of what has been said above is equally applicable to the eclectic framework situation. This is especially true if the various theoretical elements are combined a priori to a well-specified model or framework with precise predictions as to the direction, sign, and form of relationships among the included constructs. When this is the case the only difference to the theory test as described above is that the results should perhaps be regarded as more tentative, as we are now dealing with a new theory rather than one that has been proven valid before in other domains. In other cases, however, the eclectic framework is not that well specified before data collection and/or analysis starts. For sure, the sampling and data collection has been guided by theories, but rather than making our minds up as regards TBP or social learning theory we may have included items that could serve as operationalizations of either, and we may have included measures of concepts from various theories without working out beforehand how we expect these to relate to one another. We may also have included in our eclectic framework variables that eventually do not show up in our reported results, because on closer thought they were not logically compatible with other concepts in the framework, or they turned out to be unimportant. The framework in Figure 3:2 is an example of this, as is the framework in Davidsson (1995b). When the point of departure is a genuine interest in the

48

RESEARCHING ENTREPRENEURSHIP

phenomenon and the current state of theory-based knowledge about this phenomenon is short of what it could be, I see nothing wrong in such an approach. Good research is often a matter of wrestling between theory and data, hence being neither purely deductive nor purely inductive, but “abductive” (Alvesson & Sköldberg, 1994). However, the more inductive the process has been, the more tentative are the results, and the greater reason there is to give room for caution and alternative explanations in the interpretation. Importantly, when the research process has been a matter of theory-data wrestling it should be portrayed as such. Regrettably, it seems likely that it is not uncommon that what in published research is portrayed as hypotheses were little more than hazy ideas before the analysis work began. Presenting exploration as theory testing is not good scholarly practice. Neither can it be excused with reference to space limitations, convention, or the like. We should not fool ourselves—and others. The proper names for such practices are deceit, cheating and fraud. I think those who believe that packaging (partly) exploratory research as purely deductive is what it takes to get published in a good, scholarly journal should stop believing so. That is, stop believing that such a journal truly is a good and scholarly one.

Theory as Tool for Interpretation: Post Hoc Theorizing We have now approached the exploratory situation, when theory that was not used for design is used or created in the interpretation phase. This does not mean that no theory was used for designing the study. In line with my earlier reasoning, theoryfree data collection is hardly possible. What we really mean is that the theory guiding the research design was vague, unarticulated, rudimentary—or just different—relative to the theory used or created for interpretation. I have already voiced the opinion that exploratory research should be avoided, if possible. This skeptical attitude is in part formed by explicit persuasion during my research training, e.g., making me read Armstrong (1970) and his compelling example of how prone we are to find post hoc rationales for any empirical relationships—even when mistakenly based on an analysis of random numbers. More important, however, is my training in cognitive psychology. Learning about the selective nature of perception, attention and information search, the constructive nature of memory, and how easy it is to manipulate our perception, recollection and sense-making of events have a humiliating effect on a researchers belief in his/her ability to distill any form of generalizable “truth” (or inter-subjectively meaningful knowledge) from unsorted and complex data (Anderson, 1990; Goldman, 1986). There is a very real risk that the theories that actually guide the analysis are the prejudice and preconceptions that I am not consciously aware of. However, there are many situations in which the post hoc use or development of theory can be justified. The first is when we get support for the theory that was used for designing the study. In trying to understand the phenomenon, and teasing out the implications for practice, we may also want to use other theories for our interpretation. This may seem strange, but is in fact not strange at all. To illustrate this, let us return to the TBP example. We noted above that in order to get a more complete picture we might want to expand the design to include elements also from

This Thing Called “Theory”

49

other theories. However, even if we did not do that in the design phase, we can add understanding in the interpretation phase by doing so. A possible outcome of a test of TBP in an entrepreneurship context is that perceived behavioral control comes out as the strongest predictor of entrepreneurial intention, and that this variable also has a strong direct influence on behavior. We now want to know where these perceptions of behavioral control come from. Assume that TBP—like its graphical representation in Figure 3:3—does not provide sufficient explanation of this. There is then no reason for us to refrain from looking around and ask “Does some other theory, or established empirical generalization, give us a clue?” Such a search would likely lead us to Social Learning Theory (Bandura, 1982, 1986) and to noticing that the concept of self-efficacy has a large overlap with Ajzen’s concept of perceived behavioral control (Eagly & Chaiken, 1993). Bandura (1982) elaborates on the sources of self-efficacy and holds that individuals develop and strengthen such beliefs in four ways: 1) mastery experience; 2) modeling, or observational learning, 3) social persuasion, and 4) judgment of their own physiological state. Thus, by adding elements from this theory to our interpretation we can reach much farther in our understanding of the results and their practical implications. The second case also deals with the situation when we have received support for our tested theory. Even without a need for additional theory-based interpretation it is good scholarly practice in this situation to admit that even though our theory was supported, there may be alternative explanations for the results. I cannot see any reason why theories not used in the design should be banned from the discussion of alternative explanations. The third situation is when we get an unexpected result that runs counter to our hypothesis, or when a control variable2 turns out to have an effect of unexpected form or strength. Yes, we can stop at just noting that the hypothesis was not supported, or noting the effect. However, I see little reason why we should not share our after-the-fact speculations about why those unexpected effects turn up. In particular, I see little reason why it would be worse to base such speculation on previously unused theory rather than armchair reasoning or reference to possible method artifacts. However, one should clearly distinguish such tentative, exploratory findings from support for the theory that was used for study design. This is because we can always find some explanation after the fact (even for correlations among random numbers, remember?). The fourth case is when it is not the researchers’ ignorance or creative itches that make them want to go explorative, but a real lack of relevant theory. It is not the case that we already have all the theory we need, and this is especially true for a young field like entrepreneurship research (cf. the next main section of this chapter). Conceptual development has to start somewhere. In some quarters there is outright disgust for exploratory (and often qualitative) work while at the same time purely conceptual contributions are quite OK. In fact, I have heard about a colleague who got the advice to publish the very interesting theoretical ideas that emerged from an in-depth study as a purely conceptual piece rather than as empirically grounded 2

A control variable is an explanatory variable that is included not because we have a theoretical interest in its effect, but because omitting it may lead to incorrect estimation of the effects of those variables we do have a theoretical interest in (Kish, 1987).

50

RESEARCHING ENTREPRENEURSHIP

findings. I don’t see the logic. Theoretical ideas based on armchair reasoning (and necessarily affected by whatever unsystematic empirical observations the theorist ever came across) have higher scientific status than those based on at least semisystematic empirical work? This is absurd. As I see it, one has every reason to be wary about exploratory research, and when exploratory researchers turn out not to have much of conceptual skills we do have a problem. However, I don’t see the big problem with the skilled theorists leaving the armchair and basing their thoughts on exploratory empirical research.

Is It the Theory or the Data That Is Supported or Should Be Rejected? Thoughtful researchers are damned, aren’t they? I mean, shouldn’t it suffice that we get statistically significant support for our theory and say “Good; theory proven true!” or fail to get this heavenly authorization and therefore have to say, “Tough luck—theory proven wrong”? No, I don’t think that suffices. We have noted already that a given set of empirical results may be consistent with several theoretical interpretations. In addition, an inescapable problem inherent in any theory test is that the outcome we get can either be ascribed to qualities of the theory or to qualities of the data (or method). We don’t really know. For example, in the TBP example we may get support for the attitude intention relationship. What is this? Probably a correlation between two paper and pencil behaviors, conducted only a few minutes apart. Is this evidence that if we can affect people’s attitudes to entrepreneurial behavior, they will as a consequence develop more entrepreneurial intentions? This may be the reason why we obtained the result, but it could also be due to a personality-, mood-, or response style-based method artifact. Conversely, we may fail to get support for this relationship. Does this show that the theory is wrong? Possibly, but it could also be the case that poor operationalizations led to such grave measurement error that the true relationship does not emerge from the data. We have to face it: we don’t know for sure. Perhaps the theory should be rejected; perhaps we should instead conclude our data are crappy. Perhaps we are justified in strengthening our trust in a theory, but perhaps we have been misled by some peculiarity of the method. We don’t know—and probably never will. This means that such a horrible, subjective thing as judgment must have a big role in the research process. It also means that this even more horrible thing called rhetoric will have a profound role in knowledge dissemination. As consolation I offer this: perfect democracy and perfect justice are not possible to achieve. This does not mean they are not worth striving for. Similarly, there is nothing wrong with judgment when it is good, and this is easier to achieve when the judgment is based on good evidence, i.e., clear support or rejection of the theory based on good data. And rhetoric is not “just” rhetoric—its likelihood of success is related not only to the form but also to the quality of its contents. One important aspect of this is the need for replication—an issue we will return to in Chapter 9. When results are replicated in several studies using similar but slightly different samples and operationalizations, our belief or disbelief in theories will be much less contingent on skillful rhetoric.

This Thing Called “Theory”

51

DO WE NEED SPECIFIC ENTREPRENEURSHIP THEORY? In Chapter 2, I offered the following domain delineation for entrepreneurship research: Starting from assumptions of uncertainty and heterogeneity, the domain of entrepreneurship research encompasses the study of processes of (real or induced, and completed as well as terminated) emergence of new business ventures, across organizational contexts. This entails the study of the origin and characteristics of venture ideas as well as their contextual fit; of behaviors in the interrelated processes of discovery and exploitation of such ideas, and of how the ideas and behaviors link to different types of direct and indirect antecedents and outcomes on different levels of analysis.

If we examine this statement, we have to conclude that there are few contingencies of interest to entrepreneurship scholars that are not the topic of theory in at least some discipline in the social sciences (cf. Acs & Audretsch, 2003a; Delmar, 2000; Thornton, 1999). Not making full use of the tools available within the disciplines would appear to be a wasteful practice. It is not so easy, however, that all the theory entrepreneurship researchers need already exists in the disciplines. Even if it is true that there are few contingencies of interest to entrepreneurship scholars that are not the topic of theory in at least some discipline in the social sciences, it is equally true that theorizing about entrepreneurship is not the main responsibility of any discipline. I have stressed already that many of “our” questions may be peripheral to every discipline (cf. Acs & Audretsch, 2003b). Therefore, although the disciplines have developed many sophisticated tools, these tools may not always be adequate for the task at hand (cf. Davidsson & Wiklund, 2000). In relation to the above domain delineation some of the questions one should ask before applying existing theory “as is” are the following: 1. 2.

3. 4. 5.

Does the theory acknowledge uncertainty and heterogeneity? Can it be applied to the problem of emergence, or does it presuppose the existence of markets, products or organizations in a way that clashes with the research questions? Does the theory allow a process perspective? Does it apply to the preferred unit of analysis (e.g., “venture idea” or “emerging venture” rather than “firm” or “individual”)? Is it compatible with an interest in the types of outcomes that are most relevant from an entrepreneurship point of view?

Theories exist, and whenever possible, entrepreneurship research should deductively test theory from psychology, sociology and economics as well as from various branches of business research. However, as a scrutiny of some existing theories in relation to the five questions above would show, they are not always optimal for research questions addressing the processes and analysis levels of most relevance to entrepreneurship research. Therefore, the domain must allow also for

52

RESEARCHING ENTREPRENEURSHIP

adapting theories as well as for filling gaps and asking new questions through inductive, theory-building approaches. There is one additional reason why entrepreneurship research needs to build its own theory. This is that the various theoretical fragments developed within the disciplines are not likely to form a coherent whole. Again, this failure to collectively cover the entrepreneurship agenda is neither a problem nor a shortcoming on the part of the disciplines. They never had that goal or obligation. However, this state of affairs is a problem and a shortcoming from the perspective of entrepreneurship research, and our field therefore has additional needs to develop its own theories. The rich sources that disciplinary theories provide make this task somewhat more manageable.

SUMMARY AND CONCLUSION I have argued in this chapter that theories are tools that give our research more relevance and a longer life. I described two main roles for theory: guiding the design and interpreting the results. Two theory-based designs were discussed: the theory test, and combining elements from several theories into an eclectic framework. The former represents a more straightforward and manageable type of study, whereas the other—when successful—can lead to more complete understanding of the studied phenomenon. With either approach, the theory used for design should logically also be used for analysis. However, there are situations when one is justified in introducing additional theoretical tools at the interpretation stage. These include when so doing helps deepening the interpretation of positive results, or discussing the alternative explanations that should be admitted when the researcher has been lucky enough to get support for her theory. Yet another case is when we want to speculate about the reasons for an unexpected result, such as the opposite of a hypothesized relationship or a very strong effect of a control variable. As long as the interpretation is admittedly speculative, basing ones speculations on previously overlooked theory can be no great sin. Exploration-based generation of theory can also be justified when it truly is the case that no relevant theory exists. I have argued that such situations are not unlikely in a field like entrepreneurship, which is young and at the periphery of established disciplines. However, although one should ask questions about their applicability, the main rule for entrepreneurship researchers should be to use the theoretical tools already developed within psychology, sociology, economics, and various branches of business research. At the very least, before we decide not to do so we should have made an effort to learn to know what we are rejecting. Finally, we have noted that questions of theory and method are intertwined. We want to accept or reject theory on the basis of its relevance for understanding real world phenomena. However, in empirical testing we run the risk of accepting theories because of method artifacts or rejecting them because of poor sampling or measurement error. So we do not really know whether it is the theory or the data that should be rejected. In order to better justify our judgment and rhetoric about entrepreneurship theories, it is therefore critically important that the greatest care be

This Thing Called “Theory”

53

taken in the design and execution of empirical studies. This is the topic for the remainder of this book.

CHAPTER 4

GENERAL DESIGN ISSUES

GETTING STARTED AT LAST After this three-chapter warm-up, it’s about time we get to the real contents of this book: method issues. Well, the previous chapters have in fact more than touched on general design issues already. For example, Chapter 3 emphasized the need for theory-driven designs (although allowing for exploration when needed) and mentioned at least briefly the matching of theory and level of analysis. In Chapter 2 the detailed explanation of keywords in my proposed domain delineation—e.g., process, emergence, discovery/exploitation, contextual fit and antecedents and outcomes on different levels—contained many implicit and some explicit design implications. Hence, large parts of this chapter will be both a recapitulation and an elaboration of previously introduced themes. I will start, however, with an issue I have only briefly touched upon previously, namely a discussion of “qualitative” and “quantitative” approaches in entrepreneurship research. I will then return to general design implications of the keywords emergence, process and new venture. Before concluding I will also present a short discussion of where laboratory research fits in the entrepreneurship research agenda. “QUALITATIVE” AND “QUANTITATIVE” STUDIES The Need for “Qualitative” Entrepreneurship Research This book deals mainly with co called “quantitative” research. Again, this is not because of an alleged general superiority of such approaches, but a simple consequence of my lack of expertise in “qualitative” methods. Although I embrace the ideology that our knowledge development processes are incomplete without theory testing, I firmly believe that both “qualitative” and “quantitative” research is helpful for gaining insight into entrepreneurship. To illustrate this point, consider the following little anecdote. A few years ago I was present at a seminar by a strongly anti-quantitative entrepreneurship professor (quite a common species in Scandinavia). As the seminar went by it became increasingly clear that despite our different revealed preferences as regards research approaches, he and I had developed a strikingly similar understanding of many aspects of the entrepreneurship phenomenon. He had arrived at these views through extensive

56

RESEARCHING ENTREPRENEURSHIP

conceptual work and reading of others; frequent contacts with entrepreneurs, and his own “qualitative” work. I had arrived at similar views mainly through conducting large-scale quantitative research projects, including the reading and contacts that follow from this work. A reflective mind was probably the common denominator. The point is that our total knowledge development requires the combination of different types of information. The story also illustrates that there are different routes to insight (if we could claim we had any). Therefore, researchers who say or think “I cannot see any meaningful knowledge coming out of that research approach” should realize that this may reflect a deficit of “I” and not only a shortcoming of “that approach”. There are some characteristics of the entrepreneurship research domain, as I have portrayed it, that point at a need for “qualitative” research. One is the relative youth of the field. We have simply not had time enough yet to familiarize ourselves with all facets of this empirical phenomenon, or to exploratively develop all the theory we need (and which other disciplines have not provided us with). Another is the heterogeneity of the phenomenon. If we only did research at arms-length distance there are the risks that because the relationships are different for different parts of the heterogeneous population we would either come out with only weak results, or results that are “true” on average but not for most individual cases. Closeup information may be needed in order to learn about the heterogeneity, so as to assess what abstractions and generalizations we can and cannot justifiably make. Further, at least when we think about more spectacular forms of innovative entrepreneurship, we are dealing with events that are infrequent, unanticipated and/or extraordinary. Phenomena of this kind may be difficult to capture with conventional, “quantitative” approaches (cf. Baumol, 1983; Brymer, 1998). It is worth pondering that at the extreme of conventionalism, the most spectacular instances of entrepreneurship would invariably end up as disturbing and possibly deleted outliers in regression analyses (cf. Ch. 10). Another aspect that I have highlighted in my entrepreneurship research domain delineation is the process character of entrepreneurship. This may also call for “qualitative”, or close-up, approaches (cf. Brundin, 2002). An early insight I had as a researcher was, in fact, the difficulty of capturing processes in survey research. One of the cases in the pilot study for my dissertation made this particularly clear. This case was about a small firm in a shambles. The rational thing to do would have been to file for bankruptcy, but the owner-manager just couldn’t stand the thought of it. The firm was heavily in debt to suppliers and tax authorities alike. Old, uncomfortable facilities led to high personnel turnover and difficulties with recruiting. Insufficient profit margins made it impossible to catch up. At this point, a series of events led to a turnaround. The son assumed a serious management position and started by checking the profitability of different customers—an exercise that led to the conclusion that many long-term relationships were in fact unprofitable and should be reconsidered. By fortuitous coincidence, the firm got two new customers in a growing industry, which led to a reorientation that made it easier to attract—and develop special products or services for—additional, profitable and growing customers in that industry. Around the same time, the firm reached a deal with the tax authorities for a realistic plan for catching up with tax payments. Backed by this

General Design Issues

57

deal and the new customers, the firm was offered new facilities on favorable terms in the municipality’s modern industry park, which made the firm a much more attractive place of work. The reduced personnel turnover and increased job satisfaction in turn led to higher productivity and profits, and so on (Davidsson, 1986). Some of these events are causally related (although the sequence could equally well have been a different one) whereas others just happened to coincide. Clearly, virtuous circles of this kind would be very hard to capture in “quantitative” work and entirely impossible with a cross-sectional survey design. As will be further explained later I still believe it possible to capture important aspects of entrepreneurial processes in longitudinal, “quantitative” studies. I am equally convinced, however, that close-up insights from cases like the above is an indispensable input to good, “quantitative” work on entrepreneurial processes. “Quantitative” vs. “Qualitative”—a Confused Debate What entrepreneurship research does not need is the often confused and confusing debate about qualitative versus quantitative research that goes on in business studies; perhaps in Europe in particular. I call this debate confusing and confused firstly because it rarely recognizes that “quantitative” (and therefore “qualitative”) has several distinct meanings, and secondly because it often and non-justifiably equates the nature of the data with issues of philosophy of science, rigor, and depth3. As regards the first issue, critics who favor qualitative approaches rarely distinguish between the following three distinct meanings of “quantitative”: using “many cases” (census studies or large samples); applying “formal measurement” (coding data in numerical form), and “the use of statistical or mathematical analysis techniques”. These different aspects of “quantitative” each carries with it distinct advantages and shortcomings. In short, with many cases you gain generalizability and lose detail. With formal measurement and statistical techniques you gain a higher degree of objectivity and make it possible to detect patterns that are otherwise beyond our cognitive abilities, but reduce the scope of your research to what is possible to measure or estimate with currently available techniques. Further, the three aspects of “quantitative” do not necessarily go together. When a study is quantitative in the “many cases” sense it is for practical reasons also often quantitative in the second and third senses. Analyzing non-quantified data on several variables from hundreds of cases is beyond the cognitive as well as affective limits of most researchers. It is not the case, however, that the use of formal measurement requires that many cases be analyzed. Case studies may be highly quantified in this 3 Because of the failure to distinguish between the different meanings of “quantitative” and between philosophy and the nature of the data, I rarely find the criticism of quantitative approaches very illuminating. Interesting critique tends to come from those who know (some aspect of) quantitative research inside and out (cf. Cohen, 1994; McCloskey, 1998; Oakes, 1986). By the way, to my knowledge there has only existed one major branch of the social and behavioral sciences that can be called truly positivistic. This was the behaviorist paradigm that dominated academic research in psychology for several decades in the mid 1900s. Interestingly, what eventually led to its demise (or decline) were attacks carried out with the behaviorists’ own weapons: experimental evidence and mathematical proof (Baars, 1986).

58

RESEARCHING ENTREPRENEURSHIP

sense, and here already the categories “quantitative” and “qualitative” get seriously blurred—which is why I put these terms within quotation marks. Within a case study there may also be room for application of statistical or other quantitative analysis methods, such as time series regression or techniques like multi-dimensional scaling or conjoint analysis, which allow analysis of single-respondent data (Hair, Anderson, Tatham & Black, 1998). Moreover, the development of formal techniques for recording and analyzing data that have traditionally been regarded “qualitative” makes the distinction along this dimension seem inadequate and obsolete (Miles & Huberman, 1994). The second, confused and confusing part of the argumentation is the frequent assumption of a strong or even deterministic association between type of data and philosophy (or ideology) of science. Here “quantitative” is held to be almost equal to “positivistic” and “superficial” while “qualitative” data are associated with hermeneutics, phenomenology, or social constructionism, and with depth of analysis. Other parts of the researcher population would typically equate “quantitative” with “rigorous”, and consequently regard “qualitative” work as lacking rigor. Yes, I would say that it is likely that a true positivist researcher—if you could find one—would favor a quantitative approach in one or more of the above senses. It is possible that an orientation towards phenomenology or social constructionism makes the researcher more inclined to use qualitative approaches, but it is more difficult to see a logical connection here than in the case of positivism. Studying a large number of cases does not seem to come naturally for a researcher with a hermeneutic bend. However, hermeneutics does not seem fundamentally or logically (albeit perhaps by convention) opposed to the use of numerically coded data or statistical analyses. Over all, the choice of philosophical vantage point seems to have some, but far from completely deterministic, implications for the choice between “qualitative” and “quantitative”. In the other direction there is no determinism whatsoever. As I pointed out already in the preface: the data don’t know how they are going to be used. “Quantitative” data—i.e., measured and/or coming from many cases—do not make the researcher a positivist or the research deductive. There is nothing in the nature of the data that prevents the analyst from speculations about the inner meaning of the data themselves or of results that were generated inductively. To be sure, published “quantitative” research is full of exploratory findings and the use of techniques— e.g., factor analysis and cluster analysis—that a true positivist would deem unscientific. And—sorry, quantjocks and number-crunchers!—it is perfectly possible for research involving many cases and/or formal measurement and/or statistical techniques to be sloppy rather than rigorous. Likewise, the use of qualitative data does not by itself make a researcher worthy of any honorary philosophical title. Irrespective of the type of data, the researcher may confess to any philosophical congregation—or be an onto- and epistemological orphan or bastard. Again, the data don’t know what we are going to do with them. As regards rigor: yes, I would attribute what many (European) researchers regard as a bias—on the part of US-based journals, for example—against “qualitative” work instead as skepticism against research that does not show enough rigor. With rigor I then mean, roughly, well-founded motivations for the selection of cases or the like;

General Design Issues

59

systematic and transparent procedures for data collection and analysis, etc. It is perfectly possible to apply rigor in a study using data from few cases; non-numerical data, and without applying statistical methods for data analysis (but probably some systematic technique would be used). On the other hand, it is perfectly possible to be superficial with “qualitative” data. I believe there is confusion between “depth” and “detail” here. “Qualitative” data typically include more detail. Depth is another thing, and mainly an outcome of the time, effort and talent that is put into the analysis and interpretation work. I hold that there is no inherent property in “quantitative” data that prevents one from going deep in the analysis and interpretation. I also hold that “qualitative” research that is both rigorous and deep is appreciated by large parts of the entrepreneurship research community, including journal reviewers and editors. Bad Research Practice: Addressing “Quantitative” Questions with “Qualitative” Research I thus argue that entrepreneurship research needs both “qualitative” and “quantitative” approaches. However, there has to be a proper match between the research question and the chosen approach. The main problem I have with “qualitative” research is when researchers using such approaches make claims about issues their approach is fundamentally inadequate for addressing. Let me share another little anecdote to illustrate this. A couple of years ago I was at a presentation of a “qualitative” study of business founders. The cases were chosen because the founders were female and the start-ups were in a particular, recently deregulated industry. The data were collected through retrospective interviews. Several of the interviewees reported they had difficulties obtaining the bank loans they needed, and when prompted some of them ascribed this to the fact that they were women. Because of this, the researcher publicly claimed that women entrepreneurs were discriminated against by the banks4. I protested. Saying that banks systematically discriminate against a particular group is a very serious accusation, and because people have a high degree of faith in what researchers say, I get pretty upset when researchers make strong claims like this on the basis of very shaky—or in this case no—evidence. For heaven’s sake—if we want to establish that women business founders have difficulty obtaining bank loans because they are women, then for a minimum we need to a) investigate a group of subjects that is representative for the category “women business founders”, b) measure the frequency of loan refusals, or the like, c) and compare the results with another group relative to whom women are said to be discriminated against (i.e., probably male business founders). There can be absolutely no escape from these requirements. In addition, we should preferably also be able to rule out d) other substantive explanations (such as industry; venture size; size of loan application relative to own funds, etc.) and e) that the group difference we have established could easily be due to stochastic variation. In this case we had none of this. All we 4

I regret that I picked such a politically incorrect example. Those who think this proves I’m an MCP are referred to the fact that I was a proud co-supervisor of Helene Ahl’s doctoral dissertation (Ahl, 2002), which in my view is an excellent piece of feminist research.

60

RESEARCHING ENTREPRENEURSHIP

had was a few women from a judgment (or convenience) sample saying they had problems getting loans, and feeling this might have something to do with the fact that they were women. I am quite convinced there is solid research evidence elsewhere that women are discriminated against in society, and this, in conjunction with previous private experiences by these women, may have made their suspicion of discrimination a reasonable hypothesis. This is not, however, the quality of evidence needed for researchers to make strong claims about sex discrimination on the part of a specific group of actors. To make matters worse, the accusation made by the researcher was in all likelihood a false one in this case, as comprehensive and systematic research on precisely that matter was published at about the same time, arriving at the conclusion that women entrepreneurs were not discriminated against by banks (Björnsson, 2001). A review of the international entrepreneurship research literature on the issue of gender discrimination by banks suggest that the support is very limited, and when gender differences appear they do in most cases not seem to be attributable to gender per se (Ahl, 2002). More generally, the simple fact is that research questions that are inherently quantitative in nature need quantitative research to be answered. Questions about quantitative differences (more; better; stronger; more often, etc.) between groups, or about such within-group changes over time, are inherently quantitative in nature. So are questions about the form, direction and strength of relationships between variables. In order to make claims about such we need to measure the variables and estimate their relationship with some kind of analysis technique. Therefore, when a researcher makes claims like “x has a strong, positive effect on y” (e.g. “the entrepreneurs’ persuasive skills have a profound effect on their level of success”) on the basis of research where no formal measurement or estimation was involved, all I can see is extremely crude and unreliable measurement and estimation, namely the researcher’s tool-less assessments. Is this to imply, for example, that “qualitative” research can in no way inform the question of whether women entrepreneurs are discriminated against by the banks? Certainly not, but retrospective interviewing of a handful of women entrepreneurs would probably be my last choice for settling this issue. In all honesty, this was not a central question for the researcher in question, either. A piece of useful research that would be classified as “qualitative” by conventional criteria and which would get at issues that are unlikely to be within reach for a survey approach would be a participant observation study, where the loan officers’ way of talking to and about male and female loan applicants were studied. If there were discrimination, such a study would not only give strong indications of this fact, but also offer an opportunity to understand the mechanisms behind it. In order to impress a researcher of my own ilk, however, the presented evidence should not just be a number of illustrative quotes that support the researcher’s hypothesis, but convincing evidence that the loan officers’ treatment of women applicants was systematically different, and that this was to their disadvantage. As I see it, the most fruitful way forward for entrepreneurship research would be integrated research programs that included several types of research addressing different aspects of the same issues. This would make for real cross-fertilization between different approaches, rather than having different camps of researchers

General Design Issues

61

develop separate discourses that are ignored by the other camps. Because of my own balance of expertise and ignorance, respectively, I will say very little about “qualitative” designs in the remainder of this book. Again, this is not because of a lack of appreciation of such work. As previously mentioned, several of my favorite references—especially as regards the entrepreneurial process—are based on studies that were not “quantitative” (Bhave, 1994; McGrath, 1999; Sarasvathy, 2001; Shane, 2000; Van de Ven, Polley, Garud & Venkataraman, 1999).

ENTREPRENEURSHIP RESEARCH AS THE STUDY OF PROCESSES OF EMERGENCE OF NEW VENTURES What are the method consequences of a research focus as implied by this subheading (and by my previously presented domain delineation)? The keyword new ventures suggests that in order to belong in the entrepreneurship domain, the research has to meet the requirement of explicit consideration of new venturing within or associated with the studied unit of analysis. As long as this requirement is fulfilled, the research can be conducted on any level of analysis—individual, firm, industry, region, nation, or something else (cf. Davidsson & Wiklund, 2001). That is, the research design should at least include the middle box in Figure 4:1. Preferably, the research should pay attention to antecedents and outcomes as well, but this is not indispensable in the same way. On the individual level, we thus cannot confine the research to, for example, owner-managers’ personal characteristics on dimensions assumed to be entrepreneurial, as related to the size (an outcome) of their businesses. In order to qualify as entrepreneurship research there should be assessment of the middle box—new venturing activities by these individuals. On the region level, studying the relationship between structural characteristics of regions and their economic growth (or well-being) does not become entrepreneurship research until the quantity and/or quality of regional business venturing is introduced as the mechanism of such a relationship. We will return to sampling and operationalization issues on different levels of analysis in the coming two chapters.

Figure 4:1

Entrepreneurship research design possibilities

The emphasis on processes implies that we need longitudinal research, which has traditionally been short in supply in entrepreneurship research (Aldrich & Baker, 1997; Chandler & Lyon, 2001). What do we need longitudinal research for? First, in order to establish causality we need to establish for a minimum that the alleged cause precedes the ensuing effect. To take an entrepreneurship example of this problem, consider the hypothesis in early entrepreneurship researcher that entrepreneurs were characterized by a more internal Locus-of-Control (Brockhaus,

RESEARCHING ENTREPRENEURSHIP

62

1982). Having a more internal locus-of-control roughly means that you believe in your own ability to control your destiny, as opposed to it being directed by fate or powerful others. Some cross-sectional studies have supported the idea that entrepreneurs (here meaning business founders and/or owner-managers) have a more internal orientation than others. But would not such an orientation be a likely outcome of being a business owner-manager, as opposed to being bossed around by superiors within a hierarchy? Hence, a positive correlation is not enough. In the absence of longitudinal research showing that internal locus-of-control precedes business founding the hypothesis that an internal orientation causes individuals’ choices of an entrepreneurial career remains just that: a hypothesis. The study of processes involves more, however, than static comparison of a beginning state and an end state. Quite a number of things happen between the initiation of a venture start-up process and its completion/termination (Bhave, 1994; Carter et al, 1996; Davidsson & Honig, 2003; Davidsson & Klofsten, 2003; Delmar & Shane, 2002, 2004; Gartner & Carter, 2003; Katz & Gartner, 1988; Sarasvathy, 2001; Van de Ven et al, 1999). Therefore, we need longitudinal designs with repeated assessment of the ventures’ development over time in order to adequately capture those processes. The emphasis on emergence suggests that we should catch new ventures early in the process (cf. Davidsson, 2003a, 2003b). How can we study emergence? All existing business activities are eligible for retrospective studies, but such studies would be subject to severe selection and hindsight biases. For several reasons, it is preferable to study the processes as they happen, or as close to that ideal as possible. Regarding hindsight bias, it is well known in cognitive psychology that memory is constructive in nature (Anderson, 1990). This means that no matter how honest and careful a respondent is, he or she will still distort the image of what happened during the start-up process. Dead ends will likely be forgotten and certain actions will be ascribed a rationale that only fell into place afterwards. Such problems can to some extent be remedied through triangulation (second informant; written documentation), but serious distortions are likely to remain regardless of such efforts. Selection bias concerns the need to study also “unsuccessful” or prematurely terminated processes (cf. Chapter 2). For one thing, this is needed in order to acknowledge the fundamental uncertainty that we highlighted in the domain delineation. If we study only completed start-up processes we may tend to forget that completion is by no means a certain outcome for the newly initiated project. The problem of selection bias is potentially even more serious than hindsight bias. In order to illustrate this, consider the following example. Imagine that for some peculiar reason we wanted to study “factors that lead to success at betting on horses”. We design the study so that we include only those gamblers who actually won in their betting on horses, and thus left the day at the races with a net gain (cf. only those founders who actually got their venture up and running). Analyzing our data, we would arrive at the following conclusions: a. b.

Betting on horses is profitable The more you bet, the more you will win

General Design Issues c.

63

The more unlikely (higher odds) winners you bet on, the more you will win

While true for winners, these conclusions are, of course, blatantly false inferences for the entire population of gamblers. On average, gamblers do not win; the organizer of the gambling does. Likewise, ceteris paribus the expected loss increases linearly with the size of the bet, and not the other way around. And, of course, the proportion of gamblers who lose is larger among those who bet on long shots. But since we study only winners, the above are the results we will get. The scary fact is that by studying only those start-up processes that led to a successful start-up we make ourselves guilty of the same kind of error, and open up for the potential of arriving at equally biased results. We will continue the discussion of early catch when dealing with sampling issues in the next chapter (Chapter 5). It can be argued that the need for longitudinal, concurrent data is relatively more pronounced for micro-level studies than for more aggregate levels of analysis. For studies of the latter kind, a methodology like that employed by the Global Entrepreneurship Monitor (GEM) may suffice (Reynolds et al, 2001). That is, a design that aims at cross-sectional comparison of the prevalence of a) on-going startup processes and b) recently completed start-ups across countries, industries, regions, or perhaps even such a disaggregate level as firms. We noted in Chapter 2 already that there is a need for studies that use the venture idea itself as the unit of analysis, i.e., studies where “entity X” in Figure 4:1 is the new venture idea and the activity and new organization that evolves around it. For such studies, which would follow new, emerging business activities from their conception and through whatever changes in human champions and organizational home that might occur, the importance of early catch and multi-wave concurrent data collection seem particularly important. This is also a type of study where “qualitative” work in the form of longitudinal case studies is needed, not least for aiding with the difficult problems of design and interpretation of “quantitative” studies with this focus. LABORATORY RESEARCH METHODS We noted in Chapter 2 that one rationale for laboratory research in entrepreneurship is that real-world entrepreneurs or intrapreneurs, studied in the real setting, do not necessarily provide us with all the “right answers”. There is no guarantee that they have found them, and if they have found them they do not necessarily know why they are “right”. There are also additional reasons why we should welcome the increased use of “laboratory” research methods in entrepreneurship (Baron & Brush, 1999; Fiet & Migliore, 2001; Gustavsson, 2004; Sarasvathy, 1999a). One, of course, is the general strength of such methods that they can establish causality in a relatively unambiguous manner. As any standard method textbook would tell, because of the controlled situation and the possibility of manipulating the explanatory variables, it is much easier in the laboratory than in the real setting to sort out what affects what, and how much. This points at a major advantage for such

64

RESEARCHING ENTREPRENEURSHIP

methods as regards studying situational influences on the outcome of entrepreneurial processes, as well as how the outcome influences further entrepreneurial action. In connection to what has been discussed above, the process nature of entrepreneurship is an additional reason to consider simulation or experimentation. Studying real world processes is a costly and time-consuming endeavor with uncertain rewards. Although the laboratory alternatives will never completely substitute for real world studies they are a valuable complement—and an acceptable alternative when resource limitations prohibit a longitudinal study in the real setting. Laboratory methods make it possible to compress time and collect multi-period data without having to wait for ages before any serious analysis work can be done. In the extreme case we are talking about fractions of a second of computer CPU time (Fiet, Piskounov & Gustavsson, 2000)—a perspective, which can make and ethnographer or a multi-period survey jerk like myself kind of envious. However, it would be a mistake to believe that laboratory research is easy work. It is the distribution over time that is the big difference. Laboratory research tends to be heavy in the front end, in the design of the simulation or experiment, whereas actual data collection and analysis can be less of a burden relative to other methods. In Chapter 2 we portrayed entrepreneurship as consisting of two interrelated and (partially) overlapping sub-processes, which we called discovery (idea development) and exploitation (making it happen). Both of those could presumably be induced in the laboratory. However, the laboratory alternative may be particularly suitable for the earliest phases of the discovery process. Ideas for new ventures do not pop up that frequently, and even is we are lucky enough to be there we may not even see them as they happen with our real world, real time approaches (cf. Simon in Sarasvathy, 1999b, p. 52). For example, field studies of “entrepreneurs” mimicking Mintzberg’s (1974) study of managers are unlikely to capture initial discovery. Therefore, laboratory research may be better suited to cover this part of the entrepreneurship research agenda. The general shortcoming of laboratory research is that the external validity of the findings can always be questioned. What works in the laboratory does not necessarily repeat itself in the field, where loads of other influences also want to have their say. Therefore, laboratory work should preferably be integrated into programs that include also analysis of real world data, so that the field and the laboratory can inform and inspire one another. Interesting ideas along these can be found in Cialdini’s reasoning on “full cycle social psychology” (Cialdini, 1980). SUMMARY AND CONCLUSION I have argued in this chapter that knowledge development in entrepreneurship benefits from different types of research—“qualitative” as well as “quantitative”, and laboratory research as well as studies that rely on data from the real setting. Preferably, these different types of research should be combined in comprehensive programs. At least, it would be to the advantage of knowledge development if the different forms of research informed and inspired one another, rather than different methodological camps or entrepreneurship researchers developing separate and noncommunicating discourses.

General Design Issues

65

I have further argued that entrepreneurship research can, and should be conducted on different levels of analysis. However, in order to qualify as entrepreneurship research the study has to take new venturing on the studied level into explicit consideration. In empirical entrepreneurship research, the focal phenomenon should not be reduced to an assumption. Regardless of the level of analysis chosen, it is important that it be properly matched with the theory, as discussed in the previous chapter. Because of the profile of my own expertise I have given more room here—or will at least do so elsewhere—to “quantitative” studies based on primary survey data or secondary data from available registers. Regarding such studies, I have advocated that they be given a longitudinal design, so that processes can be adequately studied. Preferably the data should also be concurrent rather than retrospective, so as to emphasize the foci on emergence, uncertainty and outcome variability, and avoid biases stemming from hindsight and selection of successful cases only.

CHAPTER 5

SAMPLING ISSUES

A DIFFERENT LOOK AT SAMPLING This is not a conventional sampling chapter. A conventional sampling chapter builds on statistical inference theory and deals primarily with two issues. First, how can we create a sampling frame and a sampling mechanism that allow us do draw a statistically representative sample from the empirical population in question? Issues here are over- and under coverage of the sampling frame relative to the population; techniques for drawing a random sample or, to be more precise, one for which the sampling probability of each element in the sampling frame is known, and (possibly) techniques for minimizing non-response. Second, how large does the sample have to be for us to detect the differences and effects our theory predicts? Based on assumptions of variances and effect sizes, this involves calculating the statistical power (Cohen, 1988) of different sample sizes. The more conventional side of statistical inference theory, i.e., statistical significance testing, deals with the opposite risk, that effects found in the sample may be due to random sampling error rather than reflecting effects that are true for the population from which the sample was drawn. My agenda is different. As there already exists a plethora of books and chapters on sampling written by statistical experts (which I am not) I aim to write instead a sampling chapter based on what the statistical experts do not offer: sampling as viewed from the perspective of theory- and curiosity driven social science research, and backed with extensive practical research experience. What I want to discuss here, then, is how—on different levels of analysis—we can obtain data from a sample of cases that are theoretically relevant. By this I mean that the sample is composed of cases that reflect the theoretical unit of analysis and the theoretically relevant variance in the characteristics of these cases. I also want the sample to be workable from a practical point of view, i.e., that it is possible without breaking one’s back (or budget, although not all my suggestions will be for the most frugal research design) to obtain data from or about units in the sample. Indeed, large parts of my argumentation are not about “sampling” in the statistical sense at all, but applies equally well to studies of entire populations. What I address in this chapter is how we determine what are to be the cases in our data matrix, whereas the next chapter (on operationalization) will deal with the variables in the matrix. As will be argued below, for most research questions every accessible empirical population is a

68

RESEARCHING ENTREPRENEURSHIP

sample relative to the theoretically relevant population, i.e., the category we hope our results have some validity for. Below I will first expand on the theme just introduced (cf. Davidsson, 2003b). I will then discuss sampling problems and solutions for research on different levels of analysis—individual, venture, firm, industry, and spatial units (region, nation). As it turns out, this will be enough, or even more than enough, for a chapter. Social Science Is Not Opinion Polls Representative sampling and associated significance testing are important safeguards against ignoring relevant parts of empirical populations, giving undue weight to atypical cases, or ascribing substantive meaning to results that can easily have been generated by chance factors. However, for the statistical inference apparatus to be applicable in a strict sense the population should be well defined and the sample should reflect the composition of this population in a probabilistically known manner. These are ideals that are rarely achieved in social science research. For one thing, the painful fact is that response rates in published research typically fall in the 5-35 percent range. This alone makes application of statistical inference highly dubious. To make matters worse, statistical inference theory is a tool that is tailor made for opinion polls and industrial quality control rather than for the true needs of a social science researcher (cf. Cohen, 1994; Oakes, 1986). Consider political opinion polls. Here we have a clearly defined population, which in most countries is also reasonably reachable: all eligible voters. What we want to know are their political preferences on the day of investigation. Hence we can draw a random sample and ask the selected individuals about their preferences. Applying statistical inference theory, we can with high accuracy estimate with what uncertainty our sample results are associated, and determine whether the difference between two political parties, or the change for one party over time, deserves a substantial interpretation. Alternatively, we may conclude that these differences are likely to be the result of random sampling error. Clearly, probability sampling and significance testing are useful tools in this situation. We can say much more on the basis of this probability sample than on the basis of just any equally sized voter sample of unknown origin. There are occasions in entrepreneurship research that are very similar to this situation. The country comparisons of the prevalence of “nascent entrepreneurs” in the Global Entrepreneurship Monitor (GEM) are an example of this (Reynolds et al, 2001). Here, what we want to know are what proportions of the adult population in various countries are involved in business start-ups at a given point in time, and how uncertain are the estimates that we get from samples of a certain size from the adult populations in those countries? For the most part, however, social science research is not like opinion polls, and theories are not built by democratic vote. That is, it is not a given that every empirical case should be deemed equally important for our theory building and theory testing. What we are really after in social science research is theoretical representativeness—that the studied cases are relevant for the theory we try to test or develop. There is no way we can draw a random sample directly from the theoretical population, because that population does not exist in one place at one

Sampling Issues

69

time. This is why I argue that every empirical population, even if we investigate it in its entirety, is a non-random sample from the theoretically relevant population. The questions are: What is the theory about, and what should therefore be represented in the sample? Can we find an empirical population that is theoretically relevant (to study in its entirety or to sample from)? To illustrate the limitations of simple random sampling, consider first a firm level example. If a simple random sample of small firms, here meaning commercially active firms with less than 50 employees, were drawn in Sweden it would have the following composition of firm sizes: 62 percent self-employed without employees; just short of 35 percent micro-firms with 1-9 employees, and a remainder of less than 4 percent firms with 10-49 employees (NUTEK, 2002). I dare ask, are the solo self-employed economically and theoretically sixteen times more important just because they are sixteen times as many as the “large” small firms? I dare answer no, for most conceivable research questions they are not!5 The same goes for emerging ventures. These are a heterogeneous mix of different “types” (Bhave, 1994; Carter et al, 1996; Samuelsson, 2001; Sarasvathy, 2001), some of which may in a particular country at a particular time be larger in numbers whereas others may be economically more significant on a per capita basis. Mixing them all in one sample, and giving little weight to those that are small in numbers in the specific empirical population from which the sample was drawn, is likely to lead us to forego important findings about significant economic phenomena. As regards regions, if a country consists of ten large (population-wise) and economically growing regions and 90 small and backward ones, should all regions weigh equally in the analysis? If the size differences should be considered, by what criterion should the regions be weighted in the analysis? What we are discussing here is, of course, a special case of the general feature of heterogeneity and how it should be handled. There are more dimensions to this than those treated in conventional method texts. The upside is that this leaves room for creative judgment by the researcher and not just application of indisputable rules. There are several implications emanating from these lines of reasoning. The first is that simple random sampling is not necessarily the ideal. Stratified and deliberately “narrow” statistical samples, and even judgment samples, may on theoretical grounds be preferable in some situations. This first implication relates to the second, which I have already mentioned: that the more important issue about sampling is not statistical but theoretical representativeness. That is, it should be carefully ascertained—and communicated—that the elements in the sample represent the type of phenomenon that the theory makes statements about. This is, by the way, equally relevant for case study research. The third implication, again related to the previous ones, is that replication—not statistical significance testing— is the crucial theory test. The development and testing of sound theory requires replication in several sub-groups of analyzable size within the same study, as well as across several studies that investigate theoretically relevant samples from different

5 However, excluding the smallest firms from the design can have disastrous consequences in relation to some research questions. As will be explained in Chapter 7, the job creation prowess of new and small firm is a case in point here.

