|
Article Excerpt
We show that universities in the United States that provide stronger royalty incentives to faculty scientists generate greater license income, controlling for university characteristics. We use pre-sample data on university patenting to control for the potential endogeneity of royalty shares. Faculty responds to royalties both in the form of cash and research lab support, indicating both pecuniary and intrinsic research motivations. The impact of incentives is larger in private than in public universities, and we provide new survey evidence on the organization and objectives of university licensing offices to explain this difference. Royalty incentives work both by raising faculty effort and sorting scientists across universities. The primary impact of incentives is to increase the quality rather than the quantity of inventions.
1. Introduction
* Universities are an important source of knowledge creation. In 2006, U.S. academic institutions spent $48 billion on R&D, accounting for 56% of basic research and 33% of total research in the United States (National Science Board, 2008). Academic research has real effects by increasing productivity growth in the economy and stimulating greater private sector R&D through spillovers (Jaffe, 1989; Adams, 1990). In addition, university research contributes to the economy through the licensing of the resulting inventions to private firms. (1) Technology licensing activity has grown dramatically in the past two decades. (2) The number of U.S. patents awarded to university inventors annually increased from 500 in 1982 to 3255 in 2006. During the period 1991-2006, the annual number of licenses granted more than tripled and license revenues increased from $186 million to about $1.4 billion (Association of University Technology Managers, 2006). It is important to understand what drives academic research and technology licensing activity. It is widely accepted in the literature that academics respond to nonpecuniary incentives, such as peer recognition and advancement of science (Dasgupta and David, 1987, 1994), but is it a purely intellectual pursuit, or do monetary incentives also matter?
In this article, we take a first step to answer this question by providing econometric evidence which suggests that high-powered, pecuniary incentives strongly affect university research and licensing outcomes. We examine how cash flow rights from university inventions (the share of license royalties received by academic inventors) affect the licensing income generated by universities. In the United States, university intellectual property policies always grant the university exclusive (first-refusal) control rights over inventions, but the royalty income is shared between the inventor and the university according to specified royalty sharing schedules. We show that there is substantial variation in these royalty sharing arrangements across universities, and use this cross-sectional variation to estimate the effect of royalty sharing arrangements on license income.
To address the potentially serious problem of endogeneity of royalty shares that can arise from unobserved heterogeneity across universities, we use pre-sample information on the university's patenting activity to proxy for the university's fixed effect (following the approach developed by Blundell et al., 1999). It would be more convincing if we could control for fixed university effects, but there is not sufficient variation over time in the royalty sharing arrangements to permit this. Although the pre-sample patent control is very significant and works in the expected direction, one cannot rule out the possibility that there is some remaining unobserved heterogeneity. It is important to recognize that there are some fundamental limitations to what can be said with the available data. Although we will sometimes characterize the empirical results as describing causal relationships, these limitations should be kept in mind. To reach more definitive conclusions, we would need more time series variation in royalty shares than is available in our sample, or instrumental variables that affect royalty shares but not license income. We are not aware of the existence of such instruments and developing them would require a deeper institutional understanding of how universities determine their royalty sharing arrangements.
We develop a simple model in which a scientist makes three types of research effort: basic research, applied research devoted to starting new projects, and applied research to improve the quality of each project. Basic research generates scientific publications. The applied research efforts generate two types of outputs, projects with commercial value and scientific publications. This characterization is based on the argument that scientific research is often dual purpose, frequently referred to in the literature as "Pasteur's Quadrant" (e.g., Stokes, 1997; Murray and Stern, 2006). Scientists value both publications and royalty income. We develop sufficient conditions under which (all three types of) efforts are increasing in the inventor's royalty share. Thus the model predicts that a rise in the inventor's royalty share increases the total level of license revenues generated. We also allow for royalty incentives to affect the sorting of more productive scientists to universities. This sorting mechanism predicts that a rise in the royalty shares of "competing universities" reduces the license revenue for the university. We test these predictions with university-level data from the Association of University Technology Managers, combined with information on the distribution of royalty shares which we collected from university websites.
There are three key empirical findings. First, royalty shares affect the level of license income generated by universities. Controlling for other factors, including university size, quality, R&D funding, scientific composition, and local demand conditions, we find that universities with higher royalty shares generate higher levels of license income. Although recognizing that we cannot definitively establish causality, this finding suggests that the design of intellectual property rights, and other forms of incentives, in academic institutions can have real effects on growth and productivity. Second, the incentive effects of royalty shares appear to work both through the effort and sorting channels. Third, the response to incentives is much stronger (and more significant) in private universities than in public ones.
