The effect of federal research funding on formation of university-firm biopharmaceutical alliances

Abstract

The National Institutes of Health (NIH) is the largest funder of health and life science research in the United States. The research sponsored by the agency has continued to aid in the development of new biopharmaceutical therapies, many of which are commercialized via alliances between universities and biopharmaceutical firms. In this paper, we examine this commercialization pathway more closely, evaluating the effects of NIH research funding on US universities' alliance formation. Based on results from instrumental variables models, we estimate that, on average, producing one additional university-firm alliance requires a sustained increase of $294 million in universities' total NIH research funding over the preceding five-year period. In addition, a sustained increase in funding of $100 million over 5 years increases the probability of a university forming at least one alliance by 0.54, or 54 percentage points.

Full text

Translate

Turn on search term navigation

http://crossmark.crossref.org/dialog/?doi=10.1007/s10961-014-9374-7&domain=pdf

Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s10961-014-9374-7&domain=pdf

Web End = J Technol Transf (2015) 40:859876

DOI 10.1007/s10961-014-9374-7

Margaret E. Blume-Kohout Krishna B. Kumar

Christopher Lau Neeraj Sood

Published online: 18 November 2014 Springer Science+Business Media New York 2014

Abstract The National Institutes of Health (NIH) is the largest funder of health and life science research in the United States. The research sponsored by the agency has continued to aid in the development of new biopharmaceutical therapies, many of which are commercialized via alliances between universities and biopharmaceutical rms. In this paper, we examine this commercialization pathway more closely, evaluating the effects of NIH research funding on US universities alliance formation. Based on results from instrumental variables models, we estimate that, on average, producing one additional university-rm alliance requires a sustained increase of $294 million in universities total NIH research funding over the preceding ve-year period. In addition, a sustained increase in funding of $100 million over 5 years increases the probability of a university forming at least one alliance by 0.54, or 54 percentage points.

Keywords University-industry research partnerships Academic-industrial

collaboration Federal research funding

JEL Classication O38 O31 L24 I23

M. E. Blume-Kohout (&)

New Mexico Consortium, Albuquerque, NM, USA e-mail: [email protected]

K. B. Kumar

RAND Corporation, Santa Monica, CA, USA

C. Lau

Pardee RAND Graduate School and Genentech, Santa Monica, CA, USA

N. Sood

University of Southern California, Los Angeles, CA, USA

http://crossmark.crossref.org/dialog/?doi=10.1007/s10961-014-9374-7&domain=pdf

Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s10961-014-9374-7&domain=pdf

Web End = The effect of federal research funding on formation of university-rm biopharmaceutical alliances

123

860 M. E. Blume-Kohout et al.

1 Introduction

The US National Institutes of Health (NIH) is the largest funder of health and life sciences research in the United States. In 2012, the NIH spent over $21 billion to support research and development (R&D) projects, of which approximately $16 billion went to support extramural research projects at US universities.1 The research funded by the NIH has led to many advances in the health and life sciences, including the development of several high-impact therapeutic drugs (Ritzert and Edwards 2000; Chatterjee and Rohrbaugh 2014).

The NIH enjoys a high level of public support, with 64 % of Americans favoring increasing spending to improve and protect the nations health, and 95 % moderately or very interested in medical discoveries.2 Nonetheless, it has not been insulated from recent political pressures to rein in discretionary federal spending. In 2013, sequestration cuts authorized by the 2011 Budget Control Act (BCA) reduced the NIHs total annual budget by 5 %, or $1.55 billion (McDonough 2013). Critics from the scientic community as well as patient advocacy groups have argued that these cuts have threatened the future of the US biomedical research enterprise (Printz 2013). Due in part to these criticisms, Congress has been working to roll back budgets cuts for the NIH in 2014 and beyond. However, as of this writing, there remains considerable uncertainty surrounding the level of public research the NIH will be able to fund in the coming years.

The cuts authorized by the BCA are the latest chapter in a long-standing debate over federal research funding. At the heart of the debate is the following question: how much funding should the federal government allocate toward research? In addition to the burden placed on taxpayers, every dollar spent by the NIH on research means one less dollar for other public needs like education and infrastructure. NIH and other federal research funders thus face continuing pressure to evaluate the effectiveness of federal R&D funding, especially given the recent scal challenges and the need to identify high-return investments for scarce government dollars.

With this context in mind, we evaluated the effects of NIH extramural research grant funding on universities alliances with biopharmaceutical rms. As Stephan (2012) points out, industry support for university research grew during the 1980s and 1990s, approximately doubling its share of the total R&D funds universities expended. In many cases, these university-rm research alliances were spurred by large, established rms wanting to license a specic technology, and by far the most common category of patents issued to US university assignees are biotech, drug, and other medical inventions (Blume-Kohout 2014; Henderson et al. 1998; Guerzoni et al. 2014). In other cases, the industry partner may be a smaller rm with a more tightly dened research portfolio of its own, or a university spinoff. Particularly in the latter case, the alliance might also fund continuing collaboration between an inventors university-based lab and the new company.

Tufts Universitys Center for the Study of Drug Development centers (Milne and Malins 2012) recently identied clinical trials as the most common type of (funded) university-rm partnerships, across a sample of 22 academic medical schools and centers (Milne and Malins 2012). However, while fee-for-service clinical research remains in wide use, they also nd increasing popularity of industry-funded academic drug discovery centersanother type of alliance captured in our data.

1 According to data reported on the NIHs REPORT website: http://report.nih.gov/index.aspx

Web End =http://report.nih.gov/index.aspx .

2 Authors calculation using 2012 General Social Survey data, collected by the National Opinion Research Center (NORC) at the University of Chicago: http://www3.norc.org/gss%2bwebsite/

Web End =http://www3.norc.org/gss?website/ .

123

The effect of federal research funding 861

Thus broadly dened, alliances represent a common and potentially attractive way by which university researchers fundamental academic biomedical research can be leveraged and converted into innovative commercial healthcare products. If we assume rms seek partnerships with highly productive research groups, then alliance formation also serves as a complementary measure in establishing the total impact of NIHs extramural research grant program universities research productivity.