70

RESEARCHING ENTREPRENEURSHIP

empirical populations. This, then, further reinforces the importance of replications like those that will be presented in Chapter 9. SAMPLING INDIVIDUALS How can we obtain a theoretically relevant—and hopefully statistically representative—sample of individuals for an entrepreneurship study? “Simple”, we thought in early entrepreneurship research, “We take a sample of entrepreneurs and see what they’re like!” (cf. Hornaday, 1982). “Well,” said others, “we really need to compare the entrepreneurs to other groups—such as managers or people in general—in order to learn what’s unique about them” (cf. Brockhaus, 1982). “Worse still,” said some really nasty voices, “we don’t even know what an “entrepreneur” is, so how can we possibly sample them and investigate what they are like?” (cf. Kilby, 1971). And then came a relieving voice saying “Hey, it’s not about what they’re like, but what they do, and we know what they do: they create organizations! Let’s investigate how!” (cf. Gartner, 1988). Sampling individuals for an entrepreneurship study is not a simple matter. Consider the first backbone reaction: let’s study entrepreneurs! This is not a welldefined population but a hazy and moving target. Hence, it is not possible to create an indisputable sampling frame. Some people sometimes engage in entrepreneurial activities as we have defined them in Chapter 2. At other times they don’t—but then other people are active in entrepreneurial endeavors. Whom should we include in the sampling frame? Self-employed? Owner-managers? Current venture champions? All who have ever engaged in any behavior we define as “entrepreneurship”? By the way, how and where do we obtain contact information for these people? Assume we have defined a sampling frame of current “entrepreneurs” that we can live with, as well as a comparison group. Now we compare the two groups and find some differences. How should these differences be interpreted? As causal factors that make people engage in entrepreneurship, right? Well, so we would like to believe. The problem is that when we compare people “currently active in entrepreneurship” with those who currently are not, we confound several different factors: The propensity to engage in such behavior. Those with higher propensity should, ceteris paribus, have a higher likelihood of ending up in our “entrepreneur” sample. The ability to succeed in such behavior. Those who are successful in entrepreneurial endeavors should, ceteris paribus, have a higher likelihood of still being members of the group(s) we sample as “entrepreneurs” and therefore end up in that sample. The propensity to persist in the face of failure. Those who try again, or stay in business despite sub-standard performance (cf. Gimeno et al, 1997) should, ceteris paribus, have a higher likelihood of ending up in our “entrepreneur” sample. Situational (i.e., not person-related) factors that contribute to engaging, succeeding or persisting in entrepreneurship.

Sampling Issues

71

To put it mildly, this idea of cross-sectional comparison of “entrepreneurs” with others maybe wasn’t as great as we thought it was. In order to really sort out those different effects, we should start with a cohort of teenagers and follow who starts, succeeds, and persists in entrepreneurship over a very long period of time. And yet, along with the firm (see below), the individual is the most common level of analysis in entrepreneurship research—often meaning that the researcher has sampled a group of practicing “entrepreneurs” and investigated them in relation to some theoretical or empirical standard. Perhaps we should reconsider? The interest in the individual in early entrepreneurship research was derived from an idea that entrepreneurship was to be explained by the unique characteristics of the individuals who engaged in it. As pointed out in Chapter 2, this reflects what Ross (1977) called “the fundamental attribution error”, i.e., the tendency to look for person-related explanations also when it is in fact structural and situational factors that determine the outcome. Gartner (1988) suggested that the obvious characteristic of baseball players is that they play baseball well. The analogy is that entrepreneurs, according to Gartner, create organizations. Therefore we should look at their behavior rather than some personality characteristics. However, although Gartner took a step in the right direction, studying baseball players’ behavior does not necessarily tell us how they developed their skill or why they started playing baseball in the first place. And when people suddenly quit baseball playing—or bowling (Putnam, 1995)—on a massive scale, we are probably well advised to look beyond the individual level for explanations. That is, individual level studies tend to be under-socialized (Aldrich, 1999). But, of course, there are interesting and important questions that can—and should—be investigated on the individual level of analysis. Indeed, one of Shane and Venkataraman’s (2000) central research questions is “why, when and how some people and not others discover and exploit these opportunities”. As I see it, some particularly types of worthy studies are: 1.

2.

The study of expert (habitual; repeatedly successful) vs. novice entrepreneurs (cf. Gustavsson, 2004; Ucbasaran et al, 2001). This type of study holds some promise of generating knowledge about the teachable and learnable skills that signify successful entrepreneurship. In terms of sampling, an indisputable sampling frame is not to be hoped for, but stringent criteria for being classified as “expert” and “novice” should be employed. The use of cognitive theory on expertise may help this type of research in general, including the identification of sampling criteria. The study of fit between individual(s) and business idea(s) (Shane, 2000). This type of study will also require a sample of “practicing entrepreneurs”, and therefore the construction of the sampling frame will always be subject to potential criticism. Like (1) the researcher should try to do better than using any available list of self-employed, members of an association of owner-managers, or the like. Sometimes, it may be necessary to start from such a frame, but then

RESEARCHING ENTREPRENEURSHIP

72

3.

additional information collected from the respondents can be used to distill a sub-sample that is better qualified from a theoretical point of view. Studies of how structural and situational factors influence entrepreneurial behavior. I have argued above and elsewhere (Davidsson, 1992, 2002) that if the explanation for entrepreneurial behavior is not innate characteristics of individuals, then entrepreneurship research on the individual level can use any sample of individuals. This is true for the study of situational influences. We can here think of laboratory research where the researcher manipulates hypothetical situations, so as to induce entrepreneurial attitudes, beliefs and behaviors, or decision-making in entrepreneurial situations. Student subjects are often used for this type of research, but at least undergraduates can be seriously questioned as subjects in an entrepreneurship study. A sample from the adult population in general is preferable. In Sweden, obtaining such a sample is relatively easy as a sampling frame including all legal residents is available at reasonable cost. As I am blissfully unaware of all the problems involved in such sampling in specific other countries, the reader is advised to consult local sources on this issue. However, for laboratory research on the influence of structural and situational factors, the possibility of randomly assigning individuals to treatments is much more important than having a sample that is exactly representative for the underlying population. How structural and situational factors influence entrepreneurial behavior can also be the topic of real world studies. They would then have to be large and longitudinal in order to get sufficient variation in the dependent variable. That is, by the last wave of data collection a sufficient number of cases must have engaged in entrepreneurship as a result of the situational and structural factors under study. For this type of study a sample of the general population or some well-defined part thereof should be used.

I have here not discussed the intricate problems of sampling “nascent entrepreneurs”. This is because at the entry point, sampling individuals and sampling emerging new ventures coincide. This type of sampling will be described in some detail below. Here we may note that the study of nascent entrepreneurs can be combined with studying experts vs. novices, and fit between individual and idea, as described above. We may also sum up that drawing a theoretically relevant sample of individuals is not an easy task. SAMPLING EMERGING NEW VENTURES I have pointed out as particularly important and promising for entrepreneurship research the type of study that uses the venture idea, including the activity and organization that evolves around it, as the level of analysis. This is also a relatively neglected type of study, and a tricky one from a sampling point of view. For these

Sampling Issues

73

reasons I will devote relatively more space to this level (cf. Davidsson, 2003b; Davidsson & Wiklund, 2001). Identifying an Eligible Sample of On-Going Independent Venture Start-ups How can we identify on-going start-up processes? Some of the options would seem to be the following: 1.

2.

3.

Use informants at support agencies and the like. This would be a low cost, workable approach. Unfortunately, selection bias would severely hamper statistical representativeness and probably theoretical representativeness as well. Among independent start-ups a large proportion has not been in contact with a support agency. According to unpublished analyses of the Swedish version of the Panel Study of Entrepreneurial Dynamics (PSED) this proportion is 61 percent. Moreover, experienced business founders are less likely to take such contacts, so the self-selection is likely to really cause a bias. Use the first visible trace that the new venture leaves in some type of register, e.g., registration of a new proprietorship, partnership, or limited liability company. Although this could be a satisfactory solution in some countries and for some purposes, it would be unsatisfactory in most cases. In many countries, the smallest (and therefore the youngest) firms never enter any registers (Aldrich, Kalleberg, Marsden & Cassell, 1989). If they do, they often do so at a later stage, when they are already an established entity rather than an emerging one. Thus, they would not be eligible for a study of ongoing start-up processes. One example of the problem of severe undercoverage and late entry is the use of Dun & Bradstreet data, which are based on applications for external funding (cf. Davidsson, 1994). Using registration as identification remains problematic even when registration is a prerequisite during the process of emergence. This is so because start-up processes follow many different sequences (Bhave, 1994; Carter et al, 1996; Delmar & Shane, 2002; Sarasvathy, 2001), and though registration may in some cases be a very early step in the start-up process it is in other cases likely to be a late one. Thus, even though the same indicator is used to approximate the initiation of the process the cases will all the same be at different stages of development when they are sampled. This may make the researcher confuse “caught at late stage” with “quick to finish”. Snowball sampling (Douglas & Craig, 1983). The logic of snowball sampling is that members of small, expert populations are likely to know (about) one another. That is, the sampling strategy would be to find some on-going start-up processes through whatever means, and have those involved report on others who are also in the process of starting new ventures. Again, this would lead to a known and specific selection bias. Start-up processes led by better networked champions

74

RESEARCHING ENTREPRENEURSHIP would be more likely to be caught, and this would have a serious biasing effect as networking or social capital is known to have an important influence on making progress in the start-up process (Davidsson & Honig, 2003). In addition, the basic assumption that the relevant individuals know each other can be called in question in this case. There is research indicating that even among established business owner-managers within the same, specialized industry there can be a surprising level of ignorance about each other’s existence (Utterback & Reitberger, 1982).

The simple fact is that there is no fully satisfactory solution to the challenge of obtaining a representative sample of on-going, independent start-up processes. There is no way we will ever be able to sample strictly randomly (or probabilistically) from the universe of venture ideas. This is a problem not only for deductive, theorytesting work but also for exploratory, case-study designs. Although valuable ideas for theory development can be obtained from any empirical case, there is the risk that the resulting theory will have limited applicability. Indeed, the most important conclusion from Samuelsson’s (2001; 2004) work is that existing theories do a reasonable job at explaining the start-up process for innovative ventures, but not so for reproducing ventures—which happens to be the majority. Theorizing based on non-representative cases or samples may be the reason for this. The fact that obtaining the ideal sample is impossible does not mean that trying to approach that ideal is a futile effort. Realizing the limitations of the above approaches, the PSED and related research has adopted a two-step procedure (Gartner et al, 2004; Reynolds, 2000). The first step is to approximate as closely as possible a very large random sample of adult individuals. The sample thus obtained is the screening sample. The focal screening question posed to these individuals was: Are you, alone or with others, now trying to start a new business? (indication of being a “nascent entrepreneur”; NE)

Those who identified themselves as NE’s were—along with those identifying themselves as nascent intrapreneurs (see below), and a randomly selected comparison group—directed to a longer interview, and eligible cases were then followed up longitudinally. I consider this approach to capturing emerging ventures a giant leap forward, and I feel confident that it will in some form remain a standard tool in entrepreneurship research in the future. Nevertheless, the approach has shortcomings that have to be considered and eliminated, to the extent possible. The first problem is that the procedure is costly. Hit rates of 1-10 percent should be expected (Davidsson & Henreksson, 2002; Reynolds, Bygrave, Autio, Cox & Hay, 2002), which means that very large samples have to be screened in order to obtain a sizeable valid sample in the end. Attempts to employ techniques for more efficient, yet probabilistic sampling (see Reynolds & Miller, 1992) have not proven very successful in this context (Reynolds, personal communication). The second problem has to do with relying on the respondents’ subjective interpretations of what should and should not

Sampling Issues

75

be counted as “now trying to start a business”. People differ in what they mean by “now” as well as by “trying” and “business”. This problem may be different in different countries, and therefore the specific wording of the screening questions is crucial, in particular when conducting international comparative work. For example, work related to PSED and GEM have indicated that in Germany and Ireland, a substantial number of “no” or “don’t know” answers may occur when the researcher would have wanted a “yes” (C. O’Gorman; F. Roche, and F. Welter, personal communication). This could be due to an uncertainty among independent professionals and founders of “craft” businesses as to whether what they start is really a “business”. Also within countries or cultures people differ in their perceptions of what qualifies. As a remedy to this, the PSED research has applied more objective supplementary criteria for eligibility. Here the PSED questions about “gestation activities” are useful. Respondents were asked whether they had initiated or completed each of more than twenty activities (such as writing a business plan, talking to the bank, registering the business, talking to would-be customers, and the like (cf. Davidsson & Honig, 2003; Gartner & Carter, 2003, and Chapter 6 in this book). In addition, each affirmative answer to such activity questions was followed up with questions about what year and month the activity was initiated, undertaken, or completed. In this way all activities were time stamped by year and month, which turned out to be a very important and useful aspect of the design. A common minimum criterion employed by PSED researchers is that at least one “gestation activity” has been undertaken (Delmar & Davidsson, 2000). Sometimes stricter criteria are employed. In the Swedish PSED, it turns out that some self-appointed “nascent entrepreneurs” initiated the start-up process many, many years ago—and still have not completed it. Based on the assumption that such respondents probably are not very serious about the reported start-up, one might consider eliminating such cases entirely (cf. Delmar & Shane, 2002). Likewise, a criterion that at least x activities have been undertaken during the last y months may be considered, so as to ensure that it is really an active start-up effort (Shaver, Gartner, Crosby, Bakalarova & Gatewood, 2001). A maximum criterion is also needed, in order to establish that the case is an ongoing start-up and not an already established business. Because of the multitude of gestation activities and the many different sequences in which they are undertaken— as well as the fact that not all activities apply to all start-ups—there is no simple criterion that is fully satisfactory. PSED researchers have sometimes employed a single criterion, namely that the venture had achieved sufficient cash flow for three months to pay expenses and the owner-manager’s salary (Shaver, Gartner et al, 2001). Other PSED researchers have preferred a combination of criteria. For example, in Delmar & Davidsson (2000) we considered the venture as already started if a) money had been invested, b) a legal entity had been formed, and c) the venture had generated some income. See also Shaver, Carter, Gartner & Reynolds

76

RESEARCHING ENTREPRENEURSHIP

(2001) on the problem of determining who is eligible in a sample of “nascent entrepreneurs”6. Sampling bias. Having solved the eligibility problem, and assuming that the sampling frame of adult individuals is relatively complete as well as that there is no devastating non-response bias, the above procedure leads (approximately) to a random sample of “nascent entrepreneurs”. Although this is for many purposes far better than having a non-random sample, it is not the random (or probability) sample of emerging new ventures (or start-up processes) that we were after. This is so for several reasons. The first reason is relatively obvious whereas the others are more subtle. The easy-to-spot flaw is that the above procedure over-samples team startups, because the higher the number of team members who consider themselves NE’s, the more chances to be sampled has the venture. If information on the number of team members is available—as is the case with PSED, albeit truncated at five— this should be relatively easy to correct for, if deemed important. As was explained above I do for most purposes not find this type of statistical non-representativeness to be the most important problem researchers should worry about. A much more problematic and intriguing source of bias is that start-up processes have different duration. Assuming that the team vs. solo effect has been accounted for, the resulting sample is, in a sense, representative for the population of on-going processes of independent venture creation at the time of the empirical investigation. It is not, however, a representative sample of all such efforts that were undertaken during that year. In order to see this, consider the following example. Assume the entire population of start-up processes in a given year consists of 40 cases. Ten of those are “slow” start-ups, which are initiated on January 1 and completed on December 31. The other 30 are “quick” start-ups, which take four months from initiation to completion. Ten each are initiated on January 1, May 1, and September 1, and consequently ten “quick” start-ups are completed on April 30, August 31, and December 31. Now, although the proportion of “quick” to “slow” start-ups is three to one on a yearly basis, we will sample from a population with a 50/50 distribution no matter what date we select for our sampling. That is, the procedure over-samples “slow” start-up processes. This is a serious bias by the logic of statistical inference theory, because it is likely to distort results. On the one hand, there is the risk that the oversampling represents less committed and/or less successful start-up efforts. On the other hand, technology-based, high-potential start-ups are also likely to require longer time. Rather than hoping that these effects cancel out the analyst should be aware of this problem, and try different strategies for solving it. This might include elimination of cases that appear to be “eternal start-ups”; weighing; sub-sample analysis by length of process, and the use of process length as control variable. Yet another source of bias makes it uncertain whether even “snapshot” representativeness for “nascent entrepreneurs” is fully achieved with the suggested procedure. Bhave (1994) identified two main processes leading to independent startups. The first, which he labels “externally stimulated” starts with a wish to strike out 6

Because of their psychological vantage point, Shaver, Carter et al (2001) regard PSED as a sample of individuals. At the entry point eligible individuals are the same as eligible venture start-ups. In followups, separate decisions may have to be taken with respect to these two levels of analysis.

Sampling Issues

77

on one’s own, followed by search, screening and selection of business ideas (“opportunities”). The second, labeled “internally stimulated”, starts with identification of a personal need; continues with need fulfillment, and only then does the individual realize that this problem is general and that the solution has commercial potential. In this latter case, the individuals involved may get much further into the process before they start consider—and report—themselves as “nascent entrepreneurs”. If so, the researcher should be aware that this type of startup process will be under-sampled, and try to find remedy as fit for the research problem at hand. Again, as has been explained above, I do not think deviations from proportionate representation of the empirical population are necessarily a huge problem, as long as theoretically important categories have satisfactory representation in the sample. A much worse problem may result not from undersampling of “internally stimulated” processes but from the fact that they are further into the process when first caught. If this is not carefully controlled for in the analysis, there is a risk that differences in performance are attributed to process type when differences in starting point is the real cause. This is so because if we do not control for the fact that “internally stimulated” processes are on average further into the process when sampled, we will interpret the (probable) result that they finish quicker as showing that they are somehow “better”.

Sampling On-going Internal Venture Start-ups Emerging internal ventures may also be possible to identify through informants at support agencies or through registration procedures. For example, innovative projects may receive subsidies from government agencies, and the new activity may leave traces in the form of patent applications or registration of a new establishment or daughter company. However, by and large such sampling mechanisms are probably even more inadequate for achieving representativeness—statistically and theoretically—for emerging internal ventures than for standalone start-ups. Another option that can be—and was—used with a PSED-type study is to screen individuals also for internal ventures. That is, in PSED the respondents in the screening sample were also asked: Are you, alone or with others, now starting a new business or new venture for your employer? An effort that is part of your job assignment? (indication of being a “nascent intrapreneur”; NI).

However, I am not convinced that starting from a sample of individuals, and the PSED screening questions, are the ideal tools for sampling of internal ventures. The subjective variation among respondents concerning what qualifies here seems to be even greater than for independent start-up efforts. As a result the sample may become an arbitrary rather than a random one. We have therefore employed a slightly different strategy in our study of on-going new internal venture start-ups (Chandler et al, 2003). Instead of screening individuals, we choose to screen firms for new initiatives. One advantage of this is that it eliminates the over-sampling of team efforts. Arguably, relative to the NI item in the PSED research (cf. above) it should also reduce the problems of over- and under-reporting.

78

RESEARCHING ENTREPRENEURSHIP

We were fortunate enough to have an on-going study of a very large sample of firms in the 1994 Start-up Cohort (Dahlqvist & Davidsson, 2000; Dahlqvist et al, 2000), so we used the surviving firms (in the year 2000) in this study as the sampling frame. Researchers who start from a less fortunate situation would of course have to first obtain a suitable screening sample of firms, adding to the cost of the project. Because our firms had previously been approached with mail questionnaires (where the first few questions were mandatory data collection for a government agency, thus yielding high response rates), we choose a mail questionnaire directed at the CEO for the screening. Under other circumstances, phone interviewing would probably yield a higher response rate. The focal screening questions were the following: Have you after the start of this company in 1994, started any new venture within the company, which during some period has provided income to the company? We are interested in new business initiatives in your company, which have led or could lead to new income-generating activities. NB! Not mergers or acquisitions. Do you have a business initiative in progress now, which you or others in the company have devoted time and possibly other resources to develop, but where the new activity does not yet yield a steady income?

Additional questions asked when the new initiative in (2) was initiated, and whether the respondent had started any additional firms (separate from the sampled one) since 1994. The first question above is intended to define “new initiative” (or new internal venture) and to separate up-and-running initiatives from on-going ones. The critical screening question is (2). Those who answered this question affirmatively were later contacted for a phone interview. In the ensuing phone interview the eligibility of the initiative was double-checked with the following question: By initiative towards new business activity we mean attempts to change or expand the business, like, for example, developing new products or services, aiming for completely new customers, or entering new markets. We are interested in all such changes, which could affect your future income to a non-negligible degree. With this clarification, would you say that you today have any new initiative towards business activity in progress, which you or others in the company have devoted time and possibly other resources to develop, but where the new activity does not yet create a steady income?

If the firm had several eligible, on-going initiatives to choose from, the one deemed by the respondent to be “most important for the company right now” was chosen. As far as we can tell at the time of this writing, this strategy for sampling on-going internal ventures seems to be working satisfactorily. However, the twostep procedure, and in particular the phone contact, turned out to be very important. Because of the dual checks, many non-eligible cases could be eliminated after clarifying interaction over the phone. It should also be mentioned that our approach shares some of the problems with the PSED approach. First, the screening is costly. In our case, 4950 firms were contacted for a yield of only 250 eligible cases; a sample size which is further reduced through attrition in the longitudinal follow-ups (Chandler et al, 2003) as well as elimination of cases that in the light of additional information were not eligible after all. The firms in our sample were predominantly

Sampling Issues

79

micro-businesses. The yield would almost certainly be much higher if somewhat older and (in particular) larger firms were screened for emerging new ventures. However, increasing firm size also introduces new complications to which we shall return below. The eligibility problem is similar to the PSED and soluble through criteria for being over- or under-qualified. The problem of sampling bias because of variations in process length remains the same. There is also a parallel to the concern discussed above that externally and internally stimulated independent start-ups may be caught, on average, at different stages in the process. In Chandler et al (2003) we discuss three search processes behind ideas for new internal ventures: pro-active search, reactive search, and fortuitous discovery. It may be suspected that the latter category is less likely to be reported as a new initiative at very early stages of the process. This would lead to the same problem with under-sampling and possible confounding of effects as discussed above for internally stimulated processes in the context of independent start-ups. Time-stamping of specific behaviors may help reduce this problem. Letting the respondent choose “the most important” initiative when several eligible initiatives existed is, of course, a threat to representativeness. From a statistical point of view, a more defensible procedure would be to check the number of eligible processes in a firm, pick one randomly, and adjust in the analysis for the resulting under-sampling of new venture processes within multi-initiative firms. In our study, this problem was of little practical importance, as very few firms had more than one new initiative under way. With larger firms in the sampling frame it would be more of a problem. Indeed, above a certain firm size, almost every sampled firm would likely have more than one new internal venture under way. This calls for a more sophisticated procedure for choosing among them and—if several ventures per firm are included in the sample—techniques for adjusting for statistical dependence between cases with the same origin. Before selecting what ventures to include, however, one should ascertain that all relevant new ventures have been identified. In our cohort of young and mostly very small firms, it was reasonable to assume that the CEO/respondent be aware of any new venture initiatives going on in the firm. A study starting from a sample of large firms would either have to give up ambitions towards statistical representativeness, or develop a procedure for first locating a sufficient number of relevant informants representing different roles in the company. All of these informants would then have to be screened for information on the existence and nature of on-going internal venture initiatives. Something which is probably more of an issue with small (and independently owned) firms than with large ones is whether the new venture is going to form part of the original firm, or become a legally separate business. These two possibilities should be acknowledged in the design of the study and considered in the analysis. I personally see no reason to decide a priori to include only one type or only the other. On the contrary, this choice of “mode of exploitation” (Shane & Venkataraman, 2000) can be an interesting research question in itself. Moreover, it would be impractical to introduce such a limitation, as the respondent has not necessarily made this decision when the start-up process is first captured.

80

RESEARCHING ENTREPRENEURSHIP

In summary, this long section of the chapter has argued that sampling emerging ventures at an early stage is a cumbersome and costly endeavor loaded with difficulties and problems—but also that it is possible and important. For those researchers who wish to be different and possibly achieve more than publishable but quickly forgotten research in the mainstream there is plenty of challenging opportunity. Go for it!

SAMPLING FIRMS After this Golgotha-walk of sampling such elusive entities as emerging new ventures many a cautiously natured reader are likely to have turned a deserter, already half way to the safe haven of conventional, firm level study. Compared to “emerging new ventures”, sampling firms should be a piece of cake, right? Pah! You ain’t seen nothin’ yet! Yes, the firm is the most common sampling unit in entrepreneurship research (Chandler & Lyon, 2001; Davidsson & Wiklund, 2001). The reason for this is not, however, that the firm is the most relevant or the most unproblematic level of analysis. The reason why it is so often selected while so rarely analytically dissected is simply, I would argue, that we are blinded by our conventions. One rather common use of the firm as level of analysis in entrepreneurship studies is to sample existing firms in order to investigate their own coming-to-being. I regard this as a retrospective and—for reasons discussed above—often inferior alternative to the concurrent study of processes of emergence of new, independent ventures, as discussed above. What I have in mind in this section is instead the use of firm level study in order to study entrepreneurship within or by the established firm (“intrapreneurship”; “corporate entrepreneurship”), such as the launch of new products; entry into new markets, etc. Although usually regarded a micro-level unit, the firm is (often) already an aggregate of different individuals and business activities. In discussing the sampling of such aggregate units, starting with firms and continuing with industries and spatial units, I will organize the discussion around the relevance, size, size distribution, and heterogeneity along other dimensions of the units to be sampled. I will let the longer discussion of relevance wait till last, and start with a combined discussion of the other three criteria.

Size, Size Distribution, and Heterogeneity along Other Dimensions This entity we call “firm” comes in a variety of sizes from part time, home-based businesses with miniscule sales to multinational giants with hundreds of thousands of employees and a budget larger than the GDP of many a small nation. As researchers we have to handle this variability; otherwise we risk working with crap data and comparing apples with oranges. As regards the absolute size the question to ask about minimum size is: Will the sampled units be big enough for the entrepreneurial behavior we are investigating to have any likelihood of occurring within the studied time frame? As regards the upper limit for absolute firm size the question instead becomes: Can we obtain reliable data on entrepreneurship within this entity with the data collection method we are going to use?

Sampling Issues

81

If most of the sampled firms are so small that what we operationalize as entrepreneurship (e.g., launching a new, internal venture) almost never happens, we will end up with a dependent variable with limited and very erratic variance. As a result, we are likely to get weak and confusing results. Although the reasons for doing so were not necessarily as well articulated at the time, I have in fact used a minimum size criterion in all the firm level studies I have been involved in. For example, in my dissertation project Continued Entrepreneurship and Small Firm Growth (Davidsson, 1989a, 1989b, 1991) it was set at two employees. In the Entrepreneurship in Different Organizational Contexts study (Brown et al, 2001; Salvato, 2002) we used ten employees, and in our study of High Growth Firms (Davidsson & Delmar, 1999, 2001, 2003; Delmar et al, 2003) we set the minimum at 20 employees. In the 1994 Start-up Cohort study (Dahlqvist & Davidsson, 2000; Dahlqvist et al, 2000) the minimum criterion was that there was proof that the firm was commercially active, as indicated by registration as employer and/or for sales tax and/or corporate tax. As a result, this sample includes many very small firms. This came to illustrate the problem of insufficient minimum size when screening for cases for our New Internal Ventures study (Chandler et al, 2002; 2003). As reported above, only about 250 new internal ventures were found in a screening of close to 5000 firms. The issue of minimum size, then, overlaps with the question of relevance (cf. below). In order to have relevant variation in the dependent variable (i.e., some aspect of entrepreneurship), the firms in the sample may have to be of a certain size. Alternatively, the time span for which to report entrepreneurial behaviors can be extended, but in concurrent studies this increases time and cost, and in retrospective studies it aggravates the problem of bias from hindsight and memory decay. Let’s now return briefly to the upper limit for absolute firm size. Using secondary data, the possible choices are restricted by how the provider of the data organizes the information. Indicators of corporate entrepreneurship activity—such as filing for patents, registration of new establishments, etc—may be linked to a certain level of aggregation in a corporate hierarchy of establishments and companies (cf. below). However, as long as the ownership links between establishments and companies in a corporate hierarchy is known, the researcher can aggregate the data to any higher level than the original, according to her preferences. The problem with large absolute size is actually worse when primary (interview or questionnaire) data are collected. Not only does it become increasingly unlikely with increasing size that a single respondent can adequately report for the entire firm. With increasing size it is also increasingly unlikely that the CEO is willing to participate in the study7. For this reason I have never used a single informant for units larger than 250 employees, and I strongly recommend using multiple informants and other means of

7 1 can tell those who believe they collect mail survey data from CEOs of multinationals that probably they do not. Long before I became a researcher I learnt from my father who really filled out the questionnaires addressed to the CEO. At the time he had an idiosyncratic position as speechwriter and expert on business cycles as well as communist block barter trade—and questionnaire filler—for the CEO of a multinational (Sandvik AB). Yes, I sometimes trust samples of one!

82

RESEARCHING ENTREPRENEURSHIP

triangulation for firms this big or bigger. It is actually desirable for much smaller units as well. As regards the size distribution problem, we noted already in the introduction to this chapter that a simple random sample of “firms” would be dominated by entities that are tiny in size. In order to ascertain representation of somewhat larger firms one could stratify the sample by size, but by the conventional logic of statistical inference theory one should then weigh one’s full-sample analyses back to mirroring the size distribution of the underlying population, which means letting one’s result be totally dominated by the micro firms. The fundamental problem here seems to be that similarly to some substantive theories, statistical inference theory seems to be based on an assumption that aggregate phenomena are the result of adding up the behaviors of equally important (if not identical) micro level units. Again, the problem is that the theory does not adequately acknowledge the heterogeneity that we have taken as a fundamental assumption about the nature of the economy. The great variation in size is one aspect of this, but firms are heterogeneous along many other dimensions as well. For this reason I think that in most situations the researcher should give up aspirations to achieve exact statistical representativeness. It is simply not desirable. Instead, we should work with samples that acknowledge the heterogeneity, i.e., samples that have a reasonable and balanced representation of different kinds of valid empirical manifestations of the theoretical concept “firm”. In order to make both full-sample and sub-sample analyses meaningful we should also limit and control the heterogeneity. If the sample is too diverse, it may be difficult to find operationalizations of key concepts that work for all firms in the sample, a topic we will return to in the next chapter. It may also be difficult to arrive at strong results for the full sample, or find distinct sub-groups that are large enough for meaningful analysis. Therefore, I recommend that samples on the firm level be either narrowed down to a more homogeneous category of firm, or stratified along relevant dimensions so as to represent several such more homogenous categories. Hopefully, in the latter case the strata can collectively be regarded as a valid representation of the theoretical “firm” concept in all its richness. Stratification is the main the strategy I have followed in my own research. Returning to the studies referred to above, the Continued Entrepreneurship and Small Firm Growth study was stratified by size class (2-4; 5-9, and 10-19 employees) and industry (manufacturing of metal products and machinery; manufacturing of high-tech products; repair services, and retailing in clothing and home equipment). The Entrepreneurship in Different Organizational Contexts study was stratified along three dimensions, viz. size-class (10-49 and 50249 employees), industry (manufacturing; knowledge-intensive services; retail and wholesale, and other services), and governance (independent; part of company group with less than 250 employees in total, and ditto with 250 or more employees). In both of those cases the stratification served us well in terms of sub-sample comparison and validation of results (Brown et al, 2001; Davidsson, 1989a).

Sampling Issues

83

More for representation than for homogenization purposes, the government agency that conducted the initial survey of the 1994 Start-up Cohort stratified it according to industry, legal form and geographical location. This sample is also homogenized in the sense that it is a cohort; the firms all share the characteristics that they are independent and were started the same year. The Business Platform study (Davidsson & Klofsten, 2003; not mentioned above) used a narrow sample of technology- and knowledge-based firms located at Swedish technopoles. Being a census rather than a sample, the High Growth Firms study was not pre-stratified. To sum up, when sampling firms I have argued that the units must be large enough for the investigated aspect of entrepreneurship to have a reasonable likelihood of occurring, but not too large for obtaining reliable data through the chosen data collection method. Heterogeneity in size and along other dimensions ought to be both acknowledged and controlled, so that the resulting sample adequately reflects the theoretical concept of “firm” while at the same time it should make it possible to apply the meaningful operationalizations that are required in order to arrive at strong findings. Carefully thought through stratification, then, is the key to successful sampling of firms for entrepreneurship studies. Relevance If you thought we were done I can ensure you that the fun has just begun! What is an empirical “firm”? Well, we do have a few alternatives to choose from. Different legal forms could have been discussed above already. What should be included? Limited liability companies (of which there are several different types in many countries) only? Partnerships? Sole proprietorships? All of the above? Well, there are also foundations, cooperatives and various other types of associations that can be commercially active. What legal forms should be included may differ from country to country. In the firm level studies I have conducted I have sampled either limited liability companies solely, or added also partnerships and sole proprietorships, in order to arrive at a theoretically relevant and workable sample. Sometimes researchers use a unit that is not even a legal entity to represent “the firm”, namely the establishment. An establishment is a place of work. In a retail chain, each outlet would be an establishment whether or not it was also a separate company. In the manufacturing industry, each geographically separated plant is an establishment. The reason researchers turn to establishment data is almost without exception practical considerations rather than the question of theoretical relevance. In the first part of the Business Dynamics in Sweden project (Davidsson et al, 1994a, 1994b) we felt we had to use establishment data for data quality reasons. Establishment level data tend to be more current and reliable, and establishment codes are not changed as easily as are company codes, so establishments are in secondary data sets less subject to artificial births and deaths due to mere reregistration. In the second part of the project (Davidsson et al, 1996, 1998b) we found a way to aggregate establishments data to the theoretically more relevant firm

84

RESEARCHING ENTREPRENEURSHIP

level. However, this project used the region (and to some extent industry and nation) as the level of analysis, so the question of defining “firm” was an operationalization issue rather than a matter of sampling. I will elaborate on operationalization issues in Chapter 6 and come back to converting secondary data to fit one’s purposes in Chapter 7. There are at least two other empirical structures that researchers use when they sample “firms”. These are the company (or enterprise) and the company group (or multi-company corporation). Figure 5:1 illustrates some of the complexity of the matter—but only some. The first case, (a), is an easy one. This is the singleestablishment firm. Whether we sample establishments or companies we end up with the same entity, and there can be little doubt that this entity is an acceptable representative of the theoretical category “firm”. With exhibit (b) we start to make things complicated. Here we have an independent company with three establishments. In this case we assume that all three establishments are active in the same industry (that’s why they share index “x”). What is here the firm—the company or each establishment separately? As will be argued below, the answer can be contingent on what specific conceptualization of “firm” we are working with. Suffice it here to note that it is not a simple matter to determine what the “right” level is, because behind this sketchy representation we can find different realities. The three establishments could be three branch offices of a consultancy business, not operating under a strong brand name or common concept, but all enjoying a great deal of freedom and relying on local knowledge and contacts. Or it could be three semi-independent outlets in a retail chain, each with its own manager but working under the same brand name and the restrictions of strong company policies regarding marketing and purchasing, for example. Alternatively, it could be three production plants in a manufacturing firm, producing to order and with a minimum of decision-making discretion. Case (c) is similar to (b) but the establishments are here active in different industries, according to their industry classifications. The entire company is classified as belonging to industry X because this is the single biggest activity in the firm. In actual fact this company is a conglomerate, operating in several industries. What is the firm here? Perhaps theory can guide us? Does the theory conceive of the firm as a power structure, or as an entity whose raison d’être is an aptitude to perform a certain type of activities? We will return to theoretical conceptualization of the firm later on, and the issue of matching theoretical and empirical firm definition.

Sampling Issues

Figure 5:1

Hierarchies of possible “firms”

85

86

RESEARCHING ENTREPRENEURSHIP

With panel (d) we approach the reality of large corporations, which are in fact much more complex structures. This figure is merely a portrayal of the principles of this complexity. This structure combines the issues discussed so far. Is it the individual establishments or the companies that are firms? Should the “firm” concept be used for an entity with a logically coherent set of activities or can any disparate conglomerate under the same boss/owner(s) qualify? Is a unit like Company Y independent enough (in relation to certain theoretical conceptualizations) to deserve the “firm” label, or is it but a mechanical servant of higher or lower levels in the hierarchy? The structure in (d) also adds another possibility: to use the entire structure that appears under the same ownership as the entity called “the firm”. This may be suitable for some purposes but adds, for example, the problem of determining how to deal with partly owned units further down the hierarchy. In the High Growth Firms study the choice between company and company group level was particularly tricky. The main (policy) purpose of the project was to investigate the prevalence and job contributions of rapidly growing firms in the Swedish economy, preferably as compared to other countries. The establishment level could easily be ruled out as irrelevant. There was no policy interest in that level and claims in studies in other countries that a tiny x percent of the business population (so-called “gazelles”) created a massive y percent of all new jobs in the economy concerned firms, not establishments (Birch, Haggerty & Parsons, 1995; Birch & Medoff, 1994; Storey, 1994). The problem was that regardless of whether we choose the company or the company group level we would risk serious underestimation of the high growth firms and their job contributions. With the company level we face the problem that growing companies are likely to eventually form company groups. If the study sticks to the company level of analysis, the analyst would miss this continued growth. If, instead, we choose the company group level, we would likely miss some spectacular growth firms within corporations because other parts of the company group were shrinking at the same time. Ericsson was a case in point during the period studied (1987-96)—while mobile phones and systems skyrocketed, other parts of the corporations shrunk quite dramatically. Do you remember Ericsson PCs? I didn’t think so. Our solution was to create data sets on both levels, as well as some ability to analyze across them (Davidsson & Delmar, 2003). Panel (e) illustrates an interesting case. Here, an individual or a team owns a series of separate businesses, so like in (d) there is common ownership and ultimate control of all levels. However, there is no cross-ownership within the group, and therefore the entire group will not turn up as one “firm” in any sampling frame. Is this a problem? Well, if we accept the entire group as a firm in the (d) structure, why shouldn’t we in the (e) structure just because the owners have chosen the latter form of internal organization of their empire? Perhaps this is not a big problem in practice, but I know for sure that it exists. We noted it already when planning the Business Dynamics in Sweden and High Growth Firms data sets, and in the sampling (of companies) for the Entrepreneurship in Different Organizational Contexts study

Sampling Issues

87

there was at least a couple of very successful entrepreneurs who got into the sample repeatedly precisely because they had structured their company groups in this way8. So, there are several answers to the question: What is an empirical firm? Now let’s turn to the parallel question: What is a theoretical firm? Again, there are plenty of suggestions (see e.g., Coase, 1937; Conner & Prahalad, 1996; Cyert & March, 1963; Daft & Weick, 1984; Foss, 1993, 1996; Mueller, 1972; Seth & Thomas, 1994; Wernerfelt, 1984,1995; Williamson, 1999). There is thus no shortage of ideas about what a firm is, and some of this conceptual work has actually been undertaken from the perspective of a particular interest in entrepreneurship (Alvarez & Barney, forthcoming; Foss & Klein, 2002; Mugler, 1990; Zander, 2002). The problem is that these different conceptualizations highlight different aspects of “the firm”, and therefore only partially overlap one another—and the empirical “firm” definitions we have discussed above. Sometimes the concept “firm” is used for so different entities that it is difficult to see much common ground at all. “‘The firm’ is not a firm”, as Edith Penrose (1959) put it, referring to the difference between real world commercial organizations and “the firm” in microeconomic theory. It would certainly be pretentious of me to claim sufficient mastery of theories of the firm for providing a fully informed discussion of the matching of theoretical and empirical delineations of “the firm” in entrepreneurship studies. I hope that highlighting the problem can in itself inspire researchers to think this through more carefully. I also hope that the admittedly crude treatment below can be of some direct assistance. If we start with microeconomic theory and similar conceptualizations, we find a firm that is a portrayed as a production function, or as the cost structure of a specific production process. In the base model, the firm produces only one type of goods. In more complex formulations, there may be several products with interrelated cost structures. I have questioned above whether this type of theory is very useful at all for entrepreneurship because it does not acknowledge heterogeneity (and leaves little room for creativity). However, to the extent that it—or other and more relevant theories with a similar “firm” notion—is used, I would argue that the empirical entity that best matches this conceptualization is the establishment. This is an entity that shares with this type of conceptualization a focus on one or a limited number of outputs, and little freedom to make its own decisions. A theoretical firm concept like that employed in microeconomic theory does not seem to accommodate great variety in the program of products and services offered by individual firms, nor internal power and governance relationships. Using it in combination with some of the more complex structures in Figure 5:1, therefore, seems to be a mismatch. Conceptualizations emphasizing the firm as a unique bundle of resources, knowledge or capabilities (Barney, 1991; Kogut & Zander, 1992; Penrose, 1959; Teece et al, 1997; Wernerfelt, 1984) are problematic to match with conglomerate firms. Therefore, the more complex structures in Figure 5:1 (c, d, and e) do not seem 8 Luckily, during my student days I had worked one summer for one of these guys who had been sampled half a dozen times, many years ago when he was setting the foundations for his hotel empire to be. He remembered my name when he got the cover letter and therefore generously shared his time when he was later contacted by an interviewer. There are many odd ways to minimize non-response! However, we were sensible enough not to have him go through the same questions six times...

