16 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

The Unequal Benefits of Academic Patentingfor Science and Engineering Research

Mario Calderini, Chiara Franzoni, and Andrea Vezzulli

Abstract—The increase of university patents has raised issuesof potential conflicts of interest in Faculty activities. Nonetheless,recent empirical evidence has indicated that very productive scien-tists contribute disproportionally to academic patenting and thatinventing is likely to encourage an increase in scientific productiv-ity. This article adds to this evidence by showing that such beneficialeffects are not likely to be earned equally by every scientist. Theanalysis was run in a large sample of Italian scientists contributingto materials sciences in either chemistry or engineering of ma-terials, and makes use of several econometric techniques that aresuitable for treating unobserved heterogeneity, excess zeros, and in-cidental truncation. Results indicate that benefits are higher whenthe feedback from applied research is richer, and when regimesof secrecy are less harsh, which is more likely to be the case withengineering, as opposed to hard science research. If confirmed byfurther evidence, the findings suggest that academic policies in mat-ters of intellectual property rights should be refined and tailoredto field specificities.

Index Terms—Science and engineering research, science policy,technology transfer, university management, university patents.

I. INTRODUCTION

IN RECENT years, academic institutions have becomewidely involved with technology transfer, and academic

patenting is one of the most widespread and visible outcomes ofsuch activity. The number of universities involved with patent-ing has increased significantly during the last two decades, afterinstitutional and legal changes such as the 1980 USA Bahy–Dole Act (and similar ones undertaken in European countries)were put in place, mirroring the political consensus behind thenew role given to universities of becoming professional tradersof technologies and applications for industry.1

Manuscript received January 7, 2007; revised January 1, 2008, January 4,2008, and January 6, 2008. Current version published January 21, 2009. Reviewof this manuscript was arranged by Department Editor P. E. Bierly. The work ofM. Calderini and C. Franzoni was supported by the Ministero dell’Universitae della Ricerca (MIUR), Italian Ministry for Education, Universities, and Re-search (FIRB, Project RISC-RBNE03ZLFW_004: “Reorganization of publicresearch’s technology transfer system: governance, tools and strategies”). Thework of A. Vezzulli was supported by the Ministero dell’Universita e dellaRicerca (MIUR), Italian Ministry for Education, Universities and Research(FIRB, Project RISC-RBNE039XKA: “Research and entrepreneurship in theknowledge-based economy: the effects on the competitiveness of Italy in theEuropean Union”).

M. Calderini and C. Franzoni are with the Department of Production Systemsand Management (DISPEA), Politecnico di Torino, 10129 Torin, Italy (e-mail:chiara.franzoni@polito.it).

A. Vezzulli is with the Knowledge, Internationalization and Technology Stud-ies (KITES), Bocconi University, 20136 Milan, Italy, also with the Universityof Milan–Bicocca, 20125 Milan, Italy, and also with the University of Brescia,25121 Brescia, Italy.

Digital Object Identifier 10.1109/TEM.2008.20098891After the 1980 Bahy-Dole Act, similar laws that assign de jure IPRs from

publicly funded research to the principal investigator’s institutions were ap-proved in Canada, UK, and nearly all western European countries, except fromItaly, Finland, and Sweden [50].

A number of investigations have recently been conducted thataim at assessing the magnitude of the phenomenon and its likelyimpact on the rate of countries’ inventive activities, as well ason the productiveness of scientific research.

Current U.S. figures are impressive: according to the Associ-ation of University Technology Managers, in 2005 U.S. memberuniversities filed 15 115 patent applications (an average increaseof 55% year by year in the last 5 years), which resulted in 3278granted patents; they helped create 628 new companies (nearlytwo every day), held more than 28 000 active licenses—5000 ofwhich started in 2005—and were responsible for the launch of527 new products on the market. Although it is true that patent-ing activity has increased overall in every sector of the economy,and that such an increase was fueled by more opportunities inthe biotech and information and communication technologiessectors2 [28], [42], the surge of U.S. academic patents regis-tered after 1980 was steeper than the surge of corporate patentsin the same period [30]. Incentives set by the internal universitypolicies were effective in driving a larger number of disclosureto the local technology transfer offices (TTOs), thus producingmore patents issued and licensed directly by universities [56].3

Although comparable aggregate statistics are unfortunatelyunavailable for European countries, European universities haveexperienced a sharp increase in activities, which, however, areperhaps smaller in magnitude ([25]; see also [51]). Measuresare complicated by the fact that in many European countries,professors, rather than universities, are allowed to own the in-tellectual property rights (IPRs) of research (this is the casein Italy, Finland, Sweden, Switzerland, and Germany before2001, although universities are often free to set internal policiesof IPRs ownership). Therefore, searches need to be made bymatching names of professors in the inventors’ databases [10].Based on the Patval survey, it was estimated that between 1992and 1997 roughly 5% of all European patents were owned eitherby universities or by academic professors [59].