Using instrumental variable estimation and a linear model, we estimate that a sustained $300 million increase in total NIH research grant funding over 5 years increases by one the number of alliances universities form. Poisson models demonstrate similar positive effects: a $300 million increase in a universitys NIH research grant funding increases its rate of alliance formation by 1.2 %. The economic implications of this role of NIH funding could be substantial when one considers that, in addition to university revenue generation from licensing agreements and contracted preclinical or clinical research, these industry alliances can ultimately yield important medical advances to treat disease, improve quality of life, and extend longevity. Additional data and further research on NIH research funding and commercial alliances are needed to fully quantify these effects.

The rest of this paper proceeds as follows. In Sect. 2, we provide background for this research and compare it to previous literature. The data and econometric methodology used in our analysis are presented in Sects. 3 and 4, respectively. Section 5 presents the results, and in Sect. 6, we provide a more detailed discussion of these results, the studys limitations, and our conclusions.

2 Background and prior literature

2.1 The effects of federal research funding

Quantifying the effects of federal research funding presents a series of unique challenges. First, federal research funding can have far-reaching benets, all of which are unlikely to be captured by a single outcome measure. Moreover, the marginal increase in research productivity that results from federal funding can generate social benets that are realized many years into the future. This lag between funding and effects makes it difcult to dene an appropriate period of observation. Lastly, federal research grant funding is rarely randomly allocated across regions and individuals. The absence of random assignment limits the extent to which one can infer causal effects. Together, these challenges have given rise to a diverse set of studies using different approaches and quasi-experimental designs to examine the impact of federal research funding.

Relevant earlier attempts to address these challenges include Payne and Siow (2003), which examined the relationship between total US federal research funding at 68 research universities and their productivity in terms of publications and patents. For identication, this study used an instrumental variable based on university and geographic afliation of US Congressional members sitting on key House and Senate appropriations committees. The authors concluded that a $1 million increase in federal research funding at a university resulted in 0.2 additional patents and ten additional publications.

More often, studies of effects of NIH funding have focused either on individual academic researchers productivity. As an example of the former, a pair of studies by Jacob and Lefgren found that NIHs postdoctoral training grants and R01 research project awards yielded modest but signicant increases in publication production (Jacob and Lefgren 2011a, b). More closely related to our own research question, Bozeman and Gaughan

123

862 M. E. Blume-Kohout et al.

(2007) investigated the effects of grants and contracts from industry versus government sources on individual faculty members propensity towards subsequent closer involvement with industry, for example through co-authoring with industry personnel, serving as a paid consultant, or direct collaboration on patented work. This latter study found academic researchers with government grants were signicantly more likely to interact with industry, overall, likely reecting those faculty members unobserved quality or productivity. But, most importantly, after they control for the faculty members discipline, research center afliation, tenure status, as well as the total research funding he or she received from industry sources, Bozeman and Gaughan (2007) nd faculty members with higher total government-funded research grants are more likely to become involved with industry.

Recent studies have also examined effects of NIH research funding on productivity in the biopharmaceutical industry. Careful descriptive studies of the provenance of new molecular entities (NMEs) have demonstrated the continuing importance of public research funding in supporting commercial biopharmaceutical innovation (Sampat and Lichtenberg 2011; Stevens et al. 2011). However, such descriptive studies do not test the counterfactual, that is, what commercial innovations would have arisen in the absence of NIH funded research. Toole (2012) and Blume-Kohout (2012) both apply econometric approaches to address this issue. Using a model of knowledge stocks and ows, the former estimates a 1 % increase in public basic research funding increases NME applications by1.8 %. Blume-Kohout (2012), using a GMM IV estimation approach, exploits changes in the allocation of NIH funding across diseases to investigate effects of disease-specic funding on the number of drugs entering clinical testing. This latter study found that a sustained 10 % increase in targeted, disease-specic funding increases the number of NMEs entering Phase I clinical testing to treat that disease by 4.5 %.

2.2 University-rm biopharmaceutical alliances

Models like those noted above which connect NIH research funding to downstream biopharmaceutical industry productivity imply some means of knowledge transfer from research institutions to commercial interests. Previous qualitative studies have documented the myriad pathways by which this can occur, and numerous empirical studies since then have attempted to describe and quantify the importance of each (Manseld 1998, 1991). By providing a mechanism for sharing expertise, resources, and personnel, formation of university-rm alliances may serve as the most direct means of knowledge transfer available.

The importance of university-rm alliances to the biopharmaceutical industry is widely recognized. Past studies have suggested such alliances are closely linked to a commercial rms likelihood of success. For example, Zucker et al. (2002) found that when industry researchers co-authored publications with universities star scientists, the rms employing those researchers produced a larger number of patents. Along the same lines, George et al. (2002) found that rms engaging in university alliances were more productive, and experienced lower R&D costs.

The formation of university-rm alliances is closely tied to the productivity of university research. University-rm alliances are a common pathway by which university research is translated into innovations (Foray and Lissoni 2010). These alliances often transfer rights to research discoveries from universities to the participating companies in exchange for some form of nancial support (Stuart et al. 2007). Firms also engage in these alliances to gain access to university partners R&D facilities and expertise (Forrest and

123

The effect of federal research funding 863

Martin 1992). Consequently, one might expect that universities with higher research output and quality would form commercial alliances more frequently, all else being equal.

This expected relationship between alliance formation and university research productivity permits us to consider alliance formation as an intermediate outcome measure for the effects of NIH funding on innovative production. Because licensing agreements for universities patents are one aspect of the alliance measure, we separately estimated descriptive statistics for the effects of NIH funding on universities patents production using data from the NIHs Research Portfolio Online Reporting Tools Expenditures and Results module (NIH RePORTER), and found that 10.8 % of NIH grant funding awarded to US universities from 1998 through 2011 resulted in at least one patent. Among those grant awards that resulted in patents, the median was 1 patent per grant, with an average 2.6 patents per grant.

Building on these earlier studies, we expand the unit of observation from the individual researcher to the institutional level, examining whether an increase in the institutions total NIH-funded research increases its rate of subsequent alliance formation. Because such ventures are legally entered into at the institutional level, and institutional experience, policies, and administrative capacities for contracting and technological transfer may vary widely across institutions, we believe this level of analysis is appropriate and provides useful insight for understanding this specic collaborative pathway.

3 Data

The data for this study were drawn from three sources. Annual NIH research grant awards to individual US universities were extracted from NIH RePORTER. University-rm alliance information was extracted from Deloitte Recaps Deal Builder database. Finally, US Congressional appropriations subcommittee membership information was obtained through review of historical records maintained by the US Congress.

