Why do some US universities generate more venture-backed academic entrepreneurs than others?

Abstract

In this study, I identify academic entrepreneurs using biographical information on start-up founders contained in a comprehensive venture capital database. Multivariate analyses are conducted to investigate why some US universities generate more venture-backed academic entrepreneurs than others. I find that national academy membership and total faculty awards are the most significant variables in explaining the number of venture-backed entrepreneurs from a university. In contrast, the abundance of venture capital near the university has no significant effect, which is surprising given that this study focuses exclusively on venture-backed entrepreneurs.

Keywords:

• academic entrepreneur
• university spin-off
• venture capital

Acknowledgements

I would like to thank Nikesh Patel, whose careful and patient work on data coding has been tremendously helpful. This paper has benefited from the comments by Jon Haveman, Amy Ickowitz, Martin Kenney, Young-Choon Kim, Josh Lerner, Ting Lu, Colin Mason, David Neumark, Xue Song, Michael Teitz, Rob Valletta, Brandon Wall, Peyton Young, three anonymous referees, and seminar participants at the Public Policy Institute of California, the Center for Globalization and Information Technology at the University of California at Berkeley, the Technology Transfer Society 26th Annual Conference in Albany, New York, the 26th Annual Research Conference of the Association for Public Policy Analysis and Management (APPAM) in Atlanta, Georgia, and the Department of Economics at California State University, Hayward.

Notes

1. See Pirnay, Surlemont, and Nlemvo (2003) for a typology of university spin-offs.

2. In an early study of life scientists, Louis et al. (1989) even considered engaging in externally funded research and earning supplemental income as ‘academic entrepreneurship’.

3. Data on companies founded to exploit MIT's intellectual property during 1980–96 show that about one-third of them have the university inventor as the lead entrepreneur (Shane 2004, 6–7).

4. There is also literature that studies spin-offs from existing companies that pays more attention to the process of business creation rather than technology transfer. See, for example, Klepper (2001) and Gompers, Lerner, and Scharfstein (2005).

5. McQueen and Wallmark (1982) study spin-off companies from the Chalmers University of Technology in Sweden. Smilor, Gibson, and Dietrich (1990) examine technology start-ups from the University of Texas at Austin. Using personal interviews, Steffensen, Rogers, and Speakman (2000) analyze six spin-off companies from the University of New Mexico. Kenney and Goe (2004) use survey and Internet data to compare ‘professorial entrepreneurship’ at UC Berkeley and Stanford.

6. According to the survey conducted by Association of University Technology Managers (2005, 28), 85 (18.6%) of 458 start-ups licensing technology from US research institutions (including universities as well as research hospitals and research institutes) received venture capital financing. Data on start-ups founded to exploit MIT's intellectual property during 1980–96 indicate that venture capitalists and angel investor groups helped finance 30% of these companies (Shane 2004, 236). In both studies, the start-ups may or may not be founded by academic entrepreneurs, but these results do suggest that only a small share of academic entrepreneurs receive venture capital. However, even if venture-backed academic entrepreneurs only constitute a small proportion of academic entrepreneurs, their start-ups likely possess a higher growth potential and may have a much greater effect on the economy than their share implies.

7. All these authors count publications at the individual researcher level rather than an aggregate level, which is straightforward. However, the analysis in this paper is conducted at the university level. Aggregating the number of publications at the university level would cause many complications because publications in different academic disciplines are hardly comparable.

8. For example, Di Gregorio and Shane (2003) included a set of policy variables to explain why some universities have attracted more start-ups to license their technologies than others. They found that some of the policies, such as inventor's share of royalties and whether the university can make equity investment, do have significant effects.

9. A ‘venture-backed company’ must have received some venture capital investment from venture capital firms or corporate venture capital programs. Once in the database, VentureOne tracks the company's financing from all sources, including bank loans and initial public offerings (IPOs). While I do not count bank loans or money raised through an IPO as venture capital, I do include equity investment made by non-VC corporations or ‘angel investors’ as venture capital in my calculations.