This trend has raised mixed feelings on both sides of theAtlantic. Concerns began to grow after several comparisonsof the pre- and post-Bahy–Dole Act hinted that the quality ofuniversity patents was starting to decrease. In fact, a popularinvestigation of American data of the late 1990s showed thatwhereas the top performing institutions per number of patentswere concentrated among the best research universities both

2Among the key determinants were the U.S. supreme court’s decisions to al-low patentability of genetically modified organisms (Diamond vs. Chakrabarthi,1980), software codes (Diamond vs. Dieh, 1981), and business methods (StateStreet & ATT vs. Excel, 1998). See [28].

3Until then, inventions were frequently patented by external and partiallyindependent institutions, such as the Research Corporation and Alumni Asso-ciations [43].

0018-9391/$25.00 © 2008 IEEE

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 17

before and after the 1980s, smaller institutions, especially thenewcomers to technology transfer (i.e., those universities thathad never patented before 1980), were starting to producepatents of lower importance and generality later on [30]. Be-cause the higher quality of university patents was seen as depen-dent on the wider scope and longer decay rate of academic inven-tions as compared to firms’ inventions, this evidence provokedconcern that commercialization of science was associated to adeterioration of the breadth and basicness of academic research.

On the one hand, academic patenting and technology trans-fer at large were seen as a viable way to ease communicationbetween science and the market, encourage mutually beneficialexchanges of ideas and competencies, and unlock science froman ivory-tower position that allowed only limited impacts onsocial and economic wealth. On the other hand, there was fearthat immoderate commercial openness would come at the ex-pense of less lucrative activities, such as long-term research andstudent education (see for instance [32] and [35]). The risk—itwas stressed—was that researchers would divert their researchfocus toward applications that they could resell on the market,thus pursuing less general and less basic research strategies.This issue is clearly of crucial importance for policy makers anduniversity administrators, who certainly do not want to lower theimpact of funded scientific research for the sake of commercial-izing technology: an activity that, so far, has become profitablefor very few institutions, though certainly expensive for all [57].

To shed light on this issue, several recent empirical investiga-tions have been presented that aim at comparing the productivityof scientists in relation to their inventive activity. Those inves-tigations have brought up several unexpected scenarios, whichhave contributed to dissipating the concerns. The results of suchinvestigations will be discussed in depth in the next section ofthe paper.

A point deserving specific attention, although one that is oftenoverlooked by both policy makers and university administrators,is that so far, policies in matters of IPR have been applied in thesame way to every academic field, irrespective of any potentialdifference in the models of technology transfer and of scientificinvestigation on which they are based. Moreover, the evidenceproduced so far is disproportionately based on life sciences andbiotech research, which certainly have offered, and continue tooffer, the most influential model for inspiring policies in mattersof academic IPRs, possibly because they represent the largestshare of university revenues, and because of the contiguity ofacademic and corporate R&D.

When one looks at firms, academic research is not perceivedas equally important in every discipline and subfield [15]. Dif-ferent models of interaction take place between science andacademia, depending on several factors, such as the industry’slevel of absorptive capacity, the degree of academic knowledgecodification, the scientific and technological paradigms, the wayscientific facts are constructed, and many other factors that con-tribute overall to making interactions looser or tighter, more orless fruitful, and more or less time-consuming, thus affectingthe returns from university–industry exchanges.

One potentially important although often overlooked sourceof difference relates to the diverse models of university–industry

interactions that take place in engineering versus hard scienceresearch. This paper contributes to this issue, by offering aquantitative analysis of the postpatenting effect on productiv-ity and character of research. The hypothesis is that there is nounique impact linking inventive activity with postpatenting per-formance, but rather that this is at least partially field-dependent.Academic activity is characterized both quantitatively and qual-itatively, i.e., in terms of the amount of scientific contribu-tions, the potential impact on further research, and the level ofbasicness.

Results of the work indicate that the importance of patentingacademic research varies across different fields. In particular,materials engineers were gaining the highest returns from work-ing at applied ends, while materials chemists seemed not to beaffected, or even decreased their basic output. Interpretations ofthese findings and implications for academic research policy arediscussed.

The paper is organized as follows. Section II describes theterms of the debate on the effect of academic patenting on sci-entific publication, presents the state of the art of empiricalinvestigations and states the research questions and hypothe-ses. In Section III, we describe the dataset, in Section IV,we present the indicators and the models, along with results.Section V summarizes the findings and draws conclusion andpolicy implications.

