Scholarship and Inventive Activity in the University: Complements or Substitutes? 1

Transcription

1 Scholarship and Inventive Activy in the Universy: Complements or Substutes? 1 Brent Goldfarb Universy of Maryland Gerald Marschke Universy at Albany, State Universy of New York Amy Smh Universy at Albany, State Universy of New York First Draft: September 22, 2003 Preliminary and Incomplete Do not quote ABSTRACT: The growing involvement of industry in Universy-based scientific research has been an important subject of debate in the Uned States. Universies are engaging in more licensing and patenting activies then ever before, and the amount of research funded by industry is increasing. The lerature on the subject generally voices a concern that academics commercialization activies may inhib tradional academic scholarship. If the hypothesis is correct and the output of such scholarship is an important input into technological innovation and economic growth, then such an inhibion would be cause for concern. We test this and other hypotheses using a unique dataset on the activies and productivy of Stanford Universy s scientific faculty. 1 We are grateful for financial support of the preliminary analysis from the Center for Policy Research, Rockefeller College, Universy at Albany. All errors are the responsibily of the authors. 1

2 Introduction The growing involvement of industry in Universy-based scientific research has been an important subject of debate in the Uned States. The issue has been examined in context of the Bayh-Dole Act (Mowery, Nelson et al. 2001), as well as more generally (see, for example, Dasgupta and David 1994, or Brooks and Randazzese 1998). 2 Universies are engaging in more licensing and patenting activies then ever before (Henderson, Jaffe et al. 1998; Mowery, Nelson et al. 2001). In addion, the amount of research funded by industry is increasing (Mowery 2001). This lerature generally voices a concern that academics commercialization activies may inhib tradional academic scholarship. If the hypothesis is correct and the output of such scholarship is an important input into technological innovation and economic growth, then such an inhibion would be cause for concern. This paper describes preliminary results from a test of this hypothesis using a unique dataset on the activies and productivy of Stanford Universy s scientific faculty. To accomplish this test, we assembled a comprehensive dataset of Stanford faculty members inventive and academic output. We currently have data pertaining to faculty research support inputs and very detailed accounts of their inventive outputs, and those outputs values as measured by the market (i.e., licensing and revenues). 3 Our data are narrower but more detailed than similar data collected by other researchers (Thursby and Thursby,??). The Thursby and Thursby data allow researchers to follow several thousand academics from many different universies. These data allow the study of universy and universy-discipline specific effects. For example, using these data one is able to evaluate whether the hherto research focus on a few ele universies has created systematic biases in the current lerature s conclusions. Our work complements their contribution. While Thursby and Thursby only observe invention disclosures, our data contain information on patenting, license details (such as exclusivy, equy and royalty rates), revenue streams associated wh each invention, the identy of each licensing firm, inial internal assessment of each invention, a classification by technology category and use, as well as information describing faculty members roles in licensing firms (collected from conflict-of- 2 Of course, one should not lose sight of the fact that universy-industry ties have always been an important aspect of universy research in the U.S. (Rosenberg and Nelson 1994). 3 We are presently working to collect data on both the academic output histories, as measured by qualyadjusted publications, and that capture the indirect influence of these publications on inventive activy, as measured by patent cations to these publications. The completed database will be an 11-year panel of academics input and output activies. 2

3 interest reviews). This depth is important in light of evidence that faculty involvement in the commercialization process post-disclosure is crical for successful technology transfer (Jensen and Thursby 2001). We estimate a joint production function to directly test whether eher inventive activies, or commercialization activies, or both, and scholarly activies are complements or substutes. The distinction between inventive and commercialization activies is important in light of the above-mentioned discovery that inventive activy alone is often insufficient to successfully transfer universy inventions to the private market. It is unknown if any or all of these activies come at the expense of tradional scholarship, or contribute to them. The richness of our data allow us to associate the nature of academic inventors commercialization activies wh their academic output. It is important to consider why Stanford is an interesting universy to study. Stanford consistently ranks in the top 5 universies by several measures of inventive and commercialization activies including inventions, patents, licenses and revenue (AUTM 2002). At the same time, Stanford enjoys a leading reputation in engineering and science fields. Stanford has a history of encouraging interactions between s faculty and industry and has played an important role in the Silicon Valley phenomenon (Rowen 2002). This suggests that Stanford is on the leading edge of the commercialization phenomenon, i.e., the canary in the mineshaft. If we find, say, that at Stanford commercialization and academic outputs are complementary, we should be less concerned that increased commercialization of academia will lead to a substution of outputs at other universies whose interface wh industry is rudimentary but still developing. On the other hand, if we find that top academics are substuting inventive and commercialization activies for academic activies at Stanford, we should be concerned about similar phenomenon at other instutions. Lerature Review Universies have contributed to economic growth in the Uned States (Jaffe 1989; Nelson and Rosenberg 1993; AUTM 1998; Henderson, Jaffe et al. 1998) and this is far from a recent 3