88

RESEARCHING ENTREPRENEURSHIP

to be a good match with such theories, either. Resources, knowledge, and capabilities are relevant for performing certain types of tasks. They are usually not resources or capabilities that make the firm good at just anything. Although much closer to the realities of real world companies and corporations, these theories seem to share with microeconomics a rather narrow view on what “a firm” offers on the market. However, turning to the establishment as the matching empirical unit is probably not a good solution in this case. Strategic theories assume a degree of discretion in decision-making, which is not necessarily present within establishments that form part of larger structures. Ecological and evolutionary theory, which also emphasize distinct competence as the firm’s reason for being, put less emphasis on the deliberate actions of the decision-maker and are therefore more compatible with establishment-level analysis. The best match for strategic knowledge- and resource- based theoretical perspectives seems to be firms like (a) and (b) in Figure 5:1. For the other types of structures, one would like to cut out the different parts—perhaps “strategic business units—of the total structure that make logical units from a resource- or knowledge-based perspective. This is, of course, difficult to do in broadly based studies. Although externally rather than internally orientated, Porter’s (1980, 1985) strategic theory seems to share the same type of matching problem. Much of Porter’s theorizing is about industry attractiveness. Hence, “firms“ are assumed to operate in an industry and apply a strategy that is suitable for maintaining or improving the firm’s position in that industry. This does not seem to be a theory that works for empirical entities with a high degree of both horizontal and vertical heterogeneity. And here is the core of the matching problem for the strategic theories we are discussing. The problem arises when the theory is used for making predictions and discussing implications for the entire structure. If the internal heterogeneity of the sampled units is carefully considered in the design and derivation of hypotheses, the match between these theories and “firms” like the structures (c) and (d) in Figure 5:1 can be quite good. That is, in many cases the dependent variable should perhaps not concern the profitability or growth rate of the entire organization but instead focus on what parts of the organization will grow and yield a surplus. Theories that emphasize transaction costs or agency problems; governance and power issues; behavioral firm theory, and theories emphasizing organizational structure and incentive systems (Coase, 1937; Daft, 1983; Jensen & Meckling, 1976; Williamson, 1975) seem to be less sensitive to heterogeneity concerning the firm’s outputs and markets. They can therefore probably be meaningfully applied to different levels of “firm”—establishment, company, or multi-company corporation—as long as these different creatures are not mixed too wildly in the same sample. An issue for some of these perspectives would be whether a unit like Company Y in Figure 5:1 (d) is independent enough to qualify as a firm, and whether it can be included alongside with entities like Company X in (a) or (b) in the same figure without carefully distinguishing between the different hierarchical positions when making predictions and interpretations. Another concern that goes for these types of conceptualizations as well as for all those previously discussed, is whether it can be reasonably assumed, as the type of empirical firm grows in size and internal complexity, that the entities sampled have any consistent characteristics

Sampling Issues

89

throughout the organization along those dimension highlighted by the theory. For example, when a researcher wants to investigate whether organizational structures and incentive systems influence the occurrence and success of internal venturing in a firm, can it then be meaningful to work with “firms” that have thousands of employees working in spatially and legally separated companies that are also producing different types of products for likewise different types of customers? Without having thought it through very carefully, my spontaneous answer would be that most questions about entrepreneurship at the firm level are better researched in smaller and more homogenous units, be they independent firms or parts of a larger structure. It does not seem to be much of a problem from any specific theoretical perspective that structure (e) in Figure 5:1 does not appear as one firm in sampling frames. It is a problem, however, for studies that sample individuals (cf. above) and/or use psychological or other person-related theory to explain founders’, champions’, teams’ or owner-managers’ enterprising efforts. This is a design mismatch I became painfully aware that my dissertation study was subject to, and which I have had disciples avoid by adding information about the respondent’s other business activities (cf. Davidsson & Wiklund, 2000; Wiklund, 1998). To sum up, this section has shown that although being the most common level of analysis (or sampling unit) in entrepreneurship research, the firm level is far from unproblematic. This is an additional reason for researchers who want to make a unique contribution to consider alternative levels of analysis. If the firm level is chosen, the different theoretical and empirical definitions of “firm”, and especially the match between them, ought to be taken into careful consideration. SAMPLING INDUSTRIES (OR POPULATIONS) I cannot recall having used industry as the primary sampling unit in any of my studies. This section will therefore be much shorter than those immediately preceding it. The reason I believe I have anything to say at all on the issue is that industry has been an important secondary concern, as basis for stratification and explanatory variable, in several studies I have been involved in. In Business Dynamics in Sweden it was also one of the variables by which the data set could be—and was—aggregated. Just like with the firm level, many of the issues related to sampling industries fit under the headings relevance; size; size distribution, and heterogeneity along other dimensions. I will therefore re-use these organizing categories here. Size, Size Distribution, and Heterogeneity along Other Dimensions Industry statistics typically use a hierarchical classification system. These standards are similar across countries, but in order to make life interesting for researchers, they are not identical. In order to make life really interesting for researchers involved in longitudinal research—and perhaps to reflect real changes in the economy—the

RESEARCHING ENTREPRENEURSHIP

90

systems are also revised periodically9. At any rate, at the crudest level, these systems subdivide the economy into about ten categories, such as primary industries, manufacturing, wholesale & retail, education & health care services, and other broad categorizations like these. These industries are then subdivided to a five- or six-digit level. For example, according to the North American NAICS system, a firm that produces nuts and bolts is included in the industry aggregate on all of the following levels: 3 33 332 3327

33272 332722

Manufacturing Metal manufacturing Fabricated metal product manufacturing Machine shops; turned product; and screw, nut, and bolt manufacturing Turned product; and screw, nut, and bolt manufacturing Bolt, nut, screw, rivet, and washer manufacturing

This gives the researcher a great deal of freedom of choice as regards what level of aggregation should be used. Often the final decision does not have to be taken at the design stage. If data are collected at finer levels of disaggregation they can always be aggregated later, whereas the converse is not true. Similarly to the firm level, the design question to ask concerning minimum absolute size of the industry units in the sample is: Will the sampled units be big enough for the entrepreneurial behavior we are investigating to have any likelihood of occurring within the studied spatial unit and time frame used? For example, the total number of start-ups, patents, or other indicators of entrepreneurship per annum in Nova Scotia may not be very high in 332722. Again, if the studied units are too small, variation in the dependent variable will appear stochastic and hard to explain. If, on the other hand, too high a level of aggregation is used (i.e., too big units) we will run into other problems. First, we may simply end up with too few industries to compare in order for the analysis to yield interesting results. Second, very different types of firm will be assigned to the same industry, which is a problem of (internal) heterogeneity and therefore a threat to relevance relative to the theoretical industry concept being employed. Thirdly, expanding and shrinking sub-industries or niches may cancel out within very broadly defined industries. As regards size distribution we again run into a similar problem as on the firm level. Assume that we decide to work with industries on the two-digit level. Some of the resulting industry categories may have thousands of firms in them, whereas others only contain a handful. Should they weigh equally in the analysis? Perhaps yes; perhaps not. As noted above, the fact that a particular category is rare or numerous in a particular country at a particular time does not necessarily mean it has greater or lesser theoretical significance. However, the researcher should at least make an informed decision. If for some reason, more equally sized industries are deemed desirable, it may be worth forming categories that in some cases are on the two-digit level and in other cases on the three- or four-digit level. 9

For information on conversion between industry standards, see www.macalester.edu/research/economics/PAGE/HAVEMAN/Trade.Resources/TradeConcordances.html

Sampling Issues

91

As regards heterogeneity across industries we need to ask, again, whether we are about to compare apples with oranges. Some of the questions to ask oneself are: Does our theory apply to all industries? Do all the industries that result from application of the standard industry codes yield categories that correspond to the same theoretical “industry” (or “population”) construct? Do the operationalizations of variables that we plan to use work for all industries, and are values on those variables meaningfully comparable across those industries? As I see it, a big threat here is that our conceptualizations—sometimes explicitly, but even more so implicitly—are modeled on the manufacturing industry. Time and again in my research I have come across instances where operationalizations fit better and results were stronger for the manufacturing sub-sample, presumably because the thinking behind the research and the tools used in it were “manufacturing-biased”. So this is a pitfall worth watching out for. Relevance The questions asked above link over to the issue of relevance. Also regarding relevance, it is not a given that simple aggregation according to the industry hierarchy used in the standard system will leave us with industry categories that are maximally relevant. Depending on our theory and research questions, we may want to compare young vs. mature industries (Utterback & Abernathy, 1975); contrast growing with contracting industries; compare entrepreneurial activity for industries with high vs. low entry barriers or which are in a business-to-business vs. a businessto-consumer situation; get special insights into industries that are research- or knowledge-intensive, or investigate the effects of firm size structure and capital intensity on innovative or entrepreneurial activity on the industry level (Acs & Audretsch, 1990; Arrow, 1983). In order to arrive at industry groupings suitable for these types of contrasts, one might want to combine sub-industries that, according to the standard classification, belong to different main groups. For example, for Business Dynamics in Sweden, we created the following industry categories: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

High-tech manufacturing Wood-based; paper, and pulp manufacturing Engineering industries Mining and steel manufacturing Other manufacturing Technology-related services Other knowledge-intensive services Financial services Construction Accommodation & food services

RESEARCHING ENTREPRENEURSHIP

92 11. 12. 13. 14. 15. 16.

Wholesale & retail Transportation & communication Other services Education & health care Agriculture, forestry and fishing Government sector

Of those, only industries 9, 11, 12 and 15 are aggregated strictly by the logic of the original standard classification. All other categories were more or less customized for our purposes. Depending on the specific context, we sometimes worked with even more aggregated industry sectors. For example, industries 1-5 were combined to “manufacturing”; 6-8 to “knowledge-intensive services”, etc. My final issues concerning relevance have to do with the concept of “population” (of species of organizations; cf. Aldrich, 1999). We noted in Chapter 3 already that empirically, membership of a population is often equated with having the same industry code, and “populations” then become equal to what industrial economists call “industries”. There are two problems with this approach. If it is possible at all to achieve a good match between the theoretical and empirical concepts of “population”, it is probably only possible at very fine levels of detail (five- or six-digit groups). We may then quickly run into problems related to small size (cf. above). Second, from an entrepreneurship perspective, the most interesting questions are related to how truly new species and populations come into being. The problem with this is time lag in the industry classification system. When McDonald’s started to revolutionize the fast food industry there was probably no unique code for “(franchised) fast food restaurant belonging to a chain”, and during the Internet boom there was no unique code to assign to “dot-coms”. Therefore, other approaches may be more relevant for the study of emerging populations. For an example of a study that covers the emergence, growth, and disappearance of what can truly be called a distinct population of organizations, see Gratzer’s fascinating study of the automated restaurant industry (Gratzer, 1996). Unfortunately, only fractions of this exemplary study are available in English (Gratzer, 1999). Throughout this sub-section on sampling of industries, the assumption has been that industry codes in secondary data sets somehow be used. When collecting primary data from firms it is, of course, possible to collect any information relevant to one’s theoretical concept of industry, which can then be used for poststratification of firms into industries.

SAMPLING SPATIAL UNITS Two of the major projects I have been involved in—Business Dynamics in Sweden and the related study Culture and Entrepreneurship—used regions as the cases in the data matrix. As both of these studies were comprehensive efforts, each involving the compilation of two successive data sets, I can claim much more experience here than for industries. Figures 5:2 and 5:3 give an idea of what the main research questions were in those studies. I will use my experiences from these projects in discussing “sampling” of spatial units, while blending in observations

Sampling Issues

93

from other research as well. As this chapter is getting long already, I will try to be concise. Another measure to keep the reader awake is that here I will shift to discussing relevance first, and then turn to absolute size, size distribution and heterogeneity along other dimensions.

Figure 5:2

Region-related research questions in Business Dynamics in Sweden

Figure 5:3

Research questions in Culture and Entrepreneurship; main interest emphasized

Relevance Most countries are spatially subdivided in a number of different ways. The first to come to mind may be municipalities and counties or close equivalents to those, but there are several other subdivisions as well. The rationales for these subdivisions tend to be political (incl. legal system), military or religious administrative purposes. Apart from issues related to size and variability (below), there are important practical issues to take into account when choosing among these, such as for what units statistics are compiled, and what units have been used in other studies that we want to compare our results with? Unless such practical concerns strongly suggest administrative units of this kind, other types of subdivisions may be better suited for entrepreneurship research. In short, we want to have spatial units that make sense from an economic point of view rather than being based on some administrative criterion. For example, following Reynolds’s work in the United States, we used Labor Market Areas (LMAs) as the regional units of choice for the above mentioned

RESEARCHING ENTREPRENEURSHIP

94

studies (Reynolds & Maki, 1990; Reynolds, Miller & Maki, 1993). This type of spatial subdivision has been made also in a range of other countries. We choose LMAs because of the following, distinctive advantages: Being aggregates of municipalities, they form units for which statistics from a lower level of analysis exist and can be aggregated. Being based on travel-to-work statistics—municipalities with high levels of commuting between them are combined—it can be argued that they are natural economic entities. They are not artificially bound by county or (within-nation) state limits. When the commuting so suggests, LMAs can be defined across such borders. At least in Sweden and the US, statistics providers have clustered LMAs into region types on economic-structural criteria (Davidsson, 1995a; Reynolds et al, 1993), which gives additional input to design and interpretation of analyses. It is not a given, though, that the LMA subdivision should be adopted as is. In Business Dynamics in Sweden, for example, we did not accept Greater Stockholm as one LMA because we knew that the northern parts of the capital were “hot” at the time while the southern parts fought with a dying industrial heritage and other problems. We therefore subdivided Greater Stockholm into three spatial units: North, Central, and South. In the second part of the study we also sub-divided the other two major cities, Gothenburg and Malmö, into separate center and hinterland units because we had reason to believe these parts were distinct as regards entrepreneurial activity and the structural characteristics that might affect it. Future researchers are advised to make adaptations of this kind based on whatever knowledge they have over and above the mechanical grouping of municipalities that results from the clustering based on travel-to-work data. For the Culture and Entrepreneurship study I had the little problem that no data were available on any level of analysis concerning the types of variables—prevailing values and beliefs—that I needed. Hence, I had to collect primary data from individuals and use the mean responses per region as the regional point estimates. It is probably not hard for the reader to imagine that collecting primary data from large enough representative samples of individuals in 111 LMAs is going to be prohibitively costly, to say the least. As mentioned above, Statistics Sweden had clustered the 111 LMAs into region types based on structural criteria. The workable solution I found was to sample individuals representing these region types rather than LMAs. There were only ten region types, but this was still too much for my budget. In order to capture variability and assure relevance I drew samples from the following region types (see Davidsson, 1993, 1995a) for structural descriptions): 1. 2. 3. 4.

Greater Stockholm Regional Centers Average communities Rural LMAs

Sampling Issues 5. 6.

95

One-company (industrial) towns The Gnosjö/Gislaved industrial district

Based on the original clustering results, the non-selected types were suspected not to stand out from the selected ones on any dimensions. Category 6 is a special adaptation. Rather than sampling from the entire region type characterized by strong emphasis on small-scale manufacturing, I included what is Sweden’s most famous industrial district. The essence of its fame is its alleged “entrepreneurial spirit” (Wigren, 2003). Including this region is obviously relevant in a study on culture and entrepreneurship. Again, this is a “manual” adaptation based on knowledge outside of the mechanical subdivision of spatial units, and I think researchers should be encouraged to make such adaptations. In a regional study of entrepreneurship in any country one should, of course, try to include the regions that are the most interesting from an entrepreneurship perspective, and delineate such regions as precisely as possible. This might include also the regions thought to have the biggest problems with low levels of entrepreneurship, which was the case with the category “Onecompany towns” selected for Culture and Entrepreneurship.

Size, Size Distribution, and Heterogeneity along Other Dimensions The problems relating to size, size distribution and other heterogeneity for regions are similar to those discussed for firms and industries above. The original grouping of 111 LMAs in Sweden yielded units that ranged from a population of a couple of thousand to over a million. For the smallest units, it was easy to predict that there would not be many high-tech start-ups, for example, and that the variability over years for such start-ups would look rather haphazard. In order to increase minimum size and decrease size variability in Business Dynamics in Sweden, regions that were small (<10 000) in population and geographically adjacent were combined to larger units if they were also structurally similar. To assess structural similarity, we used the clustering into region types that was already available. In the absence of such a grouping the researcher can easily create one. At the other end of the spectrum, we split the largest agglomerations into two or three units. As discussed above, we made the latter modification also for relevance reasons. As a result of these changes we worked with 80 (83 in the second study) somewhat more equally sized regions rather than the original 111 (112). As regards other heterogeneity, we had among other things twelve different indicators of business dynamics. Relating these in a causal chain analysis to a multiplicity of regional characteristics, on the one hand, and several indicators of development of regional economic well being on the other, is a pretty complex matter (cf. Fig. 5:2). Not to say impossible, for a mortal like me. One way we dealt with this was to cluster analyze regions by “type of dynamics pattern” and relate the clusters to possible antecedents and outcomes. That is, we aggregated the cases in the data matrix from 80 regions to six clusters. The results indicated that formation and turnover rates of business establishments (whether independent or being branches of larger structures) co-varied systematically with structural characteristics that typically follow a core-periphery continuum. Formation and turnover of both

RESEARCHING ENTREPRENEURSHIP

96

independent firms and branch plants/outlets appeared important for regional development, and different region clusters represented different recipes as regards building their success on independent or corporate entrepreneurship. Indicators of growth and contraction dynamics did not have clear links to regional characteristics nor to well being (Davidsson et al, 1995). With Culture and Entrepreneurship I faced two types of heterogeneity problems. First, for budgetary reasons I had used region types rather than LMAs. Being aggregates of structurally similar but not necessarily geographically adjacent LMAs, the units may have become too large and internally heterogeneous to find any distinct cultural differences between the types. Second, with only six cases (region types) and considerable variation in three variable groups (structure, culture, and entrepreneurship), there was the (calculated) risk that the results would indicate that all the arrows in Figure 5:3 had something to them, without the possibility of sorting out relative strength or dominant causal direction. The latter is exactly what happened (Davidsson, 1993, 1995a). Overall, there seemed to be a positive relationship between the prevalence of “entrepreneurial values” and regional start-up rates. However, the same variation appeared at least as explainable by structural variation. In short, where the structural (pull) conditions for (independent) entrepreneurship were favorable, the culture also tended to favor entrepreneurship. Because the design of the original study had led to possible dilution of regionally distinct cultures as well as confounding of structural and cultural explanations, I employed a different sampling strategy for the second study. The sampling criteria for this second study were (Davidsson, 1995c): 1. 2. 3. 4.

The regional units should be small enough so that cultural variation did not cancel out within them. It should be possible to obtain data from them regarding relevant cultural and structural variables as well as on regional entrepreneurship indicators. There had to be variation in entrepreneurship among them (here operationalized as regional start-up rates for independent businesses). They should be as homogeneous as possible on variables other than the cultural variation in values and beliefs, which are the key explanatory variables.

In order to achieve (1) I selected LMAs rather than region types. Structural and entrepreneurship measures for LMAs were available from Business Dynamics in Sweden, whereas cultural variation had to be obtained through primary data collection from individuals (2). In order to fulfill (3) and (4) I cluster analyzed all LMAs in the Business Dynamics in Sweden study on those (seven) structural characteristics that according to a regression model had substantial influence on start-up rates. From the resulting clusters, I chose three matched pairs, where both LMAs in each pair belonged to the same structural cluster, whereas one had a much higher and the other a much lower start-up rate than predicted by the regression analysis. The logic was that unmeasured cultural variation might be the explanation for deviations from the values predicted by the structure model. After measuring the

Sampling Issues

97

cultural variation, the conclusions were the following. The results were more for than against a separate, causal effect of cultural variation. However, also with these more distinct regional units the cultural variation appeared small, relative to structural variation in the country. For this reason, structural variation seemed to account for relatively more of the variation in regional entrepreneurship [in Sweden during the studied period] (Davidsson, 1995c; Davidsson & Wiklund, 1997). Although the limited number of cases prohibited definitive conclusions, the structurally matched sampling procedure helped taking the analysis much further than otherwise possible. In the above example, information from samples of individuals was used to represent characteristics of spatial units. When this approach is chosen the logic of statistical inference theory applies to the full. For us to conclude that the average value in the sample is representative for the spatial unit, we have to work with an unbiased sample of individuals from the relevant (sub)population within that spatial unit. Practical and budgetary concerns may make probabilistic sampling impossible, but when the individuals in the samples are drawn from a particular company, association or educational group, and/or if non-response is high and uneven across spatial units, one should be aware that there are great risks that erroneous conclusions be drawn (cf. Hofstede, 1980; Lynn, 1991; Scheinberg & MacMillan, 1988). For example, if samples of MBA students are used across countries one should be aware that this is a group that represents an extremely small social elite in some countries, whereas in other countries having or taking an MBA is not reason for others to engage in eyebrow-raising exercises. Hence, resorting to this type of convenience sampling is likely to cause serious bias or distortions. The issues of (relative) size and heterogeneity are also important concerns for studies on the country level. Countries are very different animals, and it can be validly asked whether, for example, causes and effects of internationalization (or of national competition, cf. Porter, 1990) can be meaningfully investigated and validly generalized across spatial units that are extremely different in terms of size and internal heterogeneity (think of, e.g., Australia, Croatia, Indonesia, Japan, Luxemburg, Singapore, Switzerland, and the USA). This, of course, is a very important issue for research on “international entrepreneurship” as well as other international-comparative work in our field. In order to reduce size variability and other heterogeneity in international comparison other Swedish researchers have contrasted Sweden with Ohio rather than with the US as a whole (Braunerhjelm & Carlsson, 1999; Braunerhjelm, Carlsson, Cetindamar & Johansson, 2000; Fridh, 2002). Sweden and Ohio are relatively similar in size and share the same traditional industry structure. Therefore, comparing this “matched pair” should be a better ground than Sweden vs. US for comparing institutional factors and their effect on entrepreneurship and industrial renewal. Strategies like this appear worthy of following by researchers interested in other countries and regions as well. Summing up, I have argued in this section that ideally, regional units that make economic sense should be used rather than administrative subdivisions. If Labor Market Areas have been defined for the country in question this can be a good choice. However, considerations of size, size distribution and other heterogeneity, as

98

RESEARCHING ENTREPRENEURSHIP

well as the adding of prior knowledge about spatial variation in entrepreneurshiprelated issues, may call for adaptations of the regional subdivision offered by statistics providers. For studies comparing countries, the very large variation in size and other characteristics have to be taken into account. The more relevant comparison can sometimes be between an entire country and a region within another country. Finally, when using a particular sub-population like MBA students or IBM employees, one should carefully consider whether this subgroup is equally representative of each country.

SAMPLING OTHER UNITS OF ANALYSIS Chandler & Lyon (2001) as well as Davidsson & Wiklund (2001) show that published entrepreneurship research is dominated by studies on the individual and firm levels of analysis. Some use the aggregate levels industry or region. Very few use other levels of analysis. This does not mean they are less relevant. One very relevant but rarely used level of analysis has been treated rather elaborately above: the emerging new venture. But there are others that could be considered: the patent, the innovation, the team, the dyad, the community-of-practice, the network and the cluster, to name a few. These alternatives share the characteristic that it is difficult to obtain a sampling frame and/or secondary data on them. So why make life difficult? Well, if you ask me you could just as well ask “Why make life interesting?” Challenges are fun! Besides, why shouldn’t researchers believe their favorite strategic recipes for business success—be it Portnerian “diversification strategy” or an RBV emphasis on “sustainable competitive advantage through unique knowledge and capabilities”— have analogous applicability in research? Be different!

RESPONSE RATES When sampling is treated from the perspective of statistical inference theory it is intimately related to the issue of non-response. As non-response and remedies for it are dealt with extensively in standard method textbooks (as well as in specialized text on particular matters) I see no reason to say much about it here, especially as this lengthy chapter probably has put the reader to sleep a couple of times already. Suffice it to say that as regards response rate, it is a blessing to conduct research in Sweden, where we, for some peculiar cultural reason, repeatedly have been able to reach over 80 percent response rate in telephone surveys. Although I reached 70 percent in a mail survey to the general population in Culture and Entrepreneurship, it is today rare that we achieve much more than 50 percent (after three reminders) in Swedish mail surveys, but this is already a figure researchers in other countries can only dream about. My exposition is likely to be colored by this, and at times I may have forgotten that researchers in other countries have to live with a different reality. Regardless of the maximum attainable level in a specific country, however, there is little doubt that proper attention to, and application of the “craft” of survey research—cover letter, timing, layout, reminders, call-back schemes, etc.—will help

Sampling Issues

99

the situation (Dillman, 1978; Fink, 1995). Likewise, there are well-developed ways to deal with partially missing data (Hair et al, 1998; Little & Rubin, 1987). These, however, are general problems that we need not devote space to here.

SUMMARY AND CONCLUSIONS I opened this chapter by arguing that theoretical relevance is the most important criterion for sampling. The composition of units in the sample should match the theory used. Statistical representativeness is desirable when possible to achieve at all, but secondary to theoretical relevance. We have further observed that entrepreneurship research can be conducted on many different levels of analysis, and that each level has its problems that have to be dealt with. It is not always possible to overcome those problems—there is no such thing as “perfect” research—but it is certainly worth trying to solve as many as possible and to be aware of the remaining shortcomings of one’s sample. Importantly, the most conventional levels of analysis in entrepreneurship research—the individual and the firm—are not markedly less problematic than are other alternatives. This insight should provide incentive for researchers to consider leaving the most trodden paths and apply other levels of analyses than those that first come to mind.

CHAPTER 6

OPERATIONALIZATION ISSUES

A 90-DEGREE TURN Operationalization concerns the “translation” of theoretical concepts into measured empirical variables. It thus has to do with the columns in the data matrix, whereas the sampling issues dealt with in the previous chapter concerned the rows in the matrix. Oddly enough, there is nevertheless a certain amount of overlap between these two issues, and some important operationalization issues have therefore been dealt with already. For example, when I described above how people in the process of starting a new, independent venture were sampled in the PSED, this was at the same time a description of how the concept “nascent entrepreneur” was operationalized. Likewise, the procedure for sampling new internal ventures in existing firms can, at the same time, be regarded a firm level operationalization of entrepreneurship. However, these are just examples out of many possible ways of operationalizing entrepreneurship on the individual and firm levels. In the present chapter we will go much deeper into these issues. As usual, I will not try to be complete. I will devote much more space to the operationalization of entrepreneurship on different levels than to the operationalization of possible antecedents and outcomes. I will also devote much more space to those levels I find most interesting or important, and to those from which I have more experience. As a result, the sections on the individual and industry levels will be quite short compared to the regional level. The most comprehensive section will deal with operationalization issues when the emerging new venture is used as the level of analysis. However, before turning to operationalization problems and opportunities associated with specific levels of analysis I think we should discuss a few general operationalization issues. A NOTE ON LEVELS OF MEASUREMENT Standard method textbooks would have you believe there are four mutually exclusive and collectively exhaustive “levels of measurement”; an idea that can be traced back to Suppes & Zinnes (1963) and Stevens (1946). The types of scales corresponding to these levels of measurement are nominal, ordinal, interval and ratio scales, respectively. They can be described as follows: Nominal scales. Here, the numerical categories of the scale are merely labels. For example, we may use 1 = manufacturing firm, 2 = retailer, and 3 = service firm.

102

RESEARCHING ENTREPRENEURSHIP

We could just as well have reversed the labeling (3-2-1), used a different order (2-13), numbered the categories as 1-5-14 instead, or even used A-B-C rather than numerals. None of the alternatives would be obviously more “right” than any other. There is not much in the way of mathematical operations we can do on such a scale. We can present the respective frequencies, and we can cross-tabulate one nominal scale with another one. Few would be serious about computing a mean for such a scale, and neither can it be used (without transformation) as independent or dependent variable in, for example, a regression analysis. Ordinal scales. As the name suggests, for this type of scale the number is more than a label; it also gives information about the ordering of categories. For example, we may have 1 = most preferred alternative, 2 = second most preferred, 3 = third hand choice, etc. Here, we cannot re-shuffle or exchange the numerical symbols without changing the information—any transformation has to retain the same ordering. Unlike nominal scales, the median is a meaningful measure for an ordinal scale. All Boolean expressions (>, <, =, ~=, etc.) also apply. However, because we do not know whether the distance between 1 and 2 equals the distance between 2 and 3, we still cannot apply arithmetical operations. So computing the mean is dubious, as is using this type of variable in a regression, where the coefficient is an estimate of an effect that is assumed to be the same for one unit change, regardless of which scale step we are referring to. So we are left with very rudimentary statistical analysis methods. If we want to use this type of variable when we use a technique that requires interval scale measurement (cf. below), we are advised to transform it into dummy (dichotomous) variables for each scale step (was this scale step marked? Yes =1; No = 0). As dummy variables only have one scale step, the problem of unequal step lengths is overcome. Interval scales. For this scale, the numerical symbols are likewise ordered, and in addition we know that each scale step (interval) of the scale is equal. With this type of scale, more or less all types of mathematical and statistical operations are allowed, as long as ratios are not involved. The standard example is temperature. If it is 20 degrees Celsius today and it was 10 degrees yesterday, I can validly claim that the average was 15. However, I cannot really say it was twice as warm today compared to yesterday. Why? Because if you switch to Fahrenheit the claim is obviously false (68 vs. 50 degrees in this example). Unlike ratio scales (cf. below) interval scales have an arbitrarily chosen zero point. Now, despite talk of “economic climate”, temperature does not often turn up as a variable in entrepreneurship research. It is therefore worth mentioning that temperature is not the only example. A more likely one to turn up in our domain is birth year. What makes birth year an interval scale is the fact that while followers of various religious faiths (have come to) agree on what a “year” is they do not agree on the starting point. Ratio scales. This is the Rolls Royce of measurement—an interval scale plus an indisputable zero point. Height, weight and a number of other physical measures qualify here—as do sales and number of employees in our own domain. When you have a ratio scale measure you have few problems. It is worth noting that sometimes an interval scale variable can be converted to a ratio scale, as when a person’s age is computed by deducting the birth year from the current year.

Ope rationalization Issues

103

It is no doubt the case that the “higher” the level of measurement is for the operationalizations of a variable, the better. An advanced type of measure (sales in some currency or age in years) can be transformed to a cruder one (high – medium – low) if this for some reason serves the analyst’s purposes. The converse is not true; if you only have an ordinal measure you’re stuck with it. Nor is there any quarrel about the definitions of nominal, interval, and ratio scales. The problems concern instead the status of various kinds of rating scales that are very frequently used in survey research. Consider for example the example in Table 6:1 (from Davidsson & Klofsten, 2003).

These three statements are intended to measure the degree of completion of the business idea in an emerging venture. When three respondents tick 2, 3 and 4 respectively for one of these statements, we cannot know for sure that the difference between 2 and 3 equals the difference between 3 and 4. Therefore, this type of scale has been lumped together with pure ranking scales (which here would be “our idea x has reached a more developed stage than our idea y, which in turn is less crude than idea z) in the “ordinal scale” category. Hence, computing means would not be permissible for this type of scale according to the statistical puritan, and neither would be using them in, e.g., a regression analysis. And yet, published research is full of examples of application of arithmetic operations on this type of scale—for attitudes, intentions, entrepreneurial orientation, environmental munificence, and a range of other variables—as is often the ensuing chapters of the very method books that first classify this type of operationalization as “ordinal”. What’s going on here? There is a solution to the riddle. It has been suggested that Stevens (1946) went astray when he lumped ranking and rating scales together in the same “ordinal” category (Borgatta & Bohrnstedt, 1980; Michell, 1986). According to this latter view, a rating scale is a very crude representation of an underlying interval scale

104

RESEARCHING ENTREPRENEURSHIP

reality. There is no principal difference between a 5- or 7-point agree-disagree statement and, for example, a measuring tape, as long as both aim to measure an underlying social reality with a continuous distribution. The only difference is that the rating scale is cruder, and therefore the measurement error larger. Consider the example of the discus throwing competition in a track-and-field event. With the electronic measurement that is used today, the measurement error is very small. However, the results are rounded to whole centimeters, so there is some discrepancy between real length and reported length. In addition, it does happen on rare occasions that the length is measured from the wrong divot, adding to the measurement error. The true lengths of the throws, however, are exactly what they are regardless of our mistakes and our rounding off conventions. The same is true with measuring tape; using such instead of electronic measurement increases measurement error but does not alter the true length of the throws. Now, imagine that neither electronic measurement nor measuring tape was ever invented. Instead, we would have the athletes judge the length of their throws on a 5-point scale (1 = very short; 2 = short; 3 = average; 4 = long; 5 = very long). This is a much cruder type of measure, prone to considerable measurement error. Again, the true lengths of the throws, however, are not affected by this, and unless the athletes’ judgments are completely unreliable, there will be a substantial positive correlation between the true lengths and the reported assessments. It is, for example, highly likely (in the men’s event at world class level) that every throw longer than 70 meters would get a “5” and every throw shorter than 50 meters a “1”. Because of the crudeness of the measuring device, it would not discriminate between a throw of 70.48 and one of 72.16. Admittedly, it would sometimes occur that a throw of 67.50 was assigned a “3” while another of 65.14 got a “4”. Thus, the measurement error would be substantial, but this is the only difference between this measurement method and the electronic measurement. Both measurement methods represent an underlying continuous distribution of lengths. The example may seem funny, but this is in fact how crudely we measure attitudes and the like. As there is no difference in principle among the less-thanperfect measures in the example, this reasoning is defense for performing arithmetic operations on rating scale measures. The comparison with electronic measurement of length is also reason to reflect, however, on just how crude a type of operationalization self-reports of attitudes and the like are. In order to enhance the quality of such operationalizations, the standard recipe is to use several items that aim at measuring the same concept—as in the “degree of completion of the business idea”-example above—and sum the scores to an index. This has two advantages (which, if we are picky, reduce to one and the same). First, we get a more continuous measure and therefore presumably one that better reflects the underlying, continuous distribution. In the current example, we go from five possible scores to thirteen. Secondly, when several items are summed, random measurement errors tend to cancel out, whereas systematic measurement errors are not worsened. In order to see this, consider an example where the true score for the development stage of the idea is 3. Add to this a random measurement error component, which is +1 for

Operationalization Issues

105

the first item, ±0 for the second, and –1 for the third.10 If we sum the three items we get (4+3+2)/3 = 3, which equals the true score, whereas we would have had an expected measurement error of ±2/3 had only one item been used. If there is also a systematic measurement error (e.g., a respondent who consistently exaggerates by one scale step) this component is neither eliminated nor worsened by the summation.

VALIDITY AND RELIABILITY ISSUES With the above, we have slipped into the realm of validity and reliability issues. As there is some variance in the use of these concepts, let me explain with an example how I use them here. Validity is (relative) freedom of systematic measurement error. A measurement instrument is invalid to the extent that it yields measures that systematically deviate from the “true score”. Reliability is (relative) freedom of random measurement error. Some golf players are very consistent and, say, would never finish a round above 85 or below 75. For such players, the true result of one random round of golf would be a valid and reasonably reliable estimate of their “golfing ability”. Others are more emotional and less consistent, varying wildly between 75 and 110. For the latter group the true single round result would also be a valid measure, but unreliable11. If we asked the two golfers about their average result we would not be unlikely to get answers with low validity, as both might have a tendency to quote a score towards the lower end of their respective distributions. Validity and reliability are serious matters—much more serious than most researchers seem to realize. Why should we care about validity and reliability? Because supervisors and journal reviewers/editors so demand? NO, that is NOT the primary reason why we want our measures to be valid and reliable. We want them to be valid and reliable because if an operationalization has a substantial systematic measurement error (= low validity) we will estimate false relationships. If our operationalizations are plagued by large random measurement errors (= low reliability) we will find weak or no relationships were strong relationships actually exist. It may be possible to get false and weak results passed and/or published, but we will not have found out about how things work out there. And that is why we do research in the first place—not primarily to get passed or published, but to find out about important matters, right? I thought so. Validity is also a tricky matter, because proving validity—that our operationalization actually measures what it is intended to measure—is a neverneverland. If we had some indisputable criterion to compare our measure with, then we would use that instead of the shaky measure that we test against this unquestionably valid yardstick. So, unlike reliability, validity is something we can never prove. We can only offer incomplete support that our measures have validity. As indicated already, I think researchers (including the undersigned) could do better than we usually do as regards providing such support. When we read research, 10 The random errors would, of course, not cancel out perfectly every time, but the tendency would always be that of reducing the measurement error. 11 Sometimes validity is used with reference to the sum of systematic and random errors. Reliability is then a precondition for validity, and the single round result for the erratic golfer would not be regarded as valid.

106

RESEARCHING ENTREPRENEURSHIP

what we get is often implicit or explicitly emphasized claims about face validity— that the measures at first glance appear to be sound reflections of the theoretical concept in question. This is a weak criterion, but not a totally irrelevant one. There are examples of the opposite—measures with questionable face validity (the curious may want to check, for example, Lynn, 1969). We are frequently provided with the argument that “it has been used before”. This is actually not a bad argument— provided the measure has previously been shown to have theoretically meaningful characteristics (cf. below and the next sub-section). If we believe in the validity of a measure just because it has been used before we are victims of Francis Bacon’s idola theatri or idola scholae—the blind belief in established authorities and/or rules (Russell, 1946). Further, we are often provided implicit or explicit evidence of discriminant validity—that operationalizations of different theoretical concepts really measure different “things”—because they form separate, orthogonal (uncorrelated) factors in a factor analysis, or else have modest inter-correlations. This is another relevant piece in the validity puzzle, but far from satisfactory as a standalone criterion. And then we have the magical Cronbach’s Alpha number. With reference to Nunnally we proudly state that we have measured some tricky unobservable like an “attitude” not with a single-item measure, but with a multiple item index, which has reached the magical number 0.70 (or even 0.80) on the Cronbach’s Alpha test for internal consistency (Nunnally, 1967; Nunnally & Bernstein, 1994). Along with the 5%-rule for statistical significance, this is one of the great examples of idola scholae in our type of research. For many multiple-item measures Cronbach’s Alpha is certainly relevant, but does it prove validity? Hardly. To see why, consider the examples in Figure 6:1.

Figure 6:1

Possible relationships between theoretical concepts and operationalizations

In this figure, “X” denotes the theoretical concept, whereas “x” denotes what our operationalization actually captures. Panel a) shows the ideal situation when we have a perfectly valid measure. This is what we want to achieve. Panel b) shows what we are realistically more likely to end up with most of the time. This is an operationalization, which partly measures what it is intended to measure but also contains considerable noise—systematic or random components that do not reflect the theoretical construct we are after. When we are serious about validity we try to develop a measure that covers all of X without also capturing a lot outside of X. We

Operationalization Issues

107

can do this by starting from a rather large pool of items that according to face validity seems to cover all of the ground the theoretical concept covers. Through pre-testing we may via re-wording, elimination and addition of items approach situation a). Importantly, if we make internal consistency (Cronbach’s Alpha) the sovereign criterion, we are likely to end up with a situation like c). This is particularly likely if we want to develop a measure with a small number of items. The shortcut to a high Alpha-value with a small number of items is to have items that are very similar. They will then have high inter-correlations and therefore yield a high Cronbach’s Alpha value. However, they may be far from exhausting the theoretical concept we are after. The logic of the Alpha criterion—as well as the fit indices used when we run LISREL models—lure us to measure with high precision but a corner of the theoretical concept we’re after. No, a high Alpha does not prove validity. In fact, it is entirely possible to have a high Alpha value also in situation d), i.e., when we measure with high precision something entirely different from the theoretical concept we were after. Unless the theoretical concept is itself very narrow or easily assessed (such as age or level of education), the only way to approach X and at the same time have a high Alpha-value (which is desirable, given that x is within X) may be to have an operationalization with quite a large number of items. This is the strategy used in psychological tests of intelligence or aptitude. In survey research on complex phenomena, where we want to capture many different types of variables, we may not afford the space such “perfect” measures demand. It may then actually be preferable to have a measure that roughly captures more of X, albeit with a lower Alpha (lower reliability), than one, which very precisely covers but a fraction of X. However, the Alpha criterion should not be completely disregarded. It is worth keeping in mind that one way to achieve a low Alpha is to add up random numbers—and a sum of random numbers will not show strong relationships with anything. Researchers who are serious about validity should do more than checking factor structure and Cronbach Alpha. These are but two out of the thirteen criteria for evaluation of validity/reliability suggested by Robinson, Shaver & Wrightsman, (1991). For an application of these useful criteria, see Brown et al (2001). What we also ought to do more of, over and above checking technical criteria, is to provide evidence that our operationalization is theoretically sound. Let me give you a less than perfect—and therefore realistic—example of how this can be done. In my dissertation study (Davidsson, 1989a) I included a measure of Need for Achievement (nAch; then a much more popular concept in the entrepreneurship literature than it is today…) that I had developed myself. It was four-item measure with responses given on 5-point scales. In translation the items read: 1. 2. 3. 4.

I have always wanted to succeed and accomplish something in my life. I find it hard to understand people who always keep on striving towards new goals although they have already achieved all the success they could possibly have imagined (reversed). To face new challenges and to manage to cope with them is important to me. I am so satisfied with what I have achieved in my life that I think now I can confine myself to keeping what I already have (reversed).

108

RESEARCHING ENTREPRENEURSHIP

My research concerned the extent to which nAch influenced growth and other indicators of continued entrepreneurship in small firms. With the Grand Old Man of Achievement Motivation Theory, David McClelland (1961), claiming that nAch cannot be measured directly (hence the questionable face validity of Lynn’s, 1969, scale), and a Cronbach Alpha of just 0.55, I would seem to be on thin ice. However, neither McClelland nor Cronbach nor Nunnally (nor Davidsson) are indisputable authorities. What I did rather than just give up was the following. First, I scrutinized “The Achieving Society” (McClelland, 1961) for claims about high nAch people, for which I also happened to have data in my study. I found the following: They are moderate risk takers. They like to take some objective risks but are not attracted to games of chance. Profit is important to them as a measure of success and not for its own sake. Ownership control is not critical to them. They prefer experts over friends as workmates. I then checked the empirical patterns in my data and found that these predictions about people with high achievement motivation were, by and large, borne out. As a result, I could conclude that “it seems that the nAch-index measures a psychological difference between subjects and that labeling this difference ‘need for Achievement’ is reasonably well justified” (see Davidsson, 1989a, pp. 164-165 for details). With this support for the validity of the operationalization, the substantive relationships I hypothesized and found between nAch and entrepreneurial behavior could not as easily be disregarded as effects of method artifacts or some alternative theoretical variable. Moreover, the substantive analyses gave additional support for the validity of the measure, as I could show that nAch was positively related to growth aspirations for those managers who believed that growth would lead to increased profits, but not for those who did not hold such beliefs. This is a very logical result from the perspective of achievement motivation theory.12 Although admittedly being a far from perfect example about a far from perfect measure I hope it has effectively illustrated the underutilized principle of seeking validation for a measure through testing its construct validity regarding substantial relationship postulated by theory. For reasons discussed above in relation to Figure 6:1 it could well be argued that some evidence of this kind should always be presented before technical criteria like factor structure or Cronbach’s Alpha are even mentioned. Before closing this section it is worth mentioning what standard method textbooks tend to forget to tell you about summated index variables. This is that there are two fundamentally different philosophies we can apply when constructing indices, and that it is only for one of them that the factor structure and Cronbach’s Alpha criteria are meaningful. Figure 6:2 illustrates the two types of indices applied to the same problem: to obtain a summary measure of “Ability”. Panel a) in the 12 Although this example shows that trait explanations for entrepreneurial behavior do not completely lack backing I hope earlier chapters in this book have shown I am not a big fan of this approach to explaining entrepreneurship.

Operationalization Issues

109

figure displays the situation when the theoretical concept is regarded as an underlying non-measurable characteristic, which causes the variation in (i.e., is reflected in) the manifest variables that we use as operationalization. Because the different items have the same underlying cause, they should have high intercorrelations. We expect the person’s ability or lack thereof to show on all four tests, and the logic behind subjecting the four test measures to factor analysis and Cronbach’s Alpha testing applies to the full.

Figure 6:2 Two types of summated indices (latent variables)

However, we can also operationalize “Ability” as in panel b), arguing that both education and experience cause increased ability. In this case, the items are what build up (or form) the theoretical variable (Fornell, 1987; Fornell & Larcker, 1981; Wold, 1985). The different components that contribute to ability need not be highly correlated. In our example one could argue that long education necessarily reduces one’s chances of also having long experience. Therefore indices created via the formative logic should not be evaluated through factor analysis or Chronbach’s Alpha. Other examples of formative index variables are summated over-all evaluations of the kind that appear in Vroom’s Expectancy-Value Theory of motivation and Ajzen-Fishbein’s Reasoned-Action Theory (Ajzen & Fishbein, 1977; Vroom, 1964). According to the latter type of attitude model, we do not like, say, a particular attribute of a car because we like the car in general. Quite on the contrary, we like the car because we like several of its attributes. The fact that we dislike two out of ten salient attributes of the car (and the measures for these attributes hence have low correlations with the evaluations of the other attributes) does not make these two attributes irrelevant to our over-all assessment of the car, as index construction via mechanical application of a “maximize Alpha” strategy would suggest. When journal reviewers (or supervisors) ask you to report Alphas for your formative indices, there is only one thing you should do: gently educate them!

110

RESEARCHING ENTREPRENEURSHIP SOME BALANCING EXERCISES

When choosing or developing operationalizations, there is a number of partly contradictory interests that have to be balanced. One is that of using a previously used and validated measure versus developing a new one. Consider, for example, the items displayed in Table 6:2, which are four out of the nine items of the 1989 version of the frequently used Entrepreneurial Orientation (EO) scale (Wiklund, 1998). The EO measure can be accused of all sorts of shortcomings. For example, the items displayed seem to be a mix of preferences, past behaviors and beliefs. No wonder then that researchers have alternately labeled this measure as “entrepreneurship”, “strategic posture”, “strategic orientation”, “corporate entrepreneurship” and possibly other names as well before a consensus on calling it “entrepreneurial orientation” developed (Wiklund, 1998). Further, it has been argued that some items do not fit the three strategic dimensions the measure it intended to capture, namely innovation, risk-taking and pro-activeness. Specifically, Lumpkin & Dess (1996, 1997) argue that item (f) gauges “competitive aggressiveness” rather than pro-activeness. This item as well as the first innovation item (a) also show poor technical properties as they do no load neatly on the intended factors in a factor analysis, and lower rather than increase Cronbach’s Alpha when all items are summed to a global index (Brown et al, 2001). Finally, while the measure is intended to be a firm level operationalization of entrepreneurial tendencies it can be questioned whether it really is, as usually only one person’s answers are used to represent the firm’s views.

Operationalization Issues

111

For these reasons, it would be tempting for a researcher to develop a new and better measure for assessing firm level propensity for entrepreneurial behavior. However, using the EO scale has huge advantages. First, the measure has theoretical backing (Miller & Friesen, 1978, 1982). Second, if a couple of items are dropped, it has acceptable internal consistency both as a three-dimensional and as a onedimensional construct (Brown et al, 2001). Third, and more importantly, it has been used and shown to have theoretically meaningful relationship in a range of previous studies. In 1998, one could already review twelve studies employing this measure (Wiklund, 1998). Today, twelve is more likely the number of studies added annually that employ the EO measure. Therefore, we have now much more refined knowledge about the direct, moderating and mediating effects of EO in different settings, and rich possibilities for comparing new results with established ones. With a new measure all these advantages would be lost. My personal experience is that developing useful new measures is also harder than most people think. For example, for my two studies on (regional) culture and entrepreneurship, I tried a total of eleven different three- to five-item measures of “general values” intended to capture different dimensions of regional differences in mentality (Davidsson, 1993, 1995c). Despite varying degrees of theoretical anchoring and pre-testing efforts, it was only for two indices that I reached Cronbach Alpha values above 0.70, namely for those I borrowed from Lynn (1991). These were previously tested both in the English original and Swedish translation. Again, internal consistency is not all you should ask of a measure, but my relative failure to achieve very strong results in these studies was, in all likelihood, partly due to the questionable quality of the operationalizations. As regards EO, we have shown that it is possible to develop alternatives that have some advantages, but this is something that requires considerable effort (cf. the section about the firm level below, as well as Brown et al, 2001).