II. STATE-OF-THE-ART EMPIRICAL EVIDENCE AND THE

RESEARCH HYPOTHESES

A. Empirical Evidence: The State of the Art

Several empirical investigations were recently presented toenlighten the impact of the inventive activities over the scien-tific performance of academic scientists, on both cross sectionsand longitudinal data, based on comparisons of either entireinstitutions or individual scientists.

At the heart of those investigations is the aim of testing thehypotheses that scientific research and development of industrialapplications are rival versus complementary activities.

On the one hand, reasons to claim that such activities stand ina rival relationship refer to the well-known issue that the marketlacks adequate incentives for investing in long-term, nonfinal-ized research, because such research is subject to strong uncer-tainty, or ambiguity [49]. This is the reason why science—it isadvocated—should be free to assess its results and assign creditand resources according to its own criteria, which substantiallyrely on recognition from peers along strict scientific merits [17].If these mechanisms were not in place or were malfunctioning,nothing would ensure that scientists would produce researchthat would yield its fruits in the longer run. For instance, if themarket became a systematic source of rewards (through patentlicenses, commissioned research, consulting activities), a scien-tist might be tempted to follow only those research tracks thatwould obtain credit and visibility from industry, thus departingfrom the objectives and rules of science. The fact that scientistsare profit-motivated in selling or licensing their technologiesis well documented in managerial literature [33]. This wouldbe especially true where monitoring of activities is low [38],

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

18 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

control over academic performances is poor, and in those sys-tems where wages are fixed, as for instance in most Europeancontinental countries. On the contrary, in countries such as theUnited States where scientific merit is highly rewarded, thepressures of the market may be somehow counterbalanced [18].

A second reason that is commonly suggested to support thehypothesis of rivalry relates to the regime of secrecy that isadvisable for a patent. Although it is true that patents are of-ficially published in patent databases, and this then constitutesa form of public disclosure, often the inventor of a patent hasincentives to keep communications to the minimum, so as notto give undue advantage to potential competitors. This puts thescientist in the position of deciding about two sorts of issues.First, whether or not it is convenient for him or her to voicepatented content at scientific conferences, or with colleagues,and to publish it in scientific journals. In principle, because ca-reers in academia depend on publications and scientific—ratherthan industrial—credit, scientific publications should always besought, at least by those who are waiting for promotions andtenures. Once again, this would probably be less of the case insystems with little control on performance, or simply for thosepeople who are not pressured by career issues.

Second, if they do disclose to the scientific community, theystill have to decide when is the best timing to do so. In the UnitedStates, inventors can benefit from a law that allows publicationeven before filing a patent, under the provision (called the “gen-eral grace period”) that an author’s previous disclosures cannotbe opposed as prior art, as long as they were not made morethan 12 months prior to filing. This is not the case in Europe,where every disclosure to the public made before the date offiling bars patenting for everybody, including the scientist her-self/himself. Thus, in Europe, wishing to file a patent impliesholding up on publications, at least for the time needed to gothrough the process of submission to the TTO, the drafting ofthe patent by attorneys, and application to the office. However,because U.S. patents obtained under the special provision wouldnot receive protection outside the national borders, the issue ofdelays in publication applies to American scientists too, at leastfor all those patents for which an international market is desired[22].

On the other hand, the reasons for claiming that scientificmerits and market exploitation are not hampering each other,and can even be pursued with some degree of complementarity,can be explained in several points.

A first line of argument stresses the richness of the feedbackthat a scientist gets from engaging in applied work. The ideahere is that participation in development activities and coop-eration in firm’s projects can create so many ideas as to openup new potential areas for scientific exploration. A popular sur-vey of university and firm collaboration conducted years ago byMansfield [39] gave credit to this belief. In fact, several surveyrespondents reported that they had started conducting academicresearch on problems and ideas that they had became awareof while doing industrial consulting. Scientists reported that thecontribution of firms and research users was variable, from beingvery marginal to being fundamental in indicating the problemsand directions of research.

If we look at the scholarly contributions on the topic, we findseveral mechanisms that can serve to explain such evidence. Thefirst explanation one can think of is “serendipity,” i.e., the ideathat, in science, sudden insights and leaps in understanding canarise randomly by chance or luck, with no apparent contiguitybetween what was sought and what was found (see for instance[6] and [9]). The explanation is probably oversimplistic but it hasreceived extensive credit from the literature. Here the argumentgoes that, in a world of unpredictable payoffs, we can expect aflurry of creativity and new ideas simply by exposing scientiststo a new set of problems and topics, such as those derived fromdoing technology development or solving industrial problems[49].

A second argument to help explain the evidence is offeredby Rosemberg [54], who draws on the observation that in manycases solutions to practical problems, in the form of workingmachines, came before theoretical understanding of the lawsand principles governing their functioning. In those cases, heargued, technology provides a valuable mass of raw empiricaldata ready for scientific examination.

