ESSAYS ON TECHNOLOGY ENTREPRENEURSHIP, Study notes of Entrepreneurship

Download ESSAYS ON TECHNOLOGY ENTREPRENEURSHIP and more Study notes Entrepreneurship in PDF only on Docsity! ESSAYS ON TECHNOLOGY ENTREPRENEURSHIP A Thesis Presented to The Academic Faculty by Anak Agung Istri Shanti Dewi In Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the College of Management Georgia Institute of Technology May 2011 ESSAYS ON TECHNOLOGY ENTREPRENEURSHIP Approved by: Dr. Marie C. Thursby, Advisor College of Management Georgia Institute of Technology Dr. Jerry G. Thursby College of Management Georgia Institute of Technology Dr. Benjamin D. Herndon College of Management Georgia Institute of Technology Dr. John P. Walsh School of Public Policy Georgia Institute of Technology Dr. Frank T. Rothaermel College of Management Georgia Institute of Technology Date Approved: 31 March 2011 ‘lucky pencil’ so that each of us has an identical mechanical pencil to do analytical modeling. She introduced me to important tools that support modeling, including Mathematica, Scientific Workplace, and Winedt. Her cheerfulness makes exhaustion from all day work disappears. This graduate program would not be wonderful without Lin Jiang, Dezhi Yin (Denny), Ong-Ard, Chih-Hung Peng, Gamze Koseoglu, Wen Na (Amy), Sarah Liu, Stephen He, Wen Wen, Fabiana Vincentin and my colleagues in the strategy program. Ann Scott at the graduate office has been supportive. I am indebted to the database team in Kanayakan for the excellent assistance in construct- ing the database for this dissertation. I am grateful for the financial support from Greater Atlanta Regional Grants for the Study of Entrepreneurship and Productivity. Last but not least, I would like to thank my family who has been a constant support. Without their love and encouragement, I would not be able to undergo the ups and downs of the graduate program, and see the preciousness in both. My mother has kindly stayed with me for the last one year of the program. My father and Gung Kak are constant source of inspirations. Widhar and Djuanita are continuously enthusiastic and encouraging. Mbak Siti has greatly taken care of my family and the database team in Indonesia. During my stay in Atlanta, I am blessed for meeting wonderful people who have become my family: Yeny Hudiono, Gracy Wingkono and Robert McMannis, Mariefel Olarte, Chandana Karnati, Mario Martins and Irene Saldana, Agnes Isnawangsih and family, Adi, Lily, and Zea Karmadi, Bie Iesje and Mang Emil, and Made’s family. v TABLE OF CONTENTS DEDICATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . iv LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x I OVERALL RESEARCH GOAL AND IMPLICATIONS . . . . . 1 1.1 Scientific system and its interaction with commercial system . . . . 3 1.2 New ventures as mechanisms that transform scientific investment into commercial outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . 9 II TECHNOLOGY TRANSFER AND THE SOURCES OF RESEARCH FUNDING: IMPLICATIONS FOR THENATURE OF RESEARCH 15 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.2 Environment and Payoffs . . . . . . . . . . . . . . . . . . . . . . . . 17 2.3 Stage Two Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.4 Stage One Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.5 The University’s Shares of Research Funding . . . . . . . . . . . . . 42 2.6 Direct Benefits of Research Funding . . . . . . . . . . . . . . . . . . 44 2.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 III ACADEMIC SCIENTISTS: THE NATURE OF RESEARCH AND ENTREPRENEURIAL ACTIONS . . . . . . . . . . . . . . . . . . . 48 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.2.1 The Academic Scientist . . . . . . . . . . . . . . . . . . . . . 50 3.2.2 The University . . . . . . . . . . . . . . . . . . . . . . . . . . 52 3.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.3.1 Research Setting . . . . . . . . . . . . . . . . . . . . . . . . . 57 vi 3.3.2 Data and Sample . . . . . . . . . . . . . . . . . . . . . . . . 57 3.3.3 Variables and Measures . . . . . . . . . . . . . . . . . . . . . 58 3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 3.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 IV TEAM FORMATIONS IN TECHNOLOGY VENTURES . . . . 71 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.2 The Base Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.3 Team of n-people . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.4 Asymmetric Importance between Issues . . . . . . . . . . . . . . . . 81 4.5 Specialization and Diversity . . . . . . . . . . . . . . . . . . . . . . . 84 4.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 APPENDIX A — PROOFS . . . . . . . . . . . . . . . . . . . . . . . . 89 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 vii SUMMARY This dissertation attempts to contribute to extant discussions on how one utilizes knowledge for economic gain. In order to understand how one can bene- fit from exploiting knowledge, scholars have examined the innovation process. The mainstream view of innovation is that it is a process of knowledge recombination. Consistent with this view, two related issues are whether there is sufficient stock of knowledge to recombine, and what mechanisms there are for knowledge recombina- tion. This dissertation addresses these issues in three essays. The first essay, titled ‘Technology Transfer and the Sources of Research Funding: Implications for the Na- ture of Research’, addresses the former issue. The second essay, titled ‘Academic Scientists: Their Nature of Research and Entrepreneurial Actions’, and the third essay, titled ‘Team Formations in Technology Ventures’ undertake the latter issue. The first essay is a response to the controversies in the growing interaction between the realm of science and the realm of commercialization. One of the controversies is whether the interactions divert academic scientists research agenda toward industry interests at the expense of fundamental science. This essay considers how an academic scientist chooses the level of difficulty of a research project and its level of relevance to industry interests. The direct cost of doing science is incorporated into the scientists decision. A simple game-theoretic model between research sponsors, a government agency and a firm, and an academic scientist is constructed. The model shows that the funding decisions of research sponsors are strategic substitutes. It also shows that the academic scientists choices of project characteristics are strategic complements. The model proposes situations in which an academic scientist pursues challenging projects that are relevant to the firms interests. It also proposes situations in which x an academic scientist decides on projects that are less challenging and less relevant to the firms interests. The second essay provides insights on scientific entrepreneurs. While science- based entrepreneurship has become an increasingly important source of innovation, understanding of who these scientific entrepreneurs are is limited. Therefore, this es- say examines which academic scientists will be more likely to create new technology ventures. It is argued that the nature of scientists research, specifically the level of its commercial applicability, is an important predictor of entrepreneurial actions of academic scientists. Using data from 395 academic scientists at five top US research universities, it is observed that there is a non-linear relationship between the nature of research and entrepreneurial actions. An inverted-u shape relationship between the level of commercial applicability and the likelihood that academic scientists will create new ventures is found in the field of non-life science. In the field of life sci- ence, a decreasing relationship between the level of commercial applicability and the likelihood that academic scientist will create new ventures is observed. These results support the view that scientific human capital is heterogeneous in converting scientific result into commercial outcomes. The third essay offers insights on entrepreneurial teams. Despite the prevalence of entrepreneurial teams, insights on individual entrepreneurs are more available than understanding on entrepreneurial teams. This essay investigates mechanisms that give rise to entrepreneurial teams. A simple model is constructed. The model shows that an entrepreneur obtains less expected value from a project if the entrepreneur chooses to work solo at latter stage than working in a team. The effects of economic value, probability of failure, and cooperation cost on the timing of team formation are presented. It is also explained how asymmetry of importance between tasks in a commercialization project influences the decision of team formation and its opti- mal size. An extended model is constructed to analyze two benefits of team work: xi specialization and diversity. This model proposes that greater probability of failure does not necessarily increase propensity to form entrepreneurial teams. The situa- tions in which the likelihood of team formations increases with probability failure are discussed. xii 1.1 Scientific system and its interaction with commercial system While it is accepted that science plays an important role in technological progress and economic growth, the implications of the growing interaction between the realm of science and commercialization remain controversial (Nelson, 2004). The heart of the controversy is as follows. The workings of scientific system involve distinctive norms that are essential to the production of knowledge, such as the rule for prior- ity and communalism1. Interaction with the commercial realm exposes the scientific system to the norms associated with commercialization, including the primacy of pri- vate knowledge and to pecuniary rewards. Because these norms are contradictory to those of the scientific system (Dasgupta and David, 1994), the issue arises of whether such interaction is detrimental to the scientific system, especially in its function as the knowledge producer for society (Dasgupta and David, 1994; Siegel, Wright, and Lockett, 2007). Scholars have examined this issue from several angles. The first angle is that because economic rents depend on one’s capacity to keep information private, the growing involvement of academic scientists in commercial activity will hamper the process of generating knowledge and, hence, long-term economic growth. The ar- gument is that exclusion delays or prevents scientist access to, or use of, existing knowledge, which is an important component of potential new knowledge (e.g., Blu- menthal et al, 1996; Heller and Eisenberg, 1998; Murray and Stern, 2007; Walsh, Cohen, and Cho, 2007; Thursby and Thursby, 2008; Hong and Walsh, 2009). The second angle is that commercialization activities may divert academic scientists’ at- tention from their main mission of teaching and research. Two outcomes of diversion 1Stephan (1996) and Dasgupta and David (1994) provide insightful details regarding how the reward system in science, the priority of discovery and the winner-takes all, encourages knowledge creation and disclosure. 3 that have emerged as major concerns are effects on research output and research ori- entation. However, previous studies have not found evidence that commercialization activities result in a decline in research output. Agrawal and Henderson (2002) find no relationship between patenting activities and publication output among academic scientists in the Mechanical Engineering and Electrical Engineering departments at MIT. Moreover, empirical studies have shown that a complementary relationship ex- ists between commercialization and research output2 (e.g., Fabrizio and Di Minin, 2008; Buenstorf, 2009). Related evidence based on patenting and disclosure activities indicates that the greater involvement of academic scientists in commercialization is driven by higher research output (e.g., Azoulay, Ding, and Stuart, 2007; Thursby and Thursby, 2009a). Regarding the effects of commercialization on research orientation, it is concerned that academic scientists pursue research projects that have commer- cial applications at the expense of basic research (Vavakova, 1998; Pogayo-Theotoky, Beath, and Siegel, 2002; Campbell et al., 2005; Geuna and Nesta, 2006). Chapter 2 shares with existing studies examining the diversion of academic sci- entists’ attention. The focus of the discussion will be on research orientation rather than research output. The search for evidence of whether basic research is neglected in favor of applied research and of why this might occur is especially complicated because it is possible to pursue research problems that are both basic and applied (Stokes, 1997). Stokes argues that in such situations, known as Pasteur’s Quadrant, applied research is also fundamental in nature. Empirical studies intended to detect changes in research orientation present mixed results (Cohen et al., 1998). However, these studies have not indicated that basic research is being neglected. Some em- pirical studies demonstrate that academic scientists adapt their research to applied 2Fabrizio and Di Minin (2008) find that academic patenting correlates positively with the publi- cation output. Using licensing and spin-off activities as measures of commercial activities, Buenstorf (2009) observes a similar pattern between commercialization and research output. 