4 phenomenon. 4 For example, in an admtedly tentative analysis, Mansfield (1991) estimates that the annual social rate of return to universy research was 28% between An important theme in these studies is that the return to research varies significantly from industry to industry. While all authors find universy-research output to be most important in drug-related industries, the influence of universy-based scientific activy is pervasive in most industries. This result is robust to different methodological techniques (macro-level regression analysis: (Adams 1990; Jaffe 1989); micro-level regression analysis: (Henderson, Jaffe et al. 1998); survey analysis: (Mansfield 1991; Cohen, Florida et al. 1998; and others); and statistical descriptive analysis: (AUTM 1998; Mowery, Nelson et al. 2001)). In recent years there has been both an increase in the relevance of universy research for commercial applications as well as a growing trend for universies to try to capture a larger share of the social returns of their research output by increasing their technology commercialization activies. This is reflected by an increase in patenting activies (Henderson, Jaffe et al. 1998) and licensing activies (AUTM 1998; Mowery, Nelson et al. 2001; Mowery and Ziedonis 2001; Mowery 2001). It is thought that the observed increase in inventive activy is due to three factors: 1. An increase in Federal funding in fields that lend themselves to commercialization (Mowery, Nelson et al. 2001) 2. An increase in the propensy to commercialize inventions of marginal commercial value (Henderson, Jaffe et al. 1998; Thursby and Kemp 2001) 3. A changing allocation of effort by academics towards projects of potential commercial value (Kenney 1986; Argyres and Liebeskind 1998; Owen-Smh and Powell 2001) 5 These trends have brought to the forefront the debate as to whether universy-industry ties corrupt scientific norms in two important dimensions. First, timely disclosure of new discoveries may be inhibed. 6 This phenomenon is not a subject of this study, and hence we do not discuss further here. 4 Cohen, Florida et al. (1998) provide an excellent survey of the lerature that traces contributions of scientific research to industrial invention. 5 Brickley and Zimmerman (2001) present empirical evidence that academics will divert effort from research (in their case to teaching) in response to changes the relative rewards of alternative activies. 6 For academics, property rights over ideas are established by being first, hence there is a race to disclose and publish discoveries (see Polanyi 1962 and Dasgupta and David 1994). In industry, several methods may be used to establish intellectual property rights: secrecy and patents, control of complementary assets, 4

5 Second, is hypothesized that industry involvement has changed the character of academic research making less basic and more applied. 7 The evidence of this phenomenon is at best, weak (see Cohen, Florida et al for a survey). A key problem is that is difficult to measure whether knowledge is basic or applied. 8 For the remainder, we loosely adopt Stokes idea that knowledge can be both basic and applied simultaneously (see footnote 8). For this exercise, we will consider knowledge that is a useful input into downstream scientific research as basic, and knowledge that is of commercial value, as evidenced by disclosure, patenting and licensing activy as applied. Note that Stokes stratification concerns research motivations, our definion is one of outputs, which has the important virtue of being measurable. That said, we do know that industry-funded research tends to produce output of more short-term value (Blumenthal, et al. 1996). We also know that industry and utilarian-oriented government agencies tend not to take academic reputation into consideration when choosing which academics to fund (Mansfield 1995; Cohen, Florida et al. 1998; Goldfarb 2002). In Biotechnology, there is extensive evidence that academic stars are geographically associated wh commercial activy (Zucker, Darby et al. 1998) and are also associated wh the production of commercial value as measured by inial public offering valuation (Stephan 1998). This may indicate that the elasticy of inventive (applied) output wh respect to cation-adjusted publication (basic) output is posive in Biotechnology. However, is unknown if this is true in other fields, and indeed whether the first-mover advantage and moving down the learning curve, or some combination of these methods. Secrecy and first-mover advantage are clearly associated wh whholding information. There is evidence that these contradictory methods for establishing property rights have led universy researchers wh industrial ties to delay or whhold revelation of discoveries (Kennedy 1986; Kenney 1986, see Brooks and Randazzese 1998 for a survey). 7 There is a long history of tension between the US government and universies as to the character of research. See Goldfarb (2002), for empirical evidence and a complete review. 8 Narin (1978) has suggested a cation-based measure of basic or applied output. The intuion is that applied journals tend to ce basic journals, but not vice-versa. This characterization, which is adopted by Thursby and Thursby, does not s well wh Stokes model (1997). In this model Basic and Applied may be best thought of as orthogonal characteristics of academic activy. There are many examples of research results that are singly basic (discovery of a Jovian moon), singly applied (discovery of a better mousetrap), or both (Recombinant DNA). Stokes aptly names knowledge that is primarily basic as knowledge in Bohr s Quadrant, knowledge that is primarily applied as being in Edison s Quadrant and knowledge in both as being in Pasteur s Quadrant. It is unclear what correlation, if any, measurable academic output has wh the utilarian aspect of Stokes model. That is, is unclear if useful knowledge should be more or less publishable, or ced. Furthermore, is unclear if more basic knowledge should be more publishable or ced. Publications are ced when they are useful inputs into others research. A fundamental discovery in a field that is an intellectual backwater may be very basic, but not easy to publish or well-ced. Because of this, we will be very cautious in considering implications of our results for basic and applied research. 5