A third proposed explanation relates to resources. Contractresearch, industrial consulting, patent licenses, and so on con-tribute to increasing the attraction of resources on which a scien-tist and her/his team rely in order to carry out their research [11].It is quite normal for scientists to hide topics they want to ex-plore inside a “Trojan horse” of something more easily financedby a firm or industrial partner. Hence, industry money can bepartially reused to serve the purposes of science, with mutuallybeneficial effects.

A slightly similar argument is one that highlights the impor-tance of the “social capital” (instead of the monetary capital) towhich a scientist gains access when working with a firm [24]and the effect of the social dimension on knowledge creationand creativity [2], [16], [23]. Relations bring information (for in-stance, on the foreseeable evolution of technology and science,on the needs of firms and end-users markets, on the perceivedimportance of themes and topics) and competencies. There isquite a lot of descriptive literature that has portrayed academicinventors as highly connected individuals who are gatekeepersof information from and across different pieces of fragmentedcommunities [4], [20], [45]. As stressed by scholars assessingthe impact of collaboration and networks, the centrality of indi-viduals and their ability to connect otherwise distant communi-ties can boost creativity, as opposed to densely knit networks,whose excess cohesion may in fact limit the openness of thefield to new ideas [40], [46]. In contrast, loose and distant tieshave a stronger potential to bring in sets of resources otherwiseunavailable to the single member; hence, low-density networksmight be conveyers of higher achievements [26]. In principle,the larger and more diverse a person’s network is, the larger thepool of ideas and open possibilities that she/he could fish in is.

A number of investigations have recently been presented thatinquire into the hypothesis of the rival versus complementaryrelationship, both at the institutional level as well as that of sin-gle individuals. With regard to institutional analyses, we haveseveral pieces of evidence indicating that many top-rated uni-versities for both research and education were among the best

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 19

performers regarding the number of patents issued [30], [44],and for license income [41], and that departments in charge ofboth basic and applied research were not producing less pub-lished research than those in charge of basic research only [58].

So far, three main findings have emerged quite consistentlywith regard to individual-level analysis. When looking crosssectionally at the group of scientists who have patented versusthose who have not, all available studies show that academicinventors, despite representing a small proportion of the pop-ulation (10–15%), are disproportionately concentrated amongthose most productive in research. Fabrizio and Di Minin [21]find a positive correlation between actual and lagged numbersof papers and patents in a sample of 150 inventors and 150controls. Breschi et al. [11] show that academic inventors pub-lished on average one paper more than a matched-pair sampleof Italian academic researchers that never patented, and that thisdifference is higher for serial inventors. Stephan et al. [55] con-firm similar findings in a large sample of American doctoraterecipients with a methodology that is suitable for treating excesszeros in the dependent variable, and find that patent counts andpublication counts are positively correlated after checking forfield, seniority, and other institutional and job characteristics.Carayol [13] finds similar results for a sample of scientists atLouis Pasteur University.

With regard to longitudinal studies of the prepatent period,Azoulay et al. [7] find that in a large sample of life scientists,patents were proceeded by a flurry of publications. Calderiniet al. [12] found similar results in the materials sciences, al-though they showed a reverted effect for the (small) number ofstar scientists.

Although longitudinal evidence of the postpatent period isstill preliminary and has suffered several problems in the han-dling of data, available studies to date hint that patenting activityis also likely to produce an increase in published research in thefollowing years [7], [11], [21].4 The effect seems nonnegligiblein terms of magnitude (more than one paper increase for eachpatent), and occurs either in the year of the invention, or in thefollowing one or two years, which by and large corroboratesthe idea that patenting and publishing may be complementary,mutually sustaining activities.

With regard to the spectrum of academic fields investigated,the previous-mentioned investigations took into account chem-istry, physics, life sciences, computer sciences, and mechanicaland electronic engineering. However, at present, life scienceshappens to be the most widely analyzed field, which is causefor some caution in generalized attempts to attribute the resultsto other disciplines.

B. Research Hypotheses

The empirical evidence discussed so far has done a veryimpressive job in putting forth new issues and discarding un-supported preconceptions. At the same time, several issues havebeen left open and deserve further investigation.

4An exception is the study by Agrawal and Henderson [1] that finds nostatistically significant effect in a sample of MIT scholars.