Our initial study panel was constructed using RePORTERs administrative records of NIH funding for research project grants. In a given year, the NIH funds extramural research through newly awarded research grants and commitments from research grants that were awarded in one or more years prior. RePORTER provides a grant-year record for each scal year (FY) in which NIH actively disbursed funds for each research grant. Variables we extracted from each record include total funding disbursed, the awardee organization and principal investigator, and the NIH Institute or Center (NIC) that originated the research grant. We extracted grant-year records for all NIH funding awarded to US higher education institutions from FY 1992 through FY 2011.3 In total, we collected 13,880 grant-year records, representing research grants awarded to 682 unique academic institutions.4

Our primary outcome variable in this study is the count of university-rm alliances for each university, in each calendar year. This variable was constructed using records of pharmaceutical and biotechnology alliances contained in Deloitte Recaps Deal Builder database. Records in this database are maintained using a variety of methods including Freedom of Information Act requests and partial disclosures by private rms (Reuters

3 The records were limited to funding that was awarded as part research project grants that were not afliated with the Small Business Innovation Research (SBIR) and Non-Small Business Technology Transfer (STTR) programs. Funding tied to SBIR and STTR programs is intended to promote the development of university research spin-off companies and small businesses. The focus of this study was to examine the effect of the extramural funding that comprises the bulk NIH budget that is meant to promote.

4 The original data included 694 unique university entries. However, for 12 of the university entries, the names were spelling variants of universities that were already accounted for.

123

864 M. E. Blume-Kohout et al.

2013). Each individual record includes information that identies Licensors and Licensees that form the alliance, the year in which the alliance was formed, the stage of development of the technology that forms the basis of the alliance (if applicable), and information on the relevant disease space for the alliance. While many of the alliances captured in the dataset reect licensing agreements for technology transfers, the alliances also include collaborations.

For example, Duke Universitys three-year, $3.75 million alliance in 1998 with the Targon Corporation included funds for Dukes Comprehensive Cancer Center to conduct preclinical and clinical testing of Targons Investigational New Drug candidates, but it also included grant funding to support Duke University researchers conducting more basic drug and technology discovery, with Targon receiving right of rst, exclusive review for some inventions.5 This agreement is classied and counted as a single alliance in the dataset, and thus by extension our study. Furthermore, although some records also included information on the value or terms of the alliance, in many cases this information was not available, so we do not include nancial terms in this analysis.. To coincide with the NIH funding data, we extracted all records for alliances that were formed between the years 1992 and 2011. Our initial Recap dataset thus contained 3,810 unique pharmaceutical and/or biotechnology alliances.

3.1 Matching alliances to NIH funding

Data on pharmaceutical and biotechnology alliances were matched to universities identied in the NIH administrative records through a series of steps. We started by using a fuzzy matching algorithm to score matches between the names listed in the licensor eld for each alliance record to the set of university names contained in our set of NIH funding data. Using the scores from the fuzzy matching algorithm in combination with manual reviews, we linked the NIH reported university names to alliances that involved the same university. In situations where the alliances involved multiple universities identied in the NIH records, we attributed the alliance to all of the relevant universities. When the alliance record identied the licensor as a university system without specifying the campus, the record was attributed to the largest campus of the university system that was also listed in the NIH administrative data. Finally, alliances that were linked by name to a universitys research institutes or centers, or to its primary teaching hospital, were also assigned to the university.

In our sensitivity analyses, we also applied two separate variants of the last alliance attribution rule noted above. The rst variant was more conservative, and excluded from the universitys alliance count and alliances formed with its associated research institutes or primary teaching hospitals. The second, less conservative, allowed for greater spillover to afliated organizations by including all other identied, university-afliated research organizations and teaching hospitals.

Following the matching process, the NIH funding and biopharmaceutical alliance data were aggregated to create a university-year panel dataset. Due to changes over time in data quality and rates of alliance formation within and across universities, we restricted the analytic panel dataset to focus on alliances formed between 2001 and 2011, with up to 5 years of lagged NIH funding. Then, we excluded any NIH-funded institution that had no

5 Interestingly, with Duke being one of the eight founding institutions to receive funding from the National Cancer Institute as a Comprehensive Cancer Center since 1973, this alliance may also serve more generally as an example of potential for federal R&D funding to encourage non-federal investment in university research alliances. See Duke Universitys press release dated 20 January, 1998, last accessed 24 September 2014: http://www.eurekalert.org/pub_releases/1998-01/DU-TADC-200198.php

Web End =http://www.eurekalert.org/pub_releases/1998-01/DU-TADC-200198.php .

123

The effect of federal research funding 865

Table 1 Descriptive and summary statistics for analytic panel of US universities

Panel data description

Study period for alliances outcome 20002011

Number of university-year observations 1,560

Number of universities 130

Number of alliances 831

Mean SD

NIH Research Grant Funding ($ millions) 67.9 87.2

National Cancer Institute 8.93 12.9

National Heart, Lung, and Blood Institute 8.91 13.5

National Institute of Allergy and Infectious Diseases 8.10 12.5

National Institute of Neurological Disorders and Stroke 5.13 7.32

Mean SD Range

University-rm alliances (N) 0.53 0.97 0, 8

Cancer research 0.17 0.52 0, 7

Cardiovascular research 0.04 0.23 0, 3

Infectious diseases and autoimmune research 0.11 0.38 0, 3

Central nervous system research 0.06 0.25 0, 2

The panel for the study included universities that formed at least one industry alliance during the period 2000 and 2011. Data on alliances types were based on alliance record classications included in the RECAP dataset

biopharmaceutical rm alliance at any time during our study period. The nal analytic panel contains 130 universities, which together represent about 80 % of all NIH extramural funding awarded to US universities in 2011.

Table 1 provides the descriptive statistics for the studys analytic panel. NIH funding awarded to US domestic higher education institutions in 2012 by the National Cancer Institute (NCI) slightly exceeded that awarded by the National Institute for Allergy and Infectious Diseases (NIAID) and the National Heart, Lung, and Blood Institute (NHLBI) $2.18 billion, $2.06 billion, and $1.94 billion, respectivelybut overall, the universities in our analytic panel have tended to receive similar shares of funding from NCI and NHLBI, but slightly lower funding from NIAID. Approximately one-third of alliances formed in the study panel were for cancer-related research, with 67 % of universities forming at least one cancer-related research alliance. At the other extreme, despite the relatively high levels of research funding US universities received from NHLBI, we nd relatively few alliances formed around cardiovascular research. Only 1 in 4 (27 %) universities formed any cardiovascular-related alliance in this period, and half of those only had one such alliance. We also observe, perhaps not surprisingly, that the average number of alliances formed by a university each year is only 0.53 (with median 0), suggesting that these alliances tend to be relatively rare events. In addition, the variance in alliances formed within universities over time exceeds the variance across (between) universities.