10. See http://www.ventureone.com/products/venturesource.html (accessed January 18, 2007).

11. As noted above, a company would be captured by the VentureOne database as long as it received venture capital financing during 1992–2001. Most of these firms secured venture capital at a very early stage. On average, a company completed its first round of VC financing 16.6 months after its founding date. See Zhang (2007a) for a more detailed description of the VentureOne dataset.

12. All founders in the data are identified by VentureOne in the data-collecting process. A founder is a person who established a start-up company. In the process of venture capital financing, the founder(s) of a start-up give up a proportion of their ownership stake in exchange for equity investment from venture capital firms.

13. For an additional 387 firms, some non-biographical information about the founder is available, but these data cannot be used to identify academic entrepreneurs. For all other firms, nothing is known about their founders. There is even no information about the number of founders each firm has.

14. The availability of founder information is not entirely random, which stems from VentureOne's database management practice. A firm enters VentureOne's database once it receives equity investment from a venture capital firm. VentureOne regularly updates the information about the venture-backed firm until it ceases operation, is acquired, or goes public. Therefore, VentureOne will follow some firms longer than others. VentureOne is more likely to obtain a firm's founder information if the firm has been followed longer. VentureOne also appears to be more likely to capture founder information for firms founded in the late 1990s, possibly because these firms tend to reveal a lot of company and founder information on their websites. For example, among firms with founder information available, 20.5% were founded before 1995; for the rest, 62.4% were founded before 1995. Indeed, firms with founder information tend to be privately held, and are less likely to be out of business, to be acquired, or to complete an IPO, which is consistent with the fact that they are younger.

15. Some founders' bios indicated working experience at some research lab or research center that may or may not belong to a university. I searched the Internet to investigate whether the lab or research center is associated with a university. If it is (e.g. Lincoln Laboratory of MIT), the founder is counted as an academic entrepreneur. Otherwise (e.g. Lawrence Livermore National Laboratory), the founder is not considered an academic entrepreneur.

16. The firm data and the founder data share a common variable, ‘EntityID’, by which I can match a firm with its founder when founder information is available. The matched data then can be used to compute descriptive statistics and compare academic entrepreneurs with other venture-backed start-up founders along many dimensions (see Zhang 2007b, and Zhang forthcoming).

17. If one knows the name of a firm, one could try to use alternative information sources to identify the firm's founders and find out their working experiences. However, because of confidentiality concerns, VentureOne deleted all company names and founder names and replaced them with entity and personnel ID numbers. This makes it impossible for me to supplement the VentureOne founder data.

18. Data downloaded from http://www.census.gov/geo/www/gazetteer/places2k.html (accessed January 20, 2004).

19. The distance (D) between two points (longitude1, latitutde1) and (longitude2, latitutde2) on the earth is calculated using the formula D = R∗arccos [cos(longitude1-longitude2)∗cos(latitude1)∗cos(latitude2) + sin(latitude1)∗sin(latitude2)], where R is the radius of the earth (3961 miles). See the derivation of this formula at http://www.cs.cmu.edu/∼mws/lld.html (accessed March 12, 2004).

20. Data downloaded from http://thecenter.ufl.edu/ (accessed October 22, 2003).

21. Since most of the firms in the VentureOne data were founded in the 1990s, it is desirable to use independent variables in the same period or earlier. However, not all the university-characteristic variables are available in early years. Some of the variables, such as the national academy membership, are available for several years but not addable over time. So I chose the one in the earliest year. This hardly affects the results because university characteristics are fairly stable over time. For example, I run regressions using national academy membership in 1999, 2000, and 2001, and the differences are negligible.

22. Young-Choon Kim has helped with obtaining the data to construct these two variables.

23. Data downloaded from ftp://ftp.uspto.gov/pub/taf/ (accessed November 9, 2005).

24. This is likely the case especially when professorial entrepreneurs want to retain their academic positions.

25. Since the dependent variables are nonnegative integers, I also tried negative binomial regressions as a robustness check. Given the large number of zeros in the dependent variable, the zero-inflated negative binomial model seems appropriate. However, this model requires the specification of an extra equation determining whether the count is zero. If I want to add variables to the main equation one by one, how to re-specify the ancillary equation becomes a rather arbitrary decision. Thus I simply run the ordinary negative binomial regression on the full sample and on a truncated sample dropping all the zeros. These negative binomial regressions yield results qualitatively similar to those from the Tobit regressions, although dropping all the zeros generally gives more precise estimates (with smaller standard errors) than running the negative binomial regressions on the full sample.