Our idea is that two areas of improvement demand specificattention (and will be challenged in this paper): first, very fewanalyses encompass assessments of the character of the knowl-edge disclosed in the postpatenting period, and this mirrorsa fundamental inability of both metrics and theoretical con-cepts to characterize research beyond sheer productivity. In thefollowing, we make use of three indicators: productivity, im-pact, and basicness. To the best of the authors’ knowledge,only Azoulay et al. [7] and Breschi et al. [11] use metrics tospecifically aim at measuring qualitative features of research.Agrawal and Henderson [1] and Fabrizio and Di Minin [21]make use of citations counts, which, however, depend on thearticle age, as well as on the patterns of citations that in turnare journal-specific. As such, their use as an indicator of qual-ity is debated. Second, little consideration has been devoted tohow differences in the nature and scope of the various fieldsof science and engineering to which a researcher contributesmight affect the relation between her/his scientific and inven-tive work. In the empirical analyses mentioned before, therewas no specific attempt to separate the effect according tothe field or subfield of research, in part because the numbersof patents found were too small for further breakdowns to beadvisable.

The issue is not trivial, since so far policies in support oftechnology transfer have been equally applied to all faculties,schools, and departments, irrespective of the specificities of thedisciplines involved and of the interacting industries.

Drawing from the issues raised in the previous section, wenow make a number of explicit hypotheses to test with empiricaldata and explain each of them in detail.

Hypothesis 1: Academic patenting activities are more likely to besynergic and complementary with engineering research than withhard science research; this would result in a stronger effect of patent-ing on scientific productivity, impact, and share of basic research ofengineers, in comparison to hard scientists.

The starting point to build our hypothesis would be to considerthat not all disciplines stand in the same relation to, and gainequal benefits from serving practical ends. In fact scientific fieldsdiffer in a number of issues, which make technology transfermore or less likely to happen, more or less fruitful or costlyon either sides, and affect the returns from university–industryexchanges overall. There are at least two reasons to support thisexpectation.

1) Epistemology of Science Versus Engineering: Althoughepistemology of engineering is still regrettably quite undevel-oped, a key difference between doing research in engineering asopposed to hard science is that, whereas science is aimed at theunderstanding of phenomena, and somehow sees technologyas instrumental to that end, engineering is in its fundamentaland epistemological essence a science applied in scope, i.e.,a discipline that addresses and aims to solve problems of in-dustrial (practical) relevance, by means of a rigorous scientificmethod [60]. Saying that Engineering is “applied in scope” is ab-solutely not the same as saying that it is an “applied science,” inthe sense of being deductive, i.e., a discipline that puts findings

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

20 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

of a hierarchically-dominant scientific domain into practice.5

Rather, it is like saying that serving a practical end providesa sufficient justification to undertaking a scientific investiga-tion.6 This is not necessarily the case of (hard) science, whoseepistemology is more rigidly constructed within the borders ofthe disciplinary field, discussed in turn and agreed upon bythe scientific community of reference, which is called to definegoals, priorities, and acceptable methodologies. The problemwith acceptance encountered by “deviant” or “heterodox” spe-cialties, or simply by multidisciplinary sciences, is well docu-mented within the sociology of science as being one of goaldeviance [27]. In extreme cases, when the new goals posedby an emerging domain of inquiry are perceived as being in-compatible with the traditional domain of the discipline, the“deviant specialty” will undergo the sanction of being exiledoutside its core domain and, if successful, it will eventually be-come a new disciplinary differentiation. This mechanism, forinstance, has been described to characterize the birth of chem-ical engineering from chemistry [54], or biotechnology frombiology.

The argument we try to offer in this paper is that work-ing on industrial problems may be very beneficial in termsof new ideas to develop, and that the epistemological open-ness of engineering to considerations of scope makes it justmore likely that a scientist would be able to exploit suchopportunities.

2) Technological Regimes (Regime of Secrecy and Impor-tance of IPRs): Based on a large survey of the relationshipsbetween firms and universities, Cohen et al. [15] have reportedthat academic research is not perceived as equally important forfirms in every discipline. Different perceptions of potential im-portance relate to a number of factors, including the degree ofcodificability of scientific knowledge, the absorptive capacity ofthe related industry, the value of patents to protect innovation,and the importance of industrial secrecy.

Regarding the fields inquired in the present work, Cohenet al. [14], report that the protection offered by both secrecy andpatents is higher for all chemical products than all other non-chemical materials (with the exception of metals). If this is thecase, we once again expect that chemists (who patent chemicalproducts) might be more prone to overlook or delay disclosuresto open science journals when compared to their colleaguesworking in engineering of materials. Another strategy that isbeing reported relates to the practice of sending articles relatedto patented materials to a low-ranking journal, in order to estab-lish priority over a discovery, while at the same time avoidingmassive diffusion until the patent is secured. Such practice maycause a temporary decline in the observed impact of articles,

5An epistemological discussion of the argument would exceed the purposesof the present paper. See Walter G. Vincenti [60] and Edwin T. Layton [34] fora more comprehensive discussion.