6 elasticies of such academic output wh respect to inventive and commercial outputs are posive. 9 There is limed anecdotal evidence that production of commercially-relevant output comes at the expense of the production of academic reputation. Florida and Cohen (1999) report a study by John Servos that documents a tension at MIT s chemical engineering department. Some faculty members wished to pursue basic research, and others research of commercial relevance. The academic reputation of the department suffered when those in the basic camp chose to leave MIT. A similar tradeoff is reported in biotechnology where dissertations of students from a commercially-oriented universy lab were, some claimed, of ltle academic importance (Kenney 1986). Empirical studies of academic production functions are rare. Arora and Gambardella (1998) examine the effect of NSF grants on economists academic output. Their cross-sectional data do not allow them to control for individual effects. They find that receipt of an NSF grant does not have a large effect on academic publications, except possibly for junior faculty. Arora, David et al. (1998) study an Italian biotechnology program. Using a structural model to back out effort from information on budget size, they estimate a production function for research groups. They find that the elasticy of publication output wh respect to budget size is close to uny for highly talented groups, but closer to 0.6 for most groups. Importantly, they find that 60% of the effect of past reputation (i.e., publications weighted by importance) on future publications is due to an increased likelihood of receiving grants, and increased size of those grants. Their regressions also include a dummy variable for grant applications of projects of industrial relevance. Though not the focus of their study but important in the current context, they find that although grant applications wh claimed industrial relevance were more likely to be successful, the elasticy of output wh respect to this claim was insignificant. To the extent that these claims are truthful, this would imply that there was no direct tradeoff between the production of academically relevant output and commercially relevant output in biotechnology. However, this result differs from our study in two important ways. First, their measure of industrial relevance is an indication of the intent of the academics, while we have direct measures of inventive output. Equally if not more importantly, theirs is a study of Italian research groups. As discussed in Goldfarb and Henrekson 9 There is indirect evidence that academic activy is associated wh innovations. Cockburn and Henderson (1998) find that co-authorship of industrial scientists wh academics is associated wh productivy in drug discovery in pharmaceuticals. However, is unclear how these activies influence the output of academic scientists. 6

7 (2003) and Gtelman (2002), European academics are significantly more insulated from market incentives for commercialization of their ideas than their American counterparts. Hence, would be difficult to extrapolate from Arora, David et al. s result to the American context. An addional difference between the US and Europe is the bundling of teaching and research. In Europe, researchers are commonly excused from such responsibilies (this has been documented in many places by numerous authors, but see Henrekson and Rosenberg, 2001, for an exposion). Hence, a proper study of the U.S. would take into account the joint production nature of academic activies in this country. Data We have begun collecting comprehensive output histories of individuals from Stanford as described below. To the best of our knowledge, this is the first dataset that combines inventive, publication, and teaching output histories. Sample Creation: Tenure track faculty employed by Stanford between 1990 through 2000 from the following departments will eventually be included in our study: Electrical Engineering, Applied Physics, Medicine, Pathology, Chemistry, Mechanical Engineering, Biochemistry, the Ginzton Laboratory of Physics, Genetics, Computer Science, Radiology, Pediatrics, Microbiology & Immunology, Materials Science and Engineering, Aeronautics and Astronautics, Physics, and Biological Sciences. Each of these departments reports at least 32 innovations for the period We have obtained a comprehensive list of faculty by department and year, from past Stanford faculty directories, available in Stanford libraries. 10 For the results reported here, however, we use only the two departments for which we have complete data: the Electrical Engineering and Biochemistry departments. We hope to have data for the remaining departments very soon. Publication Output: For each academic in the sample we have complete publication histories. These data are available from the Instute of Scientific Information (ISI) s Science Cations Index (SCI). Because cations take some time to occur, measures of cations are unavailable for articles of recent vintage. Instead, as in Arora, David et al. (1998), we use ISI s journal cation reports. Journal cation measures in the reports are based upon average cation rates to articles in 10 It will also be possible to control for selection in and out of the sample. However, the importance of the selection phenomenon is unknown; hence we have not incorporated into our methodology at present. 7