26. As Harvard's website shows, it has 10,647 medical school faculty, compared to only 2497 non-medical faculty ( http://www.news.harvard.edu/glance/ (accessed January 18, 2007)).

27. The choice of the four outliers is rather arbitrary. It is based solely on the fact that they overwhelmingly outperformed all other universities and the suspicion that the entrepreneur-generating process in those institutions may be governed by a radically different model. While there are formal statistical procedures available to identify outliers in OLS regressions, they rely on the assumption of a correct model. In this study, I am trying many different model specifications, each of which points to a different set of outliers. Therefore, even if I follow such procedures, my choice of outliers seems equally arbitrary. The whole purpose here is to show that the statistical significance of the VC availability variable is sensitive to the including of a few observations. This point is valid even if the choice of outlier is somewhat arbitrary.

28. The successful stories of Stanford and MIT are almost always told in terms of total spin-off companies they generated. That is the reason why I chose to explain the total number of academic entrepreneurs. However, in many other contexts, ‘firm formation rate’ is probably a more reasonable dependent variable to use (see e.g. Reynolds, Storey, and Westhead 1994).

29. These data were hand-collected from the 13th edition of the International handbook of universities (International Association of Universities 1993). The handbook contains information on the number of faculty members in each university in year 1991 or 1992, which is almost exactly the starting sample period of the VentureOne database that was used to calculate the number of academic entrepreneurs from each university.

30. One would imagine that using distinguished scientists as a selling point should be most common in industries where it takes many years of R&D to develop a marketable product. Start-ups in those industries tend to lose money for many years. It is thus difficult for venture capitalists to sell their equity to other investors if they have nothing to show that the start-ups are promising. Having a star scientist as the founder will likely give investors confidence. Therefore, it is reasonable for venture capitalists to invest more in distinguished-scientist founders. The biopharmaceutical industry is an example of this type. And indeed, the VentureOne data show that more than half of the venture-backed biopharmaceutical start-ups were founded by academic entrepreneurs (Zhang 2007b).

31. One of the Nobel Laureates, Robert Grubbs, is claimed to have founded more than one firm although I was unable to name all of them. The entrepreneurial activities are by no means limited to the Nobel Laureates from the US. For example, I found that at least three Laureates from other countries also started businesses: Arvid Carlsson from Sweden (Nobel Prize in Medicine in 2000, founded Carlsson Research in 1998); Christiane Nüsslein-Volhard from Germany (Nobel Prize in Medicine in 1995, founded ARTEMIS Pharmaceuticals GmbHn (later acquired by Exelixis) in 1997); and Michael Smith from Canada (Nobel Prize in Chemistry in 1993, founded Zymos (now ZymoGenetics) in 1981). Although Michael Smith was associated with University of British Columbia in Canada when he won the Nobel Prize, the company he co-founded was actually located in the United States (Seattle, WA).

32. From the whole country's point of view, it does not matter whether an academic entrepreneur stays local or not. However, this may concern local policymakers who care about local economic benefits from the entrepreneurial activities at universities. Although I find that local venture capital availability does not explain the total number of academic entrepreneurs from a university, abundant venture capital may help keep university spin-offs from moving away or even attract such start-ups from other regions. This is an interesting question for future research.

33. This may explain why models in Tables 5– 6 generally have a low pseudo R² and thus low explanatory power.

34. More generally, Bercovitz and Feldman (2008) have shown that a faculty member is more likely to participate in technology transfer (by disclosing inventions) if the department chair and other faculty members at the same academic rank are active in technology transfer, clearly indicating a peer effect in entrepreneurial decisions among academics.

35. The state of Georgia provides a good example of a policy that targets potential academic entrepreneurs. The University of Georgia, Georgia Tech, and other universities in the state formed a partnership with the local government and industry, called the Georgia Research Alliance. The partnership helps these universities recruit ‘eminent scholars’ to Georgia. These scientists are expected to work as professors and entrepreneurs. They are even offered incubator space (Herper 2002).