6Quoting [60]: “I have never attempted to design an airplane in my entirecareer as a research engineer (although I participated in planning and designinglarge aeronautical research facilities). The atmosphere in which I worked, how-ever, and the knowledge I helped produce, were conditioned by the needs ofairplane designers who visited our laboratory. My colleagues and I were keenlyand continuously aware of the practical purposes we served.” [60, p. 7]

which nonetheless does not necessarily correspond to a declinein the scientific quality.

Hypothesis 2: Repeated patenting activities are likely to produceimprovements of scientific performances (quantity and quality ofpublications) at decreasing marginal rate.

This hypothesis draws on the predictions of social capital the-ory mentioned in the previous section. We expect that scientistsbenefit from networking with industrial partners because theyreceive new ideas, enlarge the spectrum of observations, andgain access to data from a larger pool of sources [2], [16], [23].This is likely to happen when a scientist starts to patent, eitherbecause she/he does so in connection to a firm with which she/hehas started to cooperate, or because she/he approaches the busi-ness sector to sell her/his own patent. This opening toward thebusiness sector, once established, will eventually produce itsbenefits, although at a decreasing rate. Networking is likely toproduce its highest beneficial outcomes when a connection isestablished across communities that would otherwise be dis-tant [26]. Established links carry less information than thosewith comparatively unrelated communities; hence, the inten-sity of interactions that comes with repeated participation innetworking is less fruitful than the initial hookup [40], [46].

In addressing Hypothesis 1, our research shares some groundswith the works by Azoulay et al. [7] and Breschi et al. [11],which also aim to test the effect of patenting over productivity.It does so by means of an entirely different set of data, fordifferent metrics of quality, in the context of materials science,and by looking at subfield-specific effects.

III. SAMPLE AND DATA

The database used for this study was based on a list of scien-tist members of an Italian association for research in materialsscience, called Consortium of Italian Universities for Scienceand Technology of Materials (INSTM). At the end of 2003,the association gathered over 1660 researchers, belonging to 42Italian universities and public research centers, who virtuallyrepresent all universities and public research units working inthe field of science of materials throughout Italy.

According to the Carnegie Mellon Survey, academic researchin materials science is perceived by firms as something thatcontributes most substantially to industrial research and devel-opment [15].

Given that admittance of researchers to the INSTM associa-tion is individual and voluntary and requires paying an annualmembership fee, scientists are self-selected as those workingin the area of materials science. We took all members at theend of 2003 who were born in 1954 or later, which resulted ina final list of 1323 names, and eliminated the laboratory engi-neers and technicians, which left us with a list of 1276 names.Materials science is a considerably homogeneous field, and itsscientific community gathers contributions from several motherdisciplines: mainly, chemistry, engineering, physics, and morerarely, mineralogy and geology. Our sample of scientists mir-rors this organization: the scientists who were observed weredistributed in the following proportions: 919 materials chemists

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 21

(72%), 309 materials engineers (34%), 35 materials physics(3%), plus 12 scientists (1%) from several other subfields.7

Our sampled scientists were tenured professors in 2003, aswell as untenured researchers, Ph.D. students, and research as-sistants, thus providing a good representation of the variety ofroles and types of professionals working for the Italian publicresearch system. To the best of our knowledge, we are not awareof any selection bias affecting stratification of our sample.

For each of the 1247 names, we collected all papers publishedin open science journals (as listed by ISI Science Citation Index)and all United States Patent and Trademark Office (USPTO) orEuropean Patent Office (EPO) patents issued (from DelphionThomson), finding 127 “academic inventors” of at least onepatent issued during the 1977–2001 period.

We took the year in which the scientist was 238 as a con-ventional starting observation time (ti0) and collected all infor-mation from that year to the end of 2001. Publication lags inmaterials science range from four weeks to six months; there-fore, we can take the publication year as a proxy of the discoverydate. Similarly, we took the patent priority date as the proxy ofthe invention date. Given that the ISI database only allows oneto query for the author’s full surname and name initials, thecase of including homonyms is highly frequent. To cope withthis problem, we filtered away those papers that were madeon clearly unrelated fields to materials science, based on theJournal of Citation Report (JCR) taxonomy of journal fields(multidisciplinary fields included).

We appraised basic/fundamental versus applied orientationof research by means of the IpIQ ranking of journal level, whichis an indicator expressed in a 1–4 ranking, where “very basic,untargeted research” is equal to level 4 [48].

In order to appraise the quality of the scientific papers, weused the impact factor (IF) of the scientific journals where thearticles were published.9,10 Usage of the journal’s IF as a proxyof quality of the published article is the same as making theassumption that good journals only publish good papers andvice versa.

Table I provides a complete explanation and summary statis-tics of the dataset variables involved in the analysis described inthe following section.

IV. METHODS AND RESULTS

In this section, we study the effect of patenting on postscien-tific performance by means of different econometric techniquesin order to account for the multiple problems of this type of anal-ysis that usually affect the characteristics of the experimentaldesign and the nature of the dependent variable under study.