112

RESEARCHING ENTREPRENEURSHIP

A partly related balancing exercise concerns the choice between the “perfect” operationalization for a specific type of venture, and the most generally applicable operationalization. This is one of the method consequences of the heterogeneity of the entrepreneurship phenomenon discussed in Chapter 2. For example, the best measure of firm size may be the number of vehicles for a taxi company, the number of seats for a restaurant operation, and the quantity of electricity delivered for a power station. However, how are we to compare the firms’ growth across these different measures? Sales and number of employees are more generally applicable, but may have other disadvantages (Bolton, 1971; Davidsson & Wiklund, 2000). A similar type of concern may be the reason for the weak quantification (“a lot”) in item (b) of the EO scale above. The specific number of innovations you would need to undertake in order to stand out as more entrepreneurial than average is highly industry-specific. Concerning operationalization of entrepreneurship, consider also the list of indicators of entrepreneurial action in Table 6:3, which we used in the Entrepreneurship in Different Organizational Contexts study (note that I would not today regard all of these as instances of entrepreneurship, cf. Ch. 1). Assume we took a simple summation of the number of yes responses as the measure of firm level entrepreneurship. Would this be an adequate or “fair” measure? For example, undertaking (b) makes sense as an expansion strategy only if the markets for inputs or outputs are such that local presence is necessary. How does a retailer respond to (g)? What does developing a new product or service mean in that context? Would not larger firms be likely to tick more affirmative responses simply because they do more things due to their sheer size; not because they are

Operationalization Issues

113

somehow “more entrepreneurial”? Although the above list is what we were left with after having screened out the less generally applicable manifestations of entrepreneurship, it is no doubt still the case that the remaining scale is sensitive to size and industry. One could hope that including several alternative manifestations of entrepreneurship reduces the industry biases, but a size effect is inescapable. In developing a scale of this kind, the less-than-perfect alternatives we are left with seem to be the following: 1.

2.

3.

Develop one operationalization that is assumed to be good for all ventures/firms. Accept that interesting manifestations of entrepreneurship that clearly apply only to narrow subsets of firms cannot be included in the measure. Also accept as a fact that larger firms and firms in some industries, on average, exercise more entrepreneurship than do smaller firms and firms in certain other industries. Develop one operationalization for all ventures/firms. However, normalize the score within industry/size class (or other) groups, and use deviation from the own class mean as the measure of entrepreneurship. This would eliminate what can be regarded as a bias against certain categories when approach (1) is applied, but this comes at the cost of assuming that all subgroups of firm are equally “entrepreneurial”. That is, only within-group and not between-group differences will be detected. Develop separate and adapted operationalizations for different subgroups (by industry, size class, or otherwise). Standardize this measure, so that comparisons can be made across different operationalizations of entrepreneurship. This would allow including the presumably most relevant indicators for each category, but involves a considerable risk of comparing apples and oranges in the analysis.

Similar issues of best vs. most generally applicable operationalizations could be discussed with regard to explanatory or outcome variables. As no obviously “right” decision can be taken on this kind of issue a good solution is often to try different approaches within the same study—if space allows. OPERATIONALIZATION ISSUES ON THE INDIVIDUAL LEVEL Concerning operationalization on the individual level, one piece of advice is probably more valuable than any other. This is the gentle reminder that there is a discipline called psychology. Go there! Researchers in psychology have already wrestled with operationalization problems concerning all kinds of person variables. They have probably made all the mistakes you don’t want to make as well. One of the lessons learnt in psychology is the distinction between distal and proximal explanatory variables (Delmar, 1996). Many person variables, such as personality and personal background are distal in nature. That is, they are rather “far

114

RESEARCHING ENTREPRENEURSHIP

away” (in an imagined causal chain) from specific behaviors. For example, one is not deterministically geared towards starting one’s own firm just because one has self-employed parents or a high need for Achievement. In particular, there is little reason to believe that such behavior would be practiced at the exact time one is contacted by a researcher. However, these dispositions may slightly increase the probability of that career choice or some other independence- or change-oriented behavior when an array of situational factors also point in that direction. This insight leads to two possible research strategies. The first is to go for more proximal explanations instead, such as goals, interests and intentions. An example would be Ajzen’s Theory of Planned Behavior (Ajzen, 1991), which we discussed in Chapter 3. However, there we also noted that if we successfully explain entrepreneurial intentions or even behavior with the help of attitudes, subjective norms, and (domain-specific) perceptions of behavioral control, we are soon hit by the frustration that we do not know where these attitudes, norms and perceptions came from. This would take us back to more fundamental, enduring, and/or tangible characteristics of individuals. As such qualities are distal to specific behaviors we have to employ the alternative research strategy. While some of what has been discussed in this sub-section so far admittedly may be called general design issues (cf. Ch. 4) rather than operationalizations issues, we are now getting to an important example of the latter. If we are to successfully explain or predict entrepreneurship with (distal) variables on the individual level, then entrepreneurship has to be broadly operationalized and/or assessed over a longer period of time. Personality and personal background characteristics cannot be assumed to explain to any considerable extent why a person right now is or isn’t the ownermanager of a firm, or involved in a business start-up. Period. If, however, we employ a summary measure of alternative manifestations of entrepreneurial behavior over a long period of time and within a group of people who all are in a position to exercise such behavior (e.g., business owner-managers), then person variables can be expected to come more to the fore as an explanation for the variations in this measure. One aspect of this is that one should avoid mismatch of different levels of analysis. If person variables are used for explaining business behavior, then all business behavior of the person has to be considered, not only that relating to one selected firm or venture (Davidsson & Wiklund, 2000). The reason a person with all the “right stuff”—if such exists—does not show a lot of entrepreneurial behavior within a specific venture may be that s/he is busy amazing the world in another venture. Somewhat related to this—and again perhaps a general design issue rather than being purely a matter of operationalization—is that if we choose an individual level of analysis, perhaps other outcomes should be considered, than the survival, growth and financial performance of specific firms or ventures. As an example, Van Gelderen, Van der Sluis & Jansen (forthcoming) used goal achievement, skill development and satisfaction as the ultimate outcome variables. As argued in Chapter 2, entrepreneurship research is not merely a sub-field of strategy research, and therefore individual, non-financial outcome variables may be the more relevant choices when a psychological perspective governs the research process.

Operationalization Issues

115

OPERATIONALIZATION ISSUES ON THE LEVEL OF THE NEW, EMERGING VENTURE A myriad of potential influences easily come to mind when trying to build a model of key relationships for the study of entrepreneurship on this (or any) level. In order to make the task manageable these have to be organized under a small number of core concepts. I have chosen Resources, Venture Idea, External Environment, Behavior and Outcomes as core concepts to be operationalized. These concepts and possible relationships among them are depicted in Figure 6:3. It may be noted that there is no separate box for “individual/the entrepreneur” or “the team” in this model. This is because from the perspective of the venture, these are aspects of the general category resources.

Operationalizing Resources Shrader & Simon (1997) define firm resources as “all assets, capabilities, organizational processes, firm attributes, knowledge, etc. controlled by a firm that enable the firm to conceive and implement strategies that improve efficiency and effectiveness.” With our new venture perspective this can be rephrased as all tangible and intangible assets that are committed to or available for the discovery and exploitation of a new venture idea. Resources can be classified in different ways. Barney (1997) divided them into four categories: financial capital, physical capital, human capital and organizational capital. Although other ways to classify resources may have stronger theoretical implications—e.g., Miller & Shamsie’s (1996) “knowledge-based” vs. “property-based” resources—I will keep Barney’s categories. With Greene, Brush and Hart (1999) I add social capital to his list (cf. Davidsson & Honig, 2003; Greene & Brown, 1997; Nahapiet & Ghosal, 1998). In Table 6:4, I summarize definitions of resource types and sub-dimensions of such. I also suggest ways to assess resources in empirical research. Clearly, there are many sub-dimensions and aspects of resources that could be important. Therefore, choices have to be made concerning what aspects to operationalize in a more elaborate way, and which will have to be excluded or included as roughly measured control variables only. For this reason, I have suggested indirect, proxy measures as well as more direct assessment of each type of resource. Throughout it is important to bear in mind that for testing hypotheses based on the resource-based view, assessing the value, rareness and non-imitability of resources and resource combinations may be more important than assessing the existence and amount of particular resources. Financial capital is perhaps the most straightforward category both as regards definition and operationalization. This does not necessarily mean it is easy to get the data, even though there is not much problem concerning what should be measured. Physical capital needs little separate explanation above what is included in the table. The inclusion of industry code as a proxy for (the value of) physical capital is based on the assumption that new ventures in some industries are generally more capital intensive. Direct assessment of specific equipment is probably not a feasible option

116

RESEARCHING ENTREPRENEURSHIP

in studies that cut across many industries. The monetary value of assets is a much more general measure. In contrast to financial and physical capital, human capital is probably the most complex category. I have therefore subdivided it further into four categories: labor, declarative knowledge, procedural knowledge, and commitment. Of these, labor is probably easy to deal with—but not necessarily very interesting. It is a recurring theme that knowledge exists in at least two distinct forms. I choose to use the terminology from cognitive psychology, where declarative knowledge is knowledge of facts and procedural knowledge is knowledge of how to carry out or accomplish a certain task.

Operationalization Issues

117

118

RESEARCHING ENTREPRENEURSHIP

Operationalization Issues

119

As regards indirect assessment the distinction between declarative and procedural knowledge is hard to uphold, so the listed suggestions are intended as proxies for both. When the knowledge dimensions are deemed central to the study, they should, as far as possible, be assessed (also) in a more direct way. The question is what knowledge to assess. Bird (1993) lists an almost endless number of knowledge-related variables that are potentially important. Some strong candidates that have been operationalized in previous research are Alertness (Busenitz, 1996; Kaish & Gilad, 1991) or “the ability to recognize, envision and act on opportunity” (Chandler & Jansen, 1992) Social skills such as social perception, expressiveness, impression management, persuasion and manipulativeness (Markman & Baron, 1998) Functional skills such as marketing, innovation, management, and financial control skills (Chen, Greene & Crick, 1998). In a survey context, it is probably impossible to distinguish in a convincing way self-reporting of declarative and procedural knowledge from measures of selfefficacy (Chandler & Hanks, 1994). Multiple informants may be a way around this problem. For example, Chen’s et al (1998) measures were developed as measures of self-efficacy but could be adapted for use with other informants. With more intensive, assessment center-like methodology both declarative and procedural knowledge could be operationalized in an even more direct and therefore, arguably, more valid way (cf. Baron & Brush, 1999; Sarasvathy, 1999a).

120

RESEARCHING ENTREPRENEURSHIP

The final category of human capital I label commitment. The inclusion of this category goes back to the fact that without human agents with preferences, motives and goals, no activity at all will be undertaken. It may seem awkward to regard perceptions and motivations as “resources”, but I argue that when the venture is the unit of analysis it is reasonable to do so. There are suggestions in the literature that psychological commitment can, to some degree, be inferred from more tangible proxies such as age (Ronen, 1983), independent vs. internal venture (Shrader and Simon 1997), legal form (Dahlqvist et al, 2000) and work hours invested (Carter et al, 1996). Start-up reasons are somewhat closer proxies for commitment, but when commitment is considered an important issue, an even more direct approach is advisable. As strong candidates for direct assessment, I suggest entrepreneurial selfefficacy with its sub-dimensions (Chen et al, 1998), “willingness and capacity to generate intense effort” (Chandler & Jansen 1992), vision content (Baum, Locke & Kirkpatrick, 1998) and implementation intentions (Gollwitzer, 1999). Social capital is relational resources embedded in personal or organizational ties (cf. Burt, 1992; Coleman, 1990; Putnam, 1995). Thus, social capital is not an attribute of someone, but rather someone has social capital to the extent that they have social relations with others, through whom they can gain access to important resources (H. Aldrich, personal communication). A venture indirectly has social capital through its champion and other team members. Location can be used as a distant proxy for social capital based on the finding that start-up rates increase with agglomeration (Reynolds, Storey & Westhead, 1994). The argument would be that the potential for networking is greater in a dense environment. As this would be a very weak measure of social capital I suggest that number and frequency of networks contacts may be better “semi-direct” measures. In the Swedish PSED study, we found strong effects of a seemingly very crude indicator of social capital (Davidsson & Honig, 2003). Otherwise, theoretical expositions of the role of networking and social capital have hitherto often been more convincing than the empirical evidence. Future research should therefore aim at explicit operationalizations of, e.g., strong and weak ties (Granovetter, 1973) and structural holes (Burt 1992) and/or at least specify what resources the network provides and the extent to which they are valuable, rare and non-imitable. Existence and frequency alone do not seem to do justice to the theories. Using organizational capital alongside with social capital leads to risk of conceptual overlap. I have largely adopted Barney’s (1997) definition of organizational capital, with the explicit addition of market channels (existing customer relationships). This type of existing tie, then, I regard as organizational capital, whereas organizational ties not directly related to the output market are regarded as social capital. For organizational capital, I feel that the indirect measures may suffice for many purposes. Explicit hypotheses can be based directly on whether the venture is an independent start-up or an internal venture (Greene et al, 1999; Katz & Gartner, 1988; Shrader and Simon 1997). There is strong theoretical input as to the liability of smallness and newness (Aldrich & Zimmer, 1986; Stinchcombe, 1965; Teece et al, 1997), so among internal ventures organizational age and size of the host firms are important (and—for once—easy) variables to operationalize.

Operationalization Issues

121

For certain purposes, direct assessment of various aspects of organizational capital may be necessary. For example, Suchman (1995) specifies several different forms of legitimacy that can not be (differentially) inferred from age and size alone. In other cases, the purpose is to attain a deeper understanding of how imperfectly imitable organizational structures and routines for discovery and exploitation are created and maintained (e.g., the 3M story; cf. Mitchell, 1991). Such understanding is of central interest from the knowledge-based perspective of the firm (Galunic & Rodan, 1998; Kogut & Zander, 1992; Teece et al, 1997) but may be beyond the scope of broadly based studies (Rouse & Daellenback, 1999). In closing, there are many aspects of resources that could enter into the design of a venture level study, and many sources of previously used operationalizations. To the best of my knowledge, there does not exist one, coherent measuring instrument that has been developed for this specific purpose. A venture level instrument that comes close—although it cuts through other issues as well, and does not address all types of resources—is our Operationalization of Kloftsten’s “Business Platform Model” (Davidsson & Klofsten, 2003). This instrument, which is suitable for repeated assessment on the venture level, captures the level of development of eight “cornerstones” of the new, emerging venture: the business idea, the product, the market, the organization, core group expertise, core group drive/motivation, customer relations and other relations. Parts of this measure would also fit well with better established theories than Klofsten’s. Operationalizing the Venture Idea For reasons explained in Chapter 2, I hold that “Venture Idea” is a better concept than “Opportunity” for denoting the conjecture about unsatisfied needs and productive possibilities upon which the entrepreneur(s) act. As venture ideas are heterogeneous and emerging phenomena it is difficult to specify what components “have to be there” for something to constitute a venture idea. Working backwards, we may conclude that when a case satisfies the sampling criteria for “nascent entrepreneur” in the PSED research, or for “new initiative” in the 1994 Start-up Cohort study, then there also exists a venture idea (cf. Ch. 5) Beyond existence, we probably also want to operationalize various characteristics of venture ideas. At the present state of empirically based knowledge, there is still a need for applying the most straightforward way to assess venture ideas, namely to have expert judges rate open-ended and as detailed as possible descriptions of them. With such an approach, several alternative ways to classify and characterize ideas can be tried on the basis of the same data. Ideally, neither the description nor its assessment should rely on a single person. For example, Baum et al (1998) used separate questionnaire data from the CEO and an employee, as well as independent raters. However, their study also illustrates some of the limitations. First, with open-ended mail questionnaire data it is unclear whether a missing attribute of the idea is not yet developed or just missing in the data. This problem can be solved with more interactive data collection techniques. Secondly, their ratings of five attributes collapsed into one factor, suggesting some kind of “halo effect” or “common method variance” problem either in the descriptions or in the

122

RESEARCHING ENTREPRENEURSHIP

ratings. In addition, post-coding of open-ended answers is always a difficult and time-consuming task. It is therefore advisable to supplement open-ended description of the venture ideas with direct ratings of specific dimensions. One obvious candidate would be how far developed or well specified the idea is. This partly overlaps asking questions about (some) “gestation behaviors” as in the PSED (see “Operationalizing behavior” sub-section below). A more narrowly focused operationalization of a venture idea’s state of development was presented above in Table 6:1. Apart from arguably having high face validity, this scale did well in terms of internal consistency and discriminant validity when tested on a sample of technology-based firms (Davidsson & Klofsten, 2003). Other properties of this measure have so far not been tested. The statements could easily be rephrased to clearly focus on the venture rather than the firm level (these two usually coincide in early stages of independent start-ups, cf. Davidsson & Wiklund, 2000). Another basic distinction here is whether the venture idea is imitative (reproductive) or innovative. When working with the Swedish PSED, Samuelsson (2001; 2004) used Latent Class Analysis to operationalize this dichotomy based on the four items in Table 6:5.

This operationalization of “venture type” yielded strong and meaningful results and has subsequently been adopted—in simplified form—into the Global Entrepreneurship Monitor design (Reynolds et al, 2002). Among innovative ideas, Aldrich (1999) discusses the important distinction between competence-enhancing versus competence-destroying innovations (cf. Anderson & Tushman, 1990). The competence-enhancing versus competence-destroying distinction may require industry expertise that the researcher does not have, and the more evaluative the assessment the more biased the entrepreneur/champion can be expected to be. Other types of experts such as venture capitalists, marketing experts or prospective customers may therefore be better informants about, e.g., market and technology novelty or as regards Rogers’ (1995) innovation attributes that increase diffusion speed. I know of no satisfactory and validated operationalization of competenceenhancing vs. competence-destroying innovation. In the New Internal Ventures study (sampled from the 1994 Start-up Cohort) we relied instead on Schumpeter’s

Operationalization Issues

123

five broad categories of innovation (or “economic development” in his terms) for classifying venture ideas.These are a) new product, b) new production method, c) new market, d) new source of supply, and e) new organization of an industry. That is, after identifying the existence of a new internal venture as described in the previous chapter, we used the package of questions displayed in Table 6:6 in order to categorize those ventures. Although this may look like too big a chew for the respondent (and possibly for the interviewer as well) our experience has been that these questions work well in the technical sense, partly because not all respondents have to answer all questions. With a detailed classification based on this set of questions it should be possible to reach much further than Samuelsson’s (2001; 2004) promising start. Whether such hopes will be borne out is too early to tell at the time of this writing. Venture ideas may also be categorized according to the source or origin of the idea. A well-known categorization here is Drucker’s (1985) eight sources of innovation, but I have as yet not seen this applied in empirical research. Based on indepth analysis of 27 cases, Bhave (1994) arrived at externally versus internally

124

RESEARCHING ENTREPRENEURSHIP

stimulated opportunity recognition as the fundamental distinction. In the PSED this was operationalized with a rather straightforward questions in the mail questionnaire about what came first: the wish to start a business venture (presumably followed by a search for specific venture ideas) or the specific idea (problem solution), which only subsequently grew into a venture idea. Presumably, these two types of process origin would be associated with continued process and outcome differences. Operationalizing The External Environment I have pointed out above that from the perspective of the venture, the properties of the individual(s) and organization(s) involved in its creation are part of the environment. Substantial parts of this have been discussed already in terms of human, social, and organizational capital in the “Resources” section. I even suggested industry code and location as proxy variables for different types of capital. This delimits to some degree what remains to be discussed under the present heading. Admittedly, however, my treatment of the external environment will be cursory. The issue of the role of the external environment opens up a vast area of research in human or economic geography (Dicken, 1998; Malecki, 1997), regional economics (Fujita, Krugman & Vennables, 1999; Isard, 1956), industrial organization economics (Acs & Audretsch, 1990; Caves, 1980; Scherer, 1980), institutional economics (Baumol, 1990; North & Thomas, 1973), evolutionary economics (Nelson & Winter, 1982) and population ecology (Hannan & Freeman, 1977). It is far beyond the scope of this chapter and the capacity of the author to go into any detail. In short, I don’t know all this stuff, but I think it would be worth going these places to find concepts and operationalizations regarding characteristics of environments, just like psychology is a good place to turn for operationalizations on the individual level. Data on characteristics of industries and spatial units have the distinct advantage of being available in official statistical registers. While accessibility and accuracy of such data vary by country, time period and level of analysis, satisfactory data for control variable purposes are likely to exist. When environmental influences are a core theoretical interest the situation is less satisfactory. Available data are likely not to capture precisely those units, characteristics or time periods that are theoretically most relevant. To make matters worse, some research questions demand, and some research approaches allow, a distinction that was not made explicit in Figure 6:3, namely the distinction between subjective and objective environment. For example, in my dissertation study, I argued that while behavior is necessarily influenced by environment as perceived, the outcomes are determined also by objective characteristics of the environment, whether they are perceived or not (cf. Davidsson, 1989a and Figure 3:2 above). In empirical micro-level studies of strategy and entrepreneurship, hostility, dynamism, and heterogeneity (Dess & Beard, 1984; Miller, 1987) have been popular environmental characteristics to include. These are largely subjective assessments of the immediate task environment that may be based on objective characteristics of both industry and location, but also on more idiosyncratic factors that are specific to the venture, the firm, or the individual—including pure misconceptions. Assessment of environmental hostility, dynamism, heterogeneity and similar dimensions require

Operationalization Issues

125

primary data. Ratings by the entrepreneur/champion may suffice if the intention really is to assess subjective perceptions of those factors. Otherwise, other relevant informants, and preferably multiple ones, need to be used. In his longitudinal study of small firm growth, Wiklund (1998) ventured another important extension of this popular set of variables. Alongside the usual operationalizations, he included also perceived change in these three dimensions—and found the dynamic versions of the variables to be at least as influential as the original ones. This “dual” operationalization of the environment (state and trend) may be well worth copying in future studies. Operationalizing Behaviors The most obvious way to assess behavior is to study it through direct observation. A fascinating approach would be to follow entrepreneurs close-up much in the same way (Mintzberg, 1974) did with managers. Access problems left aside there are fundamental differences that make such an approach much less promising for research on entrepreneurship. First, the fact that a person has assumed the entrepreneur’s role in a process does not in any way guarantee that s/he fills her entire day with “entrepreneurial” thoughts, decisions and actions. Second, the method is too resource-demanding to allow for following processes over their entire duration. Therefore, a close-up, real time, observational study of an “entrepreneur” is likely to share a very high percentage of its contents with similar studies of “managers”. Nevertheless, intense observational study of “notorious” and successful entrepreneurs through critical stages of new venture processes definitely has potential for furthering conceptual development in entrepreneurship research. As discussed already in the “Resources” section, Baron & Brush’s video-taping of business plan presentations exemplifies a more focused observational approach that retains much realism. Sarasvathy’s (1999a) verbal protocol analysis in the context of hypothetical product launch does not assess behavior directly but should be more revealing than interviews concerning past behavior. Retrospective tracking of processes is known to be biased by knowledge of the outcome of the process. This was the primary reason why the Minnesota innovation studies (Van de Ven et al, 1989) choose to design their studies as comprehensive, real time case studies with multiple types of data collection within the cases, which allows for observation of behavior. But how are we to operationalize relevant behaviors in broadly based surveys? The behaviors for which retrospection is most difficult to avoid concern the earliest stages of the discovery process, namely what type of search (or serendipity) lead to the initial realization that something might be (the seed of) a viable venture idea. Admitting this difficulty, we should not give up completely on this important issue. Busenitz (1996) as well as Kaish & Gilad (1991) represent earlier attempts to assess entrepreneurs’ information gathering behavior with questionnaires—and with some success. In the New Internal Ventures study, we cluster analyzed sixteen searchrelated items and came up with three relatively distinct and interpretable types: proactive search, reactive search, and fortuitous discovery (Chandler et al, 2003). The results are displayed in Table 6:7. Preliminary analyses suggest these types are associated with continued process and outcome differences.

126

RESEARCHING ENTREPRENEURSHIP

One of the most important parts of the questionnaires used for the PSED research is the section on “gestation behaviors”. These cover a range of discovery and exploitation behaviors (and some “events” that are not strictly behaviors). It turned out especially valuable that each behavior was time-stamped, i.e., that each

Operationalization Issues

127

affirmative answer was followed by a question about in what year and month this behavior was initiated or completed, as the case may be. Thus, the design combines “back-dating” of behaviors that had already occurred at the time of the first interview, and more precise time-stamping of behaviors that occurred in-between subsequent follow-ups. This way of assessing the accumulation of behaviors over time makes the data suitable for analysis with longitudinal methods such as Longitudinal Growth Modeling (Samuelsson, 2001, 2004) and Event History Analysis (Delmar & Shane, 2002; 2003; 2004). The gestation behaviors included in the Swedish PSED are listed in Table 6:8 (cf. Davidsson & Honig, 2003). For the most parts, the list of behaviors investigated in the US PSED is identical (cf. Gartner & Carter, 2003; Gartner et al, 2004).

128

RESEARCHING ENTREPRENEURSHIP

Operationalization Issues

129

The selection of gestation behaviors is in part inspired by Katz & Gartner (1988), to which other sources of inspiration have been added. The questions could, of course, be grouped in many different ways, and some of the behaviors could be dropped and others added. However, when it comes to exploring and testing ideas about the processes of discovery and exploitation in large samples, it is my conviction that the PSED longitudinal approach to recording gestation behaviors is a major leap in the right direction.

Operationalizing Outcomes The first rule about operationalization of outcomes, I would say, is that a credible assessment should emanate from a later point in time, and preferably from a different source than the explanatory variables. This is necessary for causal interpretation and avoidance of common method variance problems. When we use self-reports from the same cross-sectional survey in both the independent and

130

RESEARCHING ENTREPRENEURSHIP

dependent variables there is no safeguard against response style or social desirability being the true source of correlations. In Chapter 1, I defined the societal phenomenon “entrepreneurship” as a value creating activity. Therefore, it may be argued that value creation has some preeminence over other outcomes in entrepreneurship research. However, value creation can be hard to assess, and there are also many other potential outcomes that may be of interest for building a thorough understanding of the entrepreneurship phenomenon. In order to make a meaningful contribution, it may be wise to look at other outcomes than relatively short-term financial performance on the micro level. When deciding on design and operationalizations, the researcher has reason to think hard about several partly inter-related issues concerning outcomes. Some of these are level, type, direct vs. indirect, time span, and relevant yardstick. Regarding level, type and direct vs. indirect outcomes, effect variables other than venture level performance may include some of the following: learning and satisfaction among the individuals involved; firm level financial performance, cannibalization effects, and competitors’ and new entrants’ imitation and retaliation efforts. Some of these are, no doubt, harder to assess than self-reported financial performance, but studies including them would therefore also possess more uniqueness. The relevant time span is a huge but often forgotten issue in all business research on performance. When do we assume something to start take effect, and for how long should we assume that effect will last? Do we really believe the income statement three years later (to take one example) is uniformly the place to look for the financial effects of the initiation of a new internal venture? And if it were, how are we to single out this effect from the effects of all prior and subsequent events that also influenced the performance that year? It would be really sad if we all turned our brains off and settled for the first publishable correlation, wouldn’t it? There are loads and loads of organizational research trying to explain variation in performance, but not very often with impressive or indisputable results. Should we be surprised, given the simplistic manner in which this very tricky analysis problem is often treated? As I see it, there are three partial remedies to the problem, and these go for the venture level of analysis as well. The first is stronger theoretical argumentation, down to the detail of the timing of the alleged effect. The second is homogenization of the analyzed sample, so that the assumption of uniform effects has some realism to it at all. The third is greater sophistication regarding analysis methods, so that competing explanations can be justifiably ruled out. Oh, yes, there is a fourth as well: betler quality of the performance operationalization. An important final aspect of outcome assessment concerns what yardstick to use. Venkataraman (1997) argues that entrepreneurship research should not be about relative performance but about absolute performance. One reason for this is that relative financial performance has been the focus of loads of management research, and the scholarly domain of entrepreneurship research would have little to contribute if given such a focus. However, Venkataraman’s main argument seems to be a different one that is not equally easy to follow, as it is somewhat unclear what “absolute performance” stands for. The only way to assess an outcome is to relate it to something. Venkataraman’s argument seems to be that comparison across firms

Operationalization Issues

131

(or ventures) misses the point. The relevant issue is whether the outcome was positive in absolute terms and relative to the investment possibilities that were realistically open to those individuals or organizations that invested money and effort into it. This is a point worth careful consideration. If heterogeneity is accepted as a fundamental assumption, why should we employ uniform performance criteria across ventures? Imagine, for example, a low educated independent entrepreneur who succeeds in setting up a simple restaurant that provides her with a better living than she had in previous employment. That venture’s performance is not substandard just because is scores below the average for a random sample of ventures across all types of industries. An implication of this is that subjective outcome measures should not only be regarded as something one is forced to accept for lack of better alternatives. At least for some purposes, the subjective outcome assessments may actually be the most relevant. This should not be taken as a general excuse, however. Sometimes subjective measures of outcomes have higher validity as a measure of the researchers’ lack of effort than as operationalization of “true” performance.

OPERATIONALIZATION ISSUES ON THE FIRM LEVEL After this very long section about the venture level, I will keep the section on firm level issues very short. This is not just because of the length of the preceding section; there are better reasons as well. First, we have already discussed specific firm level operationalizations of entrepreneurial orientation and behavior in the section on “Some balancing exercises”. Second, many of the issues discussed in the previous section on the venture level apply to firm level studies as well. While this may be particularly true for the “resources” and “environment” sub-sections, there are aspects of the other three sub-sections (venture idea, behavior, outcomes) that are also of relevance when the firm is the level of analysis. Third, issues concerning empirical and conceptual definitions of “firm” were dealt with at length in the previous chapter. This ambiguity about what the unit “firm” is, has implications, not only for sampling, but also for operationalization. It is impossible to offer a complete treatment of those implications here, but two aspects should be mentioned. The first aspect that is general enough to deserve explicit treatment here is that it may be difficult to get agreement between the researcher and the respondent concerning what unit the interview is about. Hence, it may be difficult in phone or mail surveys to get respondents consistently report data for the same unit, and for the right unit—the one that the researcher sees as “the firm”. This, of course, is a major threat to validity, and the end result is likely to be weak or false results. Problems of this type are relatively limited when researching on-going start-ups or very young and small firms as in the Swedish PSED or the 1994 Start-up Cohort. In the study Entrepreneurship in Different Organizational Contexts, which included firms that could be subsidiaries or have daughter companies of their own, we were subjected to this issue with full force. In my opinion, the ways to deal with it are a) pre-testing— gain awareness of the problems that are likely to occur; b) in-house interviewing rather than outsourcing to a marketing research firm or the like—they may be “professional” and with “unproblematic” sampling units they can be excellent, but

132

RESEARCHING ENTREPRENEURSHIP

they are not professional academic researchers and they don’t care about the data quality as much as you do; c) training, stimulation and close supervision of interviewers—make use of their feedback in order to solve problems in time, and d) clear and repeated instructions to respondents. These measures in combination can greatly reduce the threats to validity. There is no way we can totally eliminate them. The second issue I will mention here regarding how empirical and conceptual definitions of firm—and their matching—translate to operationalization problems, concerns the assessment of growth. As this is only partly an entrepreneurship issue, I refer the reader to Davidsson & Wiklund (2000) for an elaborate treatment; see also Davidsson et al (2002); Delmar et al, (2003). Suffice it here to stipulate that a) growth is not a unidimensional phenomenon, but a collection of more or less loosely related phenomena, b) both one’s definition of “entrepreneurship” and one’s conceptualization of “firm” determine how growth should be assessed, and c) as firms—unlike biological organisms—can go through unlimited transformations it may be impossible to make meaningful statements about a firm’s growth over longer periods of time. Paradoxically, in order to say that something grows “it” at the same time has to remain “the same” in some meaningful sense. Other than that, my sly plan for this section was just to do some presumptuous advertising for an alternative to the Entrepreneurial Orientation scale (Covin & Slevin, 1986), which we developed in the Entrepreneurship in Different Organizational Contexts study (Brown et al, 2001). This work started with the observation that despite the increased popularity of Howard Stevenson’s conceptualization of entrepreneurship, his ideas had never been systematically tested in empirical research. So we set out to do it. After two rounds of pre-testing and one full scale test, we came up with the instrument displayed in Table 6:9. Factor analyses revealed that the sub-indices have very good discriminant validity, both internally and in relation to the sub-indices of Entrepreneurial Orientation (EO). The Cronbach Alpha values were satisfactory in most cases, but lower levels for Resource Orientation and Reward Philosophy bring the average down to 0.69, which is only moderately good (Brown et al, 2001). When one, global index was computed across the 20 items the Cronbach Alpha was 0.73. Early analyses show that this global index has a significant, albeit not extremely strong, influence on corporate venturing activities, which, in turn, are positively related to performance (Eliasson & Davidsson, 2003). In an independent test, the scale appears to hold up technically also in German translation (Harms & Ehrmann, 2003) and efforts are under way in other countries as well. In the German study, Entrepreneurial Management is ascribed stronger performance effects than those estimated for EO. Thus, the evidence so far suggests that this firm level operationalization of Entrepreneurial Management has some usefulness, but also that further development work is advisable for some of its sub-dimensions.

Operationalization Issues

133

134

RESEARCHING ENTREPRENEURSHIP

Operationalization Issues

135

OPERATIONALIZATION ISSUES ON AGGREGATE LEVELS I will discuss operationalization issues on the industry, regional and national levels collectively. This is for two reasons. First, I do not think the issues differ to any considerable degree between regional and cross-national studies, and to a considerable extent the issues are the same for industry level studies as well. Second, my experience from cross-national and industry level studies is limited. The exposition will therefore rely mainly on experience from regional level studies. Let me start, however, with a couple of issues that I have learnt about in a crossnational context although they may apply to some extent on other aggregate levels as well. The first has to do with the validity of operationalizations across translation to different languages. There are two cases here that should be discussed separately. The first is when you want to use a measure for a within-country study, which has previously been developed in a different language. The solution here is simple: translate—check translation through back-translation—pre-test. As this is not necessarily an issue for aggregate levels of analysis specifically we should not dwell more on it here. The second is a much bigger issue: when we use “the same” operationalization in different languages within one and the same study, how can we know that differences in the results are “real” and not just reflections of language differences? The likely answer is we can’t. I am not an expert and I haven’t seen this seriously discussed in an entrepreneurship research context, so my advice has to be to look elsewhere, where researchers have given this issue more serious thought. From what I have heard, a three-country study of attitude-type variables in three languages would be dismissed as 100 percent unreliable because language based interpretations of the results would be at least as plausible as would substantive interpretations. The comparability of the translated measures would have to have been carefully calibrated beforehand (cf. how IQ tests are carefully calibrated across population groups to avoid biases) and/or entire groups of countries—preferably not all adjacent—would have to be included in each language, so that the study includes variation both within and across languages. Other operationalization issues in cross-national studies are cultural rather than purely linguistic. Items that appear to measure “the same thing” in one culture/language may not be perceived to do so in another culture/language, even if the translation is impeccable. Also the wording within individual items may have differential applicability. I came across one example of the latter in the Culture and Entrepreneurship study, where my committing the Deadly Sin of double loading an item only became evident after trying to apply the same instrument in another country. This was one of the Autonomy items, which read “I have probably found it harder than others to let authorities like parents, teachers and superiors decide for me.” The Swedish respondents had no problem with this item, but in Estonia, respondents protested wildly against my lumping parents together with other authorities! Apparently, in the latter country obeying parents and obeying bosses have very different implications or connotations, whereas in Sweden, they can all represent the same theoretical category “authorities”. This shows how subtle—or at least non-obvious—cultural sensitivity to operationalizations often is. When it comes to selecting what individuals (or other micro-units) should represent the aggregate level, we enter a borderland between sampling and

RESEARCHING ENTREPRENEURSHIP

136

operationalization issues. If we sample individuals in order to get measures of cultural dimensions (Davidsson, 1995a; Hofstede, 1980; Lynn, 1991), then who is sampled determines how culture is ope rationalized. This is why we found reason in the previous chapter already to note that while, for example, university students may in one country not be too far off representing the “mainstream” they may in other countries represent a very particular elite. I will only mention one operationalization issue that originates specifically from industry level study (but which applies to the firm level and spatial units as well). This is the observation that results concerning innovative activity are very sensitive to whether it is measured on the input side (R&D expenditure), the output side (No. of launched innovations) or with a proxy measure in-between (e.g. patent registrations). Small firms tend to be low on (formal) R&D input, but high on output. Therefore, if innovative output is the variable of interest, using an inputbased measure will distort results by firm size, and therefore also by industry firm size distribution and regional or national industry structure (Acs, 1996; Acs & Audretsch, 1990; Kleinknecht, 1987, 1991). In the remainder of this section I will rely on experiences from Business Dynamics in Sweden (Davidsson et al, 1994a, 1994b, 1995) and Culture and Entrepreneurship (Davidsson, 1993, 1995a, 1995c; Davidsson & Wiklund, 1997), both of which applied the regional level of analysis. The issues I bring up are likely to have relevance also for other aggregate units, but I will not explicitly discuss these other levels. The model in Figure 6:4 portrays some of the highlights in graphic form.

Figure 6:4

Design and operationalization issues on the regional level

Operationalization Issues

137

To the left in this figure, we have three sets of explanatory variables: structure, culture, and macroeconomics. These, then, are our first candidates for operationalization. The small boxes to the far left denote that we need several indicators of each—probably far more than just three. Indicators of structure are the smallest problem. This type of data is often available in statistical registers, although the quality differs by country: firm size and age structure, industry structure, human population demographics and socio-economic variables, local tax level, connectivity indices, etc. The remaining issue is how the available indicators relate to theoretical concepts, and how the latter should be represented: with several separate indicators for each concept; a “best indicator” for each concept; a summated index of items arguably reflecting the concept, or with a factor score from a factor analysis? The second group of variables—macroeconomic conditions—was something I found to be lacking in our studies. We had reason to believe that certain patterns in our results were sensitive to business cycle swings, but as our studies were not longitudinal in the true sense of the word, we could not investigate this empirically in a systematic fashion. For example, we had reason to suspect that start-up rates were high for unemployment reasons in certain types of regions in downturns, and high for market pull reasons in other regions in upturns. The needed indicators of macroeconomic conditions—interest rate, unemployment, GDP growth, and what have you—are likewise relatively easy to get, but in order to get any variation in those indicators the design has to be longitudinal. Constants can be important, but they can’t explain variation! The reason I conducted the Culture and Entrepreneurship studies at all was that I felt cultural variables were lacking in Business Dynamics in Sweden, which relied solely on secondary data. From that kind of source you can only get weak and hardto-interpret indicators of culture, like election results, numbers of churchgoers, and perhaps numbers of members in associations for various leisurely activities. So, as described in the previous chapter, I decided to collect primary data from representative samples of regional populations and have their average responses to sets of attitude items represent various dimensions of the culture—understood as predominant ways of thinking—of the regions. Regrettably, despite pre-testing and reasonable theoretical underpinning I was not as successful at this as I would have wished, as discussed earlier in the section about “Some balancing exercises”. So the advice here is hard earned: that there are stacks of textbooks, chapters, and articles on attitude measurement. Make use of the literature to learn the craft thoroughly. Pre-test, and pre-test again if you have to develop new measures. But before even considering that, find and use “tried-and-true” measures that others have developed as far as possible. You are welcome to use mine if you wish (Davidsson, 1995a, 1995b; Davidsson & Wiklund, 1997), but other specific operationalizations of culture may have better measurement properties (e.g., Hofstede, 1980; Lynn, 1991). We have touched upon a couple of times already that the operationalizatio n of regional culture is contingent upon who we choose as representatives of the region. Should we sample opinion leaders or the general population? What age groups should be included—all, those in work force age, or the young who will shape the future? I chose narrow age cohorts from the general population in order rule out artificial differences due to random sampling error on age—but at the same time this

138

RESEARCHING ENTREPRENEURSHIP

design choice threatens to conceal cultural differences that are due to regional differences in the age distributions of the populations. Depending on the purpose of the analysis, I had sometimes reason to consider eliminating parts of the samples I had already collected data from. Should all respondents influence the scores for regional culture, or only those who had grown up or lived for a long time in the region in question? Should the opinions of the entrepreneurs (here = business founders) in the sample be included or excluded? If we turn now to the dependent variable there were several lessons that I learnt—or started to learn—through the regional level studies. True, our results indicated that gross dynamics in general (entry, exit, growth and contraction) had positive effects and that entry of new, independent firms in particular was the most important determinant of growth of regional economic well being. But we also learnt that this was not enough: regions that have a small firm structure need a higher birth rate merely to maintain their business population, because such regions also have higher gross death rates. Moreover, we found one region in our data that consistently over two periods representing very different business cycle conditions appeared among the top ten in the country in terms of gross start-up rates—only to reappear on the bottom ten list for net start-ups. In this case the high gross rate probably reflected heavy subsidy of doomed start-up efforts rather than sound reallocation of resources for more productive use (Davidsson et al, 1998a). Moreover, the way we have defined entrepreneurship, the reliance on independent start-ups is misplaced (cf. Ch 1). New economic activities that provide customers with new choices, stimulate incumbents to improve, and attract followers are introduced also by established firms. So indicators of dynamism among established firms are needed alongside start-ups. We included measures of entry, exit, expansion and contraction of establishments (branch plants) in existing firms, small or large. And indeed, while new firm formation appeared to be the more important factor generally, we did find groups of regions whose favorable development relied on dynamism among established firms (Davidsson et al, 1995). Other possible indicators that reflect also corporate entrepreneurship are patent registrations and industry re-classifications (presumably from stagnating to growing industries). A venture is a venture is a venture is...not true. We also learnt that we need some measure of the quality of the new firms and establishments (Davidsson, 1995a). As already touched upon above, regions can have similar entry rates for very different reasons; for example market pull versus unemployment push. Direct measures of the start-up motivations, innovativeness and growth aspirations of the founders would be handy, but if these are lacking at least the industry and average size of the start-ups could give the right hints. This issue of quality is one where the Global Entrepreneurship Monitor (Reynolds et al, 2002) has improved over the years, by distinguishing between necessity and opportunity-driven entrepreneurship, and between innovative and imitative start-ups, respectively. However, from the perspective of entrepreneurship that I use here, corporate entrepreneurship remains a weak spot in GEM.

Operationalization Issues

139

SUMMARY AND CONCLUSION This long chapter about the columns in the data matrix started with a section on levels of measurement, where the important message was that you are not a bad sinner when you use so called “ordinal” variables along with statistical techniques that assume metric properties. As long as the underlying reality is a continuous variable the “ordinal” variable is just a somewhat more error-prone measure than true interval-scale measures; not a fundamentally different type. I then turned to validity and reliability, where I emphasized that we ought to do better than mechanically adhering to the most common research rituals. Cronbach Alpha tests internal consistency but is no safeguard against measuring internally consistent garbage. We also noted that not all summated variables should have a high Alpha. A formative index—rarely discussed in the literature—can be perfectly sound without having high inter-correlations among its items. Further, validity can never be fully proven, but the measure having theoretically predicted relationships with other variables give relatively strong support. I then discussed some delicate balancing acts. The first was about trying to develop perfect-for-the-purpose ad hoc measures versus recycling others’ short-ofperfect but workable instruments. After having engaged in substantial adhocery in my (research) youth, my current bias (especially in reviewer and editor roles), is clearly towards the latter. Admittedly, as entrepreneurship is still a young field, there is room yet for justified creativity in operationalizations, but it takes a major and serious effort to come up with something really useful. The second major balancing act, which is due to the heterogeneity of entrepreneurial ventures and contexts, concerned broad applicability versus perfect suitability for narrower groups. Measures that apply to all may at the same time apply to none in the sense that they only capture a tiny fraction of each case’s manifestation of entrepreneurship. The bulk of the chapter discussed Operationalization problems and opportunities on different levels of analysis. The amount of space devoted to different levels was deliberately uneven, reflecting my own experience and interests. But not only that: the venture level has a certain centrality combined with relative virginity (if such can be relative...) that justifies devoting more space to it. Regardless of level, I have tried to give advice down to the level of actual operationalizations of particular concepts. This practice has been patchy, but then again a complete treatment is inconceivable. I hope that some of the operationalizations I have offered have inspired some readers, but also that they are sensible enough to check other sources before they rush out to collect data.