Figure 1 shows the aggregate NIH research grant funding and total commercial alliances formed, by year, across our panel of 130 US research universities. Figure 2 shows the correlation across universities in the sum total of their NIH funding and the sum total of their alliances formed, for the entire period. While no clear pattern emerges between

123

866 M. E. Blume-Kohout et al.

Fig. 1 Total annual NIH research grant funding and commercial alliances formed, 20002011. Notes: Funding and alliances formed for analytic panel of 130 US universities, each of which formed at least once alliance during the study period, 20002011. Alliance counts derived from Deloitte Recap data, and funding from administrative data in NIH RePORTER

Fig. 2 Correlation between Universities 5-year average lagged NIH research grant funding and average annual commercial alliances formed, 2000 and 2011. Notes: See Fig. 1 notes for data sources. Universities average lagged NIH funding calculated for each university as the grand average over years 2000 through 2011 of the universitys average NIH funding one to 5 years prior

contemporaneous and lagged NIH research grant funding and commercial alliance formation, as shown in Fig. 1, we do in Fig. 2 observe a relatively high correlation between the average amount of NIH research grant funding each university received during 20002011 and the average number of commercial alliances it formed each year during

123

The effect of federal research funding 867

that same period (rho = 0.644). Our aim, in the remainder of this paper, is to explore any possible causal link between higher NIH funding and higher alliance formation rates, at the university level.

4 Methodology

4.1 Model specication

We modeled the relationship between NIH research grant funding and university-rm alliances in two ways. In our rst model, we assume a linear relationship between NIH research grant funding and the number of university-rm alliances formed. The linear model is attractive in providing a simple base case for testing, and for generating easily interpretable results. However, even among universities with a track-record of alliance formation, in any given year no new alliances may be formed. Inefciency of the linear model in this case, as well as the discrete and non-negative (i.e., count) nature of our outcome variable, suggest using a Poisson model instead. For the Poisson model, we assume a log-linear relationship, such that changes in universities level of NIH research grant funding affect the rate of alliance formation rather than its level.

An important consideration for our studys specication was the appropriate lag structure. The standard argument for including multiple lags of R&D funding or expenditure is that often (depending on the specic outcome studied) outcomes of interest are not realized immediately. In the case of university-rm alliances, the delay might occur for several reasons. For example, even after a university receives notice of award for a given scal year, it takes time for the university to expend those funds in conducting research, and likely still more time would be required for those research efforts to yield tangible results that would attract industry partners. Since 1999, R01 awardsNIHs primary research grant mechanismhave had an average project duration between 4.0 and4.5 years, with durations capped at 5 years (Rockey 2013). If potential industry partners are reacting to the products of NIH-funded research projects, thenfrom the time the grant is rst awardedup to 5 years might elapse before the results of the research are seen. For various reasons, potential university or industry partners may also wish to delay formal alliance until the relevant federal award is completed. Taking these considerations into account, we therefore estimated nite distributed lag models including lags of one to 5 years for NIH research grant funding, then tested sensitivity of our results to several alternative lag specications. We found that, in most cases, statistical evidence (e.g., Akaike Information Criterion) indicates no additional predictive value for inclusion of the fth lag.

We also include in our main specication xed effects for the year and university. Fixed effects were included to account for any bias resulting from secular trends or time invariant university characteristics that are correlated with both alliance formation and receipt of federal funding. For instance, universities with more faculty members or stronger research reputations may attract both greater federal funding and have a greater ability to form alliances. These omitted variables could yield a positive association between federal funding and alliance formation, even if no causal relationship exists. On the other hand, if universities that attracted more funding tended also to conduct more behavioral or fundamental basic research with limited direct biopharmaceutical applications, this might bias our estimate downwards.

123

868 M. E. Blume-Kohout et al.

The Generalized Linear Model (GLM) link function for both the linear regression and Poisson regression are presented in Eqs. (1) and (2), respectively. The terms u and t denote the university and time for each observation. For the linear model, the left-hand side of the equation, E Allianceu;tjXu;t

, is the expected number of alliances formed conditional on our model covariates. For the Poisson model, the left-hand side of the equation, Alliance Rateu;tjXu;t; can also be interpreted as the annual rate of alliance formation. The

right hand side of each equation includes the terms NIH, t, and s, which represent NIH funding, university xed effects, and time xed effects respectively.

EAllianceu;tjXu;t a0 X

5 bn NIHu;t n t s 1

Alliance Rateu;tjXu;t EAllianceu;tjXu;t exp a0 X

" #

The marginal effects of NIH funding implied by the linear and Poisson model are presented in Equations (3) and (4), respectively. Parameter estimates for the NIH funding covariates in the linear model can be interpreted as the average effect of a dollar spent on the expected number of commercial alliances. For the Poisson model, the parameter estimates can be interpreted as the average percentage effect of a dollar spent on the rate of commercial alliance formation.

oEAllianceu;tjXu;t oNIHu;t n

bn 3

oAlliance Rateu;tjXu;t

Alliance Rateu;tjXu;t bn

oNIHu;t n

5 bn NIHu;t n t s

oE Allianceu;tjXu;t

E Allianceu;tjXu;t

oNIHu;t n

In addition to persistent university-specic differences in propensity towards alliance formation, which we account for above with university xed effects, we were also concerned that alliance formation might exhibit some path dependence. This concern is borne out, as an example, in a series of University of Texas alliances with a rm that sought rst to license a patent, and then to gain access to subsequent research produced by the inventors lab. In a sense, the rst alliance begat the second. On the other hand, because alliances are rare events (see Table 1), a university that formalized a new alliance in year t-1 could conceivably be less likely to do so again in year t, all else equal.