4 science and to commercially useful problems3 (e.g., Rahm, 1994; Van Looy et al., 2004; Gulbrandsen and Smeby, 2005; Azoulay, Ding, and Stuart, 2009). Other stud- ies find no evidence of such a change4 (e.g., Ranga, Debackere, and Von Tunzelmann, 2003; Van Looy, Callaert, and Debackere, 2006; Thursby and Thursby, 2007; Thursby and Thursby, 2009b). In general, empirical and theoretical studies of research orientation involve five positions. The first position is that the prospect of commercial profit encourages academic scientists to pursue research projects that have commercial applications regardless of the projects’ contribution to fundamental knowledge (Krimsky et al., 1996; Hane, 1999; Campbell et al., 2005). The second position is that academic 3In a survey to the top 100 US research universities, Rahm (1994) uncovers that 41% of academic scientists express the fear of negative impact of technology transfer activities on the basic research orientation of their universities. The 59% of academic scientists in their sample do not perceive such interference. Their survey also reveals that 24% of university administrators fear the interference while 76% of the university administrators do not. Van Looy et al. (2004) use division memberships as the measures of entrepreneurial involvement. Comparing academic scientists who are members of divisions and those who do not, they find that the former group published more papers in applied- oriented journals than the later. However, both groups’ publications in basic-oriented journals are comparable. In a survey on academic scientists at Norway universities, Gulbrandsen and Smeby (2005) observe that about 50% academic scientists funded by industry characterize their research as applied. About 40% of the academic scientists funded by industry characterize their research as basic. In comparison, they find that approximately 62% of academic scientists funded by non-industry (other external fund) characterize their research as basic and approximately 25% of them identify their research as applied. Azoulay, Ding, and Stuart (2009) constructed a sample of 3,862 academic scientists in life science. On these academic scientists, they find patenting activities positively correlate with commercial content of research publications. Their result is robust on three measures of commercial content: the ”patentability” of research publications, the co-authorship with industry affiliated researchers and the Journal Commercial Score (i.e., the proportion of industry affiliated authors publishing in a journal). 4Observing publications from year 1985 to 2000 of academic scientists in KU Leuven, Ranga, Debackere, and Von Tunzelmann (2003) find a small dominance of publications in basic-oriented journals to publications in applied-oriented journals. Comparing academic scientists who involved in patenting and those who did not, Van Looy, Callaert, and Debackere (2006) show that the former group published more papers in basic oriented-journals than the later. Using a dataset of academic scientists in six major US universities over a seventeen-year period, Thursby and Thursby (2007) find that there has been no change in the proportion of research published in basic-oriented journals to research published in applied-oriented journals. Based on a sample from 11 major universities in the US, Thursby and Thursby (2009b) use disclosures as the measures of academic scientists’ involvement in licensing. They observe that academic scientists who disclosed have a higher number of publications in basic-oriented journals than academic scientists who never disclose. Their findings also show that both the number of publications in basic-oriented journals and in applied-oriented journals increase with disclosure activities. The increase of the number of publications is greater for academic scientists who disclosed than those who did not. 5 (Rosset and Moore, 1998). The benefits that Novartis received in return included two out of the five seats on the department’s research committee, which gave the firm the authority to decide how the research budget would be disbursed (Press and Washburn, 2000). Nevertheless, academic scientists are not passive victims of applied research sponsors. That academic scientists exploit funding from applied sponsors is suggested in a survey of 62 major US universities conducted by Thursby, Jensen, and Thursby (2001). When asked how they measured the success of the technology transfer offices at their universities, 75% of academic scientists responded that spon- sored research is an extremely important measure of success (Thursby, Jensen, and Thursby, 2001; Jensen, Thursby, and Thursby, 2003). The approach taken in Chapter 2 builds on the literature that points out the rela- tionship between research funding and academic scientists’ decision regarding research problems (i.e., fourth approach). The essay in Chapter 2 contributes to the litera- ture in four ways. First, it complements Thursby, Thursby, and Gupta-Mukherjee (2007) and Gans and Murray (2010) in providing theoretical foundation of academic scientist’s choice of research agenda when commercial profit is plausible. The differ- ence between the essay and Thursby, Thursby, Gupta-Mukherjee (2007) is that the chapter considers the influence of funding agency on academic scientist’s decision. The chapter differs from Gans and Murray (2010) in which this essay models an aca- demic scientist as an active player in creating a research agenda while the Gans and Murray (2010) models an academic scientist as a selector of research projects. The propositions in the essay contribute to the discussion in the literature on whether aca- demic scientists’ involvement in commercialization shifts academic research agenda from objective issues toward issues of industry’s interest. Second, this essay extends our understanding on the nature of scientific work as described in Stokes’ quadrants. The essay confirms to the accepted notion that funda- mental research does not necessarily imply separation with application (Stokes, 1997). 8 Toward Stoke’s framework which depicts combinations of varying degree of fundamen- tal research and applied research, the essay suggests that the relationship between the two dimensions is complementary. Third, this essay brings industry characteris- tics into the discussion of whether there is a shift academic research agenda. These characteristics are incorporated in the level of difficulty of firm’s research problem and the equality of firm’s scientists. Inclusion of these factors is important because the quality of firm’s scientists affect the ability of the firm to solve its research prob- lem, thus its interest to fund academic scientists. The magnitude of scientists who work in the industry is not trivial. In the US, 40% of 2006 science and engineering PhD graduates in the US took employment in the industry (Sauermann, Cohen, and Stephan, 2008). Moreover, some industries are aggressive in recruiting competent scientists from academia (Washburn, 2005). 1.2 New ventures as mechanisms that transform scientific investment into commercial outcomes Chapter 3 and Chapter 4 investigate the transformation of scientific investment into commercial outcomes through new technology ventures. Chapter 3 focuses on new technology ventures founded on university research. In particular, Chapter 3 seeks to explain which academic scientists are more likely to create new technology ven- tures. Besides the magnitude of scientific investment, this question is important for two additional reasons. First, the role of scientific human capital in transforming scientific results into commercial outcomes cannot be underestimated. For instance, the tendency in the biotechnology industry for locations of new technology ventures to follow the location of prominent scientists illustrates the fact that knowledge is embedded in individuals (Zucker, Darby, and Brewer, 1998). In addition, reliance on scientists for successful technological development is intensified when inventions are in an embryonic stage (Jensen and Thursby, 2001). Furthermore, Agarwal (2006) 9 has shown that engaging inventors increases the probability of commercialization suc- cess. Second, the role of scientific human capital in transforming scientific results into commercial outcomes is not a simple input-and-output function. For example, the contribution of academic scientists to new technology ventures’ performance does not increase proportionally with their scientific productivity. Rather, their contribution to patenting productivity of the new ventures decreases as their scientific productiv- ity increases (Toole and Czarnitzki, 2009). In a related study, Gittelman and Kogut (2003) argued that research that highly impacts scientific knowledge does not neces- sarily lead to valuable inventions because different selection logic operates in science and in commercialization. This emerging literature indicates that the mechanism through which scientific human capital contributes to commercial outcomes is not homogenous, but rather heterogeneous. More importantly, the quality of science is only one part of the heterogeneity. Hence, examining which academic scientists are more likely to create new technology ventures offers a step in examining alternative sources of heterogeneity. While determinants of new ventures have been at the heart of entrepreneurship literature (e.g., Gartner, 1990; Shane and Venkataraman, 2000), extant studies have treated scientists-entrepreneurs as different from general entrepreneurs. For example, scientists engaging in commercialization activities face conflicting institutional norms in scientific and industrial community (Dasgupta and David, 1994). In addition, sci- entists’ reservation cost for leaving their laboratory bench is high because they derive utility from doing science (e.g., Stephan, 1996; Stern, 2004). Furthermore, academic scientists and industry scientists differ in their choices of the timing of commercializing an invention (Lacetera, 2009). The extant literature explaining academic scientists’ entrepreneurship has provided valuable insights, yet the majority of studies examine macro level explanations. Thus, a systematic study at the individual level is required to answer the research question. 10 provides empirical evidence at individual level. Third, the essay contributes to the discussion on the role of scientific human capital in bridging the realm of science and commercialization. Consistent with extant studies (e.g., Gittleman and Kogut, 2003; Toole and Czarnitzki, 2009), the essay show that heterogeneity of human capital explains differential outcomes of commercializing scientific results. The essay extends this literature by proposing that the nature of scientist’s research is another dimension of heterogeneity of human capital. While Chapter 3 discusses the origin of new technology ventures, Chapter 4 focuses on commercialization process once the technology ventures are founded. Specifically, Chapter 4 discusses team formation in new technology ventures, and attempts to answer the following questions: why does an entrepreneur forms a team at a partic- ular stage of a commercialization project? What are the factors that encourage or inhibit the formation of entrepreneurial teams? These questions are important for, at least, three reasons. First, there is an increasing occurrence of teamwork. In all fields of science, more research is done in teams (Wutchy, Jones, and Uzzi, 2007). In addition, the average number of inventors per patent has been steadily increasing (Jones, 2009). In the field of entrepreneurship, 40 to 50 percent of new businesses are formed by teams (Shane, 2008). Yet, most studies on entrepreneurship focus on indi- vidual entrepreneur (Forbes et al., 2006), such as entrepreneur trait and entrepreneur optimism. Second, entrepreneurial teams have been linked to the performance of new ven- tures. For instance, working in a team allows the accumulation of experience of the team members, which have been found to increase the survival and sales of the new ventures (Delmar and Shane, 2003). In addition, founding team size and its hetero- geneity are positively associated with the growth of the new ventures (Eishenhardt and Schoonhoven, 1990). Third, because new ventures are plagued with resource constraint (Stinchcombe, 1965; Rothaermel and Thursby, 2005a; Rothaermel and 13 Thursby, 2005b), it is important to carefully allocate its resource. The optimal deci- sion of team formation influences resource allocation by avoiding two types of risks. One is the risk of an early team formation is carrying unnecessary cooperation cost, hence depleting the scarce resource of the new ventures. In addition, an entrepreneur experiences the risk from a late team formation is missing higher outcome which comes from specialization in a team-project. It is found that an entrepreneur obtains less expected value from a project if the entrepreneur chooses to work solo at latter stage than working in a team. The effects of economic value, probability of failure, and cost of cooperation on the timing of team formation are presented. We also explain how asymmetry of importance between tasks in a commercialization project influences the decision of team formation. The essay in Chapter 4 contributes to the literature in four ways. First, it adds to the dearth of literature in on entrepreneurial teams. Second, the essay broadens the literature on team structure by elaborating the relation between the specialization and diversity. Third, the essay extends the existing studies on the impact of uncertainty on the propensity of working in team. It confirms to extant studies that uncertainty increases the likelihood of team formations. The essay also suggests that the likelihood of team formations declines when the uncertainty is sufficiently high. The conditions in which this pattern is reversed is analyzed. Fourth, the essay complements existing literature by evaluating how asymmetry of importance between tasks influences the propensity of team formation. 