CHAPTER 7

SPECIAL TOPIC: PREPARING A “SECONDARY” DATA SET

IF YOU DON’T HAVE IT, DON’T TRY IT... Let me start this chapter with a couple of sad stories. Once upon a time there was a team of US labor economists who got fed up with what they thought were false claims that small firms were the major providers of new jobs in the economy. So they wrote a working paper—later to become a journal article and a book chapter as well—where they used hypothetical examples and empirical analysis to show that such claims were indeed false. And they made very strong counter claims on the basis of their “results”, effectively belittling colleagues who had arrived at the opposite conclusion. There were some problems, though. The hypothetical examples concerned three method fallacies that would erroneously ascribe superior job creation to smaller firms. These were the size distribution fallacy, confusion between gross and net job creation, and the regression (to-the-mean) fallacy. For the first two it is easy to show, however, that the bias could go in any direction, and the three authors were unable to cite any studies whose results were actually contingent on these biases. As regards the regression bias, it undoubtedly skews the estimates in favor of smaller firms—but its influence on the results is likely to be very marginal (cf. next chapter and Davidsson, 1998b). On the empirical side the problems were even worse. The authors had US data from the 1973-88 period. This was a period when tens of millions of (net) new jobs were created in the US. However, they only had data for the manufacturing sector. This was a shrinking sector during the said period, and therefore perhaps not the ideal source of evidence about where in the economy new jobs were created. Moreover, the data set had incomplete coverage of the smallest size class (units with five or less employees), which is precisely the size class in which other research suggests the small firm job creation surplus occurs (Baldwin, 1995; Davidsson et al, 1994a; 1996). Again, this is perhaps not exactly the empirical material you need in order to tell how wrong others are on the issue of job creation. More generally, if restricted data availability makes your mission impossible, perhaps you should pick another mission? Yes, it is the work by Davis, Haltiwanger and Schuh (1996a, 1996b) I am talking about. These researchers were right in pointing out that if small firms are

RESEARCHING ENTREPRENEURSHIP

142

shown to create most new jobs, this does not prove that policies aimed at facilitation of small firm creation and expansion are warranted. Moreover, their argument forced other researchers to refine their tools and their interpretations, so in that sense their effort paid off. For most other purposes they were just plain wrong, and in the most arrogant way. As the next chapter is about job creation as dependent variable, we shall have reason to return to some of their claims. My second sad story concerns work that I was myself directly involved in. This was an international policy research initiative directed at the following research questions: What is the prevalence and relative job contribution of high growth firms (or “gazelles”)? How does this differ among various OECD countries? We met a couple of times in Paris to discuss how we could obtain and harmonize our data, so that we could make reliable comparisons across countries. We could agree on the time period: the ten years leading up the most current year in the data registers. We could also agree on a set of specific growth measures. Along the way, however, the following not-so-pleasant conclusions were also drawn: 1.

2.

3.

4.

We could only use data from the manufacturing sector, because complete data for other sectors were lacking in some of the participating countries. Slight problem: manufacturing accounts for some 10-20 percent of total employment, and is shrinking. We could only use firms with a minimum size of 20 employees, because in some participating countries no data were collected for smaller firms. Slight problem: research suggests small firms on average grow faster than large firms (Evans, 1987; Kumar, 1984), so many gazelles should be expected to be or at least start small. Much for the same reason, we could only include firms that were in existence for the entire ten-year period. Slight problem: Research suggests young firms on average grow faster than small firms (Evans, 1987), so one would expect to often find gazelles among young firms. We would in most countries not be able to distinguish between organic growth and growth through acquisition. Slight problem: it makes quite a difference from a job creation/policy perspective whether growth of a firm represents addition of jobs that are also new to the economy at large, or transfer of jobs that already existed within another firm.

Very obviously, data sets that are subject to these restrictions would be fundamentally inadequate for addressing the research questions that were the very rationale for the effort. What policy relevance could it possibly have to assess and compare the prevalence of high growth firms among the senior members of the manufacturing industries, when we can tell at the very outset that the design ensures that we will miss the large majority of gazelles? What credibility does it give a theory to be based on or supported by such data? Fortunately, this project was never

Preparing a “Secondary” Data Set

143

carried out as planned. I’m not convinced, however, that it was the shortcomings of the data that killed the project, as should have been the case. It may well have been lack of funds. Both of the above examples very much remind one of a joke that is probably as old in English as it is in Swedish. This is about the very drunk man who is seen rummaging about in the snow under the streetlight late at night. Another man approaches him and the following conversation takes place: Excuse me, can I help you? What are you looking for? I’m..scherscheing for my (hic)... keysch. Your keys? And you’re sure this is where you dropped them? No (hic)...not at all. I’m schure I dropped them over there (the drunk man says, pointing into the darkness, almost falling over backwards) Over there? But why then are you looking for them over here!? Well...ischn’t that obviousch? Over there it’sch scho damn dark you can’t posschibly find anything...

As researchers we do not want to be like this drunkard, right? Secondary data are, to some extent, like streetlights. They are put there for general purposes or for some other purpose than yours. They do illuminate some area but they do not necessarily cast light on the issues you are interested in. If the data cannot possible answer your research questions—as is often the case in entrepreneurship since emerging phenomena do not appear in registers—don’t try!

...OR DO THE JOB NEEDED TO MAKE IT WORK But then again, this is not advice to give up easily. Serious work with secondary data means doing more than taking an existing data set as it is, and running some analyses on it. Perhaps you can move, amplify or combine several streetlights to serve you better? That is, using secondary data to address your research questions may not be mission impossible, but it may take considerable effort and money to make the data work for you. Birch’s (1979) seminal study on job creation is actually an example of this (cf. Davidsson, 2002). Birch realized that no available data set could answer his research question. Available data only allowed analysis of changes in the aggregate size distribution of firms; they did not make it possible to follow the development of individual firms over time. So he went through the painstaking effort to create a suitable data set by matching and cleaning available data sets. That is why he was able to come up with an important finding. The quality of his data has subsequently been questioned, but those who care to read the report will find that he spends several pages explaining the remaining limitations. In the remainder of this chapter I will discuss what one can do to make secondary data work. I will do so under the headings Use prior knowledge, Combine different sources of data, Check quality and make corrections, and Other observations. My exposition relies on experiences from two big projects, namely Business Dynamics in Sweden and the High Growth Firms study. As usual, mine is an atypical account of the topic “secondary data”. This is not because I think the standard treatment is irrelevant; I just find it unnecessary to repeat it here. Likewise, some “sampling” and operationalization issues that would fit below have already

144

RESEARCHING ENTREPRENEURSHIP

been treated in previous chapters. What I try to add here is mainly what I feel is lacking in standard treatments, namely the problem of creating and developing a data set through thorough work on combining and checking data from several sources, rather than just checking, accepting or rejecting an existing set of data “as is”.

Use Prior Knowledge Business Dynamics in Sweden had a regional forerunner with a much more limited scope, but which also relied on a customized data set using data from different registers at Statistics Sweden (Olofsson, Petersson & Wahlbin, 1986). The many problems associated with the preparation of the data set for that study were a great help in our work. From this we had learnt the importance of being a competent and demanding buyer in order to get the most out of available data. Thus, we realized that close interaction with experts at Statistics Sweden was necessary. Further, we realized we had to be not just competent researchers, so we put on our team also a PhD student in management who was also a computer systems engineer. This person could bridge the world of social science research and the world of collection, storage, retrieval and combination of large amounts of data. Another aspect of valuable prior knowledge was that parts of our work built either on the work of colleagues in other countries or was done in direct collaboration with them (Birch, 1979; Keeble, Potter & Storey, 1990; Reynolds & Maki, 1990). This gave us a head start on a number of issues like over-arching models to guide the research; ideas on how to organize the data; regional subdivision into Labor Market Areas rather than administrative units; the pros and cons of establishment- vs. firm level data, as well as criteria for subdividing establishments into different types; assessment of business dynamics not only in terms of gross entry but also exit, expansion and contraction of business units; ideas on specific explanatory variables to include, etc. Although our project was novel in many ways, we were far from starting from scratch and this saved us many problems while at the same time boosting the potential yield of the project. When designing the High Growth Firms study we benefited a great deal, of course, from our experience with the previous project on business dynamics. At this stage, we already knew a great deal about the contents of different registers and the data collection behind them. More importantly, we had inside insights into whom at the receiving end we would like to have involved in the project and how we could make our project interesting for them to work with. We also knew when these people were likely to be overly busy doing other, annually recurring tasks. You don’t find much on that in method textbooks. Finally, a very important aspect of using prior knowledge is, of course, to make use of theory. This is one area where we could and should have done a better job— although it was not so bad that our efforts were completely void of theory. However, amidst the process of compiling a customized, secondary data set it is easy—perhaps easier than with other modes of data collection—to get carried away by the potential and availability of data, and start to include whatever “might be interesting”. Theoretical input helps one avoid missing really important things, but also to screen out unnecessary distractions. It also helps telling you when your mission is

Preparing a “Secondary” Data Set

145

impossible and you should refrain from going further with your project along the secondary data route.

Combine Different Sources of Data In order to create a data set that is really useful for research purposes, combining data from several existing sources is a key issue. This, of course, requires that common identifiers exist. Fortunately, they often do. Establishments, firms, industries, municipalities and counties regularly have standard codes that are used in different registers and over long periods of time. Individuals are a trickier issue. First, to match data from several sources may be prohibited—even inside the statistical agency—for integrity reasons. Second, computerized matching by name would be notoriously unreliable. In Sweden every individual has a unique ten digit “person number” that is used across registers. In other countries there may exist a “civic registration number” or a “social security number” that allows matching of several annual versions of the same register, and perhaps also matching across registers that contain different sets of variables. Depending on the purpose and the available data, there are endless examples of what kinds of combination of data one would like to make. This makes it impossible for me to craft a step by step, ready-to-use manual for the specific purposes of a particular project you might have in mind. Therefore, I choose rich examples from my own research to illustrate and inspire, knowing that it is unlikely that these particular data combinations are exactly what you need. In the Business Dynamics in Sweden project (Davidsson et al, 1994a, 1994b., 1995, 1996, 1998a, 1998b) we wanted to create a data set that made it possible to: 1.

2. 3. 4. 5. 6. 7.

monitor individual establishments and their employment figures over time in order to identify births, deaths, expansions and contractions and the associated job changes with accuracy, group the establishments according to size of employment, identify different types of establishments, i.e. autonomous single-site firms, corporate headquarters, and branch plants, separate establishments in SMEs from establishments in larger firms or corporations, group the establishments according to sectorial affiliation, relate the establishments to labor market areas, and locate geographically the ownership of branch plants, including foreign ownership.

In order to achieve this, we worked in close cooperation with register experts and programmers at Statistics Sweden (SCB; the Swedish bureau of census). Data from various data-bases and different (annual) versions thereof were combined and checked in order to achieve the highest possible quality of the input data. All together, data from four data-bases were been matched for the compilation of business dynamics data. These were: the Central Company and Establishment Register (CFAR)

146

RESEARCHING ENTREPRENEURSHIP the Regional Employment Register (ÅRSYS) the Register of Groups of Companies (KCR) the Register of Establishments and Firms with Foreign Ownership.

In addition, we used regional level data from several other registers in order to gather potential independent variables, i.e., structural regional characteristics that might explain regional variations in business dynamics (Davidsson, 1994a, 1994b). For business dynamics, however, the CFAR was our starting point. This data base contains rudimentary information such as identification code, name, address, industry classification, and employment size on the company (firm) and establishment (branch plant; outlet) levels. Theoretically, the firm level would have been more ideal, but we learnt early on that company codes were changed as soon as a company changed legal form or industry or location or ownership, which created a lot of artificial births and deaths—a core topic for our research. Therefore, we had to settle for the establishment level, as identification codes for establishments are more robust to single or minor changes such as those just mentioned. CFAR is organized as separate data sets for (November) each year. In order to make point (1) above possible, our first decision was to combine different annual versions of CFAR. This made it possible to track births, deaths, expansions and contractions. However, some delicate choices strongly affect the estimated absolute and relative magnitude of these forms of economic dynamism, which should be carefully considered in design as well as interpretation stages. For example: Which establishments should be eligible at all? We went for those classified as commercially active. In different registers activity could be based on, e.g., reaching minimum sales turnover, being registered as an employer, having paid taxes during the period, etc. We used a combination of legal form (excluding forms that are typically used for non-commercial organizations or associations) and having sales. What should be the length of the analysis period? Consider a firm that has two employees the first year and five the second year. With annual data we will record two jobs created through birth and another three through expansion. With two-year periods all five jobs will be ascribed to birth. In addition, the longer the period the larger the number of units that both enter and exit between two measurement points and hence are not recorded at all. Therefore, studies using different length of periods are not comparable. Our choice was to work with annual data. In order to achieve (2), the establishments were simply grouped into five size classes based on their number of employees: 0-4, 5-19, 20-49, 50-199, and >200. Note that this refers to the size of the establishment. With the exception of the >200 size class, units in all establishment size classes may be found in small as well as in large firms. As we learnt from the register experts that CFAR was judged less satisfactory as regards employment figures, the original size measure has in most cases been exchanged for more accurate figures from the ÅRSYS register, which tracks specific individuals and which is more frequently updated. A correction was made so that the owner-manager was counted as an employee irrespective of the

Preparing a “Secondary” Data Set

147

legal form of the firm. In the original registers, the owner-manager would typically be counted as an employee in limited liability companies, but not in partnerships and sole proprietorships. The identification of establishment types involved extensive (computer programmable) search and matching of data in CFAR, KCR, and the Register over establishments and firms with foreign ownership. Following Reynolds and Maki (1990), three types of establishments were used in the first study: Simples, which are autonomous, single-establishment firms. These are identified by matching the establishment identifier with its company identifier. If only one establishment is associated with that company code, the establishment is a suspected Simple. If the company code is not found in Register of Establishments and Firms with Foreign Ownership it is a confirmed Simple. Branches, which are either units other than headquarters in company groups, or units other than headquarters in independent multi-establishment firms. These are identified when the above procedure suggests other establishments are associated with the same company code and/or foreign ownership is confirmed. Tops, which are headquarters either in company groups or in independent, multi-establishment firms. One top was identified in each multi-establishment company or company group. This classification was done through a hierarchy of criteria such as sharing address, phone number, industry classification or zip code with the company or parent company. A major improvement in the second Business Dynamics in Sweden study was that we were able to identify the size of the larger structure (firm or company group) of which the Tops and Branches form part (for Simples the company size equals establishment size). This involved a very extensive data processing procedure in which the employment figures for each Branch that was associated with a particular Top were added to the size of the Top itself. In that way, it could be determined for each of the approximately 600 000 establishments in the data base whether it should be regarded an SME establishment or a LE establishment (4). The firm or company group employment size chosen for this dichotomization was set at 200 employees. In the second study, we thus dealt with six establishment types: SME Simples, SME Tops, SME Branches, LE Simples (very few), LE Tops and LE Branches. LE Branches can be further subdivided into those with domestic versus foreign ownership. As regards industry (5) and location (6), the CFAR contains information on sixdigit industry classification, as well as municipality and county codes. Using other information, we aggregated these classifications to categories more useful for our purposes, as described in Chapter 5. However, with longitudinal data, these classifications are not as simple as they may sound. First, we had to decide on which industries to include. In the first study, we confined our data set to the private sector except primary industries, using a combination of legal form and industry classifications as criteria. In the second study, we included the entire economy. One

RESEARCHING ENTREPRENEURSHIP

148

reason for this was the insight gained in the first study that there was an in- and outflow across sectors included and not included in the study, causing some cases of industry change to appear as exit and entry. Second, the industry classification system was changed in 1992, i.e., in the middle of the second study. This is a type of problem one should expect when working with longitudinal secondary data. We used translation keys and parallel analyses with old and new classification for one analysis year in order to minimize the problem and assess how serious distortions it might lead to. Finally, comparison between establishment and company level data in CFAR supplemented with a check against the foreign ownership register allowed the rough classification of location of ownership (7) that we wanted. The categories we used were: within same county, rest of Sweden, and foreign. Straightforward enough, isn’t it? Well, the above customizations made for weeks of work already, including several iterations with the data experts. But doing longitudinal research is more fun than that! Problems occur when establishments change categorical affiliation (region, establishment type, size class etc.) from one year to the next. This required some thinking, and absolutely impeccable decision rules are not to be found. Our solutions were the following: For establishments that changed industry or region, we assigned negative changes to the category of origin and positive changes to the new category. This gets the direction of change right, but it can be argued that their effects become somewhat exaggerated. As regards establishment type and size class, all changes were attributed to the original category. As this makes the results sensitive to the “regression fallacy” (Davis, Haltiwanger & Schuh, 1996a) we carefully checked the extent of this problem, and could conclude that the resulting overestimation of SME job creation was negligible (Davidsson et al, 1998b). For the High Growth Firms study (Davidsson & Delmar, 1997, 1998, 2000, 2003; Delmar et al, 2003) we had in part the same and in part different challenges to deal with, and hence we used both old and new data combination strategies in order to find satisfactory solutions. We used a combination of the same basic data sets as above when designing this study, plus sales data from yet another source. This time we were trying to achieve the following: i. ii. iii. iv.

identify all “high growth firms” in the Swedish economy in (then) recent years, assess their individual and collective growth in sales and employment, classify them according to industry, region, governance structure and size class in order to see where they were over- and underrepresented, respectively, and assess how they achieve their growth in terms of organic growth vs. expansion through acquisition. As has been noted above this is important from a policy perspective because growth through acquisition does not reflect creation of genuinely new jobs. For our

Preparing a “Secondary” Data Set

149

current purposes the distinction is important because it can be argued that organic growth is more likely to reflect entrepreneurial activity than is acquisition growth (cf. Chapter 2). Yes, this is the Swedish branch of the not-so-successful international collaborative effort I portrayed in the introduction to this chapter. Fortunately, the situation with data availability in Sweden was such that a meaningful study could be carried out—provided one was willing to do the job needed to make it work. Before going further, however, I should admit a (partial) fundamental problem with the study. I am today more aware than I probably was at the time (and in part as a result of being involved in this study) that, for many purposes it is better to study “firm growth” than “growth firms” since the latter is a moving target. To single out the category “growth firms” as the focus of the study almost by definition leads to the Deadly Sin of sampling on the dependent variable. But it is not such a big sin for all purposes. For example, in order to study how firms grow one has to study growth firms (Delmar et al, 2003; Penrose, 1959). At any rate, the basic design of the study was to start with firms that existed in the final year (1996) and backtrack their development each year to 1987 (or the first intermediate year they appear) through matching by identification code. Mainly for cost reasons, we delimited the study to firms that had 20 or more employees in the final year. The risk with this design is that if one uses firm characteristics in 1987 to predict subsequent development, those factors that lead to increased size change variability will be mistaken for factors that lead to growth, since both terminated firms and those that have shrunk to less than 20 employees are excluded from the data set. However, growth prediction or identification of growth factors were not our primary purposes. Identification of high growth firms and tracking of their individual development were, however, primary purposes. It is a no-brainer that the results will be sensitive how “high growth firm” is defined. Many criteria can be used and we certainly tested a whole range of them (i), effectively showing how important this choice is (Delmar et al, 2003). Among other things, in order to increase comparability (cf. Storey, 1996), we used in most analyses the ten percent of the population that grew the most. More precisely, we defined high growth firms as the top ten percent of the studied population in terms of average annual growth in absolute employment. The need to identify and follow individual firms over time meant that—unlike the Business Dynamics study—we could not use establishments as proxies for firms, or confine our analysis to the development of aggregate categories of firm. Consequently, we had two very tricky problems to solve, where the solutions were likely to strongly affect our results regarding point (ii) above. The first was that we knew that company identification codes were changed because of a change in legal form, location or ownership, also when closer scrutiny would suggest the same firm lived on. Again, this would cause artificial exits and entries. As we suspected that this was especially likely to happen with more dynamic firms, severe underestimation of the prevalence of high growth firms could result. Our remedy was to not accept continuing company identification code as the sole indicator of survival. Instead, we matched company and establishment levels in CFAR. In cases where essentially the same set of establishments appeared under

150

RESEARCHING ENTREPRENEURSHIP

different company codes over the years, we regarded it as a surviving firm in our data. More specifically, for us to accept at that firm was the same unit as firm the following had to apply: i) at least 50% of the former employment across establishments in A was now found in B; and ii) this same employment constituted at least 50% of B’s total employment. This criterion is programmable and could establish unique links in the great majority of cases. Because of mergers and splits the above criteria do not always lead to a unique and satisfactory solution. So what does one do? Put people to work! In these cases, the two foremost business register experts at Statistics Sweden used a manual procedure (on late nights and weekends—would you believe that?!) for deciding, according to their best collective judgment, which of several links should be used, or neither. Through these procedures, we ensured continuation of surviving cases that changed their company identification code, as depicted in years 1 and 2 in Figure 7:1. The longevity of 25 percent of the firms in the study was influenced by this correction. Needless to say, this has a profound effect on the results. The second problem concerned the next level up: company groups. When a firm expands it is likely to eventually start to grow not just within the limits of the original company, but add subsidiaries that are legal (albeit perhaps wholly owned) units in their own right. If we delimit the analysis to the company level only, we would again face the risk of severe underestimation of the prevalence of high growth firms, as all the growth that occurred in legally separate daughter companies would elude us. To illustrate the magnitude of the problem, it was the case with our data set that 1672 firms (14.2% of the eligible population) were parent companies in 1996. A full 1372 firms (11.7%) actually turn to parent company from something else (independent or daughter) during the ten year period. Clearly, this internal restructuring may hide many high growth business activities if employment and sales in the daughter companies are not added to those of the parent company. The ideal solution to this problem would probably have been to mix data from the Company (CFAR) and Company Group (KCR) data sets and let the company be the unit of analysis (case) when no affiliation with a group was recorded and to use the entire group when such was identified. In principle, this would have been possible. Employing this technique, one would have a data set where continuation would be ensured also for cases going through change as depicted in years 2 to 3 in Figure 7:1 (cf. Chapter 5, in particular Fig. 5:1). However, we refrained from this latter option because the company and company group registers were not fully aligned as regards the (November) dating of the data, and because even more “manual” processing and judgment would be needed in order to deal with the continued identity of cases involved in mergers and acquisitions. Instead, as a second best, we created a parallel data set on the company and company group levels so that we could ascertain that our conclusions based on company level analysis were not completely altered when we turned to the company group level instead. Fortunately, they were not. Regrettably, the main conclusion remained that a severe shortage of rapidly growing firms seems to have characterized the Swedish economy during the studied period (Davidsson & Delmar, 2000, 2003).

Preparing a “Secondary” Data Set

151

RESEARCHING ENTREPRENEURSHIP

152

With that, we have dealt with the first two points listed above: identification of high growth firms (i) and assessing their development over time (ii). Regarding the latter, it is worth mentioning that sales growth could only be assessed for about 50 percent of the firms, and a non-random 50 percent at that. As is the case with business statistics in many countries, data were less complete for smaller firms and non-manufacturing sectors. That’s why most of our empirically based knowledge is about large(r-than-the-very-smallest) and manufacturing firms. As I have indicated above, this is less than satisfactory data for many research questions and probably completely irrelevant for some. The third point about classification of the firms according to industry, region, governance structure and size class (iii) poses no really new challenges compared to what was discussed above for the Business Dynamics study. We may note in passing that by matching company and company group data we kept track of the size of the firm itself, as well as that of the entire company group. In addition, in terms of governance the same matching allowed us to classify firms as independent, parents, daughters, or both parents and daughters at the same time—plus whether they were internationally or domestically controlled. Point (iv), however, posed a novel and tricky problem. When planning the study, I was amazed how little was written—especially on an empirical basis—about the distinction between growth through acquisition and organic growth. This is particularly puzzling as the implications of these two forms of growth are likely to be significantly different for managers and policy makers alike. To the best of my knowledge, the distinction has never been done before in a data set of the kind we were developing. Then how could we do it? Doing research is sometimes similar to doing business. Entrepreneurs are occasionally blessed by their ignorance: not knowing what is impossible, they show the world that it is indeed possible. Inert, mature firms can become innovative because demanding customers push them to realize their potential. It was something like this that happened in this case. We made the unheard of demand that organic and acquisition-based growth had to be separated, and a register expert with deep knowledge of what data were available and how they were organized used his creativity to come up with a solution. Again, the key to the problem was to match establishment and company levels, and different annual versions of the register. By doing this, we could determine for each establishment in a given year whether it was: a. b. c. d. e.

new to the economy and to the firm previously existing within the same firm previously existing but new to the firm (acquired) still existing but no longer part of this firm (divested) no longer existing (closed)

We defined organic growth as total firm level growth minus the employment that already existed in establishments of type (c) at the end of the previous. This can also be double checked as the sum of employment changes across categories a, b, d and e. Acquisition growth was consequently defined as total growth minus organic growth, which equal the balance of employment in category (c) from the previous year.

Preparing a “Secondary” Data Set

153

This feature of our data set certainly paid off. In the social sciences we do not really make scientific discoveries, but the below is probably as close as I will get in my research career. For it turns out that once “high growth firms” have been singled out a breakdown analysis of their growth shows that high growth firms of different size or age pursue dramatically different growth strategies. In short, young and small firms grow almost exclusively through organic growth, while older and larger firms grow through acquisition. As displayed in tables 7:1 and 7:2 (adapted from Davidsson & Delmar, 1998) the effect is dramatic for both variables but strongest for size, as high growth firms in the largest size class actually shrink quite dramatically in organic terms!

Hopefully this example shows that toiling with the data set can pay off. An even better example of that is, of course, Birch (1979).

Check Quality and Make Corrections Hold on—we’re not done yet. That we have organized the data set as best we can does not guarantee the data are correct. There is no way we are going to be able to anticipate all systematic problems while designing the data set, so we will have to make corrections in arrears. In addition, there will be stochastic errors in the data for individual cases. These can be large and have major influence on the results, so we want to eliminate them. Still using the same two studies for my mining efforts, I will

154

RESEARCHING ENTREPRENEURSHIP

illustrate in this section, the types of quality checks and corrections that might be necessary. For the Business Dynamics in Sweden study, these are some of the measures we took to check and correct the data: We first took establishment level raw data samples from two regions of which we had good knowledge of and performed “reality checks”. That is, we checked the printout data for seemingly questionable “facts” against our prior knowledge, and we also made random as well as dataprompted calls to companies in order to check the correctness of data on starting year, industry classification, employment size, etc. For integrity reasons this could only be done with relatively “harmless” data, and with strict adherence to rules set up by Statistics Sweden. In general, the results were encouraging—the study would be worth doing. Imagine instead that this kind of check would suggest the data were fundamentally unsuitable for the purpose. What loads of useless effort could have been saved! We then ran a full scale test on one region in order to check that all register matching and aggregation to regional level worked as they should and yielded sensible data (Davidsson, Lindmark & Olofsson, 1993). This test led to some corrections but in general the outcome was again encouraging. As regards the total level of start-up activity and regional differences in this regard we ran checks against the best available benchmarks nationally and internationally (see Davidsson et al, 1994a, for details). As regards level of activity, these checks showed that we arrived at high estimates in part because of definitional differences, but in part—which is worse—also due to data quality differences compared to other sources. As regards regional patterns, the checks left no cause for concern. In the second study, we had to deal with two one-shot changes in the original registers that occurred for tax reasons during the second period studied (1989-1994). Again, this is a type of problem one is always faced with in longitudinal research based on secondary data: changing definitions or changing extent of coverage. In this particular case, the changes led to the exclusion of a substantial number of surviving, very small firms in 1991, and their re-entering the register in 1994. Through various internal and external checks, we were able to obtain reliable information on the extent of overestimation caused by these one-shot changes and correct for them (Davidsson, Lindmark & Olofsson, 1998c). Finally, and importantly, there will always be some errors in the data and it is impossible to find and correct all of them. However, it is important that all substantial errors be detected—otherwise estimated effects of variables can be completely wrong, or specific regions, industries or categories of firms are pointed out as “winners” or “losers” for all the wrong reasons. And it’s not great fun to find out afterwards that the striking results you got published in the high rank journal, or the policy advice that made your regional government invest millions, were based on an error in the data. In order to avoid such embarrassment, we

Preparing a “Secondary” Data Set

155

conducted both proactive and reactive search for erroneous data in the full data set. Proactively, we searched systematically for sizeable data punching and editing errors by looking for “funny” annual changes within relatively fine-grained breakdown categories (e.g., for a particular establishment type within one industry in one region), and corrected them as best we could. Reactively, whenever we arrived at an unexpected or particularly strong analysis results we tried to make sure that the result was not based on an outlier or some other peculiarity of the data. In the High Growth Firms study we had excluded the large population of very small firms, for which the raw data is less reliable. Hence, we started from a situation with less potential for data errors. However, we found reason to employ again benchmarking against data from other studies (see, e.g., Davidsson & Delmar, 2003) as well as proactive and reactive search for stochastic data errors, as described above. A particularly important correction we found reason to make in relation to the main purpose of the study, reflecting a systematic problem we did not foresee at the design stage, was the following. It sometimes happens that a new firm is registered with zero employees in the first year, only to appear with quite a sizable number of employees in the second year. This will in the analysis appear as instances of rapid growth. But were these units really (fully) started in the first year? There in no indisputable criterion that can tell exactly what belongs to “start-up size” and what should be ascribed to “growth”. On balance, after checking the effect of the alternatives, we choose not to accept cases with zero start size, but regarded them as started instead in the first year they had employees (Davidsson & Delmar, 2000).

Other Observations The contents of this chapter have probably made it clear enough that developing a useful and credible data set takes time. You don’t start the work now if you have to submit a full length conference paper in three months. Rather, the data set you start developing now is for next year’s submission—and for years to come. Further, it has probably come across that the degree of customization of data that we have demanded requires very close collaboration with experts within the organization providing the data. We knew pretty well what we wanted, but were pretty clueless about whether and how it was possible to achieve. The latter required other expertise, and we were lucky enough to find it. I have probably also made clear that the data are not going to be inexpensive, as method textbooks often claim secondary data are. You don’t have qualified people sit in long meetings, write customized computer programs, run extensive data quality tests and even spend weekends making “manual” matching without paying for it. I don’t recall the exact figures, but the costs for the data sets discussed in this chapter were in the EUR 40-110 thousand bracket—per study. The examples I have provided may appear idiosyncratic and one can validly ask whether they are of any help for developing data sets in other countries and for other research questions. Fortunately, for those interested, there exist other descriptions of customary data sets in the entrepreneurship or small business domains (Acs &

156

RESEARCHING ENTREPRENEURSHIP

Armington, 1998; Birch, 1979; Davis et al, 1996a; Reynolds & Maki, 1990). As regards our examples, I think there is a number of factors, some of which were more or less unique to our situation, and others that can easily be copied, that facilitated the relative success of these projects. First, business statistics in Sweden are of high quality in the first place. A lot of data are collected and stored, and with relatively high updating frequency. This certainly helped a lot. Second, data integrity laws in Sweden are stringent but more allowing for academic research than for other uses, so this did not create any insurmountable obstacles. When we encountered problems of that kind, they were always soluble. Third, a number of factors facilitated the interaction with the experts at Statistics Sweden. One was that our fundamental research questions were not the most esoteric ones, but rather clear cut and of obvious societal relevance. Another reason was that we had the authority of full professors on the team. The Business Dynamics project actually went by the unofficial name “the Professors’ Project” internally at Statistics Sweden. The fact that we had major funding from another government agency (rather than some obscure—or so perceived—private source) may also have been a factor. Further, our including a data management/computer programming expert on our team was probably a smart move. It also helped our case—like in any human interaction—that we showed due respect for the other side’s expertise. Researchers prone to attribute unsatisfactory delivery to incompetence or laziness of the other party—rather than their own lacking competence or communication ability—would not have achieved the same data quality. Finally, and importantly, our projects were different and challenging. They were, in short, much more fun to work with for the data experts than was the average project. They also pointed towards possible ways for the statistics agency to develop their own products and services. And they have; partly as a result of these projects, Statistics Sweden have developed longitudinal databases that allow dynamic analyses and included them in their standard assortment. Who knows, perhaps that was a more important result of our efforts than were our empirical findings?

SUMMARY AND CONCLUSION I started this chapter with a couple of sad stories about trying to answer research questions with secondary data that were fundamentally inadequate for the purpose. The message was that if the data do not meet the minimum requirements, please refrain from doing the research at all! The main message of the chapter, however, is that you can sometimes do much more with secondary data than you first believe possible—if you invest enough time, effort and money into it. I have tried to illustrate with examples from my own projects the kind of data matching, checking and corrections that may be necessary in developing a useful and credible data set. To be honest with you, after being through the work with these projects I sometimes think it is scary what some other researchers do—and don’t do—with their secondary data sets, and I hope that in the future you will share my skepticism when reading published research where the researchers do not seem to have thorough knowledge about the characteristics and limitations of their data. Admittedly, our work with Business Dynamics in Sweden and High Growth Firms could have been better, both in terms of theory-driven research questions and

Preparing a “Secondary Data Set

157

in terms of sophisticated analysis methods. However, I hope this chapter has shown that serious use of secondary data involves a lot more than grabbing an existing data set and subjecting it to sophisticated, theory-driven analysis. If one doesn’t invest a lot of time, effort and money in the data set in the first place, the results are likely to be uninteresting or false, regardless of theoretical and (analysis) method sophistication. Therefore, while theoretical shallowness and the use of overly simple analysis methods should also be avoided I think they represent more excusable sins than the use of crap data. GIGO rules!

CHAPTER 8

SPECIAL TOPIC: JOB CREATION AS THE DEPENDENT VARIABLE

WHY CARE ABOUT JOB CREATION? Through the short history of entrepreneurship research, job creation has been one of its central themes. In fact, Birch’s (1979) results on job creation—or, you might want to argue, the real changes in the economy that produced them—were one important reason why entrepreneurship ever got big in policy making, the media, and academic research alike. This alone is good reason for an entrepreneurship researcher to care about job creation research. In addition, I argued in the early chapters of this book that entrepreneurship is not solely a micro-level phenomenon, and that entrepreneurship research can make contributions by relating micro-level change to societal level outcomes. This, too, is good reason to take an interest in job creation. At the particular time you read this text, job creation may or may not be a hot topic in entrepreneurship research. But, if not, please believe me: it will come back with the swings of the business cycle. With rising unemployment policy makers will show an interest in the issue. The specific assigned savior that is expected to be the quick fix of the problem may shift from time to time: new firm births; high growth firms; high-tech firms; minority entrepreneurship, or something else. I’ve seen them all come and go—and sometimes come back. So if there is no better reason, opportunistic researchers will often have a fair chance to fund their research if they investigate the right job creation questions at the right time. However, regardless of the more or less noble reason for engaging in job creation research, I believe we should try to do a good job. This entails getting some of our basic bearings right. For example, not only policy makers but also researchers ought to realize that from the entrepreneur’s perspective, the creation of new jobs is normally not a goal, but perhaps a hesitantly accepted consequence of realizing ones real goals. A majority of business founders have very modest growth aspirations in general (Delmar & Davidsson, 1999; Wiklund, Davidsson & Delmar, 2003) and, in particular, as concerns employment growth (Gray, 1990). Employees are costly, and reluctance to add personnel is therefore the norm also for dynamic and expansive entrepreneurs. As remarked on stage by a very colorful entrepreneur from the notoriously underemployed northern parts of Sweden: “Every time I have a new

160

RESEARCHING ENTREPRENEURSHIP

project the local politicians ask ‘How many jobs is this going to create?’ and I answer as usual ‘As few as possible!’” Running a business is no charity. What’s a new job, anyway? It could mean an array of things. When an individual terminates her job with one employer and assumes a new post with another employer, that person feels she has a “new job”. But this does not mean that her previous job ceases to exist, or that her “new job” did not exist before, albeit performed by another individual. So from the firms’ perspectives no job losses or gains are necessarily involved in such a change. As we shall see, in job creation research the “gross” number of jobs created or lost on some level of analysis (industry, region, economy-at-large) is usually the firm or establishment level net change in the number of people employed. Such an analysis sees no difference between a firm that keeps up the same employment numbers by having exactly the same people do exactly what they did last year, and another one that changes all its people and/or their work tasks, but happens to end up with exactly the same number of people. This also means that if the researcher isn’t careful there may be a far cry between the theoretical concept of “job” he uses on the one hand, and its operationalization on the other. And perhaps it is a little narrow-sighted to just count the numbers of jobs, assuming that more is always better in this regard. I mean, if more work is what we want we can follow the old example of destroying the terrible machines that make us redundant, and get rid of the division of labor that has made us efficient. We would certainly be kept busy if we had to start looking for some iron ore every time we needed a safety pin. Yeah, one should perhaps think a little about the quality of jobs, too, and the value of leisure. For these and many other reasons we ought to be a little more thoughtful when we use job creation as the dependent variable in our research. In the following, I will discuss an array of traps and issues that one should keep in mind and deal with when doing—or reading—research of this kind. The contents of the chapter build heavily on earlier works (Davidsson, 1996; Davidsson et al, 1998b). As my insights into this issue build mainly on my work with large, secondary data sets there will be some overlap with issues discussed in the previous chapter.

DATA COVERAGE Many data sets suffer from incomplete coverage. From an entrepreneurship research point of view, it is particularly disturbing that coverage tends to be worst for the smallest and youngest firms, and for emerging industries. We have remarked already in previous chapters that it is in many countries the case that the smallest (and therefore the youngest) firms never enter any registers (Aldrich et al, 1989) and that limitations of the data may gear researchers towards confining the analysis to the manufacturing sector and to firms that have been in existence for a long time. As discussed in the first “sad story” in the previous chapter, this obviously deters any meaningful assessment of where the majority of jobs in the economy are gained and lost—as one of the three authors eventually admitted (Haltiwanger & Krizan, 1999 p. 82).

Job Creation as the Dependent Variable

161

A less obvious case of incomplete data coverage leading to a distorted view of the relative job creation prowess in different categories of firm is that when a particular cohort of firm is followed one tends to forget the jobs that are created and lost outside of that cohort. To illustrate this, consider the following example. During my sabbatical at Queensland University of Technology in 2000-2001 I did not only work on my golf handicap; I actually conducted some research as well. Together with a colleague at QUT I tracked the development of the population of 20 firms that were started as spin-offs from QUT ten years earlier, in 1991. Rudimentary archival information from the starting year exists for all of these firms. We were also able to identify nine firms that were still in operation ten years later, and to track their size development through time. The results are displayed in Table 8:1. An “X” means that the firm is no longer in operation.

The results are highly interesting. In 1991, this cohort of 20 firms employed a total of 80 people. By the end of 2000, the nine surviving firms employed a total of 153.5 people. Thus, the firms in the cohort have jointly created 73.5 new jobs through expansion (153.5-80). More interestingly, it turns out that the growth of the firms is far from evenly distributed. As few as three firms, i.e., 15 percent of the original cohort, jointly employed 75.5 people in 2000. These three firms taken together employed only 12 people during the first year in operation. Hence, their joint employment growth was 75.5-12 = 63.5 people. This means than 15 percent of

162

RESEARCHING ENTREPRENEURSHIP

the firms accounted for almost 90 percent (63.5/73.5) of all job creation subsequent to their start-up year. Well, I hope you are not entirely convinced by my analysis. The fact is that all you need in order to produce the results reported above is a pair of dice, and hence all you need in order to explain the results is a description of the entirely stochastic process that produced them. For the truth is that I did not conduct a study of QUT start-ups. What I did was a very simple simulation, which I describe below. 1. 2.

I postulated the existence of 20 firms started 10 years ago, and that each had a start size of four employees. For each firm each year, I determined their growth by first throwing one die. If it showed three or less, the firm would shrink; if it showed four or more it would grow. I then threw a pair of dice and added or subtracted their average value to/from its current size. If through this process a firm reached size zero or lower, this firm was considered “dead” (dissolved) from this point on.

From this we should learn that if there is any outcome variation at all, chance alone will always make some cases stand out from the others. The best performers need not necessarily impress us and we need not necessarily seek substantive explanations for their superior performance. Stochastic processes ascertain that a small percentage of the firms in any cohort create a large share of all jobs. The longer the time series, the more extreme will these relations be. This is important to know, because we have a tendency to look for (and make up) substantive explanations (causal patterns) even when no real pattern exists—and researchers are certainly not immune to these tendencies. Therefore, there is always reason to force oneself to consider the possibility that stochastic processes (chance) produced the results. Secondly, I fed you with quasi-quantitative talk through the neck. I proved that three firms accounted for close to 90 percent (86.4 to be precise) of the job growth of the cohort. If you apply the same computational technique to the six remaining survivors, you will find that they increased their employment by 54 people. Hence, they accounted for 54/73.5 = 73.5 percent of the total job growth of the cohort. So we have a total of 86.4+73.5 = 160 percent of the job growth accounted for! Where did I trick you? Well, we have eleven closed firms that lost a total of 44 initial jobs (-60 percent of 73.5), so the survivors have to account for more than 100 percent of the growth. Clearly, the way I presented the results was not the most informative. This also takes us back to the issue of coverage. Cohort studies saying that x percent of the firms accounted for y percent of the jobs are deceptive because what they do not tell you is that many more jobs are created outside of that cohort—by firms that were already in existence and by firms started in years t+1, t+2...t+n. If you relate the cohort elite’s contribution of jobs to the total creation of jobs in the economy, that elite will appear a lot less impressive than when you relate them only to their own cohort. In the above example, even if we make the conservative assumption that there were no QUT spin-offs that were eleven years or older, and that subsequent cohorts were exact clones of the focused one, the three heroes’ share

Job Creation as the Dependent Variable

163

of QUT spin-offs’ growth-generated jobs would be down to a mere 16.5 percent. And that is before we start considering all job creators in Brisbane that are not QUT spin-offs. It may be the case that different studies arrive at different results because of real differences between countries and time periods. However, it cannot be ruled out that differential conclusions regarding whether the main job creators are the gazelles or the mice—that is, a small minority of fast-growing firms or the large numbers of tiny, non- or slow-growing firms—are in part due to the methodological difference of following a cohort vs. the entire population.

STATIC COMPARISON VS. DYNAMIC ANALYSIS The most rudimentary analysis of where in the economy jobs are lost and gained compares the numbers of jobs that exist in some kind of categories across years. For example, if the number of jobs increases in a certain region or a certain industry it is concluded that this region or industry does well in terms of job creation. The problem with this static comparison approach is that it does not distinguish between reclassification of jobs that already existed in the economy and the creation of genuinely new ones (the issue of organic vs. acquisition-based growth, discussed in the previous chapter and below, may actually be regarded a special case of this general problem). For example, when the number of jobs in an industry increases it may be because the jobs created by new entrants and growing firms in that industry outnumber the job losses resulting from contractions and dissolutions within the same industry. Fine. But it may also reflect that existing jobs are reclassified because establishments or firms that are active in multiple industries have drifted from one dominant industry to another, causing reclassification of all jobs associated with those units, when in actual fact the change was, say, from 48 to 52 percent of the jobs. Well, no research is perfect and in this case we at least get the direction of change right. It is also reasonable to assume that as long as the level of aggregation is not too fine-grained (like narrowly defined industries in very small regions) most problems of this kind have a tendency to cancel out. In some cases, however, this type of problem leads to systematic and substantial bias. For example, this occurs if there is a systematic trend for firms in one industry to acquire or spin out establishments that by themselves have a different industry classification. As a case in point, when large manufacturers started the big outsourcing and focus-on-corebusiness trend this would easily lead to more massive losses of manufacturing jobs—and more growth of services jobs—being reported compared to the number of individuals who actually had to completely change careers or vocations as a result of these strategic changes. The same type of problem occurs for region level analysis. When firm level changes are aggregated to the regional level a region with increasing employment may appear to have such because jobs gained through entry and expansion outnumber those lost via exit and contraction. The analysis will normally not distinguish between jobs that are genuinely created in the region vs. existing jobs that are moved there, but for most purposes this would be no major issue—regions

164

RESEARCHING ENTREPRENEURSHIP

on positive and negative trajectories, respectively, would end up in the right categories. Again, however, there is a risk of systematic bias. There are periods when ownership changes lead to such bias, typically because firms based in the core acquire firms located in the periphery. If firm or company group data are aggregated to the regional level, jobs that continue to exist—or even increase in numbers—in the acquired firms in peripheral regions may now be registered as located in the acquiring firm’s region. We can now see part of the solution to the problem. When aggregating to industries or regions, it is for most purposes advantageous to aggregate from the establishment level rather than from the firm- or company group level. Establishments have their own industry classification and code for geographic location, which may differ from the main industry and the location of the headquarters of the firm that owns it. The problem of static comparison vs. dynamic analysis is most pronounced when the data are analyzed by firm or establishment (employment) size class. The change in the number of jobs in a certain firm class category is the result not only of changes within that size class, and not all changes originating in that size class will be ascribed to it. This is what Davis et al (1996a, 1996b) called “the size distribution fallacy”. Their original worry was that small firms may erroneously appear to be great job creators because large firms shrink and thus get reclassified as “small firms”. This is likely to happen in recessions and within shrinking industries. However, there is theoretical and empirical (Davidsson, 1995d) reason to believe that under normal or favorable business cycle conditions the static comparison of employment in different size classes would bias the results against small firms because growing small firms end up as large firms either due to their organic growth or because they are acquired. In fact, if all net job creation in the economy was due to the 10-49 employee size bracket, we should not expect the stock of jobs in that size class to increase one bit over time. Instead we should, of course, expect those firms to move on to larger size classes. This should be enough illustration that static comparison over time of the number of jobs in certain categories is totally inadequate for analysis of where jobs are created. In order to see what firms create jobs we need longitudinal data where the size development of individual firms or establishments can be followed over time. As I have already pointed out, it was, in fact, this insight and the non-existence at the time of longitudinal data sets that were the starting points of Birch’s (1979) seminal work.