To allow for short-run path dependence and also to sharpen our estimates of the effects of lagged funding on current alliances, we also estimated dynamic linear and generalized exponential models. These models include the rst lag of alliances, Allianceu,t-1, on the

right hand side of the estimation equations above, but exclude university xed effects. Finally, because alliances are relatively rare events, we also present results from dynamic linear probability and probit models, with the outcome variable taking on value 1 in any year the university formed one or more alliances, and 0 otherwise.

4.2 Identication strategy

The quality and characteristics of NIH funded research opportunities are important potential sources of bias for our studys models. While a xed effects model controls for

123

The effect of federal research funding 869

time-invariant unobserved heterogeneity, to generate unbiased estimates, our model covariates need to be orthogonal to any unobservable factors that affect universities rates of alliance formation over time. However, changes over time in substantive characteristics of a universitys research initiatives or fundraising strategies may violate this assumption. For example, NIH endeavors preferentially to fund high quality research opportunities (GAO 2014). If these higher quality research opportunities tend also to spur commercial alliances, the estimates of our models will be biased upwards. Similarly, initiatives pursued by universities to increase alliance formation could have occurred simultaneously with increases in federal fundraising activity, causing us to erroneously conclude that federal funding caused growth in alliance formation. On the other hand, if university investigators substitute research funding from industry for federal funds, as Blume-Kohout et al. (2014) observed among top research-intensive universities in the NIH post-doubling era, this could downward-bias our result.

Similar to Blume-Kohout et al. (2014), we used a dual instrumental variable approach to control for bias from unobserved factors that vary over time and across universities. Our rst instrument, predicted NIH funding, combines information about universities historical research specializations with relative year-on-year growth across the individual NIH institute and center (NICs) budgets, as rst presented in Blume-Kohout et al. (2009). For this study, predicted NIH funding level for a university in a given year was calculated by taking the NIH funding received by a university in a base year, in our case 1995, and projecting it forward using a weighted growth rate. The weighted growth rate takes the annual growth rate for individual NIC budgets and weights them according to the proportion of the universitys NIH funding that came from that NIC during an earlier reference period. For this study, the reference period was the years 19921994. The formula for predicted NIH funding is presented in Equation (5).

NIHu;t NIHu;b X

Center Institute Budgeti;t

Center Institute Budgeti;b Budget Shareu;i 5

To check the robustness of our study results and allow an over-identication test for exogeneity of the instruments, we also employed a second instrument inspired by the work of Payne (2003) and Hegde (2009), which demonstrated relationships between Congressional subcommittee membership and the allocation of federal research funding. Both studies found evidence that universities located in districts or states that had more representation on certain congressional subcommittees experienced higher levels of federal research funding. In a related paper, Hegde and Mowery (2008) found that universities in states with an additional representative on the Labor, Health and Human Services, and Education and Related Agencies (LHHE) subcommittee in the House of Representatives experienced a 5.910.3 % increase in NIH research funding. This sizeable effect suggests that state level representation on the LHHE subcommittee may potentially serve as an alternative instrument for NIH funding. We implemented this second instrument by extracting state-level representation on the House of Representatives LHHE subcommittee at the time each scal years budget (including individual appropriations for each NIH Institute and Center) was passed by Congress, via manual review of historical membership records maintained by the House of Representatives.

For our instrumental variable approach to be valid, the proposed instruments must satisfy two conditions: relevancy, and exogeneity. The rst is easily tested, via our rst-stage regressions of NIH research funding on our instruments, and in the earlier studies noted above these instruments have generally shown strong rst-stage predictive power.

123

870 M. E. Blume-Kohout et al.

The exogeneity condition is typically not so easy to prove, so we begin with a summary of intuitive arguments presented in earlier studies. First, for predicted NIH funding, it is important to recognize that federal budgets determined for NICs are the result of a political process with many competing stakeholders, which occurs well in advance of investigator-initiated submissions. While it is certainly the case that an increase in the quality of the research proposals faculty generate in a given year could increase a universitys subsequent portion of a given NICs total budget, it is implausible that Congress would both anticipate NIH receiving relatively higher quality proposals in a future grant cycle for research in a specic NICs domain future, and then act on that by increasing a specic NICs budget relative to other NICs. In essence, while the quality of a universitys ideas submitted in a given year can increase its total NIH funding from a NIC, the quality of a universitys ideas are unlikely to be correlated with relative changes in the NICs total budget.

Similarly, for the Congressional representation instrument, its relevance relies on the fact that universities specialize in particular areas of research, diseases or conditions that are in the domain of a specic NIC. By increasing funding for a specic NIC, a Congressional representative may thus indirectly lower the bar (decrease competition) for universities it represents to obtain NIH funding. However, to reject the exogeneity condition, we would have to argue that changes in the quality of a universitys research ideas or other measures of attractiveness to NIH and industry partners over time are somehow correlated with a political representative from their state being assigned to serve on the LHHE subcommittee. In essence, this would suggest that individual research universities with higher quality research and greater commercial focus could strategically inuence Congressional subcommittee assignments of representatives throughout their state. This scenario seems implausible.

Finally, in addition to these intuitive arguments, our use of two instrumental variables and their associated lags in our analysis permits us to employ overidentication tests for validity, and we nd no evidence to suggest rejection of the exogeneity condition in any of the models we estimated.

4.3 Estimation

The parameters for the instrumented models were estimated via two-stage generalized method of moments (GMM). The rst stage predicts universities rst through fth lags of NIH funding (and, in dynamic models, the rst lag of alliances) using the corresponding rst through fth lags of the predicted NIH funding and Congressional representation instruments. Reported standard errors are clustered on university and robust to arbitrary heteroskedasticity.6

5 Results

The main results for this study are presented in Table 2. Models (1) and (2) denote the non-instrumented linear and Poisson xed effects models, including the rst through fth lags of NIH research grant funding along with university and year xed effects. Models (3) and(4) present results from dynamic linear and Poisson models, replacing university xed

6 We also compared these results with those obtained using cluster-bootstrapped standard errors on 400 replications and found very similar results.