14 CHAPTER II TECHNOLOGY TRANSFER AND THE SOURCES OF RESEARCH FUNDING: IMPLICATIONS FOR THE NATURE OF RESEARCH 2.1 Introduction As mentioned in the first chapter, the approach taken in Chapter 2 builds on the literature that points out the relationship between research funding and academic scientists’ decisions regarding research problems. This paper also builds on the work done by Jensen, Thursby, and Thursby (2010), who detail the interaction between research sponsors and academic scientists that occurs when the potential for commer- cial profit exists. By including research funding as one of the factors that academic scientists consider in choosing a research problem, this approach results in a model that identifies mechanisms that link academic scientists and the nature of research projects. Because the focus is on the type of research project undertaken, the model abstracts from the issue of hazard in commercialization. In-depth theoretical in- vestigations of the hazard, including considerations such as disclosure, shirking and shelving, are discussed elsewhere (e.g., Jensen and Thursby, 2001; Jensen, Thursby, and Thursby, 2003; and Dechenaux, Thursby, and Thursby; 2009). The first section of the model specifies its elements. The second section addresses the funding decisions of two research sponsors, a government agency and a firm. Both research sponsors move simultaneously. It is shown that their decisions are strategic substitutes. This result differs from Jensen, Thursby, Thursby (2010) that shows sponsors’ decisions are strategic complement. The strategic substitute in this essay arises because of the concavity of effect on productivity while the strategic 15 more difficult problem is harder to solve, the probability of success decreases as the level of problem difficulty increases at an increasing rate, ∂pa ∂xa < 0 and ∂2pa ∂x2 a > 0. On the other hand, the probability of success increases with the quality of the scientist at a decreasing rate because a more capable scientist has a greater chance of solving the research problem than a less capable one, ∂pa ∂qa > 0 and ∂2pa ∂q2a < 0. In addition, research funding assists the academic scientist in searching for the solution to her chosen research problem. The more generous the research funding, the greater the chance of solving the problem, ∂pa ∂ea > 0 and ∂2pa ∂e2a < 0. It is natural to assume that research funding and research-competence are complements, ∂2ps ∂ea∂qa > 0. However, an increase in level of difficulty of a research problem decreases the marginal contribution of research funding and research-competence to the probability of success, ∂2ps ∂es∂xs < 0 and ∂2p ∂qs∂xs < 0. In conducting a research project, the academic scientist earns wages, Wi, and im- proves her reputation, Ri, where i ∈ {s, f}. Naturally, a successful project enhances a scientist’s reputation more than a failed project, Rs > Rf . A more difficult research problem has a similar effect on reputation, dR dxa > 0. If a project is successful, it has the potential to be commercialized. The chance that a successful project will entice a firm to license the academic scientist’s research output depends on the relevance of her project to the firm’s interests, l (β) ∈ [0, 1). The less relevant the academic scientist’s project is, the less likely it is that the firm will license the research output, dl dβ > 0. The firm allocates part of the profit from commercializing academic research to the university in the form of licensing revenue, L ≥ 0. That is, the academic scientist secures additional income to supplement her university salary, A ≥ 0, when a research project is successful, Ws = A + γL and Wf = A. The additional income depends on her share of the licensing fee paid to the university, γ ∈ (0, 1). An academic scientist’s utility from the research project is defined as the value that she enjoys from wages and reputation, U (R,W ). She enjoys greater utility from 18 more generous wages or a better reputation , ∂U ∂R > 0 and ∂U ∂W > 0, at a decreasing rate,∂ 2U ∂R2 < 0 and ∂2U ∂W 2 < 0. Without loss of generality, we assume that her utility is additively separable2, U (R,W ) = f (R) + g (W ). We assume that, although the marginal effect of reputation on utility when the project is successful is less than the marginal effect of reputation on utility when the project fails, a more challenging problem enhances greater additional reputation to a successful project than to a failed project such that the difference in the reputational enhancement offset the difference in utility gained between successful project and fail project, R′ s(xa) R′ f (xa) > U ′(Rf) U ′(Rs) . Given additional research funding, the academic scientist earns greater additional utility if the additional funding is allocated to the more challenging problem than if it is allocated to an easier problem3, ∂2pa ∂ea∂xa ∆U + ∂pa ∂ea ∆ ∂U ∂xa . In other words, the additional expected utility from solving the more challenging problem is larger than the opportunity cost of giving up the less challenging problem. At the same time, an academic scientist experiences disutility from the level of rel- evance of the project to the firm’s interests, V (β) ≥ 0. Problems that are relevant to firms usually encompass broader disciplines, requiring more effort to solve (Lacetera, 2009). Such disutility also arises because problems are not always equally of inter- est to firms and the scientific community, nor are they always equally of interest to firms and the particular scientist (Goldfarb, 2008). Her disutility is increasing in its argument and convex. The academic scientist’s expected utility is EUa (G,Fa, xa, β) = pa (ea, qa, xa)U (Rs,Ws) + [1− pa (ea, qa, xa)]U (Rf ,Wf )− V (β) (2.2.1) By sponsoring a research project, G ≥ 0, the government agency obtains a bet- ter reputation. That is, Rg > 0 only if G > 0. The government agency’s utility 2This approach is similar to the one in Jensen, Thursby, and Thursby (2010). 3where ∆U = U (Rs,Ws)− U (Rf ,Wf ) and ∆ ∂U ∂xa = ∂U(Rs,Ws) ∂xa − U(Rf ,Wf ) ∂xa . 19 from the academic scientist’s project is denoted as Ug (Rgi) where i ∈ {s, f}. Like the academic scientist, the government agency has more to gain in this regard from a successful project than from a failed one, Rgs > Rgf . The government agency’s utility is increasing and concave in the reputational stock. In funding the academic scien- tist’s project, the government agency forgoes the opportunity to fund other research projects. The opportunity cost of forgoing other research projects is denoted as V (G) where V (G) ≥ 0. The government agency disutility increases in its argument and convex because spending more units of funding on the academic scientist’s project means spending fewer units on other research projects. The government agency’s expected utility is EUg (G,Fa, xa, β) = pa (ea, qa, xa)Ug (Rgs) + [1− pa (ea, qa, xa)]Ug (Rgf )− V (G) (2.2.2) The firm decides the level of funding to the academic scientist, Fa ∈ (0, F ), based on its own research projects. It distributes a certain amount of the research budget, F > 0, to its own research projects and an additional amount to the academic sci- entist’s project. As with the academic scientist’s project, the probability of success of the firm’s research project, pc (ec, qc, xc) ∈ [0, 1), is contingent on the level of dif- ficulty of the research problem, the amount of research funding, and the quality of the firm’s scientists. These two research projects also have differing consequences to the firm. First, the firm naturally chooses a problem that is commercially relevant to its business when conducting its own research. Unlike the results of the academic scientist’s project, the results of the firm’s own project will absolutely be relevant for commercialization. Secondly, the firm retains all profits when commercializing its own research but shares some of the profits with the university when commercializing academic research. Thirdly, the firm’s project only receives funding from its own research budget, while the academic scientist’s project can have up to two sources of 20 will increase the amount of funding to the academic scientist until a one-unit increase of funding results in additional expected profit from the academic scientist’s research that is equal to the additional loss of expected profit from the firm’s research. That is, the firm will increase the level of funding to the academic scientist until the marginal effect of its funding on the expected profit from the academic research is offset by the marginal expected loss of profit from the firm’s research. As mentioned earlier, it is possible for the academic scientist to obtain funding from both research sponsors. Considering this likely dual source of funding, the firm decides the amount of research funds for the academic scientist, Fa, which maximizes EΠ(G,Fa, xa, β). The firm’s decision differs for different levels of government agency funding. Thus, the firm’s best response function, F̂a (G), is the level of firm funding to the academic scientist that maximizes its expected profit for any level of funding from the government agency. The government agency also chooses the amount of the grant, G, for the academic scientist, which maximizes EUg (G,Fa, xa, β). The agency’s deci- sion depends on the firm’s contribution to the academic scientist’s research. The gov- ernment agency’s best response function, Ĝ (Fa), is the level of government-provided funding for the academic scientist that maximizes the agency’s expected utility for any given level of firm funding. An example of research sponsors’ best response functions and the equilibrium level of funding is illustrated in Figure 1 below. Proposition 2.3.1 When the government agency’s and the firm’s best response are interior, Ĝ (Fa) ∈ (0, Bg) and F̂a (G) ∈ (0, F ), their best responses are negatively sloped. Proof. Available at the appendix Depicted in Figure 1, the firm’s best response is a declining function of the govern- ment agency’s funding. At the same time, the government agency’s best response is a declining function of the firm’s funding. The government agency gives the academic 23 Figure 2: Best-response functions of research sponsors scientist a smaller amount of funding when the firm provides substantial funding. The more substantial the government’s grant, the less funding the firm allocates to the academic scientist. That the government agency’s best reply is a declining function of the firm’s funding implies that the firm funding decreases the marginal effect of government funding on the its expected utility. Likewise, that the firm’s best reply is a declining function of the government agency’s funding implies that the funding from the government agency decreases the marginal effect of firm funding on its expected profit. These relationships arise because the additional chance of solving the problem using a greater amount of funding declines as the total amount of funding increases. It is more difficult to increase one’s chances of success by putting in more funding when the total research funding is abundant because there is a limit on how much funding contributes to the project’s chance of success. From the government agency’s point of view, the declining marginal chance of success generates its best response in the following way. The government agency contributes a larger marginal chance of success with each unit of its funding when 24 the agency is the only sponsor4. Consequently, the government agency receives larger marginal expected utility for each unit of its funding when it is the sole research spon- sor. This does not imply that the government agency prefers to be the sole research sponsor because the probability of success and the government agency’s expected util- ity increase along with the larger total amount of funding from the increased number of research sponsors. The government agency welcomes firm contributions and the re- sulting reduction in the government agency’s marginal expected utility as long as the reduced marginal expected utility is greater than the marginal loss from not funding alternative research projects. Recall that the equilibrium level of government agency funding is the amount of government funding for which its marginal expected util- ity is equal to its marginal loss from forgoing alternative projects. When the firm contributes to the academic scientist’s project, the equilibrium level of government agency funding is the amount of government funding for which its reduced marginal expected utility is equal to its marginal loss from not funding other projects. At the equilibrium level, additional firm funding results in a greater reduction in the marginal expected utility such that it does not compensate for the marginal loss as- sociated with forgoing alternative research projects. Hence, the government agency will adjust its contribution by decreasing the amount of funding so that its marginal expected utility is equal to the marginal expected loss. A similar explanation can be used to account for the process through which the 4To see this, we can imagine two situations. In both situations, the probability of success increases by 0.2 for the first unit of funding. The second unit of funding adds to the probability of success by 0.1. In the first situation, the government agency is the sole research sponsor. By granting one unit of funding, the government agency receives greater net-expected utility based on an increase of twenty percent in the chance of success. In the second situation, the academic scientist obtains funding from a government agency and the firm. Let us suppose that each sponsor provides one unit of funding. When the government agency grants one unit of funding, the academic scientist obtains two units of funding because the firm provides another unit of funding. Because there are two units of funding in total, the probability of success increases by 0.3. This indicates that the additional chance of success is 0.15 per unit of funding. Unlike in the first situation, in which the one unit of funding from the government agency increases the chance of success by twenty percent, the government agency receives greater net-expected utility from an increase of fifteen percent in the chance of success. 