GROSS VS. NET JOB CREATION Another issue that worried Davis et al (1996a, 1996b) was what they call the confusion between gross and net job creation. And there certainly is an issue here, although the problems I see are not exactly the same as the one they highlight. We have observed above that we can mean different things by “new job”. By “gross job creation” in the type of research discussed here we usually mean neither “the number of people who have been assigned to other work tasks than they previously had” nor “the number of one-person sets of work tasks that have been added” but

Job Creation as the Dependent Variable

165

instead the aggregate of firm or establishment level net job creation. So there is a level of analysis issue here—net on one level may be seen as gross on the next. Further, firm level fluctuations that cancel out between two measurement points do not turn up in the data. Neither does a within-firm (or within-establishment) replacement of a number of administrators with an equal number of sales people lead to any job gains or losses being recorded. This has several implications. First, total job volatility will be substantially underestimated compared with an “individuals-change-work-tasks” view on job changes. Second, inferences like “both existing and newly-created jobs are less secure at small businesses than at large businesses (...) in terms of job durability, larger employers outperform smaller ones” (Davis et al, 1996b) are strictly invalid if based on this type of data. The analysis does not capture the durability of individual sets of work tasks or employment contracts. Third, the relation between hidden and visible job changes is likely to be different for different categories of firm. For example, job turnover may be much higher in one industry than in another without this showing as differences in temporal variability of total employment size. Likewise, more internal job changes are likely to occur within large organizations than within small ones, where similar changes may instead appear visibly as transfer from one statistical entity to another. In comparison, net job creation (on a given level) is a much more straightforward concept than is gross job creation. Net job creation means that more people are active in gainful employment within the unit of analysis in question. However, this is not reason to retract to only looking at net figures. There is every indication that the gross flows that produce the net figures are informative in their own right, and that the gross job volatility has important effects per se (Davidsson et al, 1998a; Reynolds, 1994). That is, it does matter whether a region’s net job change is due to small or large underlying changes that happen to sum up to the same net figure. Another way to put this is to point out that even when the net change is zero there may be underlying gross flows that have important effects on individuals’ well-being as well as on the regional economy at large. Another problem with focusing on net figures only, is that it is possible for a category of firm to account for an impressive share of a very small total net job creation figure. When net shares are expressed as percentages without accompanying absolute numbers, this may lead to misleading conclusions. Assume, for example, an economy with a total stock of ten million jobs. In a given year, 300 000 jobs are added through entry and expansion, whereas 299 000 jobs are lost through exits and contractions. The total number of jobs thus increases by just a thousand. Further assume that your favorite category of firm accounted for a net addition of 900 jobs. That’s 90 percent. Whoopee! Perhaps not. This number of jobs represents 0.09 percent of the employment stock. There are also other complications involved in the use of net shares (Davis et al, 1996b). With slight changes in the above example we could easily arrive at negative shares and shares well over 100 percent, which are awkward ways of communicating what is going on in the economy. In addition, the same national net figure can be accounted for by several categories of firm at the same time. It may be the case, for example, that in a given year net entry (that is, entry minus exit)

166

RESEARCHING ENTREPRENEURSHIP

accounts fully for the net addition of jobs in the economy. However, this does not exclude the possibility that in that same year “net expansion of retailing firms”, “net entry of knowledge-intensive services firms in Stockholm” or even “net job change in firms with a name starting with a P” show an equally sized or even bigger surplus of new jobs. In summary, gross and net figures, as well as percentages and absolute numbers, give different versions of the story and tell different parts of it. What the researcher should strive for is maximum honesty and relevance relative to the purpose of the research, and this often involves reporting the results in several different ways.

THE REGRESSION-TO-THE-MEAN EFFECT The “regression fallacy” or “regression-to-the-mean effect” is a pitfall well-known to (but not always avoided by) researchers with a reasonably thorough statistical training. The essence is that when repeated measures are made for members of extreme categories on a scale, measurement error or random fluctuations over time can only cause changes in one direction. For example, those who answered “agree completely” on a Likert scale can, as a group, only show a lower degree of agreement in a follow up. Thus, the mean value for the extreme groups will always move back (i.e., regress) towards the grand mean. Although they were not the first to discuss this problem in the job creation context, it is undoubtedly their reasoning and empirical results concerning the regression fallacy that drew so much attention to Davis’ et al work on job creation by firm size class. The regression fallacy is an issue related to the development of existing firms, and hence it is an entrepreneurship issue only to the extent that growth is regarded an aspect of entrepreneurship (see Ch. 1 and Davidsson et al, 2002). Further, it is first and foremost an issue related to how job creation is attributed to different firm size classes, and this is therefore the perspective I take in this sub-section. Figure 8:1 illustrates the problem (cf. Davidsson, 1996). We here see three firms that fluctuate in size over time without creating any net job growth in the long run. In an analysis designed to compare job creation in different size classes we need to decide which categories the firms belong to. A common procedure has been to use one-year analysis periods and to employ base-year size, i.e. size at the beginning of the analysis year, for size classification. If we choose the first year as our analysis period all the job gains in Firm A will be ascribed to small firms. What is perhaps worse is that the same goes for Firm B, although that firm is a “large firm” for most of that analysis year (and most of the longer period as well). In addition, all job losses in Firm C will be ascribed to large firms, even though more than half of the losses took place in a firm that was already below the arbitrary cutoff for being called “small”. The important observation is that if we should at all talk about a “bias” being involved here, that bias without exception runs in favor of small firm job creation. What then about the next analysis period, when there is more decline than growth? An examination shows that the same holds true again. The job losses in Firms A and B are now ascribed solely to large firms, while no part of Firm C’s job gains is ascribed to that category. Again, the category that “benefits” from the boundary-crossing is definitely small firms. So the problem is real and there is no

Job Creation as the Dependent Variable

167

counter-balancing effect. A number of questions are raised by this analysis. For example: Is this a bias? Yes, and no. “Base-year size” is a definition of firm size. Under that definition, it is in the above example a fact that all job growth occurs in small firms and all job losses occur in large firms. So there is no bias, only a potential for misleading interpretations. Is this “bias” serious, and how should we correct for it (or, rather, what definition of firm size should be employed) ? If one suspects that using base-year size leads to misinterpretation of the results and their implications, then another definition should be used. Davis et al (1996a; 1996b) suggest current size, defined as the average size at the beginning and by the end of an analysis year, i.e. as an alternative. This size definition solves the problems we see in Figure 8:1. Firm B would always be defined as large firm while firms A and C would always be classified as small, removing the “bias”.

Figure 8:1

Hypothetical size development for three firms

And for sure, Davis et al arrived at radically different results for the smallest size class when they shifted from base-year to current size analysis, which they interpreted as evidence of serious bias caused by the regression fallacy. However, if the problem is temporary, random fluctuations like those depicted in Figure 8:1 it is solved equally well by averaging the firm’s size over the two years preceding the analysis period. To make the correction, it makes no difference if the averaging is done as or as these firms would be classified in the same manner either way. The latter measure has the advantage of not letting the

168

RESEARCHING ENTREPRENEURSHIP

change to be analyzed affect the size classification, which breaks the elementary principle that a cause must precede its ensuing effect. Baldwin & Picot (1995) tried both “current size” and the latter measure, which they call, “previous period current size”. The results show that the current size and previous period current size yield completely different results. Instead of reducing the job creation rate for the smallest size classes, shifting from base-year to previous period average size further increases that rate for the two smallest size classes. The reason for this is that it is trend growth, not temporary fluctuations that make up the lion’s share of boundarycrossing job creation in the smallest size classes (cf. Baldwin & Picot, 1995, pp 320321). The fact that trend growth is responsible for most boundary-crossing also invalidates Davis’ et al most preferred size measure, average size. This is defined as the average size of the firm during its entire existence in the data set. This size definition rests on an assumption that all firms and plants once and for all belong to one size category. To this, I can only remark that to deny the existence of any real changes in the economy is, to say the least, a high price to pay for any social scientists and an absolutely prohibitive expense for an entrepreneurship researcher.

Job Creation as the Dependent Variable

169

A more ideal correction would be to employ momentary size for size class categorization. That is, that firms and their associated job changes were distributed over size classes according to the number of employees the firm had when the job change in question occurred. In Figure 8:1, assuming that 250 employees is the arbitrary cutoff and looking at the first year, the 250th job created by Firm A would then be attributed to the small firm category, while the 251st would be attributed to large firms. Clearly, if base-year size results are interpreted from a momentary size view of firm size, there is in this example a very substantial bias involved. Therefore, if a researcher employs base-year size it is imperative that she—and her readers—understand the implications of using this definition.

170

RESEARCHING ENTREPRENEURSHIP

In Davidsson et al (1998b) we approximated momentary size in a test of the seriousness of the regression fallacy, using a data set that covered all sectors and firm size classes. The basic unit of analysis was the establishment, but for each establishment it was known whether it belonged to an SME (in this case having less than 200 employees) or a large firm (cf. Chapter 7). We did not go through the painstaking effort of checking how much of every individual boundary crossing size change occurred before and after the crossing, respectively. Instead we approximated momentary size analysis by always attributing half of the changes in such units to the original category, and the other half to the end category. The results are displayed in Table 8:2. In short, these results show that there is some “regression-to-the-mean bias” but it is small; correcting for it amounts to correcting for fractions of percentages and is far from qualitatively affecting the conclusions. In Davidsson et al (1998b) we argued that the primary reasons for the relatively small regression effect are i) that only part of total job changes represents “temporary fluctuations” and ii) that only a fraction of total job changes takes place in firms that are close to the size boundary that arbitrarily divides the business population into small and large firms. We further specified situations when boundary-crossing may cause greater problems than in this analysis, namely: when more than two size classes are used if, in a two group analysis, the cutoff is set at a smaller size than the one we used (200 employees) if the pool of medium-sized firms is relatively large if the analysis periods are longer, and if the general level of dynamism in the economy is higher However, under no circumstances is the regression “bias” likely to have the dominant effect on the results Davis et al (1996b) assumed it would have. Taken together, we can conclude from the evidence reviewed in this section that a) the “regression fallacy” systematically biases job creation estimates in favor of small firms relative to a momentary size definition of firms size; b) however, the effect is small and has not led researchers and policy makers to mistakenly believe that small firms are overrepresented as job creators, and c) the alternative size definitions advocated by Davis et al (1996a, 1996b) are actually more likely to lead to a distorted view of what goes on in the economy, than is the use of base-year employment size.

ORGANIC VS. ACQUISITION GROWTH, AND JOB CREATION VS. ECONOMIC DEVELOPMENT We noted in the preceding chapter that it is important for many purposes to keep apart the firm level expansion that results from organic growth from expansion resulting from acquisition. I argued with some emphasis that it is reasonable to assume that organic growth reflects genuine job creation to a greater extent than does acquisitions, which from a job creation point of view is mere relocation of jobs

Job Creation as the Dependent Variable

171

from one organization to another. When one firm acquires another one the first firm gets bigger, but in terms of total number of jobs the very point is often to reduce the number of people needed to carry out the firms’ operations, not to increase that number. With this reasoning, however, we still largely delimit the analysis to internal job creation by firms of different categories. But firms in different categories do not compete with one another for the title as job-creation champions. Different firms have among them and between them all sorts of relationships ranging from unidirectional dependence to symbiosis to competition to being (almost) unrelated (cf. Aldrich & Wiedenmayer, 1993). Firms do not aim at maximizing or minimizing the number of people employed, neither in their own firms nor in the economy at large. But while pursuing their real goals, their actions create (potential for) jobs— and/or enjoyable leisure—somewhere in the economy. In this perspective, chasing the truth about what category creates most jobs internally is narrow-sighted. It is also narrow-sighted in the sense that it easily leads one to conceive of contractions and firm deaths as something (always) “bad”. They are not. They are natural, necessary and efficiency-enhancing components of the workings of a market economy. Let’s assume, for example, that a category of firm is overrepresented as product innovators. They are successful on the market and they therefore experience employment growth. This is job expansion directly caused by innovation by that category of firm. It now so happens that for the industrial firms that buy their products, the “products” are part of a process innovation that dramatically increases the efficiency of those industrial firms. They therefore contract in terms of employment, and people are laid off. This is job contraction indirectly caused by innovation by the first category of firm. Those who remain in employment, however, get their fair share of the productivity gains and thus increase their purchasing power. This creates room for entry of new firms that provide, say, personal services that the employees in the industrial firms could not afford earlier on. And so forth. This is basic; perhaps even trivial. We all know it, but we do not seem able to keep our insights alive all of the time, and enter into overly simplified discussions of job creation in this vs. that category of firm.

A FEW MORE DETAILS TO CONSIDER Before we close the books on job creation research there are a few more issues that deserve at least brief mention: Choice of micro level unit: establishment vs. company vs. company group. In many data sets it is not possible to identify firms (legal units), only establishments (plants; work-places). This is a severe limitation and the implication is clear: inferences about job creation in different firm size classes (or other categories of firm) should not be made. So when a theory about firms and categories of firms guides the research, establishment level data yield distorted results. More specifically, such data exaggerate gross job volatility relative to company level data because within-firm shift of employment between establishments is included in the

172

RESEARCHING ENTREPRENEURSHIP

establishment but not the company level analysis. As we have seen already, this may, however, be a desirable feature when the theory is about industries or regions because when establishment rather than company data are aggregated to these levels relatively more to the total gross changes are captured and assigned to the right entity (industry or region code). Artificial changes. As discussed already in the previous chapter, ownership changes (including spin-offs and mergers), geographic moves, and change of legal form of the firm frequently lead to the registration of firm (and sometimes establishment) births and deaths when no genuinely new business activity has been set up or no old one has really ceased to exist. The effect is a general overestimation of gross job volatility in the economy, compared with a definition that assumes some “genuine” change. This is a problem that is more pronounced among small units compared to large ones, leading one to suspect general overestimation of job volatility among small units (and therefore within industries or regions dominated by such). However, it may well be those fewer cases where some unusual form of restructuring leads to the appearance of entirely new large units that generate the greatest—albeit detectable—errors in data for individual regions or industries in particular years. The occurrence of artificial changes may also be unevenly distributed across industries—consider for example, the special characteristics of work places in the construction industry, or the rather frequent ownership transfers that signify certain other industries. In addition, the occurrence of artificial changes may alter over time, making the researcher curse the increasing quality of their data! Length of analysis period. The length of the analysis period is critical and must be considered when, e.g., comparing results from different studies. If two-year rather than one-year periods are chosen, some firms may enter and exit within that period and therefore never be recorded. This was also mentioned in Chapter 7. At least for firm-level analysis this will lead to underestimation of job volatility among small firms, because small firms are more likely than large ones to both enter and exit within a short period of time. Possibly (but not necessarily), however, the use of longer analysis periods leads to relatively more regression fallacy problems (cf. above). The length of the analysis period also affects what part of job gains is attributed to births versus expansion, as a new firm’s second year expansion will be counted as “birth jobs” in one analysis and “expansion jobs” in the other. Assigning size class to new entrants. If base-year size definition is used, new entrants cause a special problem. These firms do not have a base-year size. For this as well as other reasons job changes associated with births and deaths should be separated from those associated with expansion and contraction among continuing firms. New entrants are typically small. Mixing them with the smallest size class of established firms may lead to the image that “small firms grow a lot” when the truth is that established firms in that category have an aggregate decline in employment. In fact, failure to make this distinction should perhaps carry the blame for much policy and research interest being directed to small firms, when it really should have been directed to the creation of new ones, i.e., to the importance of entrepreneurship.

Job Creation as the Dependent Variable

173

SUMMARY AND CONCLUSIONS Do all the methodological problems discussed above mean that any research effort to gain insight into the mechanisms of job creation is inherently futile? I would say no; in research we will always have to live with limitations and have to make assumptions and simplifications. If the alternative is to make unfounded guesses, less-than-ideal research always has a place. The important thing is that the researcher is aware of the limitations, does her best to handle them, and communicates her result with a level of confidence that accords with the qualities of the data. Do we need more research on job creation? In order to establish that small (and—in particular—new) firms have been overrepresented as job creators during the last few decades we need no more research. Few findings in the domain of social science have as solid empirical support as that. To follow the development over time should in the future rather be a task for statistical organizations. There are, however, many other reasons—some of which were stated in the introduction to this chapter— why an entrepreneurship researcher should show an interest in job creation. When we do that in the future, I think the issues discussed in this chapter suggest that we:

1. 2.

3. 4. 5.

6.

7.

Make sure the data are longitudinal and of high enough quality to make the effort worthwhile in the first place. Clarify to ourselves and readers what “new job” (and job losses) actually means on the basis of the data at hand, and that we use theory and make comparisons with other research in accordance with this notion of “new job”. Separate in the analysis the job changes that are attributable to births, deaths, expansions and contractions, respectively. Likewise separate job changes attributable to organic changes from those resulting from mergers, acquisitions and splits. Apply a size definition that comes as close as possible to momentary size. Base-year size is a defensible alternative as long as relatively short analysis periods are used and the number of size classes is relatively small. If the data permit, a correction like Davidsson’s et al (1998b) should supplement the analysis. Express the results on job changes as gross and net absolute figures and shares, and relate these to corresponding figures for the employment base. Rates are a more dubious matter when analyzing total job creation in the economy over a number of one-year analysis periods. Rates are something we want to compute for cohorts of firms, while categories like size classes, industries and regions are genuinely moving targets in the sense that they continuously change their members. However, studies of growth rates for stable cohorts of firms in different initial size classes would certainly complement the other type of study and therefore add to our understanding. Consider not only the numbers but also the quality of new jobs, and use other outcome measures alongside with job creation. We shouldn’t

174

RESEARCHING ENTREPRENEURSHIP forget that we do not live to work, and that when we do work it is nice if the work comes with both intrinsic and extrinsic rewards!

Again, doing perfect research is not possible, so it is unlikely that we can deal fully satisfactorily with all the above points at once. But thanks to forerunners and their shortcomings, it should now be possible to deal more satisfactorily with them than has been the case in earlier job creation research. If we can not, I seriously think we should not conduct the research at all. It would just be unnecessary work. More work does not always equal more well being, neither economic nor other, and neither on the individual nor the societal level.

CHAPTER 9

SPECIAL TOPIC: THE POWER OF REPLICATION

SAMPLING AND SIGNIFICANCE TESTING REVISITED In Chapter 5, I argued that there is no way we can sample probabilistically directly from the theoretically relevant population. This is because that population does not exist empirically in one place at one time. As a corollary I emphasized that statistical significance testing is not the ideal teller of truth that we would like it to be. Many researchers act as if statistical significance were definitive proof of an effect of a certain size in the theoretically relevant population, and as if lack of significance definitely proved the absence of such an effect. The fact of the matter, however, is that the outcome out a significance test is contingent also upon a range of factors other than the effect size in the population, e.g., a) the chosen risk level, b) the size of the sample, c) random sampling error—how much the effect in the sample deviates from the true effect in the population, d) model specification—the focal variable’s co-variation with variables included in and excluded from the analysis, e) the variance of the variables concerned, f) the measurement quality, and g) probably a few more things I have forgotten right now. To make matters worse, many researchers nurture severe misconceptions about what a significant result really means. For example, many believe that “significant on the 5% risk level” means that the result has a 95% chance of being replicated (Oakes, 1986). This is gross exaggeration of the power of significance testing and perhaps an explanation why business researchers underemphasize the importance of actually carrying out replications (Hubbard, Vetter & Little, 1998). The truth is that if the associated probability is exactly 0.05 and we happened to be right on target regarding effect size (i.e., the effect in the original sample exactly matches the population parameter) then the chance of successful replication with an equally sized sample is a mere fifty-fifty. Therefore, the real acid test of theory is that the theoretically postulated effect is demonstrated again and again in empirical samples drawn from theoretically relevant populations. What we can do, then, is to see to it that our empirical cases are drawn from a theoretically relevant population. Within that sample we want our

176

RESEARCHING ENTREPRENEURSHIP

theory to hold up in the form of statistically significant results, but we have to realize that sometimes true predictions are not borne out due to random sampling error or insufficient statistical power (Cohen, 1988). In other cases statistically significant results—theoretically predicted or not—appear in the sample although they are not true for the population. Therefore, the real power of a theory is demonstrated through replication. If a theory is any good, it will show its effects in several, slightly different empirical samples, and also be robust against (valid) variations in operationalizations. Hopefully, medical doctors don’t treat patients a certain way just because a substance was ascribed a statistically significant effect in a small, single study—and hopefully they don’t stop giving treatment that works just because one study was too small for the theoretically predicted effect to reach statistical significance. Similarly, why should we tell students and business practitioners “truths” that have not been shown to be replicable? If they do not stand the test of academic replication our recipes are unlikely to work in these recipients business practice, either. In the remainder of this chapter I will use examples from my own work to demonstrate how various forms or replication can boost one’s justified confidence in a theoretically proposed relationship.

REPLICATING OTHERS What first comes to mind when you think about replication is to copy what somebody else has done in already published work in order to either confirm or question the findings. This is what my first two examples are about. However, my examples are not pure replications because the studies were not designed with that main purpose and hence the operationalizations differ. This decreases the value of the replication as test of internal validity while it actually increases its value as a test of external validity (Hubbard et al, 1998). That is, the theory test is a tougher one because if the support for the theory in the original study was to some part due to an artifact of the specific operationalizations this “benefit” does not carry over to the replication. My first example was briefly described in Chapter 3 already, when we discussed theory. This example was also part of my dissertation study (Davidsson, 1989a) and my very first conference presentation on the other side of the Atlantic (Davidsson, 1988). The starting point for this research was that I realized when reading Smith’s (1967) then oft-cited study that I had data in my study to test many of his propositions about types of entrepreneurs and the firms they create. More specifically, Smith made claims that Opportunistic Entrepreneurs—relative to Craftsman Entrepreneurs—had the following characteristics, for which I also had data available (cf. Davidsson, 1989a, pp. 143-144 and 159-160): 1. 2. 3.

They are more likely to have run (an)other firm(s) prior to the sampled one. They are more likely to currently run more than one firm. They are more likely to have management experience prior to starting their own firm.

The Power of Replication 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.

177

They are less likely to have considerable experience from the specific industry prior to becoming CEO of the sampled firm. They have higher average level of general education. They have higher average level of business education. They have more internal Locus-of-Control. They have higher Need for Achievement. They have more self-confidence. They are less concerned that their firm might become overly dependent on a small number of customers, suppliers or investors. Personal control and surveillance of the firm’s activities are relatively less important to them. Ownership control is relatively less important to them. They find recruiting easier. They are less likely to have their spouse employed in the firm. They have more positive attitudes towards growth.

Further, Smith (1967) postulated that Opportunistic Entrepreneurs create Adaptive Firms, implying the following testable (in my study) characteristics relative to the Rigid Firms that Craftsman Entrepreneurs create. 16. 17. 18. 19. 20. 21. 22.

The firm is less likely to be wholly owned by the respondent. A smaller share of the firm’s sales is generated within the home county. The share of the firm’s sales that is generated on export markets is more likely to be above industry average. The share of the firm’s sales that is generated by products developed “in house” is higher. They are more likely to be involved in product development currently. The firm has higher historical growth rate. The growth aspirations for the future are also higher.

This list of characteristics can be regarded a complex hypothesis saying that if we divided the sample into the two most homogeneous groups we can find, these two groups would be split in accordance with the 22 statements above. So how do we find the two most homogeneous subgroups? With cluster analysis, of course. And the results were largely supportive. It turned out that in a two-group hierarchical cluster analysis using Ward’s method 20 of the 22 differences were in the predicted direction. For all the firm variables the differences were substantial, which was also the case for the educational and psychological characteristics of the individuals. The differences were in the right direction but rather unimpressive for habitual entrepreneurship and the experience variables. The two instances of differences in the “wrong” direction—ownership and supervisory control—were also very small and best regarded as indicating “no difference”. Smith derived his taxonomy from in-depth study of a relatively small (52 cases), all-male, all-manufacturing sample in Michigan in the 1960s. Nonetheless, my test on a much larger mixed-gender, mixed-industry sample in Sweden 20 years later

178

RESEARCHING ENTREPRENEURSHIP

suggested his taxonomy was a meaningful way to distinguish conceptually and empirically between groups of business owner-managers. My conclusion was that “Despite temporal, cultural, sampling and operationalization differences, the groups that emerge in a cluster analysis of this new sample show considerable resemblance to the entrepreneurial groups suggested by Smith (1967). This result provides fairly strong support for the usefulness of his typology” [sic] (Davidsson, 1989a, p. 155). This example illustrates the mutually beneficial nature of replications. Smith’s theory gains credibility and generalizability by being replicated in a different empirical context. My study becomes a much more meaningful contribution by being framed as a test of Smiths existing taxonomy rather than as a stand-alone, inductive attempt to find distinct subgroups among business founders. However, perhaps Smith’s two-group delineation is a bit too catchy. My own analysis showed there was no bimodality in the data, suggesting our delineation into types represents a convenient but arbitrary cutoff in a continuous distribution. That is, the two types are not empirically distinct. Moreover, at the very same conference Woo, Cooper & Dunkelberg (1988, 2000) presented the first version of an article that came to drastically reduce the popularity of taxonomies and typologies in entrepreneurship. Using a clever but different form of replication they showed that two groups interpretable as “Craftsmen” vs. “Opportunists” tended to emerge regardless of which variable sub-set were used. However, although the different analyses were done on the same sample, the overlap between, for example, the different “Opportunist” groups that were distilled was not impressive. As a consequence the authors justifiably concluded that their study “challenges the conclusion that we [i.e., the entrepreneurship research community] have succeeded in classifying the population of entrepreneurs into two robust categories with by and large identical attributes across samples.” (Woo et al, 1988, p. 173). As we can see, replication enhances knowledge development—either by supporting or questioning established “truths”. Incidentally, my second example of “replicating others” also involves work by Arnold Cooper and Carolyn Woo. Based on human capital theory as well as previous empirical research on entrepreneurial performance, Cooper, GimenoGascon & Woo (1994) derived ten hypotheses about how initial conditions influence new venture performance. The hypotheses predicted effects of four broad categories of initial capital: general human capital, management know-how, industry-specific know-how and financial capital. They tested their hypotheses on a large, longitudinal data set representing new ventures across all US industries and regions in the mid 1980s. They got very limited support for the effect of management knowhow; otherwise the results largely confirmed their hypotheses. We realized that with The 1994 Start-up Cohort Study we had access to a similarly composed but even larger, Swedish sample from the mid 1990s. Although we did not always have access to the exact same operationalizations we did have indicators for all four groups of predictors that Cooper et al (1994) used, and we could also add a fifth category: access to markets and resources. We were also able to follow their use of three outcome categories (failure; marginal survival, and growth) although we used “high performance” rather than “growth” for the best performing group. As Cooper et al (1994) did not get support for all their hypotheses

The Power of Replication

179

we could use either their theorizing or their results as the basis for our own hypotheses. We choose the former. Our results revealed the following as regards similarities (cf. Dahlqvist et al, 2000). In both studies, indicators of general human capital contributed positively to marginal survival and high performance. In both studies, indicators of management know-how contributed positively to marginal survival. In both studies, indicators of financial capital contributed positively to high performance. In both studies, ventures in retailing and personal services had lower probabilities of marginal survival and high performance. The following differences between the studies stand out relatively clearly: While our model was much stronger in predicting high performance than marginal survival, their results appear more balanced in this regard. While they found effects of industry-specific knowledge on both survival and high growth, our analysis confirmed neither of these effects. Our study also lacked the following effects obtained by Cooper et al (1994): a) a positive effect of management know-how on high performance, and b) a positive effect of financial capital on marginal survival. Again, it may be argued that the replication adds value to both studies. For example, without our follow-up readers of the Cooper et al article can choose rather freely to interpret lack of a specific effect as a real lack of such an effect or as a shortcoming of their operationalizations. It becomes more difficult to argue that way when our results—based on the same theoretical logic but using other operationalizations—point in the same direction. Likewise, both studies arrive in some instances at the same rather subtle differential influence on marginal survival and high performance, respectively. This is the case with the gender effect, where according to both studies ventures run by women show a lower probability of high performance, but not a higher probability of failure. Further, Cooper et al (1994) found that presence of a parental role model (vicarious learning) increases the probability of marginal survival but not of high performance. In a similar fashion we found that previous start-up experience (experiential learning) is positively associated with survival but not with high performance. When such patterns are repeated in more than one, large-scale study they achieve a much higher level of credibility than when they are afforded a couple of asterisks in a single study.

180

RESEARCHING ENTREPRENEURSHIP REPLICATING ONE ANOTHER: HARMONIZED RESEARCH COLLABORATION

Another type of replication—whether or not it is thought of and presented as such— is when researchers collaborate on conducting several parallel studies aimed at addressing the same research questions. One of my most gratifying professional experiences ever was when as a young researcher I got involved in a seven country international collaborative project on the regional determinants of firm start-up rates, under the competent leadership of Paul Reynolds and David Storey. Sure, we had our differences within the group—and cats from some countries turned out to be less easily herded than others. In addition, data limitations in some countries—similar to those discussed in Chapter 7—led to certain frustrations. Overall, however, our meetings were joyful and rewarding because everybody was working on the same problem and therefore up to speed with the relevant theoretical and methodological issues that we had to deal with. Thus, unlike some sessions at broader meetings the discussion could start on a very high level—and climb from there. In short, we collaborated in the design phase in order to harmonize the data collection (or, rather, compilation of secondary data) as far as possible. In many cases it proved impossible to get exactly the same indicators, but then we tried to assure that each country study included at least some indicator(s) of the following regional characteristics: demand growth, urbanization/agglomeration; unemployment; personal/household wealth; small firms/economic specialization; political ethos and government spending/policies. These characteristics were related to subsequent firm start-up rates in manufacturing only as well as across all sectors, and relative to the size of the firm population as well as relative to the size of the workforce. The country teams could conduct and present whatever analyses they liked, wherever they liked, but the main country reports as well as harmonized comparative analyses were published jointly in a special issue of Regional Studies (Vol. 28; No. 4, 1994).

Those who care to read the individual country reports may get a rather confused picture of what determines the regional rate of business start-ups. Different studies used different indicators and did not necessarily present them as indicators of higher

The Power of Replication

181

order theoretical constructs (cf. Chapter 3). The French study (Guesnier, 1994) reported unstandardized regression coefficients and positive effects of unemployment rate small firm density. The German study (Audretsch & Fritsch, 1994) reported standardized regression coefficients and negative effects of seemingly the same variables. The Irish study (Hart & Gudgin, 1994) included only the manufacturing sector. The Italian (Garofoli, 1994) and Swedish (Davidsson et al, 1994b) studies reported results for start-ups relative to the size of the workforce only and stubbornly refused to relate the number of start-ups also to the number of organizations in the economy. The British study (Keeble & Walker, 1994) included predictive models also for small firm growth and death, which no other study matched. The US study (Reynolds, 1994) included 15 times as many regions as did the Irish study, so effects deemed “statistically significant” for the US may not be so in Ireland, even if the estimated magnitudes were exactly the same. And so on. However, the harmonized analysis in the final article (Reynolds et al, 1994) nevertheless arrived at powerful, generalizable conclusions. Once the analyses were harmonized and the proper level of abstraction applied, the authors were able to convert the above mess to the summary in Table 9:1. In verbal terms they summarized the findings as follows: Analysis of the processes associated with new firm births across seven advanced market economies…indicates three processes having a positive impact on firm birth rates: growth in demand, indicated by population growth and growth in income a population of business organizations dominated by small firms a dense, urbanized context, reflecting the advantages of agglomeration (...) Other processes—related to unemployment, personal wealth, liberal political climate or government actions—had weak or mixed impact. (Reynolds et al, 1994, p. 453)

Few conclusions about entrepreneurship—or, indeed, in social science—have as solid empirical backing as these. It would not be possible to achieve the same in a single study, no matter how comprehensive and well designed. Replication rules! Convinced yet? Ready for another one? My second case of “replicating one another” exemplifies temporal rather than spatial replication. As part of my doctoral dissertation project I developed a package of questions concerning small firm owner-managers’ expected consequences of growth. Again, I saw growth as an entrepreneurship issue at the time; subsequently I have refined that view (cf. Davidsson et al, 2002 and Chapter 1 of this book). I used my pilot study and the extant literature to find aspects of small firm owner-managers’ work environment that were important for their “job satisfaction”, and which could be suspected to be improved or worsened as a consequence of expansion. I came up with the following dimensions (Davidsson, 1989b): 1. 2. 3.

Workload—would the owner-manager have to work more or less if the firm were twice as big? Work tasks—would the owner-manager get to spend a smaller or larger share of his/her time doing the most preferred work tasks? Employee well-being—would the firm be a better or worse place of work in the eyes of the employees?

182 4. 5. 6. 7. 8.

RESEARCHING ENTREPRENEURSHIP Personal income—would the owner-manager make more or less money, were the firm twice as big? Control—would it be easier or more difficult for the owner-manager to survey and control the operations of the firm? Independence—would the owner-manager enjoy a greater or lesser feeling of independence relative to important external stakeholders? Vulnerability—would it be easier or more difficult for the larger firm to survive a severe crisis? Product/service quality—would it be easier or more difficult to keep a high quality in the firm’s products and/or services?

Albeit a side issue in all three projects, the same package of questions were reused in Frédéric Delmar’s and Johan Wiklund’s respective dissertation projects in the mid to late 1990s (Delmar, 1996; Wiklund, 1998). In Wiklund et al (2003) we finally wrapped up the three studies and related the owner-managers’ expected consequences of growth to their overall growth attitude. That is, we tried to answer the question “What specific expectations determine small business owner-managers’ general positive-negative inclination towards growing their firms?” Tables 9:2 and 9:3 summarize the results.

From Table 9:2 we learn that in each study expectations concerning the effect of growth on Employee well-being comes out as the most important determinant of growth attitude. Had this result appeared in a single study sceptics could have regarded it a peculiarity of little consequence. When replicated in three studies the suggestion that this non-financial concern may be more important than financial ones (Personal income) in determining overall growth attitude has to be taken seriously. From Table 9:2 we also learn an important lesson regarding “relative importance” of not-very-strong predictors. Other than the dominance of Employee well-being the estimated relative importance of different expectations appears to be extremely sensitive to random sampling error or other stochastic factors. In a single

The Power of Replication

183

study researchers and readers alike are often lured to over interpret small differences like these. Moreover, we here learn important lessons about statistical significance. First, we can clearly see the non-replicability of results that are significant on the 5% risk level in fairly sized samples. Take, for example, the statistically significant and “second most important” effect of Workload in the 1986 sample, or the likewise statistically significant and “third most important” effect of Vulnerability in the 1994 sample. Neither of these comes out significant on the 5% level in the other two studies. However, the lesson in this case is not that we should disregard these effects. Instead, the results across studies illuminate how stupid it is to regard lack of significant result as proof of non-existence of an effect’, i.e., as firm evidence against the theory. In the analyses in Table 9:2 it is true that we have in each study statistically significant support for only three or four out of eight proposed relationships. We would be in serious error all the same if we concluded from either of the first two studies that “Small firm owner-managers willingness to grow their firms has nothing to do with the expected effects of growth on their personal income.” As we can see from the last column of the table the likelihood of finding positive effects this large or larger in three separate studies of this size is actually less that seven in a million. Quite impressive for a “non-existing” relationship, wouldn’t you say? See now that we need replications? The fact is that all the results in Table 9:2 support the theory. Every single one of the 24 coefficients is positive as predicted (including the miniscule effect of Work tasks in the 1996 study). Many of the effects are small, but no less real.

In Table 9:3 the data from the three studies have been pooled in order to make possible breakdown analyses on large enough sub-samples. This is another form of replication of results that bridge over to the next section about “Replicating Yourself”. After seeing these results our confidence in the importance of employee well-being concerns should be further enhanced. Moreover, we can say with some

184

RESEARCHING ENTREPRENEURSHIP

confidence that no important industry differences seem to exist regarding the influence of expected consequences of growth on overall growth attitude, other than perhaps a somewhat lesser suitability of the model for explaining variance in the retailing industry (lower R2). With the smaller sub-samples we would typically have in a single study we would probably have had stochastic industry differences that we were lured to over interpret. We do not see in the table any dramatic sub-sample differences for size or age, either. However, given the rather sizable sub-samples we should perhaps dare to conclude that the issues of Control and Independence come to the fore only as the firms have already grown out of the smallest size class. Conversely, concerns about Workload and Work tasks do not seem to be a factor at all for the “largest” firms, which explains the modest overall results for these dimensions. I hope the above examples have demonstrated that replicating one another and publishing the results jointly is not a bad idea. Those who hunt for maximum numbers of publications may think otherwise, but if you ask me the above are two examples that are much more worthy of being published at all (and read, and cited) than are single studies of the same kind. Seriously, what do we need them for, all the published single studies that nobody reads, cites or replicates? For getting tenure? It’s really sad—I thought we were in this business for something more than just making a living for ourselves…

REPLICATING YOURSELF Regrettably, it is difficult to get a straight replication of somebody else’s work published in a highly ranked journal. If you ask me this shouldn’t be the case—the outlets that published the original work should also make room for succinct replication manuscripts that support or question it. But such is life, and we are unlikely to see any dramatic change soon. Also regrettably, you don’t always have friends who want to play your game. So what to do, when editors don’t appreciate straight replications as they should, when you can’t get colleagues to conduct parallel studies, and this annoying Davidsson guy claims single studies aren’t worth that much, anyway? Replicate yourself! That is, make your study large enough so that you can prove your theses on several, separate sub-samples. The following examples are intended to show how this can dramatically increase the value and credibility of stand-alone studies. My first example originates from the most influential and cited of my nonpublished studies, namely a paper on determinants of entrepreneurial intentions from the mid 1990s (Davidsson, 1995b). Why non-published, you would ask? Well, it is not much of a shock for a Scandinavian or other Europeans that ten years ago—and still today—some perfectly publishable work is never sent to journals. It happened/happens all the time and has much more to do with the incentive structure than with the quality of work. Well, at any rate, in this specific case one factor was that the paper was “just” a by-product of my Culture and Entrepreneurship study (Davidsson, 1995a; Davidsson & Wiklund, 1997). For that project I had to sample attitude data from representative samples of the adult population in several regions. I realized that the data were useful for individual level analysis as well, so I read the

The Power of Replication

185

literature on determinants of entrepreneurial intentions, compiled a conceptual model, performed some analysis, and wrote a conference paper on the topic. A journal editor caught sight of it and offered to publish it more or less “as is”, but I wanted to strengthen the theory side of the paper and redo the analysis with structural equations modelling techniques. And then, predictably, I got sucked into other things and never got down to it. Well, anyway, writing this chapter gives me a new chance to share some of the results with a perhaps slightly broader audience. According to my model the most direct influences on a people’s intentions to start their own firms should be a) their degree of conviction that this is a suitable career choice for them and b) their current employment status, with those in permanent employment being less inclined to strike out on their own any time soon. As indirect antecedents I included certain general attitudes as well as some domain-specific beliefs. As the most distal influences I modelled certain personal background factors like gender and access to role models. Thus, the regressions in Table 9:4 represent a quick and dirty way of exploring the proposed relationships, as entrepreneurial intention is here regressed directly on all proximal as well as distal antecedents. Further, they display only the antecedents that turned out significant in the full sample analysis—a few additional ones were included in the conceptual model.

Again, the results tell us a great deal about what kind of results are and which are not replicable in a “normally” sized study. The overall explanatory power is fairly stable, as is the strong effect of Conviction. The positive effect of Know-how

186

RESEARCHING ENTREPRENEURSHIP

and the negative effect of Lacks role model appear consistently in every sub-sample but meet conventional statistical significance criteria only in two cases out of six. The other variables, while significant in the full sample analysis, show an even more erratic pattern, including weak effects in the wrong direction is some cases. A more normal situation would be to have only one of these sub-samples as the basis for one’s research contribution. It is telling and somewhat scary to see how erratic small effects in stand-alone studies really are. Apart from the effect of Conviction it is clear that we would highlight different factors—and devote discussion section space to completely different implications—depending on which of the above samples we happened to capture. For example, had we worked with sample 2 or 3 we would have discussed the gender effect; had we instead analyzed sample 4 or 6 we would perhaps instead have made a number out of its absence. Based on sample I we might have argued for influencing people’s attitudes: they will become more willing to start firms if they become more change-oriented in general and understand that new firms make valuable contributions to the economy at large. Based on sample 5 we would instead have argued forcefully that people need practical knowledge about how to obtain the resources (etc.) needed get a firm up and running. Scary. How worthwhile are really our elaborate discussions of practical implications derived from normally sized single studies? The full sample analysis shows that all the included antecedents have some effect, but also that these effects are small and somewhat uncertain, and—again—that statistical significance has a great deal to do with the size of the sample. Some non-believers might argue that given that we actually have access to the large, full sample, performing the sub-sample analysis—the “replications—in this case actually makes the results weaker. 1 cannot concur. The point is that we learn more when we see also the shaky results of the sub-sample analysis than when we refrain from doing so. If we want everything to fit perfectly and logically all the time we probably shouldn’t engage in empirical research at all. I take the final example of “Replicating Yourself” from the other end of my publication spectrum, namely from our piece in the Strategic Management Journal on operationalizing Howard Stevenson’s conceptualization of entrepreneurship (Brown et al, 2001). You may recall this piece of research from our discussion of operationalizations in Chapter 6. The rationale behind the research, which was originally Terrence Brown’s idea, was that despite its popularity Stevenson’s conceptualization (Stevenson, 1984; Stevenson & Gumpert, 1991; Stevenson & Jarillo, 1990) had never been systematically tested. In order to make it testable someone had to first operationalize his dimensions of entrepreneurial management. Terrence and I took on the challenge when designing the project Entrepreneurship in Different Organizational Contexts. Johan Wiklund joined us as co-author at a later stage and did a great job not least on validation issues, which relates to our current emphasis. Table 9:5 displays our (i.e., Brown ’s et al, 2001) main results. We were pleased that it turned out possible after some addition and deletion of items to arrive at a factor analysis solution where different dimensions that accorded with Stevenson’s explicit and implicit dimensions of entrepreneurial management came out quite clearly.

The Power of Replication

187

In most cases the corresponding computed indices also show a satisfactory degree of internal consistency. To be honest we were also somewhat surprised; as all of the dimensions aim at capturing some aspect of organizational entrepreneurship they should not necessarily come out this clearly in an orthogonally rotated factor analysis. But they did. So far so good. However, were our results stable or had we just been lucky and/or used stochastic variation to the max when we dropped and added items until we arrived at this clean factor structure? Again, we could use the fact that we had a large and stratified sample. As internal replication we re-ran the factor analysis for different strata for a total of ten sub-sample analyses. Displaying factor loadings from ten separate analyses is a bit over the limit, so in Table 9:6 I have summarized the essence of the results.

188

RESEARCHING ENTREPRENEURSHIP

By and large the results of the sub-sample analyses are very encouraging. The cumulative variance explained is very similar in every analysis; the right number of factors is extracted in all analyses but one, and—importantly—the extracted factors remain the same across sub-samples. There are some problems with the Resource Orientation factor but otherwise our operationalization appears successful at least from a technical point of view. With the internal replication on different groups of firms I would argue we have made a much, much stronger case for the validity of this operationalization than if we had only had access to, say, the 399 independent firms, or the 372 manufacturers.

SUMMARY AND CONCLUSION In this chapter I have argued for more replication and tried to support that view with examples from actual empirical research. I have so much praised the virtue of replication that I will not burden the reader with much more canonizing of if here. Suffice it to repeat that replication provides us with much better truth criteria than other tools at our disposal. Replication therefore facilitates the building of cumulative knowledge, which is what research is all about (if you ask me). Also importantly, replication has a sound, humbling effect that may make us less prone to over interpret single study results regarding relative importance of explanatory variables; prematurely disregard antecedents that do not turn out significant in an individual study, or show an undue level of confidence in a result that happens to be (marginally) statistically significant in one study.

CHAPTER 10

A QUICK LOOK AT ANALYSIS METHOD

LET’S MAKE THIS A SHORT ONE I’m close to exceeding the publisher’s maximum number of pages here, so let’s keep this brief. Ever heard a worse excuse? OK, I want this over with and, I confess: I have not cultivated my expertise in analysis methods as much as I should in recent years. But collaborators and doctoral students around me have! With some help of their work I will in this short, concluding chapter discuss some analysis method implications of the perspectives on the entrepreneurship phenomenon, and the field of entrepreneurship research, that were developed in the early parts of this book. Unlike previous chapters, however, I will refrain from elaborate examples and instead confine my exposition to giving hints about general problems that have to be dealt with, and references to works by scholars in possession of deeper methods knowledge than mine regarding solutions to these problems.