123

The effect of federal research funding 871

Table 2 Effects of NIH research grant funding on university-rm alliance formation counts and rates, linear and poisson models

Fixed effects Dynamic Dynamic GMM IV

Linear (1) Poisson (2) Linear (3) Poisson (4) Linear (5) Poisson (6)

NIH R&D funding

Lag = 1 0.286 0.185 0.337 0.384 0.765 0.943***

(0.335) (0.352) (0.255) (0.276) (2.09) (0.0775)

Lag = 2 0.792 0.932** 0.780 0.882* 0.881 0.213

(0.505) (0.472) (0.521) (0.520) (5.69) (0.177)

Lag = 3 -1.16** -1.39** -1.30** -1.81** -1.62 -1.32***

(0.562) (0.629) (0.642) (0.761) (6.13) (0.175)

Lag = 4 0.204 0.447 0.420 1.062 -1.46 0.347*

(0.376) (0.487) (0.438) (0.675) (5.72) (0.193)

Lag = 5 0.131 0.0515 0.0898 -0.121 1.80 0.294***

(0.320) (0.298) (0.256) (0.292) (2.71) (0.111)

Sum of the lags 0.254 0.223 0.325*** 0.398*** 0.341*** 0.472***

(0.266) (0.264) (0.0541) (0.0636) (0.0077) (0.0208)

Alliances, rst lag 0.239*** 0.256*** 0.214 0.161***

(0.0603) (0.0520) (0.685) (0.0154)

Observations 1,560 1,560 1,560 1,560 1,560 1,560

Results from generalized linear and log-linear models, and from general method of moments instrumental variables estimation, for analytic panel of 130 US universities that formed at least one biopharmaceutical industry alliance between 2000 and 2011. Marginal effects reported for Poisson-type models to aid in comparison of results. Outcome variable for all models is the number of alliances formed by university u, in year t. NIH R&D funding is expressed in hundreds of millions, $USD. All models include year xed effects. Models (1) and (2) include university xed effects. Standard errors reported in parentheses below each estimate are clustered on university and robust to arbitrary heteroskedasticity. Instrumental variables for Models (5) and (6) include the rst through fth lags of predicted NIH R&D funding, and the second through fth lags of Congressional representation

* p \ 0.1; ** p \ 0.05; *** p \ 0.01

effects with the rst lag of the outcome variable. In both sets of non-instrumented models, there is general agreement with respect to the signs and magnitudes of the estimates produced. However, the dynamic models substantially improve precision of the estimates, resulting in highly signicant sums of the lags in both the linear and Poisson models.

The descriptive ndings of the dynamic, non-instrumented models suggest that the receipt of NIH funding is associated with a modest increase in the number and rate of commercial alliances formed over a 5 year period. The sum of the lags in Model (3) is0.325 (p \ 0.001), indicating that a $300 million increase in universities NIH funding over 5 years is associated with about one additional alliance at the end of that period, for a1.2 % increase in the rate of alliance formation.

However, we also observe a signicant temporary reduction in alliance formation

3 years after an increase in NIH funding is received. Our non-instrumented models predict that receiving a temporary increase of $77$104 million in NIH research grant funding in a given year, all else equal, would reduce by one the number of commercial alliances the university forms 3 years later (p \ 0.05).

123

872 M. E. Blume-Kohout et al.

Using predicted NIH funding and Congressional representation as instruments to account for time-varying changes in universities research funding strategies, quality, or commercial focus, we nd similar results, though the coefcient estimates for the individual lags are no longer statistically signicant in the linear model. Our estimates from the instrumented linear model suggest that universities would require, on average, a sustained $294 million increase over 5 years in NIH research grant funding to generate one additional university-rm alliance (p \ 0.001). The GMM IV Poisson-type model indicates that a similar increase in NIH research funding would increase the rate of alliance formation by 1.4 % after 5 years.7

The key results reported in Table 2 are robust to alternative alliance attribution procedures, including both a more conservative rule that excluded alliances from a universitys research institutions, primary teaching hospitals, or research organizations, as well as a less conservative rule that included all potential university afliates, including non-primary teaching hospitals. Table 3 presents results for the alternative linear models, with the magnitude of the ve-year funding increase required to generate one additional university-rm alliance ranging from $304 million for main campus alliances to $262 million when considering all afliates.

Table 3 also presents results from an alternative specication, in which we predict the probability of any alliance for university u in year t. Model (9) is the non-instrumented linear probability model (LPM), and shows that a $100 million increase over 5 years in a universitys NIH research funding is associated with an increase of 0.17, or 17 percentage points in the probability of their forming at least one alliance at the end of that period. Model (10) re-estimates the LPM via GMM IV, with the same instruments used previously, and similarly nds that a sustained $100 million increase over 5 years in NIH research funding increases probability of at least one alliance at the end of that period by 0.18, or 18 percentage points. However, as is often the case when modeling rare events, we nd a substantial difference using an IV probit model versus the LPM. The probit model indicates that a sustained $100 million increase in NIH research funding over 5 years increases the probability of a university forming a biopharmaceutical alliance by over 0.54, or 54 percentage points.

6 Discussion

The primary mission of the NIH is two-fold: (1) to seek fundamental knowledge about the nature and behavior of living systems, and (2) to seek the application of that knowledge to enhance health, lengthen life, and reduce illness and disability.8 Consequently, to evaluate the effectiveness of the NIHs extramural research grant program one needs to consider not only the effects of NIH funding on overall production of knowledge, but also, to what degree does this knowledge contribute to improvements in population health? Our study provides a bridge towards evaluation of the second question, by focusing on a critical pathway for converting basic knowledge produced into commercial technologies: via university-rm alliances.

Our results suggest that NIH research grant funding promotes subsequent formation of commercial alliances at individual universities. Assuming that our models are correctly

7 Marginal effects for the Poisson regression are presented as percentages because the link function for the Poisson model involves the exponentiation of the linear model.

8 See NIH website: http://www.nih.gov/about/mission.htm

Web End =http://www.nih.gov/about/mission.htm .