25 in the level of firm’s funding. In summary, the government agency is exposed to two conflicting forces when the academic scientist’s problem becomes more challenging. One is a decrease in agency willingness based on its lower marginal expected utility. The other is an increase in agency willingness based on its lowered contribution. Like the government agency, the firm experiences two conflicting forces when the academic scientist chooses a more challenging problem. On one hand, the firm’s willingness to fund the academic scientist declines because of the lowered marginal expected profit from the academic research. On the other hand, the firm’s willingness to contribute to the academic research increases because the government agency is less willing to fund the academic research. The government agency will experience an increase in willingness based on its lowered contribution that is larger than the decrease in willingness based on its lower marginal expected utility. Hence, the government agency provides a larger amount of funding and the firm reduces its funding. An increase in the relevance of her research problem to the firm’s field of interest decreases the level of funding from the government agency and increases the level of funding from the firm. This adjustment occurs because the firm’s best response shifts upward. Meanwhile, the government agency’s best response remains unaffected. The firm’s best response shifts upward because the more relevant the academic scientist’s problem is to the firm’s interest, the more likely it is that successful academic research will be beneficial to the firm’s business. Accordingly, the firm’s marginal expected profit from academic research increases. When the research-competence of the academic scientist increases, the government will reduce its level of funding and the firm will increase its level of funding. These adjustments take place because a unit of funding in the hands of a high-quality scientist contributes more to the chances of solving the research problem than does the same unit of funding in the hands of a lower-quality scientist because the more 28 competent scientist is more capable of finding a solution to the problem. Accordingly, an increase in the quality of the academic scientist will result in a greater marginal expected utility of the government agency. The government agency is willing to contribute more to the scientist’s project for any given level of firm funding. Thus, the government agency’s best response shifts to the right. An increase in the quality of the academic scientist also enhances the firm’s marginal expected profit from academic research. The firm is willing to allocate more of its research budget to the academic scientist for any given level of govern- ment agency funding. Thus, the firm best response shifts upward. Because the government agency’s best response is declining in the level of firm funding, the gov- ernment is less willing to provide funding. In short, an increase in the quality of the academic scientist creates two opposing forces that influence the government agency. The government agency’s increased willingness to contribute because of the greater marginal expected utility is less than the government agency’s decreased willingness to contribute because of the firm’s greater interest in the academic project. There- fore, the government agency reduces its funding and the firm provides larger amount of funding. When the firm’s research problem becomes more challenging, the government agency reduces its funding, whereas the firm provides the academic scientist with a larger amount of funding. The reason is that the firm’s best reply shifts upward and the government agency’s best reply is unchanged. The firm is willing to allocate a greater proportion of its research budget to the academic scientist because the more challenging its research problem is, the lower the chance that the firm’s scientist will solve the problem and the lower the firm’s opportunity cost as associated with diverting its research funds to the academic project. In contrast, the firm reduces its funding to the academic scientist and the government grants a larger amount of funding to her when the firm’s scientist is more competent. This adjustment is 29 attributed to the downward shift in the firm’s best reply. Meanwhile, the government’s best reply is not affected. With a higher quality scientist, the firm gains larger marginal expected profit from its research. Thus, the firm’s loss of marginal expected profit as associated with diverting the research budget away from its own research increases, and the firm is less willing to fund the academic scientist given any level of government agency funding. Any changes in the academic scientist’s share of the licensing paid to the university have no effect on the level of funding from both sponsors. However, an increase in the licensing paid to the university results in a larger amount of funding from the government agency and a smaller amount of funding from the firm. This adjustment occurs because the firm’s best reply shifts downward and the government agency’s best reply does not change. A larger licensing payment to the university reduces the firm’s profit from commercializing academic research and the associated expected profit. Therefore, the firm is willing to provide less funding to the academic scientist. Consequently, the government agency gives more funding because the government agency’s best reply is decreasing in the level of firm funding. 2.4 Stage One Equilibrium In the first stage, the academic scientist decides on the level of difficulty of her project, xa, and its level of relevance, β. We assume that the academic scientists chooses these two characteristics based on the equilibrium decisions of research sponsors in the stage two equilibrium, G∗ (xa, β) and F ∗ a (xa, β). Thus, the academic scientist’s objective function is max {xa,β} EUa (G ∗ (xa, β) , F ∗ a (xa, β) , xa, β) (2.4.1) The first-order conditions are 30 An academic scientist chooses a more challenging problem when her chosen prob- lem is more relevant to the firm’s interests. She selects a problem that is less relevant to the firm’s interests if the problem is less challenging. In other words, the diffi- culty best-choice function is an increasing function of the level of relevance, and the relevance best-choice function is an increasing function of the level of difficulty. An example of difficulty and relevance best-choices is illustrated in the figures below. Figure 3: Best-choice functions of an academic scientist The mechanism underlying these best-choice functions is the following. As previ- ously mentioned, the academic scientist increases the level of problem difficulty until the marginal gain in her expected utility because of the greater increase in her repu- tation is offset by the marginal decrease in the expected utility because of her lowered chance of success. The extra funding that the problem relevance inspires increases both the scientist’s marginal gain in expected utility from the more challenging prob- lem and her marginal loss in expected utility from the easier problem. This implies that an academic scientist earns a larger marginal expected utility from problem diffi- culty when she aligns her research project with the firm’s interests. The scientist also 33 receives a larger marginal expected utility from relevance when the scientist works on a more challenging problem. The scientist experiences an increase in the marginal gain that is greater than the increase in the marginal loss because of two reasons. Receiving more generous funding based on increased relevance, the academic scientist has a greater chance of achieving the additional utility from the more challenging problem if she allocates the extra funding to the difficult project. The extra funding also attenuates the impact of the reduction in funding that occurs when the academic scientist increases the level of difficulty of the problem. Because she will suffer fewer consequences from increasing the level of difficulty, the academic scientist will be willing to take on a more challenging problem. We can obtain comparative statics under the following reasonable assumptions: A1. ∂2p∗a ∂xa∂i < 0 for i = qa, xc, L and ∂2p∗a ∂xa∂i > 0 for i = qc A2. ∂2p∗a ∂xa∂i ∆U + ∂p∗a ∂xa∂i ∆ ∂U ∂xa < 0 for i = qc and ∂2p∗a ∂xa∂i ∆U + ∂p∗a ∂xa∂i ∆ ∂U ∂xa > 0 for i = qa, xc A3. ∂2pa(ea(G∗(xa,β),F ∗ a (xa,β)),qa,xa) ∂β∂qa > 0 where ∂2p∗a ∂xa∂i = ∂2pa(ea(G∗(xa,β),F ∗ a (xa,β)),qa,xa) ∂xa∂i and ∂p∗a ∂xa∂i = ∂pa(ea(G∗(xa,β),F∗ a (xa,β)),qa,xa) ∂i These conditions state that an increase in the quality of the academic scientist, in difficulty of the firm’s own project, or in the licensing payment to the university decreases the marginal effect of the level of difficulty on the stage-two equilibrium probability of success of the academic scientist’s project. An increase in the quality of firm’s scientist increases the marginal effect of the level of difficulty on the stage- two equilibrium probability of success. Furthermore, an academic scientist who works on a difficult project enjoys a higher marginal expected utility when she has more 34 than less funding. This situation occurs when the firm has low opportunity cost associated with low research-competence of its scientists. The situation also happens when the firm has to work on a difficult problem. In addition, an academic scientist who has higher research-competence enjoys larger additional expected utility from a more challenging problem than a less competent scientist. Meanwhile, an increase in the research-competence of an academic scientist increases the marginal effect of relevance, thus the additional funding brought into the project, in the stage-two equilibrium probability of success. Proposition 2.4.2 Assume that second-order effects on the equilibrium funding from government agency and from the firm are negligible, ∂2G∗ ∂i∂j ≈ 0 and ∂2F ∗ a ∂i∂j ≈ 0, for all parameters i and j. Then: 1. An increase in the research-competence of academic scientist, qa, increase the level of problem difficulty and the level of alignment to the firm’s interest. 2. An increase in the research-competence of firm’s scientist, qc, decreases the level of problem difficulty and the level of alignment to the firm’s interest 3. An increase in the level of difficulty of the firm’s research problem, xc, increases the level of problem difficulty and the alignment to the firm’s interest 4. If ∂2EUa/∂β2 ∂2EUa/∂x2 a > δγ ,an increase in the academic scientist’s share of licensing paid to the university, γ, decreases the level of problem difficulty and the level of alignment to the firm’s interest when ∂2EUa ∂xa∂β < Mγ, but it increases the level of problem difficulty while decreasing the level of alignment when ∂2EUa ∂xa∂β > Mγ. If ∂2EUa/∂β2 ∂2EUa/∂x2 a < δγ, an increase in the academic scientist’s share of licensing paid to the university, γ, increases the level of problem difficulty and the level of alignment when ∂2EUa ∂xa∂β < M̄γ, but it increases the level of problem difficulty while increasing the level of alignment when ∂2EUa ∂xa∂β > M̄γ. 35 of difficulty. Thus, the best-choice function for relevance shifts upward. The implica- tion of this upward shift is that the academic scientist is more willing to undertake a difficult problem because the scientist’s best-choice function for difficulty is upward sloping. The academic scientist will select a less challenging and less relevant prob- lem because the encouragement caused by the increasing relevance of the problem is smaller than the disincentive caused by the decline in firm funding. An academic scientist will select a more challenging and more relevant problem when the firm’s research problem becomes more difficult. These adjustments arise from the following process. As its own research problem becomes more difficult, the chance that the firm will develop a solution to that problem declines. This means that the firm will incur lower opportunity costs if it allocates less funding to the problem. Thus, the firm is more willing to provide research funding to the academic scientist8. In receiving more funding, an academic scientist obtains larger marginal expected utility if she works on a more difficult problem. As previously explained, extra funding attenuates the impact of reductions in funding when the academic scientist increases the level of difficulty. In addition, the academic scientist has a better chance of achieving the additional utility, a result of the increased difficulty level, because the project has more funding, a result of the decline in the firm’s opportunity cost. Thus, an academic scientist will favor a more difficult problem for any level of relevance. It indicates that the best-choice function for difficulty shifts to the right. Another implication of the increased difficulty of the firm’s problem is that it reduces the marginal gains of the expected utility that result from increasing the level of relevance. As previously mentioned, the firm provides a larger amount of funding to 8In the previous section, we note that the firm provides a higher equilibrium level of funding when the difficulty of its own research increases. In response, the government agency then decreases its equilibrium level of funding. The additional funding from the firm is greater than the reduction in funding by the government agency. Therefore, the total funding increases. 