Under a Bertrand assumption that universities do not expect a strategic reaction from their competitors, we find that in most private universities, and in about half the public ones, the incentive effect is strong enough to produce a Laffer effect, where raising the inventor's royalty share would increase the license revenue retained by the university (net of payments to inventors). However, if universities expect competitors to match changes in their royalty share, this Laffer effect holds for a much smaller subset of universities.
We also show that technology licensing offices (TLOs) are more productive in private universities, on average, suggesting that private institutions have more effective, commercially oriented technology transfer activity. We argue that differences in TLO effectiveness help explain why there is a larger response to royalty incentives in private universities. Because universities retain the control rights over inventions, the TLO has exclusive rights to commercialize inventions disclosed by the faculty (unless expressly waived). As the "gatekeeper," the TLO's effectiveness in licensing activity directly affects the monetary returns to the faculty scientist. Raising the royalty share will have a smaller effect on incentives if the faculty scientist anticipates that the TLO will be ineffective at commercializing her inventions. We provide new survey evidence which shows that TLOs in private universities are more likely to use performance-based pay, are less constrained in their freedom of operation by state laws and regulations, and are more focused on generating license income rather than "social" objectives such as promoting local and regional development. The survey evidence is consistent with our findings that private university TLOs are more effective at generating license income, on average, and that royalty incentives have a larger impact in private universities.
We emphasize that this article is not a normative analysis of university technology licensing activity. Greater commercialization has both benefits and costs. We show that private benefits to universities, in the form of license income, appear to be strongly affected by royalty incentives. The potential costs of commercialization include the reallocation of scientists' effort from basic to more applied research and less "open science" in universities. Although the public debate has focused heavily on such costs, economic research in this area is only just beginning. (3) We do not address these costs in this article.
The article is organized as follows. Section 2 describes the data. Section 3 presents a simple theoretical model of academic research that establishes a relationship between royalty incentives and scientists' research effort. In Section 4, we present the empirical specification and address the empirical issues that arise in testing the main theoretical implications. Section 5 presents the empirical results and their implications, as well as a variety of robustness checks. Brief concluding remarks follow.
2. Data
* The data assembled for this project came from three main sources: (i) the Annual Licensing Surveys for the years 1991-1999 published by the Association of University Technology Managers (AUTM), (ii) the 1993 National Survey of Graduate Faculty conducted by the National Research Council (NRC), and (iii) royalty sharing arrangements downloaded from technology licensing offices' websites. Details of the variables and the sample selection are provided in Appendix A.
The AUTM surveys provide information on licensing income, number of licenses, number of inventions reported to the TLO (invention disclosures), characteristics of the technology licensing office (TLO), and R&D funding from external sources in universities.
To control for differences across universities in faculty size (in the hard sciences) and scholarly quality, we use data from the 1993 NRC Survey. For each university, we have information on faculty size and on three measures of quality for doctoral programs in 23 different fields of science, which we aggregate to the university level using faculty size weights. The primary quality measure we use is the number of citations per faculty during the period 1988-1992. (4)
Table 1 reports descriptive statistics for private and public universities separately. The universities in our sample account for 68.1% of total license income in 1999, as reported by AUTM. These universities generate an average of $3.6 million of license income per year. Not surprisingly, this income is unevenly distributed across universities: the median license income is just $868,000 for private and $539,000 for public universities, but the top 10% of private universities earn over $11.5 million per year ($5.8 million for public). Normalizing by the number of active licenses (row 2) does not eliminate this variation. The median revenue per license is $28,000 for private and $17,000 for public universities, although the top 10% of universities have mean license income above $99,000 and $65,000, respectively. In short, the distribution of license income is very skewed: only a few universities produce very valuable inventions.
Citations per faculty reflect both the quantity and quality of publications and exhibit the highest dispersion across universities. The three measures of quality are highly correlated (with correlations above 0.76). Technology licensing offices at most universities are quite small, with a mean of about three full-time professionals. The average age of TLOs in 1999 was 16, reflecting the stimulus to commercialize university inventions given by the 1980 Bayh-Dole Act. Except for the quality measures--private universities are of higher quality on average--there are no statistically significant differences among the two groups in the other university characteristics.
Our third source of data was information on the distribution of licensing income between faculty scientists and the university, that is, on the arrangements for sharing the royalties generated by the licensed inventions. This information was downloaded from the websites of individual technology licensing offices during the summer of 2001 and it constitutes the novel aspect of our data.
The intellectual property policies of the universities usually state that a percentage of the net income received by the university from licensing an invention is retained by the inventor and the rest is allocated to the inventor's lab, department, college, and to the university. The criterion we used for identifying the inventor share is that the inventor must gain either cash flow rights or direct control rights over the income. Thus, when the university's intellectual property policy states that the share accruing to the lab was under the control of the inventor, we added it to the inventor's share, but otherwise we did not. We call this the inventor royalty share. In Section 5, we examine whether cash payments to the inventor and to her research lab have different incentive effects. This allows us to say something about the relative importance of monetary and intrinsic (research-oriented) motivations.