In general, the estimation of the causal effect of a treatment(patenting in this case) on a variable of outcome (quantity, ba-sicness, and quality of scientific production) can be difficult in

7Based on classifications of the Italian Ministry of Research (http://sito.cineca.it/murst-daus/docenti/docenti.shtml).

8The minimum age at which one can gain an M.S. degree is 23 in the Italianeducation system.

9IF figures were taken from the 2002 edition of JCR.10For general information on the index and on citation-based indicators,

see [19] and [47].

noncontrolled studies due to the presence of confounding vari-ables (or confounders) that affect both the outcome of interestand the probability of being treated.

We then followed the inverse probability of treatment weights(henceforth IPTW) approach [8], [11], a method that is widelyaccepted in biostatistics for estimating average treatment effectsin observational studies [31], [52], which tackles the problem ofendogenous selection into treatment in a way which is similarto other propensity-scores matching techniques.11

This method relies on the crucial assumptions that the se-lection into treatment is based on observables variables andthat the modeling structure of the selection process is correctlyspecified.12 It nonetheless has the considerable advantage of notrequiring exclusion restrictions for identification, unlike in theinstrumental variable approach, so that there is no need for spe-cific instruments that are not easy to find in this context. Withthe IPTW methodology, the role of confounders is neutralizedby weighting each observation with its inverse probability oftreatment and it can be interpreted as the inverse of a subject’sconditional probability of receiving her/his treatment history upto time t, given past treatment history and other “prognostic”factors.

We implemented this procedure by estimating a logit modelon the probability of applying for a patent for the first time. Asin Azoulay et al. [8], we adopted the “stabilized” version of theIPTW methodology; hence, we estimated separate regressionswith different specifications for the numerator and the denom-inator of the weights. The predictions obtained from the logitestimates are then used to weigh each observation when regress-ing the outcome variable of interest Y on the set of covariates Xand on the treatment variables Z.

The set of covariates X include seniorityit , which measuresthe number of years a scientist has spent in academia up to yeart, expttomait , which proxies the experience of the institutionin patenting, and hence captures environmental effects (mea-sured as the total number of patents granted to the institutionin the previous five years), and a dummy variable for gender(genderit). The set of treatment variables Z includes postpatit(regime treatment indicator), codified as a dummy variable thatequals 1 if scientist i has at least one patent up to year t (where tis the year of priority of the first application for patents that hasscientist i among the inventors) and cumpatit (cumulative treat-ment indicator) codified as the total number of patents grantedto scientist i up to year t.

In addition to the aforementioned variables, a set of calendartime dummy variables has been included in all the regressions, toaccount for time trends on each dependent variable considered.These dummy variables have been codified into 5-year timeintervals from 1975 to 1994, with the most recent period (1995–2001) set as the baseline. Results show an upward time trendfor all the scientific productivity indicators considered in everyregression, and are not reported in the tables for the sake ofbrevity. See Table II for the main descriptive statistics of all thevariables involved in the analysis.

11[53].12see [8] for details.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

22 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

TABLE IDESCRIPTION OF THE VARIABLES INVOLVED IN THE ANALYSIS

The logit model formulation and parameter estimates are re-ported in Table III. Columns 1–3 report the estimates of theIPTW denominator and columns 4–6 of the numerator. SpecificIPTW have been estimated for the subsets of chemists and en-gineers, and the reason for this task will be clear to the reader inthe next section. We found a positive (although nonlinearly de-creasing) influence of the years of experience on the probability

of issuing a patent for the first time. Male scientists tend to patentmore than female ones, especially in chemistry. Moreover, theaverage number of scientific publications significantly increasesthe probability to patent in the next year only in engineering(column 2); this finding is consistent with our Hp. 1 which im-plies that “selling” the academic scientific findings via patentingactivity is easier and less time delayed than in hard sciences.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 23

TABLE IIDESCRIPTIVE STATISTICS OF THE MAIN VARIABLES INVOLVED IN THE

ANALYSIS (SEPARATED FOR ENGINEERS AND CHEMISTS)

A. Scientific Productivity: Quantity of Scientific Production

The first relation we wanted to investigate is the effect ofpatenting on the quantity of scientific production measured bythe number of articles published in a year.

We first considered the raw number of (authored and coau-thored) scientific papers published by scientist i in year t(publ mit) as a dependent variable. Since this is a count variableshowing a disproportional amount of zeroes (more than 40%),the natural choice for modeling it is a zero-inflated negativebinomial (ZINB) model. This model entails two different re-gressions because it assumes two different processes governingthe dependent variable: one for the inflation part (zero outcome)and the other for the count outcome (without extra zeroes).Moreover, it allows for unobserved heterogeneity among sub-jects by assuming individual gamma-distributed random effects.This model is estimated via iterative maximum likelihood (ML)techniques [61] with robust standard errors clustered across sub-jects. Table IV shows the resulting estimates.13 Looking at thewhole sample (column 1), no statistically significant effect isfound by either the lagged patent regime variable (postpatit-1)nor by the cumulative number of past patents (cumpatit-1).