38 the academic scientist. The academic scientist experiences a lower additional chance of success because, as the total funding increases, it is harder for the extra funding associated with the greater relevance of the problem to increase the scientist’s chances of success. Thus, the academic scientist prefers a lower level of relevance for any level of difficulty, and the best-choice function for relevance shifts downward. In the light of this downward shift, an academic scientist will prefer a less challenging problem because the scientist’s best-choice function is upward sloping. The academic scientist will decide on a more difficult and more relevant problem because the inducement caused by additional funding is larger than the disincentive caused by decreasing relevance. The implications of an increase in the licensing paid to the university are not straightforward, but rather contingent upon the extent of complementarity between problem difficulty and its relevance. When the licensing paid to the university in- creases, the firm retains a smaller profit from commercializing a successful academic research project. Consequently, the firm will have less interest in the academic sci- entist’s research and will reduce its support9. However, the academic scientist will obtain more additional income if her research project is successful. If the academic scientist increases the difficulty of the research problem, she will receive a smaller marginal expected utility. First, an academic scientist has less of a chance of achiev- ing additional income both because the problem is harder to solve and because the research sponsors react by reducing the amount of funding. Secondly, the academic scientist suffers more from a reduction in funding based on increasing problem dif- ficulty because she already receives less funding because of the increased licensing 9As explained in the earlier section, the firm cuts down the amount of equilibrium level of funding when the licensing paid to the university increases. In response, the government agency provides a larger amount of equilibrium level of funding. The reduction of funding from the firm is larger than the additional funding from the government agency. Hence, the total funding declines. 39 revenue. This implies that in increasing difficulty of her research problem, the aca- demic scientist experiences a larger decline in her probability of success when the licensing revenue increases. Thirdly, the academic scientist has less of a chance to ob- tain the additional reputation by working a more challenging problem because of her lowered funding (which, again, results from the increase in licensing revenue). Thus, the academic scientist prefers a less challenging problem for any level of relevance. That is, the best-choice for difficulty level shifts to the left. In addition to its influence on the effect of difficulty, an increase in the licensing paid to the university changes the effect of problem relevance. An academic scientist gains larger net-marginal expected utility if she chooses a problem that is more rel- evant to the firm’s interests. As previously discussed, the firm will allocate a larger amount of funding to an academic scientist whose problem is more relevant. This extra funding attenuates the decline in the academic scientist’s total funding when the firm cuts down its contribution because of the increase in licensing revenue. Fur- thermore, the extra funding gives the scientist a better chance of solving the problem. Accordingly, the scientist will be more likely to earn additional income from increased licensing revenue. Hence, the academic scientist will select a problem with greater relevance at any level of difficulty, and the best-choice function for relevance shifts upward. When the best-choice function for relevance shifts upward, a more challeng- ing problem becomes more attractive because the academic scientist’s best-choice function for difficulty is upward sloping. The steepness of the slopes of best-choice functions indicates the extent of com- plementarity between the level of difficulty and the level of relevance. When licensing paid to the university increases, the academic scientist will choose a more challeng- ing but less relevant problem if the level of difficulty and the level of relevance are highly complementary as shown by steep slopes. If the level of difficulty and level of relevance are low to moderate in complementarity, the academic scientist prefers 40 ea = δGG+ δFFa (2.5.1) The first-order conditions in the second stage are ∂EUg (G,Fa, xa, β) ∂G = ∂pa ∂ea δG [Ug (Rgs)− Ug (Rgf )]− V ′ (G) = 0 (2.5.2) and ∂EΠ(G,Fa, xa, β) ∂Fa = ∂pa ∂ea δF l (β) (Πu − L)− ∂pc ∂ec Πc − 1 = 0 (2.5.3) In the first stage, the first-order conditions are ∂EUa ∂xa = ( ∂pa ∂ea ∂G∗ ∂xa δG + ∂pa ∂ea ∂F ∗ a ∂xa δF + ∂pa ∂xa ) (U (Rs,Ws)− U (Rf ,Wf )) + ( pa ∂U(Rs,Ws) ∂xa + (1− pa) ∂U(Rf ,Wf) ∂xa ) = 0 (2.5.4) and ∂EUa ∂β = ( ∂pa ∂ea ∂G∗ ∂β δG + ∂pa ∂ea ∂F ∗ a ∂β δF ) (U (Rs,Ws)− U (Rf ,Wf ))− V ′ (β) = 0 (2.5.5) As before, the government agency increases the level of funding until the marginal effect of government funding on its expected utility is offset by the marginal loss; and the firm will increase the level of funding to the academic scientist until the marginal effect of its funding on the expected profit from the academic research is offset by the marginal expected loss of profit from the firm’s research. In the first stage, the process of decision making remains. That is, the academic scientist will increase the difficulty of her research problem until the marginal increase in her expected utility because of the greater improvement in her reputation is offset by the marginal decrease in her expected utility because of her reduced chance of success; and she increases the 43 relevance of her research problem to the firm’s field of interest until the marginal gain in her expected utility is equalized by a marginal increase in disutility. Stage-two equilibrium and extant comparative statics do not change by the inclu- sion of the academic scientist’s and the university’s shares of research funding. In the same way, the equilibrium and extant comparative statics in stage-one equilib- rium remains despite the presence of academic scientist’s and the university’s shares of research funding. However, the effects of these shares on the level of government funding, the level of firm funding, and the academic scientist’s choice of research project are ambiguous. 2.6 Direct Benefits of Research Funding In the earlier sections, we consider indirect effects of research funding on the utility of an academic scientist. Research funding indirectly influence the utility of an aca- demic scientist through the effect of resource on the probability of solving a research problem. Besides indirect effects, research funding directly influence the utility of an academic scientist. The direct effect is positive and independent of research output. For example, research funding enhances the power of an academic scientist in the department or university (Pfeffer and Salancik, 1974). The academic scientist may use this power to obtain a bigger office, to obtain nicer equipments, to provide fellow- ships for students, and to avoid doing committee work. Research funding also allows the faculty to buy out teaching, and it relieves any possible disutility associated with teaching. For simplicity, we abstract from academic scientist’s and the university’s share of research funding. We define Ud (ea) that is the direct benefit of research funding on academic scientist’s utility. It is increasing in its argument and concave. The academic scientist’s expected utility, EUa (G,Fa, xa, β), is pa (ea, qa, xa)U (Rs,Ws) + [1− pa (ea, qa, xa)]U (Rf ,Wf ) + Ud (ea)− V (β) (2.6.1) 44 Stage-two equilibrium and its related comparative statics do not change by the inclusion of the direct benefit, Ud (ea). In stage-one equilibrium, the first-order con- ditions are ∂EUa ∂xa = ( ∂pa ∂ea ∂G∗ ∂xa + ∂pa ∂ea ∂F ∗ a ∂xa + ∂pa ∂xa ) (U (Rs,Ws)− U (Rf ,Wf )) + ( pa ∂U(Rs,Ws) ∂xa + (1− pa) ∂U(Rf ,Wf) ∂xa ) + ∂Ud ∂ea ( ∂G∗ ∂xa + ∂F ∗ a ∂xa ) = 0 (2.6.2) and ∂EUa ∂β = ( ∂pa ∂ea ∂G∗ ∂β + ∂pa ∂ea ∂F ∗ a ∂β ) (U (Rs,Ws)− U (Rf ,Wf )) +∂Ud ∂ea ( ∂G∗ ∂β + ∂F ∗ a ∂β ) − V ′ (β) = 0 (2.6.3) When deciding the level of difficulty, the academic scientist will increase the dif- ficulty of her research problem until the marginal increase in her expected utility because of the greater improvement in her reputation is offset by the marginal de- crease in her expected utility because of her reduced chance of success and because of lower direct benefits. When choosing the level of relevance, the academic scientist increases the relevance of her research problem to the firm’s field of interest until the marginal gain in her expected utility is equalized by a marginal increase in disutility. The marginal gain in her expected utility arises because additional funding from the firm improve the chance of solving the problem and because of the larger utility from direct benefits. Similar to the section where only indirect effect of research funding is considered, the difficulty best-choice function and the relevance best-choice function are positively sloped. Moreover, inclusion of direct benefits does not affect the comparative statics involving the level of difficulty of firm’s research problem, the amount of licensing paid to the university, and the academic scientist’s share of licensing paid to the university. It is reasonable to assume that, in an academic community, the utility related with 45 CHAPTER III ACADEMIC SCIENTISTS: THE NATURE OF RESEARCH AND ENTREPRENEURIAL ACTIONS 3.1 Introduction This chapter discusses how academic scientists’ decision to create new ventures is influenced by the nature of research, specifically the level of commercial applicabil- ity. The approach taken builds on extant studies that argue that the opportunity cost of engaging in non-research activities is less time spent on scientists’ labora- tory work (e.g., Levin and Stephan, 1991; Thursby, Thursby and Gupta-Mukherjee, 2007; Jensen, Thursby and Thursby, 2010). The model shows that the opportunity cost is attenuated because knowledge is transferred from successful entrepreneurial actions to scientists’ research agenda. The attenuation is larger for some academic scientists than for others, depending on the nature of their respective research. The model explains academic scientist’s nature of research in one dimension: the level of commercial applicability. This dimension spans a continuum from low to high. Within the continuum of commercial applicability, there is a point after which a scientist is willing to create a new venture. Before reaching that point on the con- tinuum, the scientist will not create a new venture. Within this continuum, there is another point starting from which an established firm is willing to license the scien- tist’s invention. If the point after which a scientist deems an invention commercially applicable, and is thus willing to create a new venture, is less than the point at which an established firm believes it is commercially applicable, and is thus willing to li- cense the invention, we will observe that scientists who have the highest probability of creating new ventures are those whose level of commercial applicability is medium. 48 Scientists whose nature of research is low and high in the dimension of level of com- mercial applicability will have lower probability than scientists with medium level of commercial applicability. In this case, there is an inverted-u shape relationship between the level of commercial applicability and the likelihood that an academic scientist creates a new venture. When the level of commercial applicability which an academic scientist decides to create a new venture is higher than the level of com- mercial applicability which an established firm is interested to license the invention, we predict a decreasing relationship between level of commercial applicability and the likelihood that an academic scientist creates a new venture. Empirical estimation of the model is performed on a sample of 395 academic scientists at five top U.S. research universities. 