8 the journal. In later drafts, we will explore various weighting schemes based upon these measures. The Journal cation measures have been collected. Although the SCI is not a comprehensive database of all publications, does cover the most important and influential publications in any field. The ISI builds s database wh an explic bias towards basic work. Garfield (1996) observes that only 150 journals account for half of what is ced and that a core of 2000 journals account for 85% of published articles and 95% of ced articles. The small number of cations to journals not covered by the database suggests that most academically important results are contained whin the journals that the SCI covers. 11 In later drafts, to detect if publications are becoming more commercially relevant we will check for patent cations to articles. These data are available from ISI as well. Teaching Output: We have obtained records of courses taught from course catalogues located in Stanford libraries. Inventive Output: Comprehensive data of inventive output come from Stanford s Office of Technology Licensing (OTL). Stanford has kept comprehensive, computerized records of all inventive and licensing activy since 1996, partial computerized records from and complete archival hard-copy (paper-based) records from the 1970s through the present. The earlier the year, the less likely information will be included in the OTL s computerized database. Because of this, we exhaustively examined paper files of all inventions between 1990 and The data include a comprehensive list of all Stanford inventions that were disclosed to the OTL and deemed of some commercial value. OTL licensing associates estimate that roughly 1/3 of disclosures do not reach a minimal bar of estimated commercial value and a docket, which is a record of an invention disclosure, is not opened for these innovations. Stanford had 2,147 invention disclosures from 1990 through 2001, corresponding to 5,104 unique inventors. From , Stanford was issued 772 US patents; that is, roughly one third of disclosures lead to a patent, although at times an individual disclosure may lead to multiple patents. A patent is a significant hurdle, as Stanford generally does not pursue a patent unless 11 This also suggests that measures of publications and cations from the SCI are measures of knowledge in Bohr s or Pasteur s quadrants. See footnote 8. 8

9 has identified a licensee. 12 We also observe how each disclosure was licensed (that is, exclusively, non-exclusively, wh or whout equy), the term of license, whether or not a product was sold, the amount of royalties earned, and any complementary relationships between faculty members and licensing firms. Hence we have comprehensive measures of inventive output of Stanford faculty from 1990 through Venture Capal: We have collected venture capal outlays by broad industrial category from the SBC database, available through the Robert H. Smh School of Business at the Universy of Maryland. We will use this as a proxy for demand for inventive activy (see below). Method This section of the paper outlines some preliminary analysis as well as our proposals to extend the analysis. Our un of analysis is the academic scientist and the object of our study is his or her scholarly and inventive productivy. We assume academic scientists exert two kinds of effort, effort whose primary purpose is publication, and effort that yields laboratory inventions. 14 not have direct measures of these efforts. We do observe, however, their results. Let We do P and I be respectively the number of publications and number of inventions produced by scientist i in a given period. P and I are assumed a function of abily and effort. We examine whether efforts devoted to publication and research are complements or substutes. On the one hand, effort is subject to a budget constraint---loosely, that efforts must add up to an available effort endowment---so that efforts expended on P and I may be negative inputs in each other s production. This is the crowding-out concern expressed in the introduction. Efforts across activies may also be complementary, however. Effort in the laboratory directed at producing an innovation may complement the production of scholarship because may suggest 12 This information is from personal correspondence wh Kathy Ku, Head of Stanford s Office of Technology Licensing. 13 It is important to note that these measures miss inventive output that cannot be protected by legal intellectual property mechanisms and are transferred by consulting or other methods. 14 Of course, academic scientists also devote time to teaching, administration, consulting activies, and leisure. We plan to obtain include the teaching load datas of onstanford faculty (obtained from course catalogs) in future analyses. We do not have data on administration consulting activies, or leisure, however, so we exclude them from our analysis. 9