123

The effect of federal research funding 873

Table 3 Effects of NIH research grant funding on university-rm alliance formation, alternate specications

Main campus All afliates Pr (any alliance)

Linear (7) Linear (8) No IV LPM (9) GMM IV LPM (10) IV Probit (11)

NIH R&D funding

Lag = 1 0.656 0.989 0.0887 0.368 0.828

(0.946) (1.04) (0.122) (0.610) (2.28)

Lag = 2 1.57 0.0672 0.0804 0.0224 0.541

(2.62) (2.92) (0.210) (1.51) (5.24)

Lag = 3 -3.00 -0.602 -0.334 -0.291 -0.979

(4.08) (4.61) (0.243) (1.86) (6.36)

Lag = 4 -0.298 -1.92 0.401* -0.578 -2.29

(4.26) (4.80) (0.233) (1.86) (6.47)

Lag = 5 1.40 1.85 -0.0637 0.663 2.45

(1.85) (2.05) (0.154) (0.824) (2.93)

Sum of the lags 0.329*** 0.382*** 0.172*** 0.184*** 0.543***

(0.0657) (0.0770) (0.0213) (0.0193) (0.0787)

Alliances, rst lag 0.266*** 0.254*** 0.110*** 0.0986*** 0.309***

(0.0540) (0.0572) (0.0342) (0.0432) (0.0794)

Observations 1,560 1,560 1,560 1,560 1,560

Results from two-stage instrumental variables estimations (models (7), (8), (10), and (11)) and a non-instrumented least-squares linear probability model (9), for analytic panel of 130 US universities that formed at least one biopharmaceutical industry alliance between 2000 and 2011. NIH R&D funding is expressed in hundreds of millions, $USD. All models include year xed effects. Average marginal effects reported for probit model (11) lieu of coefcients. Outcome variable for models (7) and (8) is the count of alliances formed by university u, in year t. Model (7) restricts alliance counts to universities main campus. Model (8) expands alliance counts to include all university afliated hospitals, research institutes and centers. Outcome variable for models (9)(11) takes on value 1 if any alliance was formed by university u in year t, and 0 otherwise. Standard errors reported in parentheses below each estimate are clustered on university and robust to arbitrary heteroskedasticity. Instrumental variables for Models (7), (8), (10), and (11) include the rst through fth lags of predicted NIH R&D funding and Congressional representation

* p \ 0.1; ** p \ 0.05; *** p \ 0.01

specied and our instruments are valid, we can interpret our ndings as saying that, among universities that formed one or more biopharmaceutical alliances in the last decade, a sustained increase of about $300 million over 5 years yields, on average, one additional research alliance with a biopharmaceutical rm. Moreover, a sustained $100 million increase in NIH funding over 5 years may increase the probability of a university forming any alliance by more than 50 % points. Though the magnitude of this effect appears small, each alliance has potential to yield one or more disease treatments, potentially affecting the lives and health of hundreds or thousands of individuals. In addition, by encouraging greater interaction between universities and industry, alliances can facilitate placement of trained graduate students and postdocs, potentially improving the correspondence between these training experiences and trainees subsequent careers (Bozeman and Gaughan 2007; Tilghman et al. 2012).

The relatively small effect of a sustained increase in funding we observe based on the sums of the lag coefcients seems consistent with two possible explanations. First, a sustained increase in NIH research funding may serve as a substitute for university revenue generation from other sources, especially given existing short run limitations on facilities,

123

874 M. E. Blume-Kohout et al.

capital equipment, and administrative capacity (including principal investigators time). Second, as our descriptive statistics demonstrate, NIH-funded research is heterogeneous in its effects on alliances. Just as universities differ in their research specializations, so too do NIH Institutes and Centers differ in their emphasis on translational research for funding decisions. Universities also differ in their relative propensities to exploit inventions, as shown in prior studies documenting the relatively higher unexplained variation in researcher-level outcomes and the importance of other institutional characteristics in inuencing exploitation of universities intellectual property (Bozeman and Gaughan 2007; Di Gregorio and Shane 2003).

With respect to policy implications, the results from this study should be considered with several caveats. Research supported by the NIH can advance medical science and improve health in a number of ways, and many important scientic discoveries do not result in technologies that can be commercialized. Our study examines just one of many pathways by which NIH-sponsored research might produce economic and social returns. To the extent that NIH-funded research increases the production of knowledge on the margins, this knowledge may improve health and reduce illness by improving healthcare delivery systems, facilitating the creation of new start-ups, and inducing existing rms to explore new avenues of research, independent of universities. Indeed, past research has found that university scientists engage in a wide variety of informal technology transfer activities (Link et al. 2007). All of these factors suggest that the results of our study alone offer a lower bound for the total social and economic effects of NIH research grant funding.

There are several important limitations in the data and models we employ as well, which may be addressed in future research. First, the alliance data used in our study lacks information to identify the nancial value of each alliance. The nancial value of an alliance would provide information one could use in evaluating the relative importance or economic impact of NIH research grant awards. Second, this analysis also ignores possible relationships between or across commercial alliances, including whether multiple alliances in a given year represent a single commercialization effort around one technology, or multiple innovative technologies each potentially to be developed in its own right. Finally, alternative econometric specications and estimation techniques might improve precision of our estimates. In particular, the instrumental variables used in this paper rely on variation across NIH institutes and centers in their respective shares of funding awarded to each university, each year. Future research may nd alternative approaches that permit closer investigation of funding and alliances within specic therapeutic areas, for example the relationship between universities NCI funding and their subsequent alliances with industry to develop treatments for cancer.

Evaluating the downstream effects of NIH research grant funding presents several challenges. Our study sheds light on one aspect of these effects, by evaluating how changes in NIH funding levels impact formation of university-rm alliances. The results we have presented suggest that NIH research funding promotes the formation of university-rm commercial alliances, and by extension, may promote the development of new health-related innovations. However, more research is needed to draw stronger conclusions about the magnitude of this effect, and also to investigate further potential sources of heterogeneity in these effects not only due to differences across universities, but also across NIH funding mechanisms. Future studies evaluating the effect of NIH funding on alternative commercialization pathways may help to complete our perspective of the economic and social impact of the NIHs research funding programs.

123

The effect of federal research funding 875

Acknowledgments This material is based upon work supported by the National Science Foundations Science of Science and Innovation Policy (SciSIP) program, Grant awards 1064215 and 1355279.

References

Blume-Kohout, M. E. (2012). Does targeted, disease-specic public research funding inuence pharmaceutical innovation? Journal of Policy Analysis and Management, 31(3), 641660.

Blume-Kohout, M. E. (2014). Understanding the gender gap in STEM elds entrepreneurship. SBA Ofce of Advocacy Report No. 424. Washington, DC: U.S. Small business administration.

Blume-Kohout, M. E., Kumar, K. B., & Sood, N. (2009). Federal life sciences funding and university R&D. NBER Working Paper No. w15146. Cambridge, MA: National Bureau of Economic Research. Blume-Kohout, M. E., Kumar, K. B., & Sood, N. (2014). University R&D funding strategies in a changing federal funding environment. Science and Public Policy. doi:http://dx.doi.org/10.1093/scipol/scu054

Web End =10.1093/scipol/scu054 .