3.2 The Model This section presents a theoretical analysis of an academic scientist’s decision to create a new venture. As the starting point of the game, an academic scientist has disclosed his invention to the university. At this first stage, the university evaluates the invention and has two choices: to shelve (i.e. give the invention to the academic scientist) or not to shelve the invention. If the university shelves the invention, the academic scientist can choose either to create a new venture based on the shelved invention or to work on another research project instead. If the university decides not to shelve the invention, at the second stage, the university can either search for a licensee or offer the academic scientist the option to create a new venture based on the invention licensed by the university. Facing such an offer at this stage, the academic scientist chooses whether or not to create a new venture. Figure 4 depicts the extensive form of the game. 49 Figure 4: Academic entrepreneurship: Extensive form of the game 3.2.1 The Academic Scientist An academic scientist is associated with his level of scientific prominence, q. His scientific prominence determines his probability of success should he found a new venture, p(q). The probability of success, a function of q, is both increasing and concave because more prominent scientists are more likely to get resources. For instance, prominent scientists are better able to attract partners or signal to investors that they perform exceptional assessment of the technology (Higgins, Stephan, and Thursby, 2008). For simplicity, we assume risk neutrality. In the case of successful commercialization, the academic scientist gains utility from non-scientific return, B, such as money and satisfaction from having a practical impact (e.g., reaching people). He also obtains utility from scientific return (i.e. knowledge), K. The extent to which his scientific return from commercialization activity is valuable for his research at the university depends on a, which is the level of commercialization applicability of the academic scientist’s research orientation. Where a > 0, this transferability 50 Thursby, and Thursby, 2003). Since the extent of the firm’s effort (including invest- ment in capital and human resources) depends on the level of commercial applicability of the invention, the university will not search for a licensee unless pF (a) ≥ VF (a) /R where pF (a) is the probability of commercialization success if the invention is further developed by the firm. The probability of success by the firm licensee is increasing in its argument and concave. We denote VF (a) as the firm’s disutility from the com- mercialization effort. It is increasing in its argument and convex. We denote R as the return to the firm licensee if the commercialization is successful. We also assume that p (q) > pF (a) when min(a); dp(q) dq < dpF (a) da such that p (q) < pF (a) when max(a) for all q. These specifications show that the probability of commercialization success by an academic scientist is larger than that of commercialization success by licensee firms when the research orientation of academic scientist, hence his invention, is low in the level of commercial applicability; this is true since the academic scientist has the tacit knowledge to further develop the embryonic invention (Jensen and Thursby, 2001). The specifications also indicate that the probability of success by a firm li- censee is larger than the probability of success by the academic scientist when the research orientation of the academic scientist, hence his invention, is high in the level of commercial applicability. This illustrates the fact that the higher an invention’s level of commercial applicability, the better a firm licensee can utilize its existing as- sets to successfully commercialize it. In other words, the higher the invention’s level of commercial applicability, the greater is a firm licensee’s advantage from possessing complementary assets. If the university searches for a licensee, it incurs disutility from searching, VUl. The university’s expected utility from licensing is thus: EUUl = pF (a)BU + L− VUl (3.2.3) While the university’s payoff from supporting an academic scientist’s venture is 53 positive only when the commercialization is successful, it is possible that the uni- versity searches for a licensee even if the commercialization will not be successful, EUUl = L − VUl > 0. These two conditions reflect higher disutility from supporting academic scientists. Unlike licensing to an established firm, the university responsi- bility does not end by the signing of licensing contracts, but rather it entails providing the academic scientist supports such as preparing business plans and connecting aca- demic entrepreneurs to potential partners (e.g., surrogate entrepreneurs or VCs). If the university licenses the invention to a firm licensee, the academic scientist is in- volved in the further development, and thus the non-academic benefit from successful commercialization to established firm is normalized to zero. The former is in line with the fact that university invention is embryonic such that the involvement of an academic scientist is needed for successful commercialization (Jensen and Thursby, 2001). The latter is justified by the magnitude of difference between the satisfac- tion that comes from creating a successful new venture and that which comes from consulting, respectively . Thus, the academic scientist’s expected utility when the invention is licensed to a firm is EUIl = pF (a) (aK) + (1− pF (a)) (0) (3.2.4) Proposition 3.2.2 In the equilibrium, there are as = r(q) p(q) − B K and pF (af ) = VF (af) R 1. When af > as, the likelihood that the academic scientist engages in creating a new venture increases until a cut-off point, ā , after which the likelihood of creating a new venture decreases. 2. When af > as, an increase in the academic scientist’s prominence, q, increases the cutoff point, ā. 54 3. When af ≤ as, the likelihood that the academic scientist engages in creating a new venture decreases as variable a increases. Proof. Available at the appendix First consider the situation in which the scientist’s willingness to create a new venture occurs earlier in the continuum of level of commercial applicability than the point on the continuum where the established firm becomes interested in licensing the invention. Academic scientists whose research orientation entails a low level of commercial applicability do not find it worthwhile to found new ventures because the expected return does not compensate for the opportunity cost of ceasing to focus on university research. Increasing commercial applicability increases the incentive for academic scientists to engage in entrepreneurship because of the greater expected entrepreneurial return, which comes from the greater scientific benefit of the activity. As long as there is no established firm interested in licensing the invention and its net expected return from supporting the scientist entrepreneur is positive, the university does not shelve the invention and, instead, will support the scientist’s venture. However, as commercial applicability increases, the established firm’s expected return from licensing the scientist’s invention also increases. At an equal expected re- turn from either licensing to an established firm or supporting a scientist entrepreneur, the university incurs higher cost from the latter, which makes supporting a scientist entrepreneur the less attractive option. That is, although the scientist’s willingness to create new ventures increases in the level of commercial applicability, the university’s willingness to support the scientist’s venture decreases in the level of commercial applicability. Hence, once the established firms are interested in licensing the scien- tist’s invention, there is a decline in the likelihood that the scientist will create a new venture. The point of decline, however, depends on the scientific prominence of the scientist. That is, it is easier for scientists of higher prominence to attract the re- sources required for a successful venture. Their prominence both lures talented team 55 whether the academic scientists started new ventures. We identified academic en- trepreneurs from CVs and web searches (i.e. google search). In the first part of the third step, we matched academic scientists’ publication lists from ISI Web of Science with information in their CVs, such as their statements of research interests, prior affiliations, and selected publication lists. The second part of this third step entailed a similar procedure performed on academic scientists’ patent lists from USPTO and Delphion. In the fourth step, we matched the resulting publication list from the second step with the National Science Foundation-IpiQ 2007 journal classification system (i.e. NSF/IpiQ classification). IpiQ classification categorizes journals into four categories: 1 indicates applied technology, 2 indicate engineering and technolog- ical science, 3 indicates applied research, targeted basic research, and 4 indicate basic scientific research. For ease of interpretation of econometric analysis, we recoded the categories such that 1 indicates basic scientific research, 2 indicates applied research, targeted basic research, 3 indicates engineering and technological science, targeted basic research, and 4 indicates applied technology. The four steps result in 395 academic scientists whose CVs are available on-line and for whom the matching process (i.e. the third step) was possible. Out of 395 academic scientists, 101 scientists created new ventures. These scientists published a total of 35,840 publications. In addition, they are listed as inventors in a total of 6,558 patents. 3.3.3 Variables and Measures Dependent variable: Startup Our dependent variable, startup, was binary, with 1 indicating academic scientists who founded new technology ventures. As our dependent was binary, we applied a logistic regression model estimating how the nature of research and scientific promi- nence affect the probability of an academic scientist engaging in entrepreneurship. 58 Independent variables In order to capture the nature of research, an independent variable research orien- tation was constructed based on NSF/iPiQ classification. We calculated the research orientation variable as follow: research orientation = 1− number of publication rated by NSF/IpiQ as basic scientific research Total number of publication rated by NSF/IpiQ Academic scientists whose research is low in the level of commercial applicability are associated with a low value of research orientation variable. In order to capture the non-linear relationship between the nature of research and the creation of new technology ventures, we constructed an independent variable research orientation square. The independent variable for scientific prominence is measured by an average number of publications per year. We calculated the average number of publications is calculated as follow: average number of publication per year = Total number of publications Number of active year We define ’number of active years’ as the number of years from PhD completion until retirement. We consider the length of an academic career to be approximately 35 years. The calculation of ’number of active years’ is as follows: if the sum of year of PhD completion and thirty five is less than 2008, ’number of active years’ is 35; if the sum of year of PhD completion and thirty five is greater than 2008, ’number of active years’ is 2008 minus the year of PhD completion. In order to test proposition 2.3, we construct a variable bio x research orientation. The variable bio is 1 for academic scientists in the life sciences discipline. Since the research of the life sciences is characteristically basic in nature yet also relevant to practical problems (i.e. Pasteur Quadrant’s, Stokes (1997)), the research output of scientists in the life sciences is close to the interest of existing firms in the industry. Control variables 59 Since literature in entrepreneurship indicates that women are less likely than men to start new businesses (Shane, 2004) and that woman scientists are less likely to be involved in the commercialization of their research (Ding, Murray, and Stuart, 2006), we included the control variable gender with 1 indicating women. In addition, recent studies argue that an academic scientist’s choice of research project may be influ- enced by the commercial potential of that project (e.g., Lacetera, 2009). To address this possibility, we include patenting variables which portray academic scientists’ in- clination towards commercialization. These variables are commercial patent, which indicates the average number of patents assigned to non-research institutions and university patent which indicates the average number of patents assigned to research institutions. Its calculation is: average number of commercial patents per year = Cummulative number of commercial patents Cummulative years since PhD completion average number of university patents per year = Cummulative number of university patents Cummulative years since PhD completion We define ’cumulative number of patents’ in two ways. If the academic scientist was involved in founding a new technology venture, the cumulative number of patents is the number of patents accumulated until the year prior to the founding of the new technology venture. If the academic scientist does not found a new venture, the cumulative number of patents is the number of patents accumulated until the year 2008. ’Cumulative years since PhD completion’ is computed in a similar way. It is the number of years elapsed from PhD completion to the year of founding the first new venture if the academic scientist was also an entrepreneur. It is the number of years elapsed from PhD completion to the year 2008 if the academic scientist member is not an entrepreneur. If then the academic scientist has retired and did not found a new venture, we use 35 years as the cumulative years since PhD completion. Moreover, we control for the possibility of social influence on the academic scientist’s choice of research orientation. To take into account such a possibility, we create a variable dep fresor, which is the average research orientation in the scientist’s department. 