10 new avenues of academic scholarship and generate results that can be the basis of academic scholarship. Based on the above discussion, we assume over a given period P is determined according to: (1) P = X b + I d + a i + e where X is a vector that includes a constant, the number of years since receiving a Ph.D., s square, a variable indicating whether i is tenured, lagged output and year dummies. Tenure status and number of years since receiving a Ph.D. reflect i's experience and skills. They may also reflect i's incentives. Number of years since Ph.D. also picks up life cycle effects on productivy; some researchers have found that age is negatively correlated wh a scientist s output (see Levin and Stephan, 1991 and the references ced therein). Tenure status and years since Ph.D. are measured as of the year of output. a i is a fixed effect that captures i's abily. We include lagged output proxies to capture changes in abily, experience or reputation over the 11-year period. e is an error term, independently and identically distributed across individuals and time. While (1) is a reduced form model, s estimation can shed light on whether the estimation of a structural model is likely to be successful. Our parameter of interest is d, which reflects the effect of inventive effort on publication output. Of course, I is endogenous and determined by effort allocations between academic and inventive and/or commercialization activies. The key to the identification strategy is to find demand shifters that affect relative preferences for inventive and commercialization activy as opposed to publication activy. Because we use fixed-effects regressions, is crical that these shifters vary throughout the period , and further that scientists preferences are influenced by these changes. We propose two measures. The amount of venture capal (VC) disbursed in a scientist s broad area of interest. During this period, the size of total national VC disbursements increased 5-fold from a rate of $12B annually in 1990 to close to $60B at the height of the Bubble in As shown, this trend is echoed in the information technology and biological sciences fields (see Table 1). While we do not believe individual scientists spend time following venture capal disbursements, we do feel these activies are correlated wh information flows that would influence beliefs as to the potential returns to inventive activy as 10

11 reported in the daily press, and various activies around campus. (It is clear that this perception was apparent around the country considering the mass influx in workers and capal into the Bay Area during the latter part of the 1990s). When new and fertile technologies are developed, venture capal is assumed to flow towards. These dramatic year-to-year changes in venture capal should reflect changes in the (perceived) relative returns to invention whin a discipline. We also try addional measures that would likely change the perceived relative returns to inventive activies. We use measures of the number and revenue of inventions by colleagues in the scientist s department. There is evidence that the individuals weight strongly the experiences of those in their near viciny when formulating expectations of returns to particular events (Rabin 1998). 15 In this paper, we estimate equation (1) using members of the Electrical Engineering and Biochemistry departments. These departments are two of the more innovative departments at Stanford. For this exercise faculty members in Biochemistry are identified from course catalogues. Thus we observe changes in the department s composion throughout the period. Due to the preliminary nature of the analysis, faculty members in Electrical Engineering were identified from the departmental webse in June Table 2 shows the means of the variables we use in our analysis for all scientist-years. (An academic appears in these data as many times as her years at Stanford between 1990 and 2000). Note that the over the period in these departments, the average number of inventions per researcher was between.3 and.4. Our measures of inventive output are highly skewed, however. The median faculty member generated 1 invention in the Biochemistry department and 0 inventions in Electrical Engineering over the period. The median faculty member in these departments produced no licenses and no royalty income over the period. Results Separate analyses are conducted for the two departments. The estimation results are reported in Table 3. Models 1 and 2 show respectively the results of a least squares and instrumental 15 To identify the parameters in an equation like (1) for describing the determination of inventive output, we would need exogenous shifters of the returns to publication activies. While we have not as yet worked out the details, we believe there are measures available to serve this purpose. 11