Bozeman, B., & Gaughan, M. (2007). Impacts of grants and contracts on academic researchers interactions with industry. Research Policy, 36(5), 694707.

Chatterjee, S. K., & Rohrbaugh, M. L. (2014). NIH inventions translate into drugs and biologics with high public health impact. Nature Biotechnology, 32(1), 5258.

Di Gregorio, D., & Shane, S. (2003). Why do some universities generate more start-ups than others?Research Policy, 32(2), 209227.

Foray, D., & Lissoni, F. (2010). University research and public-private interaction. Handbook of the Economics of Innovation, 1, 276314.

Forrest, J. E., & Martin, M. (1992). Strategic alliances between large and small research intensive organizations: Experiences in the biotechnology industry. R&D Management, 22(1), 041054.

GAO. (2014). Research priority setting and funding allocations across selected diseases and conditions.Washington DC: US Government Accountability Ofce.

George, G., Zahra, S. A., & Wood, D. R, Jr. (2002). The effects of businessuniversity alliances on innovative output and nancial performance: A study of publicly traded biotechnology companies. Journal of Business Venturing, 17(6), 577609.

Guerzoni, M., Aldridge, T. T., Audretsch, D. B., & Desai, S. (2014). A new industry creation and originality: Insight from the funding sources of university patents. Research Policy, forthcoming. doi:http://dx.doi.org/10.1016/j.respol.2014.07.009

Web End =10.1016/j. http://dx.doi.org/10.1016/j.respol.2014.07.009

Web End =respol.2014.07.009 .

Hegde, D. (2009). Political inuence behind the veil of peer review: An analysis of public biomedical research funding in the US. Journal of Law and Economics, 52(4), 665690.

Hegde, D., & Mowery, D. (2008). Politics and funding in the US biomedical research system. Science, 322,17971798.

Henderson, R., Jaffe, A. B., & Trajtenberg, M. (1998). Universities as a source of commercial technology: A detailed analysis of university patenting, 19651988. Review of Economics and Statistics, 80(1), 119127.

Jacob, B. A., & Lefgren, L. (2011a). The impact of NIH postdoctoral training grants on scientic productivity. Research Policy, 40(6), 864874.

Jacob, B. A., & Lefgren, L. (2011b). The impact of research grant funding on scientic productivity. Journal of Public Economics, 95(9), 11681177.

Link, A. N., Siegel, D. S., & Bozeman, B. (2007). An empirical analysis of the propensity of academics to engage in informal university technology transfer. Industrial and Corporate Change, 16(4), 641655. Manseld, E. (1991). Academic research and industrial innovation. Research Policy, 20(1), 112. Manseld, E. (1998). Academic research and industrial innovation: An update of empirical ndings.

Research Policy, 26(78), 773776.

McDonough, J. E. (2013). Budget sequestration and the US health sector. New England Journal of Medicine, 368(14), 12691271.

Milne, C.-P., & Malins, A. (2012). Academic-industry partnerships for biopharmaceutical research & development: Advancing medical science in the US. Boston, MA: Tufts Center for the Study of Drug Development, Tufts University School of Medicine.

Payne, A. A. (2003). The effects of congressional appropriation committee membership on the distribution of federal research funding to universities. Economic Inquiry, 41(2), 325345.

Payne, A. A., & Siow, A. (2003). Does federal research funding increase university research output?Advances in Economic Analysis & Policy, 3(1), 124.

Printz, C. (2013). Federal budget woes continue to imperil cancer research. Cancer, 119(23), 40554056. doi:http://dx.doi.org/10.1002/cncr.28473

Web End =10.1002/cncr.28473 .

123

876 M. E. Blume-Kohout et al.

Reuters, T. (2013). Recap Factsheet. http://www.recap.com/sites/rc/files/pdf/recap-iq-deal-builder-select-factsheet.pdf

Web End =http://www.recap.com/sites/rc/les/pdf/recap-iq-deal-builder-select- http://www.recap.com/sites/rc/files/pdf/recap-iq-deal-builder-select-factsheet.pdf

Web End =factsheet.pdf .

Ritzert, A., & Edwards, C. (2000). The benets of medical research and the role of the NIH. Washington,DC: U.S. Senate Joint Economic Committee.

Rockey, S. (2013). How long is an R01? Extramural Nexus. http://nexus.od.nih.gov/all/2013/11/07/how-long-is-an-r01/

Web End =http://nexus.od.nih.gov/all/2013/11/07/how- http://nexus.od.nih.gov/all/2013/11/07/how-long-is-an-r01/

Web End =long-is-an-r01/ . Accessed 12 Nov 2014.

Sampat, B. N., & Lichtenberg, F. R. (2011). What are the respective roles of the public and private sectors in pharmaceutical innovation? Health Affairs, 30(2), 332339. doi:http://dx.doi.org/10.1377/hlthaff.2009.0917

Web End =10.1377/hlthaff.2009.0917 . Stephan, P. E. (2012). Funding for research. In How economics shapes science (pp. 111150). Cambridge,

MA: Harvard University Press.

Stevens, A. J., Jensen, J. J., Wyller, K., Kilgore, P. C., Chatterjee, S., & Rohrbaugh, M. L. (2011). The role of public-sector research in the discovery of drugs and vaccines. New England Journal of Medicine, 364(6), 535541. doi:http://dx.doi.org/10.1056/NEJMsa1008268

Web End =10.1056/NEJMsa1008268 .

Stuart, T. E., Ozdemir, S. Z., & Ding, W. W. (2007). Vertical alliance networks: The case of university biotechnologypharmaceutical alliance chains. Research Policy, 36(4), 477498.

Tilghman, S., Rockey, S., Degen, S., Forese, L., Ginther, D., Gutierrez-Hartmann, A., et al. (2012). Biomedical research workforce working group report. Bethesda, MD: National Institutes of Health. Toole, A. A. (2012). The impact of public basic research on industrial innovation: Evidence from the pharmaceutical industry. Research Policy, 41(1), 112.

Zucker, L. G., Darby, M. R., & Armstrong, J. S. (2002). Commercializing knowledge: University science, knowledge capture, and rm performance in biotechnology. Management Science, 48(1), 138153.

123

Word count: 8949

Show less

Springer Science+Business Media New York 2015

The effect of federal research funding on formation of university-firm biopharmaceutical alliances

Content area

Abstract

Full text