The observed royalty shares were those in effect (and posted on the web) in 2001. Because we study the impact of royalty shares on licensing outcomes during the period 1991-1999, we wanted to identify any changes that occurred during these years. We sent an email inquiry to the directors of the TLOs in the sample, and found that 70% of the universities did not change their royalty distribution during the sample period. In fact, in many cases the arrangements were set in the early 1980s and never changed. In the universities where royalty shares changed, and where the pre- and post-change levels were available, we assigned the reported values of the royalty shares to the relevant years. (5)
In 58 universities, the inventor royalty share is a fixed percentage of the license income generated by an invention (hereafter, linear royalty schedules). Interestingly, in the other 44 universities, these royalty shares vary with the level of license income generated by an invention (nonlinear royalty schedules). Because the income intervals differ across universities, we divided the license income into seven intervals based on the most frequently observed structure (in US$): 0-10,000, 10,000-50,000, 50,000-100,000, 100,000-300,000, 300,000-0.5 million, 0.5-1.0 million, and over 1 million. (6) For these universities, we compute an expected royalty share by weighting the average share in each income interval by the probability of observing license income in that interval. These probabilities were estimated nonparametrically from the distribution of license revenue per invention over all years in the AUTM sample. Let [v.sub.it] denote license income per invention disclosure in university i in year t. We first estimated the density f([v.sub.it)] by kernel methods at these values. We then computed an average royalty share for each value of v, [bar.s](v), using the royalty schedule for each university, taking into account the varying marginal royalty rates. (7) The expected royalty share is then s [equivalent to] [[summation].sub.v][bar.s](v)[??](v). (8,9)
Table 2 summarizes the main features of the royalty share data. The average inventor share is 39% and 42% for private and public universities using linear royalty schedules, but there is substantial cross-sectional variation within each group. Average royalty shares in the universities with nonlinear schedules is 51%, higher than for the linear schedules, and displaying even larger cross-sectional variation. The striking variation in inventor royalty shares is shown in Figure 1, where the histogram and a nonparametric estimate of the density of the expected royalty share are displayed for private and public universities separately. We exploit this cross-sectional variation to estimate the effect of monetary incentives on license revenue from invention.
[FIGURE 1 OMITTED]
Another striking feature of Table 2 is that inventor royalty shares are either constant or decline in the level of license income per invention--royalty retention is regressive (i.e., the university "tax" on inventors is progressive). On average, they start at 53% in the lowest interval and decline to 30% for inventions generating over $1 million. This feature holds in every quartile of the cross-sectional distribution and, in fact, it holds for every university in our sample with nonlinear royalty schedules. (10)
In order to get some understanding of the determinants of the variation in royalty shares across universities, we split the sample into four quartiles defined by a variety of university characteristics and computed the mean royalty share in each quartile. Table 3 summarizes the results, separately, for private and public universities. Royalty shares are not systematically related to faculty size, the number of citations per faculty (our measure of academic quality), the size of the TLO (measured by the number of TLO professionals per faculty), the age of the TLO, or the shares of the faculty in biomedical sciences and engineering. As the last row in the table shows, we cannot reject the hypothesis that the mean royalty rate is the same across the four quartiles of the distribution for each characteristic. Apparently there is no significant correlation between royalty shares and these university characteristics, taken individually.
These simple bivariate comparisons also hold in a regression context. In a regression of the royalty share on three binary indicators for each of the quartiles of the six characteristics (not reported for brevity), we find that these characteristics are not jointly significant (the p values for significance of the regression are 0.59 and 0.17 for private and public universities, respectively). We also cannot reject the hypothesis that each of the characteristics is individually insignificant. In addition, we ran a probit regression on the choice between linear and nonlinear royalty sharing against the same six characteristics. Again we do not reject the null that these characteristics are jointly insignificant (p values are 0.30 and 0.94 for private and public universities, respectively). Finally, it is worth noting that the top-quality and leading licensing performers are not the universities that offer the highest royalty shares. For example, the royalty shares are 33% for Stanford and MIT, 34% for Harvard, and 49% for Columbia and the University of California System. The overall mean royalty share in the sample is 45%.
To summarize, the two salient features of observed royalty sharing arrangements are their variability across universities and their regressiveness in the level of license income. Moreover, the evidence suggests that neither the form of the royalty sharing arrangement (linear versus nonlinear) nor the level of inventor royalty shares is significantly related to observed university characteristics. Although it is important to study the...
|