12 variables estimation of the determinants of weighted publications, where each publication is weighted by the average cations of the journal in which the publication appears. Both types of regressions are estimated wh fixed effects. In model 1, for both departments, the estimated coefficient for the inventions variable is posive and significant, suggesting complementary of invention effort in the production of publications. The signs on the academic age variables imply that publications increase wh the number of years since Ph.D. but at a decreasing rate. (These estimated coefficients on age in the Electrical Engineering regressions are not statistically significant, however.) The estimated coefficient corresponding to the tenure dummy is always insignificant. In Electrical Engineering, publications appear to predict greater publications in the succeeding year, however. In model 2 we use venture capal (and s lags) as well as the average inventions of the scientist s colleagues (and s lags) as instruments. We find similar results in this specification: an invention in Electrical Engineering increases the number of (weighted) publications by about.7. The coefficient estimate for the biochemistry department is almost 11. That is, an addional invention increases weighted publications by 1/3 in biochemistry (mean weighted publications is 32.3) compared to 1/7 in electrical engineering (mean weighted publications is 5.2). This suggests, as hypothesized, that there is significant variation in the effect across disciplines. In model 3, we repeat the exercise using licenses of an inventor s inventions as opposed to inventions as the endogenous independent variable. 16 We fail to measure a significant coefficient on licenses in Electrical Engineering, suggesting that commercial value, as measured by the decision to license, is not related to academic output. Moreover, if we hypothesize that successful licensing is associated wh inventors commercialization effort this results suggests that such effort is less likely to be complementary to the production of publication output. However, in Biochemistry we find a strong effect relating licenses to qualy-weighted publications. Following the assumption that licensing is associated wh inventor effort, this suggests that commercializing activies complement academic publication activies. This result is consistent wh Zucker and Darby (1998). 16 In these regressions, we use venture capal (and s lags) and the average licenses of the scientist s colleagues (and s lags) as instruments. 12

13 Discussion These results are intriguing. For scientists in the Electrical Engineering and Biochemistry departments, we find a posive relationship between inventive activy, as measured by inventions, and publications, after controlling for researchers characteristics such as age, rank, and department as well as researcher fixed-effects. The estimation procedure employs novel instruments used to proxy for demand-shifters for inventive activy. Moreover, the magnude of the effects are different in the two departments and furthermore the posive effect disappears when we use licenses as a measure of inventive activy as opposed to straight invention counts. This may indicate that academics are allocating effort towards the licensing process and this allocation causes the result to disappear. While intriguing, our results are only preliminary. In the future, we plan to expand our analysis to the remaining departments, and thus explore the disciplinary variation of the crowding out hypothesis. In addion, because the publication variable is an integer, we will employ statistical methods appropriate for estimating models in a panel setting when the dependent variables are count variables and some regressors are endogenous. We propose a generalized method of moments (GMM) estimation of the publication-innovation relationship that controls for the endogeney of the output measures when they appear on the right hand side. We follow Wooldridge (1991) and Windmeijer (2000). Faculty i's publication count in year t, P, is related to the explanatory variablesætaking some liberties wh the notationæaccording to (32) P = exp( X b + I d ) w. We assume dependence among the errors, w, whin a panel. The vector X is assumed exogenous and I, i's inventions in year t, is assumed endogenous, that is, correlated wh I might be correlated wh w if w. w includes, for example, unmeasured abily that contributes both to the production of publications and inventions. These and several addional assumptions give us the following moment condions 17 : 17 The assumptions in total are: (i) w, is equal to b s i s square are uncorrelated wh s, (iii) s is uncorrelated wh, where (ii) the person-specific fixed effect, b i, and s, for t t, (iv) the X process is 13

14 (43) E ( Z l -ldw ) = 0, for each element of Z - z and all l =...,-3,-2,-1,0,1,2,3,... (54) E ( I Dw ) = 0, for alll 2, -l where Dw = w - w- 1and Z is a vector that includes exogenous explanatory variables the strictly exogenous instruments. 18 X and In future work, we also plan to investigate addional questions. For example, the longudinal nature of the data will allow us to investigate whether inventive output exhibs the same lifecycle effects that have been found for publication output. Our data on the fate of universy inventions will allow us to study the determinants of their commercial success and in particular the academic researcher s role in their success. uncorrelated wh b i and s + l for all l, and (v) the I process is assumed correlated wh b i and s - l, l 0, and uncorrelated for l < In our empirical work the lags and the instruments we use produce more moment condions than the number of parameters we wish to estimate. Our estimates of the parameters minimize a quadratic function formed by the weighted sample moment condions corresponding to (4) and (5) and the data. Note that we have made no more assumptions on the moments of w than those embodied in (i) through (v) in the previous footnote. For example, we need not assume homoscedasticy or normaly on the term w. 14

15 Year Table 1 Venture Capal Disbursements by Industrial Category Information Technology (thousands year 2000 USD) Table 2 Variable Means Observations are Faculty-Years Biochemistry Mean Biological Sciences (thousands year 2000 USD) Electrical Engineering Mean (Std. Dev.) (Std. Dev.) Publications (3.935) (3.181) Weighted Publications (36.411) (8.216) Inventions (.871) (.959) Exclusive Licenses (.814) (.547) Non-exclusive Licenses (.767) (.968) License Revenue ( ) ( ) Tenure (.403) (.419) Years since Ph.D (16.141) (12.377) Observations