60 In proposition 3.2.2.1, we predicted that the relationship between academic sci- entists’ nature of research and the creation of new technology venture is non-linear such that the likelihood of creating new ventures increases as the research orientation increases in its commercial applicability until a cut-off point; after this cut-off point, the likelihood of creating new ventures decreases as the level commercial applicability further increases. Model 2 shows that the research orientation variable is positive and significant (p<0.01)2. In addition, the research orientation square variable is negative and significant (p<0.05), confirming the non-linear relationship. Plotting the result, as shown in figure 5, we observe that the cut-off point is approximately at a research orientation of 0.9. Figure 5 depicts the effect of research orientation on the probabil- ity of creating new technology ventures while holding other variables at their mean values. Figure 5: Probability of starting new ventures by the nature of research for academic scientists in non-life science 2The author is working on the interpretation of interaction terms in non-linear models as sug- gested by Hoetker (2007) and Wiersema and Bowen (2009). In addition, Linear Probability Model of the specification does not change the significance and signs. 63 We split the sample into academic scientists who have an average number of publi- cations below the median (i.e. in the lower 50% of the average number of publications) and academic scientists who have an average number of publications above the me- dian (i.e. in the upper 50% of the average number of publications). The resulting logit regression is depicted in model 3 and model 4, respectively. As shown in model 3, the research orientation variable is positive and significant (p<0.1) while the re- search orientation square variable is negative and significant (p< 0.05), confirming the non-linear relation between the research orientation and creation of new technology ventures. Figure 6 depicts the effect of the research orientation on the probability of creating new technology ventures while holding other variables at their mean values for academic scientists whose average publication number is below the median. It shows that the cut-off point is approximately at a research orientation of 0.6. Figure 6: Probability of starting new ventures by the nature of research for academic scientists in non-life science whose average numbers of publications are below the median Model 4 shows that the research orientation variable is positive and significant 64 (p<0.05). The research orientation square is negative but not significant. Figure 7 depicts the effect of research orientation on the probability of creating new technology ventures while holding other variables at their means for scientists whose average publication number is above the median. Taken together, results of model 3 and model 4 indicate a weak support of proposition 3.2.2.2. In proposition 3.2.2.3, we predict a decreasing relationship between the level of commercial applicability and the likelihood that an academic scientist creates a new venture when the point on the continuum after which a scientist deems an invention commercially applicable (i.e. is willing to create a new venture) is higher than the point at which an established firm believes it is commercially applicable (i.e. is willing to license the invention). Model 2 shows that the coefficient of bio x research orien- tation is negative and significant (p<0.05). Figure 8 depicts the effect the research orientation on the probability of creating new technology ventures while holding other variables at their mean values for academic scientists in the life sciences, confirming proposition 3.2.2.3. Our findings also show that a number of control variables were significant pre- dictors of academic scientists’ entrepreneurial activity. Model 2 shows that the more recently an academic scientist graduated from his doctoral program, the more likely that scientist is to engage in new venture formation. This is depicted by the positive and significant coefficient of PhD year variable (p<0.001). In addition, women aca- demic scientists are less likely to create new ventures (p<0.05). We also found that academic scientists who graduated from MIT and Stanford are less likely to create new ventures (p<0.05). This may indicate that these scientists are more selective in deciding whether to create new ventures, given their exposure to entrepreneurial activities during their graduate studies. Furthermore, those who graduated from public universities are less likely to en- gage in entrepreneurship than graduates from private universities (p<0.05). This 65 the literature of science-driven entrepreneurship by concluding that academic scien- tists’ nature of research matters in predicting new venture creation. This resonates with entrepreneurship literature which emphasizes a ‘knowledge corridor’ to recognize entrepreneurial opportunities (Shane, 2000). However, these results are not without limitations. For example, entrepreneurship literature has illuminated varying individ- ual characteristics that explain entrepreneurial entry, such as risk-taking propensities and a taste for variety. But all of these characteristics are difficult to obtain from publicly available data. Future research that captures these characteristics in the context of university entrepreneurship will provides valuable insights. 68 T a b le 1 : D es cr ip ti ve st at is ti cs an d co rr el at io n m at ri x M e a n s .d . 1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 S t a r t u p 0 .2 5 6 0 .4 3 7 1 .0 0 0 R e s e a r c h o r ie n t a t io n 0 .5 8 8 0 .4 1 3 0 .1 3 4 1 .0 0 0 B io 0 .4 0 0 0 .4 9 1 -0 .1 0 0 -0 .9 1 3 1 .0 0 0 S c ie n t ifi c p r o m in e n c e 2 .9 2 8 2 .6 6 5 0 .1 5 1 -0 .3 6 6 0 .3 5 4 1 .0 0 0 C o m m e r c ia l a b il it y 2 .9 5 4 8 .9 5 5 0 .2 4 1 0 .1 7 2 -0 .1 5 3 0 .0 6 0 1 .0 0 0 C o m m e r c ia l P a t e n t 0 .0 4 7 0 .1 2 6 0 .0 2 3 0 .2 1 7 -0 .1 8 7 -0 .0 2 6 0 .3 3 1 1 .0 0 0 U n iv e r s it y P a t e n t 0 .0 8 6 0 .1 7 8 0 .0 9 9 0 .0 3 2 0 .0 1 1 0 .3 2 8 0 .4 6 1 0 .2 3 4 1 .0 0 0 G e n d e r 0 .0 9 1 0 .2 8 8 -0 .1 0 5 -0 .0 8 2 0 .1 0 1 -0 .0 3 8 -0 .0 7 4 -0 .0 4 9 -0 .0 3 4 1 .0 0 0 P h D S t a n f 0 .1 1 1 0 .3 1 5 0 .0 5 1 0 .0 9 0 -0 .1 2 5 0 .0 2 2 0 .2 2 1 0 .0 7 4 0 .1 2 5 0 .0 0 0 1 .0 0 0 P h D M I T 0 .1 7 0 0 .3 7 6 0 .0 2 9 0 .2 1 6 -0 .2 5 9 -0 .1 0 8 0 .0 1 9 0 .0 4 3 0 .0 0 3 -0 .0 4 9 -0 .1 6 0 1 .0 0 0 P h D P u b li c 0 .4 4 1 0 .4 9 7 -0 .1 2 3 0 .0 0 5 0 .0 3 5 -0 .0 5 6 -0 .0 7 7 0 .0 1 5 -0 .0 5 6 0 .0 0 3 -0 .3 1 4 -0 .4 0 1 1 .0 0 0 P h D N o n U S 0 .0 8 1 0 .2 7 3 0 .0 1 7 0 .0 3 3 -0 .0 5 3 0 .0 3 8 -0 .0 6 7 -0 .0 3 4 -0 .0 7 1 -0 .0 6 2 -0 .1 0 5 -0 .1 3 4 0 .2 9 7 1 .0 0 0 Y e a r P h D 1 9 7 4 .6 3 0 1 0 .1 2 9 0 .0 8 5 0 .1 2 1 -0 .0 9 0 -0 .2 8 2 0 .0 3 8 0 .1 0 6 -0 .0 5 7 0 .1 5 2 -0 .0 2 7 -0 .1 0 1 0 .1 8 0 -0 .0 2 1 1 .0 0 0 V C 3 0 5 9 .1 5 5 3 4 1 1 .1 0 0 0 .1 7 2 0 .2 3 4 -0 .2 9 5 0 .0 7 2 0 .0 8 0 0 .0 5 4 0 .0 0 9 -0 .0 6 6 0 .2 8 0 0 .0 6 0 -0 .2 0 0 0 .0 4 1 -0 .2 0 4 1 .0 0 0 D e p R e s o r ie n t a t io n 0 .5 8 8 0 .3 7 9 0 .0 9 8 0 .9 1 7 -0 .9 9 6 -0 .3 6 9 0 .1 4 8 0 .1 9 1 -0 .0 1 4 -0 .0 9 5 0 .1 1 6 0 .2 3 8 -0 .0 1 3 0 .0 5 6 0 .1 1 2 0 .2 5 6 1 .0 0 0 D e p S t a r t u p 0 .2 5 6 0 .1 5 3 0 .3 5 0 0 .2 5 7 -0 .2 8 4 0 .1 2 3 0 .1 4 0 0 .0 4 1 0 .0 7 0 0 .0 2 2 0 .1 7 2 0 .2 7 7 -0 .2 7 6 0 .0 1 5 -0 .1 4 8 0 .4 9 2 0 .2 8 0 1 .0 0 0 R o y a lt y 0 .2 7 3 0 .0 5 5 0 .0 8 4 0 .2 3 4 -0 .3 0 1 0 .0 0 0 0 .0 4 1 -0 .0 2 7 -0 .0 3 0 -0 .0 9 0 0 .1 2 2 0 .1 5 4 -0 .1 2 4 0 .0 3 1 -0 .1 4 3 0 .5 9 7 0 .2 5 5 0 .2 4 0 1 .0 0 0 I n d u s t r y F u n d 0 .0 5 9 0 .0 3 7 0 .1 9 5 0 .1 2 2 -0 .1 6 7 0 .0 8 8 0 .1 1 6 -0 .0 3 6 0 .0 4 3 0 .0 4 0 -0 .0 1 8 0 .4 4 9 -0 .2 9 4 -0 .0 2 6 -0 .1 1 5 -0 .0 2 9 0 .1 3 3 0 .5 5 7 0 .3 2 2 1 .0 0 0 T T O 1 7 .3 5 4 3 .3 9 8 -0 .2 6 5 -0 .2 8 0 0 .3 5 3 -0 .1 0 0 -0 .1 3 6 -0 .0 2 8 -0 .0 3 2 0 .0 5 5 -0 .2 1 5 -0 .2 8 8 0 .3 1 0 -0 .0 1 7 0 .2 1 9 -0 .7 2 3 -0 .3 0 6 -0 .7 5 7 -0 .7 0 1 -0 .5 7 4 1 .0 0 0 69 T a b le 2 : R es u lt s of L og it re gr es si on an al y si s p re d ic ti n g ac ad em ic sc ie n ti st s tr an si ti on to en tr ep re n eu rs h ip M o d e l 1 M o d e l 2 M o d e l 3 M o d e l 4 b e t a s .e . b e t a s .e . b e t a s .e . b e t a s .e . C o n s t a n t 2 .0 1 5 0 (3 .3 3 8 8 ) -0 .0 2 8 4 (3 .3 6 3 9 ) -0 .7 1 8 3 (4 .6 4 8 4 ) 2 3 .1 7 7 2 * (1 1 .1 5 9 0 ) R e s e a r c h o r ie n t a t io n 3 .1 2 5 1 * * (1 .3 0 6 1 ) 3 .9 5 1 8 † (2 .4 8 7 0 ) 2 .6 5 5 2 * (1 .3 5 2 8 ) R e s e a r c h o r ie n t a t io n 2 -1 .8 2 6 9 * (1 .0 1 3 9 ) -3 .4 8 6 6 * (1 .9 9 3 0 ) -0 .5 8 2 6 (0 .7 4 5 3 ) S c ie n t ifi c p r o m in e n c e 0 .5 2 2 3 * * * (0 .1 6 6 6 ) 1 .7 2 0 6 (1 .5 6 0 6 ) 0 .4 2 6 3 * (0 .2 0 1 7 ) C o m m e r c ia l im p a c t 0 .8 9 1 5 * * (0 .3 0 4 2 ) 1 .0 1 2 0 * * * (0 .2 9 6 9 ) 2 .4 8 7 2 * * (0 .8 7 3 0 ) 0 .8 2 7 6 * * * (0 .2 6 1 9 ) C o m m e r c ia l P a t e n t -0 .2 7 1 0 † (0 .1 9 2 9 ) -0 .3 2 6 5 * (0 .1 8 8 5 ) -0 .6 0 2 8 * (0 .3 0 5 7 ) -0 .2 3 5 4 (0 .2 3 5 8 ) U n iv e r s it y P a t e n t -0 .1 5 9 5 (0 .2 4 8 8 ) -0 .3 5 5 0 * (0 .1 8 2 8 ) -0 .8 2 3 4 † (0 .6 1 5 9 ) -0 .2 9 6 6 * (0 .1 6 3 5 ) G e n d e r -1 .4 9 3 0 * * (0 .6 2 1 5 ) -1 .2 8 0 8 * (0 .6 2 6 5 ) -0 .9 6 0 0 (0 .9 1 1 4 ) -1 .1 4 9 7 † (0 .8 5 6 9 ) P h D S t a n f -1 .0 6 9 0 * (0 .5 2 5 3 ) -1 .0 0 8 9 * (0 .5 9 0 7 ) -0 .3 9 6 8 (0 .8 3 0 9 ) -1 .6 2 7 1 * (0 .8 0 6 3 ) P h D M I T -0 .8 8 8 7 * (0 .4 5 5 6 ) -0 .7 2 7 7 † (0 .4 6 4 9 ) -1 .1 3 7 0 † (0 .7 6 3 3 ) -0 .5 6 5 9 (0 .6 6 6 8 ) P h D P u b li c -0 .7 0 4 2 * (0 .3 4 7 5 ) -0 .7 8 7 5 * (0 .3 5 4 6 ) -1 .1 4 8 0 * (0 .6 5 2 6 ) -0 .7 3 3 1 † (0 .4 8 8 6 ) P h D N o n U S 0 .3 0 5 2 (0 .5 2 2 3 ) 0 .3 1 4 8 (0 .5 3 1 8 ) 0 .9 4 9 4 (1 .0 3 4 3 ) 0 .0 9 2 1 (0 .7 4 5 4 ) Y e a r P h D 0 .4 6 9 1 * * * (0 .1 4 7 7 ) 0 .5 5 3 8 * * * (0 .1 5 1 2 ) 0 .5 9 9 3 * (0 .2 9 3 0 ) 0 .4 2 6 9 * (0 .2 0 0 7 ) B io -6 .6 8 6 3 (8 .3 4 3 0 ) -4 .1 5 1 2 (8 .2 1 1 4 ) -0 .1 8 1 1 (1 1 .7 7 7 2 ) -6 3 .3 7 6 5 * (2 8 .2 2 2 4 ) B io x R e s e a r c h o r ie n t a t io n -6 .0 8 4 7 * (2 .8 8 5 1 ) -9 .2 1 8 8 * (5 .5 6 6 8 ) -3 .5 0 7 5 * (2 .0 7 1 9 ) V C 0 .0 8 8 0 (0 .4 7 0 3 ) 0 .1 3 6 6 (0 .4 5 1 7 ) 0 .5 0 5 3 (0 .7 2 0 3 ) -3 .2 6 7 9 * (1 .5 5 1 8 ) D e p R e s o r ie n t a t io n -3 .2 8 9 5 (4 .0 6 5 0 ) -1 .9 7 2 0 (4 .0 0 8 4 ) 0 .7 7 2 6 (5 .8 1 2 5 ) -3 1 .9 3 1 1 * (1 3 .7 9 1 7 ) D e p S t a r t u p 1 .4 1 2 3 * * (0 .5 9 4 1 ) 1 .2 0 4 7 * (0 .5 9 0 2 ) 0 .3 5 1 1 (0 .7 9 7 9 ) 5 .7 5 6 0 * * (2 .2 6 4 0 ) R o y a lt y 0 .1 6 0 7 (0 .3 4 5 7 ) 0 .0 4 6 8 (0 .3 3 4 8 ) -0 .6 4 6 0 (0 .5 0 8 3 ) 2 .0 4 3 6 * (0 .9 7 6 9 ) I n d u s t r y F u n d 0 .0 3 5 6 (0 .4 1 4 8 ) 0 .0 8 5 0 (0 .4 0 4 1 ) 0 .1 1 4 6 (0 .6 3 0 9 ) -3 .0 8 7 6 * (1 .4 6 8 0 ) T T O 0 .6 6 8 9 (0 .6 1 4 7 ) 0 .5 6 9 3 (0 .6 4 9 2 ) -0 .3 2 7 8 (1 .1 4 8 9 ) 2 .3 9 4 5 * (1 .2 5 4 6 ) L o g li k e li h o o d -1 7 8 .2 5 5 -1 6 8 .2 3 1 -6 5 .0 7 4 -9 0 .8 9 4 P s e u d o R 2 0 .2 0 6 2 0 .2 5 1 0 .3 2 1 0 .2 6 4 C h i S q u a r e ( χ 2 ) 0 .0 0 0 0 * * * 0 .0 0 0 0 * * * 0 .0 0 3 6 * * 0 .0 0 9 4 * * 70 the problem in a project that comprises of two stages: Stage 1 and Stage 21. At the end of Stage 2, an economic value, V , is generated, where V > 0. At each stage, the project involves two issues: Issue 1, I1, and Issue 2, I2. For example, we can think of Issue 1 as technology related issues and Issue 2 as market related issues. In this section, both issues are equally important in every stage of the project. The entrepreneur obtains an outcome at the end of a stage only if both issues are successful. Otherwise, the entrepreneur obtains nothing. At the beginning of each stage, the entrepreneur decides whether to work alone (i.e., solo-project) or work in a team (i.e., team-project). When a team is employed, the project entails cooperation cost2, c, where c > 0. For simplicity, we start with a team of two-people. In other words, if the entrepreneur decides to work in a team, she adds only one other person into the project. If the entrepreneur works on both issues, the individual probability of failure on each issue increases because the person’s attention is divided. We denote q as the probability of failure on an issue when the person works on both issues, where q ∈ [0, 1). By specializing in one issue, the individual probability of failure on each issue decreases (Sine, Mitsuhashi, and Kirsch, 2006). We denote α as the coefficient of reduction in the probability of failure, where α ∈ (0, 1). We assume that, in a team-project, individuals specialize. We also assume that an individual’s chance of success is independent of each other. The decision of the entrepreneur is depicted in the figure 9. Consider Stage 2, the expected return of a solo-project, E ( ΠS 2 ) , is the probability that both issues are successful multiplied by the final economic value plus the proba- bility that one of the issues is successful multiplied by zero plus the probability that both issues are unsuccessful multiplied by zero. The probability of success of an issue 1For simplicity, we define that a project contains two stages. By induction, it is shown the results do not change if the project comprises of any number of stages. 