Heterogeneity and Analysis Method I have suggested that heterogeneity be accepted as a fundamental factor in entrepreneurship research. This has implications not only for sampling strategy (Chapter 5) but also for analysis. Ever thought about why it is that in work on the micro level you almost never see a statistical analysis where more than 50 percent of the variance is explained? The usual suspects are model misspecification and measurement error. That is, omitted variables are suspected to account for the unexplained variance, or error variance in independent as well as dependent variables is assumed to dilute the true explanatory power of the model. However, this does not explain why it is that even in the most comprehensive and well designed work, based on sound and state-of-the art theoretical backing as well as the best available measuring instruments, you still don’t reach anything near full explanation. A plausible additional reason for this is that heterogeneity concerns not only variance across cases in the values of the included (and non-included) variables, but also the size of the effects. That is, it is reasonable to assume—unlike underlying assumptions of microeconomic theory and various statistical techniques—that the effect of variable x on variable y is different for different cases (individuals, firms, regions, etc.). This was one of George Katona’s hobby-horses when establishing the field of Economic Psychology/Behavioral Economics as an alternative to

190

RESEARCHING ENTREPRENEURSHIP

conventional economic approaches (Katona, 1974, 1975). In other words, the reason why we reached only 20-23 percent explanation in the analysis in Table 9:2 is not that a range of other expected consequences of growth, which we were too stupid to consider, explain close to 80 percent of the variance. Likewise, the reason why the eight variables come out with the coefficients they do is not that all eight factors influence overall growth attitude exactly this much for every small business ownermanager. Instead, regression coefficients as we use them normally represent average effects for members of the investigated population. Around this average there is considerable variation, leaving variance unexplained (cf. Davidsson, 1991). The reason why Employee well-being seems more important than Independence is not necessarily that for every respondent the first factor is more important than the second. It may just as well be the case that relatively more managers consider the first factor at all in their evaluation of the prospect of growth. For a distinct minority, concerns about independence may be the defining growth motivator or growth deterrent. There are different ways in which we can deal with heterogeneity in the analysis. One is, of course, to accept it and its effect on explanatory power. Another obvious way to deal with it is to conduct sub-sample analyses in order to see what the relationships look like for different, more homogeneous, sub-groups (Brown et al, 2001; Davidsson, 1991). A particularly relevant example here is Samuelsson’s (2001; 2004) work, where he first uses Latent Class Analysis to demonstrate the empirical existence of two theoretically derived groups of venture ideas—innovative and reproducing—and then goes on to show considerable differences in the venture creation processes for these two groups. Yet another way to deal with the heterogeneity of the effect of predictor variables, which has become rather common in recent years, is to explicitly model it in a moderated regression analysis (Brown, 1996). There are also more creative, non-standard approaches that can be used. A particularly exemplary demonstration is Gimeno’s et al (1997) careful adaptation of analysis tools to the analysis problem. In particular, their study is exceptional in its attention to heterogeneity regarding what is deemed an acceptable level of success. This mirrors Venkataraman’s (1997) argument that performance relative to other ventures may not be the most relevant outcome variable for entrepreneurship research.

Analysis Implications of the Minority Nature of Entrepreneurship Entrepreneurship research often takes an interest in the exceptional. Often the analysis is aimed at learning about a small minority that stands out from a larger population: individuals currently involved in a start-up; venture-capital backed ventures; IPOs; rapidly growing firms, or some other select minority. Again, this has implications not only for sampling but also for analysis method. For example, one major research question in PSED and GEM deals with understanding who is more likely to be a nascent entrepreneur. We are then talking about a single digit minority of the general population. The typical analysis method for this type of research question, logistic regression analysis, does not provide unbiased estimates when the group sizes are this uneven. In addition, with standard

A Quick Look at Analysis Method

191

application of logistic regression or discriminant analysis you may end up with a function that reaches a high overall correct classification rate by putting almost all cases in the less entrepreneurial group, i.e., by performing poorly with respect to what was the researcher’s key interest. In the PSED research a similarly sized representative sample of the general population was interviewed as a comparison group. By using this comparison group we avoided the problem of uneven group sizes in our application of logistic regression (Davidsson & Honig, 2003). Analyzing German GEM data, Wagner (2004) recently applied Rare Events Logistic Regression (King & Zeng, 2001a, 2001b) in order to get unbiased estimates. So there are method guys out there who continuously work on solving our problems, if we only make the effort to find their work. Although my perspective acknowledges also relatively mundane, imitative venture start-ups as instances of entrepreneurship, it is no doubt the case that the most interesting (and infrequent) cases are found at the other outskirt of the distribution. This points at a very fundamental problem of the standard package of statistical methods—you know, this method knowledge that we have all invested sweat equity in—as applied to entrepreneurship research problems. These varianceexplaining techniques typically focus on central tendencies, preferably for normally distributed variables. Outliers are technical problems to be eliminated; not thrilling empirical phenomena of the highest societal import. In sharp contrast, our key interest may rest with the rare cases at the high end of a highly skewed distribution. This is where we must have the guts and the energy to disregard our vested interest in the conventional. Much like the problem of secondary data that are fundamentally inadequate for the purpose (Ch. 7) we have to face it when the methods we have learnt don’t do the job properly. We then have to look elsewhere. There are alternatives to opening the tin can with a hammer.

Analysis Implications of Entrepreneurship as Process This book has advocated a process perspective on entrepreneurship. Again, this has implications not only for data collection but also for choice of analysis techniques. It is possible to attain new insights about entrepreneurial processes by applying conventional techniques to longitudinal data (Carter et al, 1996; Davidsson & Honig, 2003). However, this is the way of us old fossils and not the hope for the future. In order to better deal with the specific data challenges, and to make full use of the longitudinal aspects of the data, other techniques may have to be learned and applied. With my limited knowledge there are two (sets of) techniques that I find especially promising, namely Event History Analysis (Blossfeld & Rohwer, 2002) and Longitudinal Growth Modeling (Muthén & Curran, 1997; Muthén & Khoo, 1999). In Event History Analysis the data set is organized as monthly (or weekly, bimonthly, etc.) spells. The technique makes use of the longitudinal aspect of the dependent as well as independent variables. Independent variables can be entered as time invariant or time variant. In the latter case the value of the independent variable is allowed to change over time. The dependent variable changes its value in the

192

RESEARCHING ENTREPRENEURSHIP

month when the event to be predicted has occurred. Cases where the event has still not occurred when the last data collection is made are treated as right censored – a problem the technique is designed to deal with. The logic of the technique makes it especially suited for predicting abandonment (vs. continuation) of the start-up processes, but it can also be applied for analyzing, e.g., ‘up and running’ vs. ‘still trying’. See Delmar & Shane (2002, 2003a, 2003b, 2004) for relevant applications. Independent variables in Event History Analysis can be either continuous or dichotomous. The dependent variable, however, is always dichotomous. Thus, the technique can in this sense be regarded a longitudinal alternative to logistic regression. When the dependent variable is continuous there exist special regression methods, but I find Longitudinal Growth Modeling (LGM) a particularly interesting alternative for this situation. In the context of new venture emergence the dependent variable may be, for example, the accumulation of gestation activities in PSED-like research; the gradual attainment of the cornerstones of Klofsten’s Business Platform Model (cf. Davidsson & Klofsten, 2003), or any other variable that is analogous to growth. LGM is a longitudinal cousin to structural equation modeling techniques like LISREL and thus has the advantages of being applicable to models with latent variables and indirect as well as direct relationships. As the technique aims at predicting both initial situation and development over time it can at least to some extent handle problems like the cases being first sampled at different stages of development. A shortcoming of LGM is that unlike Event History models it cannot include cases that dissolve during the studied period in the analysis. In order to avoid erroneous conclusions based on success bias the LGM analysis should therefore be supplemented with other types of analyses of the discontinued cases, so as to make sure that these do not share the characteristics that appear as success factors in LGM. For a relevant application see Samuelsson (2001; 2004). Finally, while partially missing data (internal non-response) is always a problem that has to be dealt with in data analysis, it is aggravated when the data are longitudinal. With multiple waves of data the likelihood that a case has complete information on every variable we want to include in the analysis asymptotically approaches zero. The problem of attrition—that some cases are lost entirely over time—is bad enough; when we add loss of cases due to partially missing data we may end up having nothing left to analyze. So, we must find ways to include cases with missing information. The quick and dirty tricks like replacing missing data with the mean or with a predicted value form a regression may be defensible when but a tiny percentage of the cases are manipulated in this way. However, such techniques reduce the error variance, and therefore it amounts to cheating to apply them when we have a lot of partially missing data. Luckily, method experts have developed more sophisticated techniques for data imputation that can be applied. See for example Little (1987), Fichman & Cummings (2003) or other contributions to the same special issue of the Organization Research Methods journal (Vol. 6; Iss. 3).

SUMMARY AND CONCLUSION In this short and incomplete treatment I have pointed at some ways in which the perspective on entrepreneurship I advocate influences the choice and application of

A Quick Look at Analysis Method

193

analysis methods. We dealt in particular with three aspects: the heterogeneity of entrepreneurship; its minority character, and its process nature. Regarding heterogeneity I argued that this can be dealt with in the analysis through sub-sample analysis or by explicitly modeling differential influence of the predictor variables. I mentioned the Gimeno et al (1997) study as a particularly impressive example of a way of modeling heterogeneity that seems highly relevant for entrepreneurship studies. The toughest-to-accept conclusion emanates from the discussion of the minority characteristic of entrepreneurship. This is that the standard set of varianceexplaining techniques, which focus on central tendencies, assume normal distributions, and regard outliers a problem, may be fundamentally inadequate tools for many analysis tasks in entrepreneurship research. We also noted here that for some techniques the unevenness of group sizes may lead to biased estimates, but also that we are likely to find remedies to these problems if we care to look around. In discussing the process nature of entrepreneurship I highlighted Event History Analysis and Longitudinal Growth Modeling as promising alternatives for making fuller use of longitudinal data. I also mentioned the availability of data imputation techniques; something we need more of when we work with longitudinal rather than cross-sectional data. Have fun squeezing the most out of your data!

NOW THAT WE’RE DONE…

Was it any good? Enjoyable? Useful? Worthwhile? Bearable? Since you’re part of the biased sub-sample who is still with me you probably found some reason to read on. I can only hope it was a good one. I regret it if I have annoyed people with my patchy, biased and incomplete coverage of method issues in entrepreneurship research, or with the numerous references to my own work. But, then again, perhaps you misunderstood my intentions, possibly because you skipped the preface? This book was an attempt to codify the knowledge—if I may so name it—that I have built up over the years; it was decidedly not an attempt to review and present the most central themes of the extant collective knowledge of entrepreneurship scholars. I’m neither smart nor energetic enough to take on the latter task. Some may have been turned off by the non-academic, chat-like writing style that I apply in places. I regret it if some feel that form deducted from the contents, but I feel confident that there are other readers who appreciate that we do not always have to keep a dead serious tone even if we are dead serious about the message. My intention was to make the contents more digestible. There may be a need for that— people have remarked more than once that my texts tend to be jam-packed with information. So even if I used a catchy phrase here and there I trust you did not find a lot of empty BS. I have put a lot of time and intellectual effort into the first two chapters of the book—those on the entrepreneurship phenomenon and the corresponding field of research. While respecting that some researchers could not care less, or find this type of effort futile, I hope others will appreciate this attempt to reconcile some of the issues that plague the field and feel inspired to continue to contribute to this debate and development. I got pleasantly surprised when I re-read Chapter 3, on theory, and hope researchers who are still in the formative stages of their career will find it useful. I realize and accept, however, that others might feel that I could have left the more general aspects of theory use to the experts—whoever they are. I also realize there is the risk that the more general discussion about qualitative and quantitative research in Chapter 4 either confused or appeared self-evident for those who are themselves deeply into a quantitative paradigm, and did not reach at all the people I would really like to contemplate these issues in a truly academic rather than religious way. In the long chapters 5 and 6 I also dealt with issues of importance far outside the realm of entrepreneurship research. This was because I find them missing in most standard treatments. However, chapters 4, 5 and 6 also include a lot of practical, hands on design advice that in a sense is the core of this book. Although I have tried to give advice that has applicability for a broad range of possible entrepreneurship questions it is unavoidable—especially when it comes to operationalizations—that the examples become so specific that they best serve as inspiration rather than being of immediate use for the reader. The chapters on sampling and operationalizations are long and may be heavy, with fall-asleep

196

RESEARCHING ENTREPRENEURSHIP

potential if attempted for a leisurely reading. I hope, however, that they contain useful advice for those who are in the middle of dealing with design problems. Chapters 7 and 8 are perhaps the narrowest. Being niche products I hope they prove very useful for those in the niche, but I realize they are less enjoyable parts of a general-purpose, cover-to-cover reading. So I don’t blame those who just browsed or skipped entirely these chapters. Chapter 9 is one of my own favorites and I hope it will inspire readers to appreciate more the value of replications. Finally, Chapter 10 was merely a briefing on analysis method implications—although there is a message in there that at least I find very important. In order not to make a fool of myself, or step on too many toes, 1 should, of course, have made trusted colleagues preview this manuscript, or parts thereof, to a much greater extent than I have done. However, although I nowadays appreciate the quality-enhancing effect of the peer review process much more than I used to, I maintain that it also makes a work lose some of its individual voice. I’m neither cautiously natured nor non-tenured, and in this case I really wanted to keep my voice. Albeit shaped in part by the environment like for any social animal, what you found in this book were my ideas, my experiences, my advice, my convictions, my misconceptions, my omissions, my shortcomings, my rudeness and my outright errors. Thank you for bearing with me! P.D.

REFERENCES

Acs, Z. J. (1996). Small firms and economic growth. In Z. Acs, B. Carlsson & R. Thurik (Eds.), Small Business in the Modern Economy. Oxford, UK: Blackwell. Acs, Z. J. & Armington, C. (1998). Longitudinal establishment and enterprise microdata (LEEM) documentation. Working Paper 1998-10. Washington, DC: Center for Economic Studies, U.S. Bureau of the Census. Acs, Z. J. & Audretsch, D. B. (01990). Innovation and Small Firms. Cambridge, MA: MIT Press. Acs, Z. J. & Audretsch, D. B. (2003b). Editor’s introduction. In Z. J. Acs & D. B. Audretsch (Eds.), Handbook of entrepreneurship research (Vol. Kluwer). Dordrecht, NL. Acs, Z. J. & Audretsch, D. B. (Eds.). (2003a). Handbook of Entrepreneurship Research: An Interdisciplinary Survey and Introduction. Dordrecht, NL: Kluwer. Ahl, H. J. (2002). The Making of the Female Entrepreneur. Doctoral dissertation, Jönköping International Business School, Jönköping. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179-211. Ajzen, I. & Fishbein, M. (1977). Attitude-behavior relations: A theoretical analysis and review of empirical research. Psychological Bulletin, 84, 888-918. Aldrich, H. (1999). Organizations Evolving. Newbury Park, CA: Sage Publications. Aldrich, H. & Baker, T. (1997). Blinded by the cites? Has there been progress in the entrepreneurship field? In D. Sexton & R. Smilor (Eds.), Entrepreneurship 2000 (pp. 377-400). Chicago, IL: Upstart Publishing Company. Aldrich, H., Kalleberg, A. L., Marsden, P. V. & Cassell, J. (1989). In pursuit of evidence: strategies for locating new businesses. Journal of Business Venturing, 4(6), 367-386. Aldrich, H. & Martinez, M. E. (2001). Many are called but few are chosen: an evolutionary perspective for the study of entrepreneurship. Entrepreneurship Theory & Practice, 25(4, Summer), 41-56. Aldrich, H. & Wiedenmayer, G. (1993). From traits to rates: an ecological perspective on organizational foundings. In J. Katz & R. Brockhaus (Eds.), Advances in Entrepreneurship, Firm Emergence, and Growth (Vol. 1, pp. 145-196.). Greenwich, CT: JAI Press. Aldrich, H. & Zimmer, C. (1986). Entrepreneurship through social networks. In D. Sexton & R. Smilor (Eds.), The Art and Science of Entrepreneurship (pp. 3-23). Cambridge, MA: Ballinger. Alvarez, S. A. & Barney, J. B. (forthcoming). Organizing Rent Generation and Appropriation: Toward a Theory of The Entrepreneurial Firm. journal of Business Venturing. Alvesson, M. & Sköldberg, K. (1994). Tolkning och reflektion. Lund: Studentlitteratur. Amit, R. & Zott, C. (2001). Value drivers in e-business. Strategic Management Journal, 22, 493-520. Anderson, J. R. (1990). Cognitive Psychology and its Implications. New York: Freeman. Anderson, P. & Tushman, M. (1990). Technological discontinuities and dominant designs: a cyclical model of technological change. Administrative Science Quarterly, 35, 604-633. Armstrong, J. S. (1970). How to avoid exploratory research. Journal of Advertising Research, 10(4), 2730. Arrow, K. (1983). Innovation in small and large firms. In J. Ronen (Ed.), Entrepreneurship (pp. 15-28). Lexington, MA: Lexington Books. Audretsch, D. & Fritsch, M. (1994). The geography of firm births in Germany. Regional Studies, 28(4), 359-365. Baars, B. J. (1986). The Cognitive Revolution in Psychology. New York: The Guildford Press. Baldwin, J. & Picot, G. (1995). Employment generation by small producers in the Canadian manufacturing sector. Small Business Economics, 7, 317-331. Bandura, A. (1982). Self-efficacy mechanism in human agency. American Psychologist, 37, 122-147. Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory. Englewood Cliffs: Prentice-Hall. Barney, J. B. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17(1), 99-120. Barney, J. B. (1997). Gaining and Sustaining Competitive Advantage. Menlo Park, CA: Addison Wesley.

198

RESEARCHING ENTREPRENEURSHIP

Baron, R. A. & Brush, C. G. (1999). The role of social skills in entrepreneurs’ success: evidence from videotapes of entrepreneurs’ presentations. In P. D. Reynolds, W. D. Bygrave, N. M. Carter, S. Manigart, C. M. Mason, G. D. Meyer & K. G. Shaver (Eds.), Frontiers of Entrepreneurship 1999. Wellesley, MA: Babson College. Baum, J. R., Locke, E. A. & Kirkpatrick, S. A. (1998). A longitudinal study of the relation of vision and vision communication to venture growth and performance. Journal of Applied Psychology, 83(1), 43-54. Baumol, W. J. (1983). Toward operational models of entrepreneurship. In J. Ronen (Ed.), Entrepreneurship (pp. 29-47). Lexington, MA: Lexington Books. Baumol, W. J. (1990). Entrepreneurship: Productive, unproductive and destructive. Journal of Political Economy, 98(5), 893-921. Baumol, W. J. (1993). Entrepreneurship, Management and the Structure of Payoffs. Cambridge, MA: MIT Press. Bhave, M. P. (1994). A process model of entrepreneurial venture creation. Journal of Business Venturing, 9, 223-242. Birch, D. L. (1979). The Job Generating Process. Cambridge, MA: MIT Program on Neighborhood and Regional Change. Birch, D. L., Haggerty, A. & Parsons, W. (1995). Who’s Creating Jobs? Boston: Cognetics, Inc. Birch, D. L. & Medoff, J. (1994). Gazelles. In L. C. Solomon & A. R. Levenson (Eds.), Labor Markets, Employment policy and Job Creation. Boulder and London: Westview Press. Bird, B. (1993). Demographic approaches to entrepreneurship: The role of experience and background. In J. A. Katz & R. H. Brockhaus Sr. (Eds.), Advances in Entrepreneurship, Firm Emergence, and Growth (Vol. 1, pp. 11-48). Greenwich: JAI Press. Birley, S. & Westhead, P. (1994). A taxonomy of business start-up reasons and their impact on firm growth and size. Journal of Business Venturing, 9, 7-31. Björnsson, B. (2001). En bank även för kvinnor? Småföretagares erfarenheter av rådgivningsmöten i bank (A Bank Also for Women? Small-Business Customers’ Perceptions of Counselling Encounters in Banks). Doctoral dissertation, Göteborgs universitet., Göteborg. Blossfeld, H.-P. & Rohwer, G. (2002). Techniques of event history modeling: new approaches to causal analysis. Mahwah, NJ: Lawrence Erlbaum Associates. Bolton, J. E. (1971). Small Firms. Report of the Committee of Inquiry on Small Firms. London: Her Majesty’s Stationery Office. Borgatta, E. F. & Bohrnstedt, G. W. (1980). Level of measurement: once over again. Sociological Methods and Research, 9(2), 147-160. Braunerhjelm, P. & Carlsson, B. (1999). Industry clusters in Ohio and in Sweden. Small Business Economics, 1(4), 279-293. Braunerhjelm, P., Carlsson, B., Cetindamar, D. & Johansson, D. (2000). The old and the new: The evolution of polymer and biomedical clusters in Ohio and Sweden. Journal of Evolutionary Economics, 10(5), 471-488. Brockhaus, R. H. (1982). The psychology of the entrepreneur. In C. A. Kent, D. L. Sexton & K. H. Vesper (Eds.), Encyclopedia of Entrepreneurship (pp. 39-71). Englewood Cliffs, NJ: Prentice Hall. Brown, T. (1996). Resource Orientation, Entrepreneurial Orientation and Growth: How the Perception of Resource Availability Affects Small Firm Growth. Unpublished Doctoral dissertation, Rutgers University, Newark, NJ. Brown, T., Davidsson, P. & Wiklund, J. (2001). An operationalization of Stevenson’s conceptualization of entrepreneurship as opportunity-based firm behavior. Strategic Management Journal, 22(10), 953968. Brundin, E. (2002). Emotions in Motion. Leadership during Radical Change. Doctoral dissertation. Jönköping: Jönköping International Business School. Bryman, A. & Cramer, D. (1999). Quantitative Data Analysis with SPPS Release 8 for Windows: A Guide for Social Scientists. London: Routledge. Brymer, R. (1998). Hanging out with the good ’ole boys, gangsters, and other disreputable characters: field research, quantitative research, and exceptional events. In S. Grills (Ed.), Doing Ethnographic Research: Fieldwork Settings (pp. 143-162). Thousand Oaks: Sage. Bull, I. & Willard, G. E. (1993). Towards a theory of entrepreneurship. Journal of Business Venturing, 8, 183-195. Burt, R. S. (1992). Structural Holes: The Social Structure of Competition. Cambridge, MA: Harvard University Press.

References

199

Busenitz, L. W. (1996). Research on entrepreneurial alertness. Journal of Small Business Management, 34, 35-44. Carter, N. M., Gartner, W. B. & Reynolds, P. D. (1996). Exploring start-up event sequences. Journal of Business Venturing, 11, 151-166. Casson, M. (1982). The Entrepreneur. Totowa, NJ: Barnes & Noble Books. Caves, R. E. (1980). Industrial organization, corporate strategy and structure. Journal of Economic Literature, 58, 64-92. Chandler, G. N., Dahlqvist, J. & Davidsson, P. (2002). The opportunity discovery process: Resource and performance implications. In W. Bygrave, P. Davidsson, J. Fiet, P. G. Greene, R. Harrison, M. Lerner, G. D. Meyer, J. Sohl & A. Zacharakis (Eds.), Frontiers of Entrepreneurship Research 2002. Wellesley, MA. Babson College. Chandler, G. N., Dahlqvist, J. & Davidsson, P. (2003). Opportunity recognition processes: A taxonomic classification and outcome implications, Academy of Management Meeting. Seattle. Chandler, G. N. & Hanks, S. H. (1994). Market attractiveness, resource-based capabilities, venture strategies, and venture performance. Journal of Business Venturing, 9, 331-349. Chandler, G. N. & Jansen, E. (1992). The founder’s self-assessed competence and venture performance. Journal of Business Venturing, 7, 223-236. Chandler, G. N. & Lyon, D. W. (2001). Methodological issues in entrepreneurship research: the past decade. Entrepreneurship Theory & Practice, 25(4, Summer), 101-113. Chen, C. C., Greene, P. G. & Crick, A. (1998). Does entrepreneurial self-efficacy distinguish entrepreneurs from managers? Journal of Business Venturing, 13, 295-316. Cialdini, R. B. (1980). Full cycle social psychology. In L. Beckman (Ed.), Applied Social Psychology Annual (Vol. 1). Beverly Hills, CA: Sage. Coase, R. H. (1937). The nature of the firm. Economica, 4, 386-405. Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2 ed.). Hillsdale, NJ: Lawrence Erlbaum Associates. Cohen, J. (1994). The earth is round (p<.05). American Psychologist, 47(12), 997-1003. Cohen, W. M. & Levinthal, D. A. (1990). Absorptive capacity: a new perspective on learning and innovation. Administrative Science Quarterly, 35, 128-152. Cole, A. H. (1949). Entrepreneurship and entrepreneurial history. Change and the Entrepreneur, 88-107. Coleman, J. (1990). Social capital and the creation of human capital. American Journal of Sociology, 94, 95-120. Collins, D. J. & Montgomery, C. A. (1995). Competing on resources: strategy in the 1990s. Harvard Business Review, 73(4), 117-128. Conner, K. & Prahalad, C. K. (1996). A Resource-based theory of the firm: knowledge vs. opportunism. Organization Science, 7(5), 477-501. Cooper, A. C., Folta, T. B. & Woo, C. Y. (1995). Entrepreneurial information search. Journal of Business Venturing, 10, 107-120. Cooper, A. C., Gimeno-Gascon, F. J. & Woo, C. Y. (1994). Initial human and financial capital as predictors of new venture performance. Journal of Business Venturing, 9(5), 371-395. Covin, J. G. & Slevin, D. P. (1986). The development and testing of an organizational-level entrepreneurship scale. In R. Ronstadt, J. A. Hornaday, R. Peterson & K. H. Vesper (Eds.), Frontiers of Entrepreneurship Research 1986 (pp. 628-639). Wellesley, MA: Babson College. Cyert, R. M. & March, J. G. (1963). A Behavioral Theory of the Firm. Englewood Cliffs, NJ.: PrenticeHall. Daft, R.: (1983). Organization Theory and Design. New York: West. Daft, R. L. & Weick, K. E. (1984). Toward a model of organizations as interpretations systems. Academy of Management Review, 9(2), 284-295. Dahlqvist, J. & Davidsson, P. (2000). Business start-up reasons and firm performance. In P. Reynolds, E. Autio, C. Brush, W. Bygrave, S. Manigart, H. Sapienza & K. Shaver (Eds.), Frontiers of Entrepreneurship Research 2000. Wellesley, MA: Babson College. Dahlqvist, J., Davidsson, P. & Wiklund, J. (2000). Initial Conditions as Predictors of New Venture Performance: a Replication and Extension of the Cooper et al study. Enterprise and Innovation Management Studies, 1(1), 1-17. Davidsson, P. (1986). Tillväxt i små företag: en pilotstudie om tillväxtvilja och tillväxtförutsättningar i små företag (Small Firm Growth: A Pilot Study on Growth Willingness and Opportunity for Growth in Small Firms) (Studies in Economic Psychology No. 120). Stockholm: Stockholm School of Economics.

200

RESEARCHING ENTREPRENEURSHIP

Davidsson, P. (1988). Type of Man and Type of Company Revisited: A Confirmatory Cluster Analysis Approach. In B. Kirchhoff, W. Long, W. McMullan, K. Vesper & W. Wetzel (Eds.), Frontiers of Entrepreneurship Research 1988 (pp. 88-105). Wellesley, MA: Babson College. Davidsson, P. (1989a). Continued Entrepreneurship and Small Firm Growth. Doctoral dissertation, Stockholm School of Economics, Stockholm. Davidsson, P. (1989b). Entrepreneurship -- and after? A study of growth willingness in small firms. Journal of Business Venturing, 4, 211-226. Davidsson, P. (1991). Continued entrepreneurship: Ability, need, and opportunity as determinants of small firm growth. Journal of Business Venturing, 6, 405-429. Davidsson, P. (1992). Entrepreneurship and Small Business Research: How Do We Get Further? (BApublications No. 126). Umeå: Umeå Business School. Davidsson, P. (1993). Kultur och entreprenörskap - om orsaker till regional variation i nyföretagande (Culture and Entrepreneurship - On the Determinants of Regional Variation in New Firm Formation). Stockholm: NUTEK. Davidsson, P. (1994). Job creation in small firms: Myth or fact. Paper presented at the RENT VIII Workshop, Tampere. Davidsson, P. (1995a). Culture, structure and regional levels of entrepreneurship. Entrepreneurship & Regional Development, 7, 41-62. Davidsson, P. (1995b). Determinants of Entrepreneurial Intentions, Working Paper 1995:1. Jönköping: Jönköping International Business School. Davidsson, P. (1995c). Kultur och Entreprenörskap - en uppföljning (Culture and Entrepreneurship - A Follow-up). Örebro: Stiftelsen Forum för Småföretagsforskning. Davidsson, P. (1995d). Small firms: Has their role as job creators been exaggerated? Paper presented at the ICSB 40th World Conference, Sydney. Davidsson, P. (1996). Methodological concerns in the estimation of job creation in different firm size classes. Paper presented at Rencontres St. Gall, Gottlieben, Switzerland. Davidsson, P. (2002). What entrepreneurship research can do for business and policy practice. International Journal of Entrepreneurship Education, 1(1), 5-24. Davidsson, P. (2003a). The domain of entrepreneurship research: Some suggestions. In J. Katz & D. Shepherd (Eds.), Advances in Entrepreneurship, Firm Emergence and Growth (Vol. 6, pp. 315372). Oxford, UK: Elsevier/JAI Press. Davidsson, P. (2003b). Method issues in the study of new venture start-up processes. Paper presented at the NFF Conference, Reykjavik. Davidsson, P. & Delmar, F. (1992). Cultural values and entrepreneurship. In N. C. Churchill, S. Birley, W. D. Bygrave, C. Wahlbin & W. E. J. Wetzel (Eds.), Frontiers of Entrepreneurship Research 1992 (pp. 444-458). Wellesley, MA: Babson College. Davidsson, P. & Delmar, F. (1997). High-growth firms and their contribution to employment: The case of Sweden 1987-96. Paris: OECD Working Party on SMEs. Davidsson, P. & Delmar, F. (1998). Some important observations concerning job creation by firm size and age. Paper presented at the Rencontres St. Gall, Elm, Switzerland. Davidsson, P. & Delmar, F. (1999). Zur Schaffung von Arbeitsplätzen nach Firmengrösse und -alter. Internationales Gewerbearchiv, 46(219-233). Davidsson, P. & Delmar, F. (2000). Tillväxtföretag: var de finns, hur de växer, och vilka jobb de skapar. (Growth Firms: Where They Are Found, How They Grow, and the Jobs They Create). Report No. 19, Nya jobb & företag. Stockholm: Rådet för Arbetslivsforskning. Davidsson, P. & Delmar, F. (2001). Les entreprises à forte croissance et leur contribution à 1’emploi: le cas de la Suéde 1987-1996. Revue Internationale PME, 14(3-4), 164-187. Davidsson, P. & Delmar, F. (2003). Hunting for new employment: the role of high-growth firms. In D. Kirby & A. Watson (Eds.), Small Firms and Economic Development in Developed and Transition Economies: A Reader (pp. 7-20). Aldershot, UK: Ashgate. Davidsson, P., Delmar, F. & Wiklund, J. (2002). Entrepreneurship as growth; growth as entrepreneurship. In M. A. Hitt, R. D. Ireland, S. M. Camp & D. L. Sexton (Eds.), Strategic Entrepreneurship: Creating a New Mindset (pp. 328-342). Oxford, UK: Blackwell. Davidsson, P. & Henreksson, M. (2002). Institutional determinants of the prevalence of start-ups and high-growth firms: Evidence from Sweden. Small Business Economics, 19(2), 81-104. Davidsson, P. & Honig, B. (2003). The role of social and human capital among nascent entrepreneurs. Journal of Business Venturing, 18(3), 301-331.

References

201

Davidsson, P., Kirchhoff, B., Hatemi-J, A. & Gustavsson, H. (2002). Empirical analysis of growth factors using Swedish data. Journal of Small Business Management, 40(4), 332-349. Davidsson, P. & Klofsten, M. (2003). The Business Platform: Developing an instrument to gauge and assist the development of young firms. Journal of Small Business Management, 41(1), 1-26. Davidsson, P., Lindmark, L. & Olofsson, C. (1993). Regional characteristics, business dynamics, and economic development - a project presentation and some preliminary findings. In C. Karlsson, B. Johannisson & D. Storey (Eds.), Small Business Dynamics: International, National and Regional Perspectives (pp. 145-172). London: Routledge. Davidsson, P., Lindmark, L. & Olofsson, C. (1994a). Dynamiken i svenskt näringsliv (Business Dynamics in Sweden). Lund, Sweden: Studentlitteratur. Davidsson, P., Lindmark, L. & Olofsson, C. (1994b). New firm formation and regional development in Sweden. Regional Studies, 28, 395-410. Davidsson, P., Lindmark, L. & Olofsson, C. (1995). Small firms, business dynamics and differential development of economic well-being. Small Business Economics, 7, 301-315. Davidsson, P., Lindmark, L. & Olofsson, C. (1996). Näringslivsdynamik under 90-talet (Business Dynamics in the 90s), Stockholm: Nutek. Davidsson, P., Lindmark, L. & Olofsson, C. (1998a). Smallness, newness and regional development. Swedish Journal of Agricultural Research, 28, 57-71. Davidsson, P., Lindmark, L. & Olofsson, C. (1998b). The extent of overestimation of small firm job creation: an empirical examination of the ‘regression bias’. Small Business Economics, 10, 87-100. Davidsson, P., Lindmark, L. & Olofsson, C. (1998c). Small firms and job creation during a recession and recovery. In Z. Acs, B. Carlsson & C. Karlsson (Eds.), Entrepreneurship, Small and Medium-sized Enterprises and the Macroeconomy (pp. 286-309). Cambridge, MA: Cambridge University Press. Davidsson, P. & Wahlund, R. (1992). A Note on the Failure to Use Negative Information. Journal of Economic Psychology, 13, 343-353. Davidsson, P. & Wiklund, J. (1997). Values, beliefs and regional variations in new firm formation rates. Journal of Economic Psychology, 18, 179-199. Davidsson, P. & Wiklund, J. (2000). Conceptual and empirical challenges in the study of firm growth. In D. Sexton & H. Landström (Eds.), The Blackwell Handbook of Entrepreneurship (pp. 26-44). Oxford, MA: Blackwell Business. Davidsson, P. & Wiklund, J. (2001). Levels of analysis in entrepreneurship research: current practice and suggestions for the future. Entrepreneurship Theory & Practice, 25(4, Summer), 81-99. Davis, S. J., Haltiwanger, J. & Schuh, S. (1996a). Job Creation and Destruction. Boston, MA: MIT Press. Davis, S. J., Haltiwanger, J. & Schuh, S. (1996b). Small business and job creation: Dissecting the myth and reassessing the facts. Small Business Economics, 8(297-315). de Koning, A. (1999a). Opportunity formation from a socio-cognitive perspective. Paper presented at the Babson/Kaufman Entrepreneurship Research Conference, Columbia, SC., May 1999. de Koning, A. (1999b). Conceptualising Opportunity Formation as a Socio-Cognitive Process. Unpublished Doctoral dissertation, INSEAD, Fontainebleau, France. Delmar, F. (1996). Entrepreneurial Behavior and Business Performance. Stockholm: Stockholm School of Economics. Delmar, F. (2000). The psychology of the entrepreneur. In S. Carter & D. Jones-Evans (Eds.), Enterprise & Small Business: Principles, Practice and Policy. (pp. 132-154). Harlow: Financial Times. Delmar, F. & Davidsson, P. (1999). Firm size expectations of nascent entrepreneurs. In P. D. Reynolds, W. D. Bygrave, S. Manigart, C. Mason, G. D. Meyer, H. J. Sapienza & K. G. Shaver (Eds.), Frontiers of Entrepreneurship Research 1999 (Vol. 19, pp. 90-104). Wellesley, MA: Babson College. Delmar, F. & Davidsson, P. (2000). Where do they come from? Prevalence and characteristics of nascent entrepreneurs. Entrepreneurship & Regional Development, 12, 1-23. Delmar, F., Davidsson, P. & Gartner, W. (2003). Arriving at the high-growth firm. Journal of Business Venturing, 18(2), 189-216. Delmar, F. & Shane, S. (2002). What founders do: A longitudinal study of the start-up process. In P. D. Reynolds et al (Ed.), Frontiers of Entrepreneurship Research 2002 (pp. 632-645). Wellesley, MA. Delmar, F. & Shane, S. (2003a). Does business planning facilitate the development of new ventures? Strategic Management Journal 24, 1165-1185. Delmar, F. & Shane, S. (2003b). Does the order of organizing activities matter for new venture performance? In P. D. Reynolds et al (Ed.), Frontiers of Entrepreneurship 2003. Wellesley, MA.: Babson College.

202

RESEARCHING ENTREPRENEURSHIP

Delmar, F. & Shane, S. (2004). Legitimating first: Organizing activities and the survival of new ventures. Journal of Business Venturing, 19, 385-410. Dess, G. G. & Beard, D. W. (1984). Dimensions of organizational task environments. Administrative Science Quarterly, 29, 52-73. Dicken, P. (1998). Global Shift - Transforming the World Economy (3rd ed.). London: Paul Chapman Publishing. Dillman, D. A. (1978). Mail and Telephone Surveys: The Total Design Method. New York: WileyInterscience. Douglas, S. R. & Craig, C. S. (1983). International Marketing Rresearch. Englewood Cliffs, NJ: Prentice-Hall. Drucker, P. (1985). Innovation and Entrepreneurship. New York: Harper & Row. Eagly, A. H. & Chaiken, S. (1993). The Psychology of Attitudes. Orlando, FLA: Harcourt Brace Jonanovich, Inc. Eliasson, C. & Davidsson, P. (2003). Entrepreneurial management, corporate venturing, and financial performance. Paper presented at the 22nd Babson College/Kauffman Foundation Entrepreneurship Research Conference, Wellesley, MA. Eliasson, G. (1991). Modeling the experimentally organized economy: complex dynamics in an empirical micro-macro model of endogenous economic growth. Journal of Economic Behavior and Organization, 16, 153-182. Evans, D. S. (1987). Test of alternative theories of firm growth. Journal of Political Economy, 95, 657674. Fichman, M. & Cummings, J. N. (2003). Multiple imputation for missing data: Making the most of what you know. Organizational Reserach Methods, 6(3), 282-308. Fiet, J. O. (1999). Entrepreneurial competence as knowledge. Jönköping, Sweden: Jönköping International Business School. Fiet, J. O. (2002). The Search for Entrepreneurial Discoveries. Westport, CT.: Quorum Books. Fiet, J. O. & Migliore, P. J. (2001). The testing of a model of entrepreneurial discovery by aspiring entrepreneurs. In W. D. Bygrave, E. Autio, C. G. Brush, P. Davidsson, P. G. Green, P. D. Reynolds & H. J. Sapienza (Eds.), Frontiers of Entrepreneurship Research 2001 (pp. 1-12). Wellesley, MA. Fiet, J. O., Piskounov, A. & Gustavsson, V. (2000). How to decide how to search for entrepreneurial discoveries. Paper presented at the Babson/Kauffman Foundation Entrepreneurship Research Conference, Babson College. Fink, A. (Ed.). (1995). The Survey Kit. Thousands Oaks, CA.: Sage Publicatons. Fornell, C. (1987). A second generation of multivariate analysis: Classification of methods and implications for marketing research. In M. Houston (Ed.), The Review of Marketing: American Marketing Association. Fornell, C. & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18, 39-50. Foss, N. J. (1993). Theories of the firm: contractual and competence perspectives. Journal of Evolutionary Economics, 127-144. Foss, N. J. (1996). Knowledge-based approaches to the theory of the firm: some critical comments. Organization Science, 7(5), 470-476. Foss, N. J. & Klein, P. G. (Eds.). (2002). Entrepreneurship and the Firm: Austrian Perspectives on Economic Organization. Chletenham: Edward Elgar. Fridh, A.-C. (2002). Dynamics and Growth: The Health Care Industry. Doctoral dissertation, The Royal Institute of Technology, Stockholm. Fujita, M., Krugman, P. & Vennables, M. J. (1999). The Spatial Economy: Cities, Regions and International Trade. Cambridge, MA.: MIT Press. Gaglio, C. M. (1997). Opportunity identification: Review, critique and suggested research directions. In J. Katz & J. Brockhaus (Eds.), Advances in Entrepreneurship, Firm Emergence, and Growth (Vol. 3, pp. 139-202). Greenwich, CT: JAI Press. Galunic, D. C. & Rodan, S. (1998). Resource combinations in the firm: Knowledge structures and the potential for Schumpeterian innovation. Strategic Management Journal, 19, 1193-1201. Garofoli, F. (1994). New firm formation and regional development: the Italian case. Regional Studies, 28(4), 381-393. Gartner, W. B. (1988). “Who is an Entrepreneur” is the wrong question. American Small Business Journal(Spring), 11-31.

References

203

Gartner, W. B. (1990). What are we talking about when we are talking about entrepreneurship? Journal of Business Venturing, 5, 15-28. Gartner, W. B. (1993). Words lead to deeds: Towards an organizational emergence vocabulary. Journal of Business Venturing, 8, 231-239. Gartner, W. B. (2001). Is there an elephant in entrepreneurship research? Blind assumptions in theory development. Entrepreneurship Theory & Practice, 25(4, Summer), 27-39. Gartner, W. B. & Brush, C. B. (1999). Entrepreneurship as organizing: Emergence, newness, and transformation. Academy of Management Entrepreneurship Division Doctoral Consortium. Gartner, W. B. & Carter, N. (2003). Entrepreneurial behavior and firm organising processes. In Z. J. Acs & D. B. Audretsch (Eds.), Handbook of Entrepreneurship Research. Dordrecht, NL: Kluwer. Gartner, W. B., Shaver, K. G., Carter, N. M. & Reynolds, P. D. (2004). Handbook of Entrepreneurial Dynamics: The Process of Business Creation. Thousand Oaks, CA: Sage. Gimeno, J., Folta, T. B., Cooper, A. C. & Woo, C. Y. (1997). Survival of the fittest? Entrepreneurial human capital and the persistence of underperforming firms. Administrative Science Quarterly, 42, 750-783. Goldman, A. I. (1986). Epistemology and cognition. Cambridge, MA.: Harvard University Press. Gollwitzer, P. M. (1999). Implementation intentions: strong effects of simple plans. American Psychologist, 54(7), 493-503. Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360-1380. Gratzer, K. (1996). Småföretagandets villkor. Automatrestauranger under 1900-talet (Conditions for Small Firms. Automated Restaurants During the Twentieth Century). Stockholm: Almqvist & Wicksell. Gratzer, K. (1999). The making of a new industry -- the introduction of fast food in Sweden. In B. Johannisson & H. Landström (Eds.), Images of Entrepreneurship Research -- Emergent Swedish Contributions to Academic Research (pp. 82-114). Lund, Sweden: Studentlitteratur. Gray, C. (1990, July 4-7). Entrepreneurial motivation and the smaller business. Paper presented at the 15th IAREP Colloquium, Exeter, UK. Greene, P. G., Brown, T. (1997). Resource needs and the dynamic capitalism typology. Journal of Business Venturing, 12, 161-173. Greene, P. G., Brush, C. G. & Hart, M. M. (1999). The corporate venture champion: a resource-based approach to role and process. Entrepreneurship Theory & Practice, 23(Spring), 103-122. Guesnier, B. (1994). Regional variations in new firm formation in France. Regional Studies, 28(4), 347358. Gustavsson, V. (2004). Entrepreneurial Decision-Making. Doctoral dissertation, Jönköping International Business School, Jönköping. Hair, J. F., Anderson, R. E., Tatham, R. L. & Black, W. C. (1998). Multivariate Data Analysis (5 ed.). Upper Saddle River, NJ: Prentice Hall. Haltiwanger, J. & Krizan, C. J. (1999). Small business and job creation in the United States: the role of new and young businesses. In Z. J. Acs (Ed.), Are Small Firms Important? Their Role and Impact (pp. 79-98). Dordrecht, NL: Kluwer. Hannan, M. T. & Freeman, J. H. (1977). The population ecology of organizations. American Journal of Sociology, 82(5), 929-964. Harms, R. & Ehrmann, T. (2003). The performance implications of Entrepreneurial Management: Linking Stevenson’s and Miller’s conceptualizations to growth. Paper presented at the Babson College/Kauffman Foundation Entrepreneurship Research Conference: Babson College. Hart, M. & Gudgin, G. (1994). Spatial variations in new firm formation in the republic of Ireland, 198090. Regional Studies, 28(4), 367-380. Hisrisch, R. D. & Peters, M. P. (1989). Entrepreneurship. Starting, Developing and Managing a New Enterprise. Homewood, IL: Irwin. Hofstede, G. (1980). Culture’s Consequences: International Differences in Work-related Values. Beverly Hills, CA: Sage Publications. Hornaday, J. A. (1982). Research about living entrepreneurs. In C. A. Kent, D. L. Sexton & K. H. Vesper (Eds.), Encyclopedia of Entrepreneurship (pp. 20-38). Englewood Cliffs: Prentice-Hall. Hubbard, R., Vetter, D. E. & Little, E. L. (1998). Replication in strategic management: Scientific testing for validity, generalizability, and usefulness. Strategic Management Journal, 19, 243-254. Isard, W. (1956). Location and Space-Economy. Cambridge, MA: MIT Press. Jensen, M. C. & Meckling, W. H. (1976). Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics, 3, 305-360.