2We can interpret cooperation cost to include coordination cost, cost of conflict, or the cost of maintaining a team, such as compensating partners for the opportunity cost of joining the team. 73 Figure 9: Decision tree of team formations in a solo-project is (1− q). Thus, E ( ΠS 2 ) = (1− q)2 V (4.2.1) The expected return of a team-project, E ( ΠT 2 ) , is the probability that both issues are successful multiplied by the final economic value plus the probability that one of the issues is successful multiplied by zero plus the probability that both issues are unsuccessful multiplied by zero minus the cooperation cost. Because individuals specialize when working in a team, the probability of success of an issue in a team- project is higher than the probability of success in a solo-project. The probability of success of an issue in a team-project is (1− αq). Thus, E ( ΠT 2 ) = (1− αq)2 V − c (4.2.2) Let the highest expected return be Π2, where Π2 = max { E ( ΠS 2 ) , E ( ΠT 2 )} . Nat- urally, the entrepreneur will choose the mode of work that delivers the higher expected return. Folding back to Stage 1, the expected return of a solo-project, E ( ΠS 1 ) is 74 E ( ΠS 1 ) = (1− q)2 Π2 (4.2.3) and the expected return of a team-project is E ( ΠT 1 ) = (1− αq)2Π2 − c (4.2.4) For Stage 2 to be a team-project, it must be that E ( ΠT 2 ) > E ( ΠS 2 ) (1− αq)2 V − c > (1− q)2 V ⇔ [ (1− αq)2 − (1− q)2 ] V > c (1− α) q (2− (1 + α) q)V > c (4.2.5) For Stage 1 to be a team-project, it must be that E ( ΠT 1 ) > E ( ΠS 1 ) (1− αq)2Π2 − c > (1− q)2Π2 ⇔ [ (1− αq)2 − (1− q)2 ] Π2 > c (1− α) q (2− (1 + α) q)Π2 > c (4.2.6) The advantage of the team-project over a solo-project is contingent upon the specialization effect on each issue as well as the complementarity between the two issues. The entrepreneur favors a team-project if the cooperation cost is less than the expected return when only one of the issues has higher probability of success or when both issues have higher probability of success because of specialization. Proposition 4.2.1 It cannot be value maximizing to have a solo-project operates at a latter stage than a team-project. Proof. Available at the appendix It is possible that the entrepreneur chooses a solo-project at both stages (i.e., SS), or a team-project at both stages (i.e., TT). She may choose a solo-project at Stage 1 and a team-project at Stage 2. However, she will not choose a team-project at Stage 1 and a solo-project at Stage 2 (i.e., TS). The reason is as follows. The outcome of 75 When the probability of failure of an issue is low, the former dominates. In these situations, the advantage of a team-project to a solo-project improves as the probability of failure an issue increases. When the probability of failure of an issue is low, the latter dominate. In other words, the advantage of a team-project to a solo-project diminishes as the probability of failure of an issue increases. In Stage 1 the region where the advantage of a team-project to a solo-project is smaller than such region in Stage 2 because the outcome at the end of Stage 1, the value of a work in progress, is smaller than the outcome at the end of Stage 2, the value of a completed work. A team-project is preferable to a solo-project when the probability of failure of an issue is not too low or not too high. When the probability of failure of an issue is too low, the probability that the entrepreneur solves each issue by herself is high. In this situation, the return from specialization is low. Given the low return from specialization, a solo-project is preferable because it does not entail cooperation cost although the attractiveness of a team-project increases as the probability of failure rises. When the probability of failure of an issue is too high, the probability that both issues are successful is too low despite specialization. Therefore, the entrepreneur favors a solo-project to a team-project. The larger return from specialization, as indicated by the smaller coefficient of specialization, improves the advantage of a team-project to a solo-project because the probability that each issue is successful increases. In addition, the larger return from specialization enlarges the region where increasing probability of failure improves the advantage of a team-project to a solo-project. At the same time, it reduces the region where increasing probability of failure decreases the advantage of a team-project to a solo-project. 78 4.3 Team of n-people In this section we relax the assumption that a team consists of two people. If the entrepreneur chooses a team-project, she adds n−1 people into the project and creates a team of n-people. The reduction in the probability of failure of an issue depends on the team size. Specifically, the coefficient of specialization, α (n), is decreasing in its argument and concave where α ∈ (0, 1]. The larger the team size is the better team members specialize on an issue. The larger team size also demand greater cooperation cost. That is, the cooperation cost, c (n), is increasing in its argument and convex where c (n) ≥ 0. When the team size is one (i.e., a solo-project), the coefficient of specialization is α (1) = 1 and the cooperation cost is c (1) = 0. We begin with the assumption that both issues are equally important at every stage of the project. Consider Stage 2, the probability of success of an issue is (1− α (n2) q). The entrepreneur’s objective function is max n2 E (Π2) = (1− α (n2) q) 2 V − c (n2) (4.3.1) The first-order condition is dE (Π2) dn2 = −2α′ (n2) qV (1− α (n2) q)− c′ (n2) = 0 (4.3.2) Let Π∗ 2 be the highest expected return at Stage 2. Folding back to Stage 1, the entrepreneur’s objective function is max n1 E (Π1) = (1− α (n1) q) 2Π∗ 2 − c (n1) (4.3.3) The first-order condition is dE (Π1) dn1 = −2α′ (n1) qΠ ∗ 2 (1− α (n1) q)− c′ (n1) = 0 (4.3.4) 79 When the entrepreneur adds another person into the project, team members can better specialize, thus increasing the chance of solving each issue. However, a new member increases the cooperation cost. If the looses from cooperation cost is too high, the entrepreneur will choose a solo-project. Otherwise, she will increase the team size until the marginal expected return from larger chance of solving both issues are offset by the marginal increase in the cooperation cost. As noted in Proposition 1, it cannot be value maximizing to have a solo-project operates at a later stage than a team-project. The effect of changes in the economic value on the advantage of a team-project to a solo-project remains the same. That is, a greater economic value enhances the advantage of a team-project. Proposition 4.3.1 1. If a team-project is optimal at both stages, it is not value maximizing to have a smaller team operate at latter stage than a larger team. 2. An increase in probability of failure, q, decreases the optimal team size at Stage k if q > qk but increases the optimal team size if q < qk, where q1 = 1 2α(n∗ 1) − Π2 Π′ 2 − 1 2 ( 1 α(n∗ 1) 2 + ( 2Π∗ 2 Π∗′ 2 )2) 1 2 and q2 = 1 2α(n∗ 2) . Proof. Available at the appendix As explained in the earlier section, the value of a work in progress is less than the value of a completed work because the latter contains uncertainty of successful completion while the former is successfully completed with certainty. Consequently, it is harder for a team-project to generate pay-off at Stage 1 than at Stage 2. Because larger team size comes with greater cooperation cost, it is impossible for a larger team-project to outperform a smaller team-project at Stage 1 if, at Stage 2, the smaller team-project outperforms the larger team-project. Similar to situations when an increase in the probability of failure of an issue does not necessarily enhance the advantage of a team-project to a solo-project, an increase 80 successful. Therefore, an increase in the probability of failure of an issue does not always improve the advantage of a team-project to a solo-project. Similar intuition applies on the optimal team size. That is, a greater probability of failure of an issue increases the optimal team size when the probability of failure is not too high. At the border between the region where greater probability of failure enhances the advantage of a team-project and the region where greater probability of failure lessens the advantage of a team-project is a cut-off point. The cut-off point is the level of probability of failure where the two counter forces are equal. The effect of a decrease in the importance of an issue on the region where greater probability of failure enhances the advantage of a team-project to a solo-project is contingent upon whether the issue is the more important. If the issue is the more important of the two, a decrease in the importance reduces the region where greater probability of failure enhances the advantage of a team-project. If the issue is the less important of the two, a decrease in the importance expands the region where greater probability of failure enhances the advantage of a team-project. The reason is as follows. A decrease in the importance of an issue implies an increase of importance of the other issue. Consequently, the success of the other issue becomes more crucial and the success of the focal issue becomes less crucial for the completion of the stage. If the issue is the less important of the two, an increase in the importance reduces the asymmetry of importance between issues. In contrast, the asymmetry of importance between issues is widened by an increase in the importance of an issue if the issue is the more crucial of the two. Regardless the asymmetry, completion of a stage requires that both issues are successful. When the asymmetry is large, the less important issue becomes a costly necessity in a team-project because specialization on the less critical issue nevertheless involves expense in cooperation. Therefore, an increase in asymmetry reduces the advantage of a team-project to a solo-project. The region where greater probability of failure enhances the advantage 83 of a team-project is at its greatest extent when the issues are equally important. Similar intuition applies on the optimal team size. If the issue is the more im- portant of the two, a decrease of importance reduces the asymmetry of importance between issues. In contrast, if the issue is the less crucial of the two, a decrease of importance enlarges the asymmetry of importance between issues. Because of the asymmetry, the less important issue becomes a costly necessity for completing a stage. Greater team size exacerbates the problem because it involves larger cooperation cost. Consequently, an increase in asymmetry reduces the attractiveness of a larger team- project to a smaller team-project. As in the earlier section, the implication is that the region where greater probability of failure increases the optimal team size is at its greatest extent when both issues are equally important. 4.5 Specialization and Diversity In this section, we consider diversity as one of the benefits from working in a team. Building on literature that argues team diversity enhances the quality of solution (e.g., Milliken and Martins, 1996; Cannella, Park, and Lee, 2008), we specify the economic value of the project as a function of diversity. Because increasing team size allows greater diversity, we capture the effect of diversity on economic value through team size. That is, V (n) = A+k(n) where A is the initial value without diversity and k(n) is the additional economic value because diversity enhances the quality of solution. The additional economic value, k(n), is increasing in its argument and concave. For simplicity, we assume that there is only one stage of development and that both issues are equally important. The entrepreneur objective function is max n E (Π) = (1− α (n) q)2 V (n)− c (n) (4.5.1) The first-order condition is 84 dE (Π) dn = −2α′ (n) qV (n) (1− α (n) q) + (1− α (n) q)2 V ′ (n)− c′ (n) = 0 (4.5.2) Team size influences the expected return from the project in two ways: special- ization and diversity. First, increasing team size allows team members to better specialize. The greater return from specialization is reflected in the smaller prob- ability of failure of an issue. It follows that the probability of a successful project increases. The reason is that the reduction of the probability of failure of an issue is accompanied by a reduction of the probability of failure of the other issue. Sec- ond, increasing team size allows diversity that improves the final economic value of the project. Therefore, it improves the expected return of a project. Specialization increases the chance of obtaining the larger economic value resulting from diversity. However, a new member increases the cooperation cost. If the looses from cooper- ation is too high, the entrepreneur will choose a solo-project. Otherwise, she will increase the team size until the marginal expected return from the larger chance of solving issues and from the larger economic value is offset by the marginal loss due to the cooperation cost. Proposition 4.5.1 1. An increase in the initial value of the project increases team size and team diversity. 2. If α′ (n∗) < ᾱ, an increase in the probability of failure decreases team size and team diversity if q < q̄ . Otherwise, an increase in the probability of failure can increases or decreases team size and team diversity. If α′ (n∗) > ᾱ, an increase in the probability of failure increases team size and team diversity if q < q̄ . Otherwise, an increase in the probability of failure can increases or decreases 85