204

RESEARCHING ENTREPRENEURSHIP

Kaish, S. & Gilad, B. (1991). Characteristics of opportunities search of entrepreneurs: sources, interests, general alertness. Journal of Business Venturing, 6, 45-61. Katona, G. (1974). Psychology and consumer economics. Journal of Consumer Research, 1, 1-8. Katona, G. (1975). Psychological Economics. New York: Elsevier. Katz, J. & Gartner, W. B. (1988). Properties of emerging organizations. Academy of Management Review, 13(3), 429-441, Keeble, D., Potter, J. & Storey, D. J. (1990). Cross National Comparisons of the Role of SMEs in Regional Economic Growth in the European Community (Working Paper). Cambridge: University of Cambridge & University of Warwick. Keeble, D. & Walker, S. (1994). New firms, small firms and dead firms: Spatial patterns and determinants in the United Kingdom. Regional Studies, 28(4), 411-442. Kilby, P. (1971). Hunting the Heffalump. In P. Kilby (Ed.), Entrepreneurship and Economic Development (pp. 1-40). New York: Free Press. King, G. & Zeng, L. (2001a). Logistic regression in rare events data. Political Analysis, 9(2), 137-163. King, G. & Zeng, L. (2001b). Explaining rare events in international relations. International Organization, 55(3), 693-715. Kirzner, I. M. (1973). Competition and Entrepreneurship. Chicago, IL: University of Chicago Press. Kirzner, I. M. (1983). Entrepreneurs and the entrepreneurial function: a commentary. In J. Ronen (Ed.), Entrepreneurship. Lexington, MA: Lexington Books. Kish, L. (1987). Statistical design for research. New York: Wiley. Kleinknecht, A. (1987). Measuring R&D in small firms: How much are we missing? Journal of Industrial Economics, 36, 253-256. Kleinknecht, A. (1991). Firm size and innovation. Reply to Scheirer. Small Business Economics, 3, 157158. Knight, F. (1921). Risk, Uncertainty and Profit. New York: Hougthon Mifflin. Kogut, B. & Zander, U. (1992). Knowledge of the firm, combinative capabilities, and the replication of technology. Organization Science, 3(3), 383-397. Krueger, N. F. (Ed.). (2002). Entrepreneurship : critical perspectives on business and management. (Vol. 2). London: Routledge. Krueger, N. F., Reilly, M. D. & Carsrud, A. L. (2000). Competing models of entrepreneurial intentions. Journal of Business Venturing, 15(5/6), 411-432. Kufver, B. (1995). Hur definieras entreprenörskap och hur kan det mätas (How is entrepreneurship defined and how can it be measured). Unpublished Master thesis, Jönköping International Business School, Jönköping. Kumar, M. S. (1984). Growth, Acquisition, and Investment. (Occasional Paper). Cambridge, UK: Department of Applied Economics. Little, R. J. A. & Rubin, D. B. (1987). Statistical Analysis with Missing Data. New York: Wiley. Low, M. B. (2001). The adolescence of entrepreneurship research: specification of purpose. Entrepreneurship Theory & Practice, 25(4, Summer), 17-25. Low, M. B. & MacMillan, I. C. (1988). Entrepreneurship: past research and future challenges. Journal of Management(14), 139-161. Lumpkin, G. T. & Dess, G. G. (1996). Clarifying the entrepreneurial orientation construct and linking it to performance. Academy of Management Review, 21(1), 135-172. Lumpkin, G. T. & Dess, G. G. (1997). Proactiveness versus competitive aggressiveness: Teasing apart key dimension of an entrepreneurial orientation. In P. D. Reynolds, W. D. Bygrave, N. M. Carter, P. Davidsson, W. B. Gartner, C. M. Mason & P. P. McDougall (Eds.), Frontiers of Entrepreneurship Research 1997. Wellesley, MA: Babson College. Lynn, R. (1969). An achievement motivation questionnaire. British Journal of Psychology, 60(529-534). Lynn, R. (1991). The Secret of the Miracle Economy. Different National Attitudes to Competitiveness and Money. London: The Social Affaires Unit. Malecki, E. J. (1997). Technology & Economic Development (2nd ed.). Harlow: Addison Wesley Longman. March, J. G. & Sevón, G. (1988). Behavioral perspectives on theories of the firm. In W. F. Van Raaij, G. M. van Veldhoven & K. E. Wärneryd (Eds.), Handbook of Economic Psychology (pp. 368-403). Dordrecht, NL: Kluwer. Markman, G. D. & Baron, R. A. (1998). Social skills and entrepreneurss’ financial success: evidence that the ability to get along with others really matters. In P. D. Reynolds, W. D. Bygrave, N. M. Carter,

References

205

S. Manigart, M. D.M., G. D. Meyer & K. G. Shaver (Eds.), Frontiers of Entrepreneurship Research 1998. Wellesley, MA.: Babson College. McClelland, D. C. (1961). The Achieving Society. Princeton, NJ: Van Nostrand. McCloskey, D. (1998). The Rhetoric of Economics (2 ed.). Madison, WI: The University of Wisconsin Press. McGrath, R. G. (1996). Options and the entrepreneur: Towards a strategic theory of entrepreneurial wealth creation. Paper presented at the Academy of Management, Cincinnati. McGrath, R. G. (1999). Falling forward: Real options reasoning and entrepreneurial failure. Academy of Management Review, 24(1), 13-30. Michell, J. (1986). Measurement scales and statistics: A clash of paradigms. Psychological Bulletin, 100(3), 398-407. Miles, M. B. & Huberman, A. M. (1994). Qualitative Data Analysis (2 ed.). Thousand Oakes: Sage. Miller, D. (1987). The structural and environmental correlates of business strategy. Strategic Management Journal, 8, 55-76. Miller, D. & Friesen, P. H. (1978). Archetypes of strategy formulation. Management Science, 24(9), 921933. Miller, D. & Friesen, P. H. (1982). Innovation in conservative and entrepreneurial firms: two models of strategic momentum. Strategic Management Journal, 3, 1-25. Miller, D. & Shamsie, J. (1996). The resource-based view of the firm in two environments: The Hollywood film studios from 1936 to 1965. Academy of Management Journal, 39(3), 519-543. Miner, A. S. & Mezias, S. J. (1996). Ugly duckling no more: Organizational learning research. Organization Science, 7(1), 88-99. Mintzberg, H. (1974). The Nature of Managerial Work. New York: Harper & Row. Mises, L. (1949). Human Action. New Haven: Yale University Press. Mitchell, R. (1991). Masters of innovation: how 3M keeps its new products coming. In J. Henry & D. Walker (Eds.), Managing Innovation (pp. 171-181). London: Sage. Morris, M. H., Lewis, P. L. & Sexton, D. L. (1994). Reconceptualizing entrepreneurship: An input-output perspective. Advanced Management Journal, 9(Winter), 21-31. Mueller, D. C. (1972). A life cycle theory of the firm. Journal of Industrial Economics, 199-219. Mugler, J. (1990). Entrepreneurship and the theory of the firm. In R. Donckels & A. Miettinen (Eds.), New findings and perspectives in entrepreneurship (pp. 3-15). Aldershot: Avebury. Musgrave, A. (1981). Unreal assumptions in economic theory: the F-twist untwisted. Kyklos, 34, 377-387. Muthén, B. O. & Curran, P. J. (1997). General longitudinal modeling of individual differences in experimental designs: A latent variable framework for analysis and power estimation. Psychological Methods, 2(4), 371-402. Muthén, B. O. & Khoo, S.-T. (1999). Longitudinal studies of achievement growth using latent variable modeling. Learning and Individual Differences, 10, 73-101. Nahapiet, J. & Ghosal, S. (1998). Social capital, intellectual capital, and the organizational advantage. Academy of Management Review, 23(2), 242-266. Nelson, R. R. & Winter, S. G. (1982). An Evolutionary Theory of Economic Change. Cambridge, MA.: Belknap Press. North, C. D. & Thomas, P. R. (1973). The Rise of the Western World. Cambridge: Cambridge University Press. Nunnally, J. C. (1967). Psychometric Theory. New York: McGraw-Hill. Nunnally, J. C. & Bernstein, I. H. (1994). Psychometric Theory (3 ed.). New York, NY: McGraw-Hill. NUTEK (2002). Företagens villkor och verklighet 2002. Dokumentation och svarsöversikt (Conditions and Reality of Small Firms 2002. Documentation and Overview of Responses). Stockholm: NUTEK. Oakes, M. (1986). Statistical Inference: A Commentary for the Social and Behavioural Sciences. Chichester: Wiley. Olofsson, C., Petersson, G. & Wahlbin, C. (1986). Opportunities and obstacles -- a study of start-ups and their development. In R. Ronstadt, J. A. Hornaday, R. Peterson & K. H. Vesper (Eds.), Frontiers of Entrepreneurship Research 1986. Wellesley, MA: Babson College. Penrose, E. (1959). The Theory of the Growth of the Firm. Oxford: Oxford University Press. Pickering, J. F. (1981). A behavioural model of the demand for consumer durables. Journal of Economic Psychology, 1, 59-77. Popper, K. (1992). The Logic of Scientific Discovery (4th ed.). London: Routledge. Porter, M. E. (1980). Competitive Strategy. New York, NY: Free Press.

206

RESEARCHING ENTREPRENEURSHIP

Porter, M. E. (1985). Competitive Advantage. New York, NY: Free Press. Porter, M. E. (1990). The Competitive Advantage of Nations. London: Macmillan. Putnam, R. D. (1995). Bowling alone: America’s declining social capital. Journal of Democracy, 6(1), 6578. Raffa, M., Zollo, G. & Caponi, R. (1996). The development process of small firms. Entrepreneurship and Regional Development. 8, 359-372. Reynolds, P. D. (1994). Autonomous firm dynamics and economic growth in the United States 19861990. Small Business Economics, 28(4), 429-442. Reynolds, P. D. (1999). Creative destruction: source or symptom of economic growth. In Z. J. Acs, B. Carlsson & K. Karlsson (Eds.), Entrepreneurship, Small and Medium-sized Firms and the Macroeconomy (pp. 97-136). Cambridge: Cambridge University Press. Reynolds, P. D. (2000). National panel study of US business start-ups. Background and methodology. In J. A. Katz (Ed.), Advances in Entrepreneurship, Firm Emergence and Growth (Vol. 4, pp. 153-227). Stamford, CT: JAI Press. Reynolds, P. D., Bygrave, W. D., Autio, E., Cox, L. & Hay, M. (2002). GEM Global 2002 Executive Report. Kansas MO.: Kauffman Foundation. Reynolds, P. D., Camp, S. M., Bygrave, W. D., Autio, E. & Hay, M. (2001). Global Entrepreneurship Monitor. 2001 Executive Report. Kansas MO.: Kauffman Foundation. Reynolds, P. D. & Maki, W. R. (1990). Business Volatility and Economic Growth. Final Report. Washington, D.C.: Small Business Administration. Reynolds, P. D. & Miller, B. (1992). New firm gestation: conception, birth and implications for research. Journal of Business Venturing, 7(405-417). Reynolds, P. D., Miller, B. & Maki, W. R. (1993). Regional characteristics affecting business volatility in the United States, 1980-94. In C. Karlsson, B. Johannisson & D. Storey (Eds.), Small Business Dynamics. International, National and Regional Perspectives. (pp. 78-114). London: Routledge. Reynolds, P. D., Storey, D. J. & Westhead, P. (1994). Cross-national comparisons of the variation in new firm formation rates. Regional Studies, 28(4), 443-456. Robinson, J. P., Shaver, P. R. & Wrightsman, L. S. (1991). Criteria for scale selection and evaluation. In J. P. Robinson, P. R. Shaver & L. S. Wrightsman (Eds.), Measures of Personality and Social Psychological Attitudes (pp. 1-16). San Diego, CA: Academic Press. Rogers, E. M. (1995). Diffusion of Innovations (4th ed.). New York: The Free Press. Ronen, J. (1983). Some insights into the entrepreneurial process. In J. Ronen (Ed.), Entrepreneurship (pp. 137-174). Lexington, MA.: Lexington Books. Ross, L. (1977). The intuitive psychologist and his shortcomings: Distortions in the attribution process. In m. A. i. E. S. P. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 10). New York: Academic Press. Rouse, M. J. & Daellenback, U. S. (1999). Rethinking research on methods for the resource-based perspective: Isolating sources of sustainable competitive advantage. Strategic Management Journal, 20, 487-494. Russell, B. (1946). History of Western Philosophy. London: Allen & Unwin. Salvato, C. 2002. Antecedents of Entrepreneurship in Three Types of Family Firms, JIBS Research Reports No. 2002-1. Jönköping: Jönköping International Business School Samuelsson, M. (2001). Modelling the nascent venture opportunity exploitation process across time. In W. D. Bygrave, E. Autio, C. G. Brush, P. Davidsson, P. G. Green, P. D. Reynolds & H. J. Sapienza (Eds.), Frontiers of Entrepreneurship Research 2001 (pp. 66-79). Wellesley, MA. Samuelsson, M. (2004). Creating New Ventures: A Longitudinal Investigation of the Nascent Venturing Process. Doctoral dissertation, Jönköping International Business School, Jönköping. Sarasvathy, S. (1999a), Decision making in the absence of markets: An empirically grounded model of entrepreneurial expertise. School of Business: University of Washington. Sarasvathy, S. (1999b). Seminar on research perspectives in entrepreneurship. Journal of Business Venturing, 15, 1-57. Sarasvathy, S. (2001). Causation and effectuation: towards a theoretical shift from economic inevitability to entrepreneurial contingency. Academy of Management Review, 26(2), 243-288. Sarasvathy, S., Dew, N., Velamuri, R. & Venkataraman, S. (2003). Three views of entrepreneurial opportunity. In A. Z.J. & D. B. Audretsch (Eds.), Handbook of Entrepreneurship Research. Dordrecht, NL: Kluwer. Schafer, D. S. (1990). Level of entrepreneurship and scanning source usage by very small businesses. Entrepreneurship Theory & Practice, 15(2), 19-31.

References

207

Scheinberg, S. & MacMillan, I. C. (1988). An 11 country study of motivations to start a business. In B. A. Kirchhoff, W. A. Long, W. E. McMullan, K. H. Vesper & W. E. Wetzel (Eds.), Frontiers of Entrepreneurship Research 1988 (pp. 669-687). Wellesley, MA: Babson College. Scherer, F. M. (1980). Industrial Market Structure and Economic Performance. Boston: Houghton Mifflin Company. Schumpeter, J. A. (1934). The Theory of Economic Development. Cambridge: MA: Harvard University Press. Schumpeter, J. A. (1954). History of Economic Analysis. London: Allen & Unwin. Seth, A. & Thomas, H. (1994). Theories of the firm: Implications for strategy research. Journal of Management Studies. 3, 165-191. Shane, S. (2000). Prior knowledge and the discovery of entrepreneurial opportunities. Organization Science, 11(4), 448-469. Shane, S. & Eckhardt, J. (2003). The individual-opportunity nexus. In Z. J. Acs & D. B. Audretsch (Eds.), Handbook of Entrepreneurship Research (pp. 161-194). Dordrecht, NL: Kluwer. Shane, S. & Venkataraman, S. (2000). The promise of entrepreneurship as a field of research. Academy of Management Review, 25(1), 217-226. Shane, S. & Venkataraman, S. (2001). Entrepreneurship as a field of research: a response to Zahra and Dess, Singh, and Erikson. Academy of Management Review, 26(1), 13-16. Shapero, A. & Sokol, L. (1982). The social dimension of entrepreneurship. In C. A. Kent, D. L. Sexton & K. H. Vesper (Eds.), The Encyclopedia of Entrepreneurship (pp. 72-90). Englewood Cliffs: NJ: Prentice-Hall. Sharma, P. & Chrisman, J. J. (1999). Toward a reconciliation of the definitional issues in the field of corporate entrepreneurship. Entrepreneurship Theory & Practice, 23(3, Spring), 11-27. Shaver, K. G., Carter, N. M., Gartner, W. B. & Reynolds, P. D. (2001). Who is a nascent entrepreneur? Decision rules for identifying and selecting entrepreneurs in the panel study of entrepreneurial dynamics (PSED) [summary]. In W. D. Bygrave, E. Autio, C. G. Brush, P. Davidsson, P. G. Green, P. D. Reynolds & H. J. Sapienza (Eds.), Frontiers of Entrepreneurship Research 2001 (pp. 122). Wellesley, MA: Babson College. Shaver, K. G., Gartner, W. B., Crosby, E., Bakalarova, K. & Gatewood, E. J. (2001). Attributions about entrepreneurship: A framework for analyzing reasons for starting a business. Entrepreneurship Theory and Practice, 26(2), 5-32. Shrader, R. C. & Simon, M. (1997). Corporate vs. independent new ventures: Resource, strategy and performance differences. Journal of Business Venturing, 12, 47-66. Singh, R. P. (2001). A comment on developing the field of entrepreneurship through the study of opportunity recognition and exploitation. Academy of Management Review, 26(1), 10-12. Smith, N. R. (1967). The Entrepreneur and His Firm: The Relationship between Type of Man and Type of Company. East Lansing, MI: Michigan State University. Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 667-680. Stevenson, H. H. (1984). A perspective of entrepreneurship. In H. H. Stevenson, M. J. Roberts & H. Grousebeck (Eds.), New Business Venture and the Entrepreneur. Boston, MA: Harvard Business School. Stevenson, H. H. & Gumpert, D. E. (1991). The heart of entrepreneurship. In W. A. Sahlman & H. H. Stevenson (Eds.), The Entrepreneurial Venture (pp. 71-80). Boston, MA: Harvard Business School. Stevenson, H. H. & Jarillo, J. C. (1986). Preserving entrepreneurship as companies grow. Journal of Business Strategy, 6, 10-23. Stevenson, H. H. & Jarillo, J. C. (1990). A paradigm of entrepreneurship: Entrepreneurial management. Strategic Management Journal, 11, 17-27. Stinchcombe, A. L. (1965). Social structure and organizations. In J. D. March (Ed.), Handbook of Organizations (pp. 142-193). Chicago, IL: Rand McNally. Storey, D. J. (1994). Understanding the Small Business Sector. London: Routledge. Storey, D. J. (1996). The Ten Percenters. London: Deliotte & Touche. Storey, D. J. (Ed.). (2000). Small Business: Critical Perspectives on Business and Management. Cheltenham: Edward Elgar. Suchman, M. C. (1995). Managing legitimacy: Strategic and institutional approaches. Academy of Management Review, 20, 571-610. Suppes, P. & Zinnes, J. L. (1963). Basic measurement theory. In R. D. Luce, R. R. Bush & E. Galanter (Eds.), Handbook of Mathematical Psychology (Vol. 1, pp. 3-76). New York: Wiley.

208

RESEARCHING ENTREPRENEURSHIP

Tay, R. S. (1998). Degree of entrepreneurship: an econometric model using the ordinal probit model. Journal of Small Business and Entrepreneurship, 15(1), 83-99. Teece, D. J., Pisano, G. & Shuen, A. (1997). Dynamic capabilities and strategic management. Strategic Management Journal, 18(7), 509-533. Thornton, P. H. (1999). The sociology of entrepreneurship. Annual Review of Sociology, 25, 19-46. Ucbasaran, D., Westhead, P. & Wright, M. (2001). The focus of entrepreneurship research: Contextual and process issues. Entrepreneurship Theory & Practice, 25(4, Summer), 57-80. Utterback, J. M. & Abernathy, W. J. (1975). A dynamic model of product and process innovation. Omega, 3, 639-656. Utterback, J. M. & Reitberger, G. (1982). Technology and industrial innovation in Sweden: A study of new technology based firms. Stockholm: CPA/STU. Wagner, J. (2004). Are Young and Small Firms Hothouses for Nascent Entrepreneurs? Bonn, Germany: Forschungsinstitut zur Zukunft der Arbeit. Van de Ven, A. H. (1996). The business creation journey in different organizational settings. Symposium paper presented at the Academy of Management meeting, Cincinnati, August 1996. Van de Ven, A. H., Angle, H. L. & Poole, M. S. (1989). Research on the Management of Innovation: The Minnesota Studies. New York: Harper & Row. Van de Ven, A. H., Polley, D., Garud, R. & Venkataraman, S. (1999). The Innovation Journey. Oxford: Oxford University Press. Van Gelderen, M., Van der Sluis, L. & Jansen, P. (forthcoming). Learning opportunities and learning behaviors of small business starters: relations with goal achievement, skill development, and satisfaction. Small Business Economics. Venkataraman, S. (1996). Some methodological challenges for entrepreneurial process research. Symposium paper presented at the Academy of Management meeting, Cincinnati, August 1996. Venkataraman, S. (1997). The distinctive domain of entrepreneurship research: An editor’s perspective. In J. Katz & J. Brockhaus (Eds.), Advances in Entrepreneurship, Firm Emergence, and Growth (pp. 119-138). Greenwich, CT: JAI Press. Wernerfelt, B. (1984). A resource based view of the firm. Strategic Management Journal, 5, 171-180. Wernerfelt, B. (1995). The resource-based view of the firm: Ten years after. Strategic Management Journal, 16, 171-174. Vesper, K. H. (1991). New venture ideas: do not overlook the experience factor. In W. A. Sahlman & H. H. Stevenson (Eds.), The Entrepreneurial Venture (pp. 73-80). Boston: Harvard Business School. Westhead, P. & Wright, M. (Eds.). (2000). Advances in Entrepreneurship. Cheltenham: Edward Elgar. Wigren, C. (2003). The spirit of Gnosjö. The Grand Narrative and Beyond. Doctoral dissertation, Jönköping International Business School, Jönköping. Wiklund, J. (1998). Small firm growth and performance: Entrepreneurship and Beyond. Doctoral dissertation, Jönköping International Business School, Jönköping. Wiklund, J., Davidsson, P. & Delmar, F. (2003). Expected consequences of growth and their effect on growth willingness in different samples of small firms. Entrepreneurship Theory & Practice, 27(Spring), 247-269. Wiklund, J., Eliasson, C. & Davidsson, P. (2002). Entrepreneurial management and Schumpeterian resource recombination. In W. Bygrave, P. Davidsson, J. Fiet, P. G. Greene, R. Harrison, M. Lerner, G. D. Meyer, J. Sohl & A. Zacharakis (Eds.), Frontiers of Entrepreneurship Research 2002. Wellesley, MA.: Babson College. Williamson, O. E. (1975). Markets and Hierarchies. New York: The Free Press. Williamson, O. E. (1999). Strategy research: governance and competence perspectives. Strategic Management Journal, 20, 1087-1108. Wold, H. (1985). Partial Least Squares. In S. Kotz & N. L. Johnson (Eds.), Encyclopaedia of Statistical Sciences (Vol. 6, pp. 581-591). New York: Wiley. Woo, C. Y., Cooper, A. C. & Dunkelberg, W. C. (1988). Entrepreneurial typologies: Definitions and implications. In B. A. Kirchhoff, W. A. Long, W. E. McMullan, K. H. Vesper & W. E. J. Wetzel (Eds.), Frontiers of Entrepreneurship Research 1988 (pp. 165-176). Wellesley, MA: Babson College. Woo, C. Y., Cooper, A. C. & Dunkelberg, W. C. (2000). The development and interpretation of entrepreneurial typologies. In P. Westhead & M. Wright (Eds.), Advances in Entrepreneurship. (Vol. 2, pp. 143-161). Cheltenham: Edward Elgar. Vroom, V. (1964). Work and Motivation. New York: Wiley.

References

209

Zahra, S. & Dess, G. G. (2001). Entrepreneurship as a field of research: Encouraging dialogue and debate. Academy of Management Review, 26(1), 8-10. Zander, I. (2002). Why does the firm come into existence? Towards an entrepreneurial understanding of the nature of the firm. Working paper. Stockholm: Institute of International Business; Stockholm School of Economics.

INDEX Abstraction, in research theory, 34–39 Acquisition growth, organic growth, contrasted, 170–171 Activity, purposeful, as entrepreneurship, 3 Aggregate levels, operationalization on, 135–138 Ajzen, I., theory of planned behavior, 40, 44 Analysis, levels of, in entrepreneurship research, 28–30 Analysis method, 189–194 employee well-being, importance of, 189–192 event history analysis, 191–192 heterogeneity and, 189–190 independence, importance of, 189–192 latent class analysis, 190 longitudinal growth modeling, 191–192 minority nature of entrepreneurship, analysis implications, 190–191 process, entrepreneurship as, analysis implications, 191–192 Arbitrageur, entrepreneur as, 2 Assessment of resources, operationalization, 118–119 Assumption of risk, as role of entrepreneur, 2 Behavior entrepreneurship as type of, 3 operationalizing, 125–129 Beliefs of individuals, as human capital, 116, 118 Bias, sampling, 76 Business Dynamics in Sweden project, 83, 86, 92, 96, 143, 145, 149, 152, 155, 156 Business Platform study, 83 Capital. See under specific type of capital Change in marketplace, introduction of new economic activity leading to, entrepreneurship as, 8 Change-inducing micro-level initiatives, contexts in which taken, 7 Changes of firm’s legal form, to establishment/acquisition of daughter company, 151

Checking quality in preparation of secondary data set, 153–155 Chronbach Alpha, 106–109 Collaboration, harmonized research, 180–184 Combination, of different sources of data, in preparation of secondary data set, 145–153 Commercial sector, entrepreneurship as restricted to, 3 Commitment, as capital, 116, 118 Company, sampling, 84 Company group, sampling, 84 Competence-destroying innovations, 122 Competitive behaviors driving market process, entrepreneurship as, 6, 7 Competitor, new, as entrepreneurship, 9–10 Concept compatibility, abstraction of, 35 Continued Entrepreneurship and Small Firm Growth study, 81, 82 Continuing firm, through changes of legal form to establishment/acquisition of daughter company, 151 Contraction job losses, expansion job gains, by firm size, annual, 169 Control internal locus of, 61–62 sense of, with expansion, job satisfaction and, 182 Coordinator, entrepreneur as, 2 Corrections, in preparation of secondary data set, 153–155 Creating value, through entrepreneurship, 2 Creation of opportunities, entrepreneurship as involving, 3 Culture, market channels, as organizational capital, 119 Culture and Entrepreneurship study, 92–98, 135–138 research questions, 93 Data set, secondary, preparation of, 141–158 branch establishment type, 147 changes, continuing firm through, 151 corrections, 153–155

212

RESEARCHING ENTREPRENEURSHIP

development of data set, 142–156 different sources of data, combination of, 145–153 establishment types, identification of, 147 growth, total, organic for high growth firms of different age, 153 for high growth firms of different size, 153 method fallacies, 141 prior knowledge, usage of, 144–145 quality, checking of, 153–155 simple establishment type, 147 top establishment type, 147 Declarative knowledge, as capital, 116, 118 Definition, entrepreneurship, 1–16 Definitional keywords, entrepreneurship, 2 Degrees of entrepreneurship, 14–15 Delineations of entrepreneurship phenomenon, 5 Dependence variable, job creation as, 158–174 Design, 55–66 emergence of new ventures, study of processes of, 61–63 initiation of research, 55 laboratory research methods, 63–64 qualitative entrepreneurship research, 55–61 need for, 55–57 quantitative entrepreneurship research, compared, 57–59 quantitative entrepreneurship research, 55–61 quantitative questions, addressing with qualitative research, 59–61 scope of design possibilities, 61 Determinants of firm start-up rates, regional, 180 Different sources of data combination of, in preparation of secondary data set, 145–153 combining, in preparation of secondary data set, 145–153 Differing views of entrepreneurship, 3 Direct impact on economic system, entrepreneurship and, 15 Discovery, exploitation, interrelation between, 24 Discovery of opportunities, entrepreneurship as involving, 3 Discriminant validity, 106

Distal explanatory variables, proximal, distinction between, 113 Domain delineation, entrepreneurship research, 18–21 Dynamic analysis, static comparison, contrasted, 163–164 Eclecticframework as tool for interpretation, 47–48 understanding through, 44–46 Economic development, job creation, distinguished, 170–171 Economic theorists, roles assigned to entrepreneur by, 2 Emerging new ventures operationalization, 115–131 sampling, 72–80 on–going independent venture start-ups, 73–77 on-going internal venture start-ups, 77–80 sampling bias, 76 study of processes of, research design, 61–63 Empirical firm, defined, 83 Empirical generalization, 34 Employee well-being with expansion, job satisfaction and, 181 importance of, 189–192 Employer, entrepreneur as, 2 Enterprise, sampling, 84 Entrepreneurial orientation scale, 110, 132 sample items from, 110 Entrepreneurship analysis method, 189–194 defined, 1–16 variety of definitions, 1–6 design of, 55–66 job creation, 158–174 operationalization issues, 101–140 replication, 175–188 as research domain, 17–32 sampling, 67–100 “secondary” data set, 141–158 theoretical analysis of, 33–54 view of phenomenon, 6 Entrepreneurship in Different Organizational Contexts study, 81, 82, 86, 112, 132–134, 186 Environment, working, 181 Establishment types, identification of, in preparing data set, 147 Event history analysis, 191–192

Index Expansion, effect on job satisfaction, 181 Expansion job gains, contraction job losses, by firm size, annual, 169 Expectation-value theory, of motivation, 109 Expected consequences, effect on growth attitude, 182 Expert vs. novice entrepreneurs, study of, 71 Exploitation discovery, interrelation between, 24 entrepreneurship and, 23–24 Exploitation of opportunities, entrepreneurship as involving, 3 External environment, operationalizing, 124–125 Factor analysis results entrepreneurial management, 187 stability across sampling strata, 188 Fallacies, method, in preparing secondary data set, 141 Financial capital, 115, 118 Firm level, operationalization on, 131–134 Firm start-up rates, regional determinants, 180 Firms, sampling, 80–89 company, 84 company group, 84 empirical firm, defined, 83 heterogeneity, 80–83 hierarchies of, 85 legal forms, 83 limited liability companies, 83 multi-company corporation, 84 partnerships, 83 proprietorships, 83 relevance issues, 83–89 size, 80–83 size distribution, 80–83 Gains in expansion jobs, contraction job losses, by firm size, annual, 169 Gap-filler, entrepreneur as, 2 Geographical market expansion as entrepreneurship, 10 Goods, tangible. See Physical capital Goods controlled by entrepreneur. See Physical capital Gross job creation net, distinguishing, 141, 164–166 net job creation, distinguishing, 141, 164–166 Growth, total, organic for high growth firms of different age, 153

213

for high growth firms of different size, 153 Growth attitude, effect of expected consequences, 182 Harmonized research collaboration, 180–184 Heterogeneity analysis method and, 189–190 entrepreneurship research, 22–23 Hierarchies of firms, 85 High Growth Firms study, 81, 83, 86, 143, 144, 148, 155, 156 Human capital, 116, 118–119 Idea, venture, 121–124 classification items, 122 state of development, 103 Identification of establishment types, in preparing data set, 147 Impact on economic system, degree of, 15 Income, with expansion, job satisfaction and, 182 Independence importance of, 189–192 job satisfaction and, 189–192 with expansion, 182 Independent venture start-ups on-going, sampling, 73–77 sampling, 73–77 Indicators of entrepreneurial activities, 112 Indirect impact on economic system, entrepreneurship and, 15 Individual level, operationalization on, 113–114 Individuals, sampling, 70–72 business idea fit, 71–72 expert vs. novice entrepreneurs, study of, 71 structural, situational factors, influence on entrepreneurial behavior, 72 types of studies, 71–72 Industries, sampling, 89–92 relevance issues, 91–92 size, 89–91 Information, source of, entrepreneur acting as, 2 Inherent meanings, entrepreneurship, 2. See also Definition, entrepreneurship Initiation of research, 55 Innovation competence-destroying, 122 by entrepreneur, 2 Innovator, entrepreneur as, 2 Insight, as human capital, 116, 118

214

RESEARCHING ENTREPRENEURSHIP

Intelligence. See also Human capital as human capital, 116, 118 Intentions, entrepreneurial, determinants of, 185 Internal locus of control, 61–62 Internal venture start-ups on-going, sampling, 77–80 sampling, 77–80 Internal ventures, search processes leading to, operationalization of, 126 Interrelated process of behaviors, entrepreneurship as, 23–24 Interval scales, operationalization, 102 Intrinsic nature of entrepreneurship, 3 Introduction of new economic activity, leading to change in marketplace, entrepreneurship as, 8 Issues on which views on entrepreneurship differ, 3

declarative, as capital, 116, 118 as human capital, 116, 118 procedural, as capital, 116, 118 Labor, as capital, 116, 118 Labor market area, as regional unit, 93–97 Labor service, entrepreneur’s role in, 2 Laboratory research methods, 63–64 Latent class analysis, 190 Latent variables, in operationalizations, 109 Leadership, provision of, by entrepreneur, 2 Legal form of firm, to establishment/acquisition of daughter company, 151 Legal forms of firms, 83 Levels of analysis in entrepreneurship research, 28–30 Limited liability companies, sampling, 83 Locus of control, 61–62 Longitudinal growth modeling, 191–192

Job creation, 158–174 artificial changes, 172 data coverage, 160–163 economic development, distinguished, 170–171 expansion job gains, contraction job losses, by firm size, annual, 169 gross, net, distinguishing, 141, 164–166 gross job creation, net job creation, distinguishing, 141, 164–166 length of analysis period, 172 micro level unit, choice of, 171–172 organic growth, vs. acquisition growth, contrasted, 170–171 regression (to-the-mean) effect, 141, 166–170 significance of issue, 159–160 simulation results, 161 size class, assignment to new entrants, 172 size development for firms, hypothetical, 167 static comparison, dynamic analysis, contrasted, 163–164 Job satisfaction, 181 effect of expansion, 181 independence, 189–192 with expansion, 182 Judgment. See also Human capital as human capital, 116, 118

Manager, entrepreneur acting as, 2 Market channels, as organizational capital, 119 Market context, entrepreneurship in, 6–7 Market process competitive behaviors driving, entrepreneurship as, 6 entrepreneurship driving, 7 McClelland, David, 108 Meanings, inherent, entrepreneurship, 2 Measurement levels, operationalization, 101–105 scales corresponding to, 101–102 Method fallacies, in preparing secondary data set, 141 Micro level unit, choice of, 171–172 Micro-level behavior change-inducing, contexts in which taken, 7 entrepreneurship as, 6 with macro-level implications, entrepreneurship as, 12–14 Micro-level phenomenon, entrepreneurship as, 3 Minority nature of entrepreneurship, analysis implications, 190–191 Money resources. See Financial capital Multi-company corporations, sampling, 84

Katona, George, 189 Keywords, definitional, entrepreneurship, 2 Knowledge. See also Human capital

Nature of entrepreneurship, 3 Need for entrepreneurship theory, 51–52 Net job creation

Index gross job creation, distinguishing, 141, 164–166 New firms, entrepreneurship as restricted to, 3 New Internal Ventures study, 81, 122, 125 New ventures, sampling, 72–80 on-going independent venture start-ups, 73–77 on-going internal venture start-ups, 77–80 sampling bias, 76 Newness of economic activities, firm/market, 8 1994 Start-up Cohort study, 121, 131 Nominal scales, 101–102 Non-entrepreneurial growth, 12 Novelty to market/actor, entrepreneurship and, 15 Novice entrepreneurs, vs. expert, study of, 71 Offer, new, as entrepreneurship, 9 Operationalization, 101–140 on aggregate levels, 135–138 assessment of resources, 118–119 balancing exercises, 110–113 behaviors, operationalizing, 125–129 commitment, as capital, 116, 118 competence-destroying innovations, 122 declarative knowledge, as capital, 116, 118 distal, proximal explanatory variables, distinction between, 113 entrepreneurial orientation scale, 110, 132 Entrepreneurship in Different Organizational Contexts study, 81, 82, 86, 112, 132–134, 186 external environment, operationalizing, 124–125 financial capital, 115, 118 on firm level, 131–134 human capital, 116, 118–119 indicators of entrepreneurial activities, 112 on individual level, 113–114 interval scales, 102 labor, as capital, 116, 118 on level of new, emerging venture, 115–131 measurement levels, 101–105 scales corresponding to, 101–102 nominal scales, 101–102 ordinal scales, 102 organizational capital, 119, 120 outcomes, operationalizing, 129–131 physical capital, 115, 118 procedural knowledge, as capital, 116, 118

215

ratio scales, 102 on regional level, 136 relational resources, embedded in personal/organizational ties, 119 reliability issues, 105–109 resources operationalizing, 115–121 types of, 118–119 search processes leading to new, internal ventures, operationalization of, 126 social capital, 119, 120 Stevenson, Howard, 133–134 summated indices, types of, 109 theoretical concepts, relationships between, 106 validity issues, 105–109 venture idea, 103, 121–124 classification items, 122 venture level model, entrepreneurship concepts, 117 Opinion polls, social science, contrasted, 68–70 Opportunities, entrepreneur’s alertness to, 2 Ordinal scales, 102 Organic growth acquisition growth, contrasted, 170–171 vs. acquisition growth, contrasted, 170–171 Organization creation, entrepreneurship as, 2 Organizational capital, 119, 120 Organizational changes, entrepreneurship, contrasted, 11–12 Organizational contexts, 25–26 Organizational/personal ties, relational resources embedded in, 119 Organizer, entrepreneur as, 2 Outcome, entrepreneurship as type of, 3 Outcomes different levels, for new ventures, 13 operationalizing, 129–131 Owner-managed firms, entrepreneurship as restricted to, 3 Owner-manager, entrepreneur characterized as, 2 Ownership changes, entrepreneurship, contrasted, 11–12 Panel Study of Entrepreneurial Dynamics, 73–79, 131 Partnerships, sampling, 83 Perceptive abilities, as human capital, 116, 118

216

RESEARCHING ENTREPRENEURSHIP

Personal income, with expansion, job satisfaction and, 182 Personal/organizational ties, relational resources embedded in, 119 Phenomenon, distinguished, 17–18 Physical capital, 115, 118 Planned behavior, theory of, graphical representation, 40 Populations, sampling, 89–92 heterogeneity, 89–91 relevance issues, 91–92 size, 89–91 size distribution, 89–91 Post hoc theorizing, 48–50 Power of replication, 175–188. See also Replication Premises controlled by entrepreneur. See Physical capital Preparation of secondary data set, 141–158 changes, continuing firm through, 151 corrections, 153–155 development of data set, 142–156 establishment types, identification of, 147 Prior knowledge, usage of, in preparation of secondary data set, 144–145 Procedural knowledge, as capital, 116,118 Process, entrepreneurship as, analysis implications, 191–192 Processes of emergence, in entrepreneurship research, 23–24 Product quality, with expansion, job satisfaction and, 182 Proprietorships, sampling, 83 Proximal explanatory variables, distal, distinction between, 113 PSED. See Panel Study of Entrepreneurial Dynamics Pure theory test, 39 Purposeful activity, as entrepreneurship, 3 Qualitative entrepreneurship research, 55–61 addressing quantitative questions with, 59–61 need for, 55–57 quantitative entrepreneurship research, compared, 57–59 Quality checking of, in preparation of secondary data set, 153–155 with expansion, job satisfaction and, 182 Quantitative entrepreneurship research, 55–61

addressing with qualitative research, 59–61 qualitative entrepreneurship research, compared, 57–59 Reasoned-action theory, 109 Regional determinants, of firm start-up rates, 180 Regional level, operationalization on, 136 Regression (to-the-mean) effect, 141, 166–170 Regression (to-the-mean) fallacy, 141, 166–170 Rejection of research theory, 50 Relational resources, embedded in personal/organizational ties, 119 Reliability issues, operationalization, 105–109 Replication, 175–188 entrepreneurial intentions, determinants of, 185 expansion, effect on job satisfaction, 181 factor analysis results entrepreneurial management, 187 stability across sampling strata, 188 growth attitude, effect of expected consequences, 182 harmonized research collaboration, 180–184 job satisfaction, 181 of one another, 180–184 of others, 176–179 regional determinants, firm start-up rates, 180 sampling, 175–176 of self, 184–188 significance testing, 175–176 work environment, 181 Reputation of organization, as organizational capital, 119 Research, in entrepreneurship analysis method, 189–194 defining entrepreneurship, 1–16 variety of definitions, 1–6 delineation of domain, 18–21 design, 55–66 discovery behaviors in interrelated process, 23–24 exploitation, interrelation between, 24 domain of entrepreneurship, 17–32 emergence, processes of, 23–24 exploitation, behaviors in interrelated process, 23–24

Index heterogeneity, 22–23 job creation, 158–174 levels of analysis, 28–30 operationalization issues, 101–140 organizational contexts, 25–26 phenomenon, distinguished, 17–18 replication, 175–188 research domain, entrepreneurship as, 17–32 sampling, 67–100 “secondary” data set, 141–158 theoretical analysis, 33–54 uncertainty, 22–23 venture ideas, contextual fit, 26–28 Resources operationalizing, 115–121 types of, 118–119 Response rates, in sampling, 98–99 Reynolds, Paul, 180 Risk, assumption of, as role of entrepreneu, 2 Risk-taking, as component of entrepreneurship, 3 Role of theory in research, 39–50 Roles assigned to entrepreneur by economi theorists, 2 Routines, in which new venture evolves, as organizational capital, 119 Sampling, 67–100, 175–176 bias, emerging new ventures, sampling, 76 emerging new ventures, 72–80 bias in sampling, 76 on-going independent venture start-ups, 73–77 on-going internal venture start-ups, 77–80 firms, 80–89 company, 84 company group, 84 empirical firm, defined, 83 heterogeneity, 80–83 hierarchies of, 85 legal forms, 83 limited liability companies, 83 multi-company corporation, 84 partnerships, 83 proprietorships, 83 relevance issues, 83–89 size, 80–83 size distribution, 80–83 individuals, 70–72 business idea fit, 71–72

217

expert vs. novice entrepreneurs, study of, 71 structural, situational factors, influence on entrepreneurial behavior, 72 types of studies, 71–72 industries, 89–92 relevance issues, 91–92 size, 89–91 populations, 89–92 heterogeneity, 89–91 relevance issues, 91–92 size, 89–91 size distribution, 89–91 response rates, 98–99 social science, opinion polls, contrasted, 68–70 spatial units, 92–98 culture and, 93–96 research questions, 93 heterogeneity, 95–98 relevance issues, 93–95 size, 95–98 size distribution, 95–98 Satisfaction with job, effect of expansion, 181 Scope of research design possibilities, 61 Screening sample, 74 Search processes leading to new, internal ventures, operationalization of, 126 Secondary data set preparation, 141–158 branch establishment type, 147 changes, continuing firm through, 151 corrections, 153–155 development of data set, 142–156 different sources of data, combination of, 145–153 establishment types, identification of, 147 growth, total, organic for high growth firms of different age, 153 for high growth firms of different size, 153 method fallacies, 141 prior knowledge, usage of, 144–145 quality, checking of, 153–155 simple establishment type, 147 top establishment type, 147 Self replication, 184–188 Sense of control. See also Locus of control with expansion, job satisfaction and, 182 Service, labor, entrepreneur’s role in, 2 Service quality, with expansion, job satisfaction and, 182

218

RESEARCHING ENTREPRENEURSHIP

Significance testing, 175–176 Simple establishment type, 147 Size class, assignment to new entrants, 172 Size development for firms, hypothetical, 167 Size distribution, effect of, 82 Size distribution fallacy, 141 Small firms, entrepreneurship as restricted to, 3 Social capital, 119, 120 Social learning theory, 49 Social realities, as themes in entrepreneurship definitions, 3–5 Social science, opinion polls, contrasted, 68–70 Sorted array, factors possibly affecting entrepreneurship, 38 Source of information, entrepreneur acting as, 2 Sources of data, combining, in preparation of secondary data set, 145–153 Spatial units, sampling, 92–98 culture and, 93–96 research questions, 93 heterogeneity, 95–98 relevance issues, 93–95 size, 95–98 size distribution, 95–98 Speculator, entrepreneur acting as, 2 Start-up Cohort study, 81, 83 Start-up rates, regional determinants, 180 Static comparison, dynamic analysis, contrasted, 163–164 Stevenson, Howard, 133–134 Storey, David, 180 Structural, situational factors, influence on entrepreneurial behavior, 72 Subjective interpretation, of respondents, influence on data, 74–75 Summated indices, types of, 109 Superintendent, entrepreneur acting as, 2 Support of research theory, 50 Tangible goods. See Physical capital Tasks, with expansion, job satisfaction and, 181 Theory in entrepreneurship research, 33–54 abstraction, need for, 34–39 factors possibly affecting entrepreneurship sorted array, 38 unsorted array, 37

as guide to research design eclectic framework, understanding through, 44–46 theory test, 39–44 need for entrepreneurship theory, 51–52 overview of theory, 33–34 rejection of, 50 role of theory in research, 39–50 support of, 50 theory of planned behavior, 40, 44 as tool for interpretation eclectic framework approach, 47–48 post hoc theorizing, 48–50 theory test, 46–47 Theory of planned behavior, 40, 44 graphical representation, 40, 44 Theory test, 39–44, 46–47 Ties, organizational/personal, relational resources embedded in, 119 Uncertainty, impact, 22–23 Uniqueness, as characteristic of entrepreneur, 2 Unsorted array, factors affecting entrepreneurship, 37 Validity issues, operationalization, 105–109 Values of individuals, as human capital, 116, 118 Variety of definitions of entrepreneur, 1–6 Venture idea, 121–124 classification items, 122 contextual fit, 26–28 state of development, 103 Venture level model, operationalization, entrepreneurship concepts, 117 View of entrepreneurship phenomenon, 6 Views of entrepreneurship, differences among, 3 Vulnerability, with expansion, job satisfaction and, 182 Well-being of employee, with expansion, job satisfaction and, 181 Will, exercise of, by entrepreneur, 2 Work environment, 181. See also Job satisfaction Work tasks, with expansion, job satisfaction and, 181 Workload, with expansion, job satisfaction and, 181