However, since the average number of annual scientific pub-lications depends largely on the researcher’s scientific field,14

we ran separated regressions for the two subsamples (engineersand chemists). The estimated coefficients and marginal effects(Table IV, columns 2 and 3) show that after the first patent, engi-neers tend to have a greater yearly number of publications thannonpatenters (although the positive coefficient of postpatit-1 issignificant only at 10% in the outcome equation). Conversely,for chemists, we find a negative impact (significant at 5% level inthe outcome equation). Besides, we also find a positive effect ofthe number of past cumulated patents (cumpatit) on article pro-ductivity, which tends to overwhelm the former negative effectafter the third to fourth patent is granted, although this counter-effect of cumpatit-1 is relevant only for a small proportion ofthe observations.15

The estimated coefficients of the controls are quite intuitive:the number of published articles first increases with seniority(seniorityit) and eventually declines at a later stage of career(senioritysqit). Men tend to have higher productivity than theirfemale colleagues, while the overall number of past patentsowned by the institution of affiliation (expttomait), which cap-tures the institutional/environmental effects (including the avail-ability of financial resources), has a positive impact on a scien-tist’s productivity.

We also considered a different measure of scientific produc-tivity that takes into account coauthorships (i.e., shared articles)and created an alternative weighted indicator (wpubl mit) bydividing the number of yearly publications (publ mit) of scien-tist i in year t by the average number of authors (aut mit). Sincethis new dependent variable is no longer a count, we will relyon ordinary least squares (OLS) methodology for estimatingstandard linear models. We partially recovered its skewedness

13For each of the regressions in columns 1–6, the likelihood ratio test (forα = 0) rejected the zero-inflated Poisson model in favor of the ZINB model,whereas the Vuong test statistic rejected the standard negative binomial modelin favor of the ZINB model.

14In our sample, scientists in chemistry show an average number of yearlyper capita number of publications of about 1.46 (Std. Dev. 2.38) versus 0.77(Std. Dev. 1.76) of engineering and 1.01 (Std. Dev. 1.2) of physics.

15For instance, in 2001, only 2% of chemists had more than three patentsgranted.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

24 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

TABLE IIILOGIT-ML MODELS ESTIMATED USING, FOR EACH SCIENTIST, THE OBSERVATIONS UP TO THE PRIORITY YEAR OF THE FIRST GRANTED PATENT

(also due to the excess of zeros) by means of the followinglog-transformation lwpubl mit = log(wpubl mit + 1). For thesake of comparison, we also reestimated a similar linear modelfor the log-transformation of the original unweighted depen-dent variable lpubl mit = log(publ mit + 1). This “lineariza-tion” of the former count model has the advantage of allowingthe application of a linear fixed-effect (FE) estimation methodthat is more robust to functional form misspecification (althoughless efficient) than ML methods. The results are reported inTable V (columns 1–6): the estimates of the linear model withboth the unweighted and weighted dependent variables basi-cally confirm the previous findings of the ZINB model for theengineers (columns 2 and 5) with a stronger evidence of sta-tistical significance, whereas for the chemists the previouslyestimated U-shaped relation between the number of publica-tions and patents is confirmed only with the weighted dependentvariable (column 6). The patenting–publishing enforcing mech-anism is thus confirmed for the engineers, consistently with ourHypothesis 1, whereas for the chemists no clear path in thepatenting–publishing relationship seems to emerge.

B. Scientific Productivity: Basicness of Scientific Production

The second question we wanted to investigate is whetherpatenting hampers or boosts basic scientific research. We fol-lowed an approach similar to the previous section, but took asnew dependent variables the raw number of scientific papers au-thored (or coauthored) by scientist i in year t (publbas4it) thatresulted in being ranked as “very basic” (level 4) in the IpIQ

classification, its log-linear transformation (lpublbas4it) and itscoauthoring weighted version (lwpublbas4it) following a simi-lar model specification with the previous section. Here too, theanalysis is disentangled between engineers and chemists and themodels are estimated by ZINB-ML (Table IV, columns 4–6) andstandard linear fixed effects techniques (Table V, columns 7–9and 10–12, respectively). Concerning patenting, the only no-table and significant effect relates to the subsample of chemists.While in fact patenting does not seem to have a significant im-pact on the basic scientific output of engineers, for chemistswe find a negative impact of postpatit-1 on the dependent vari-able in all the models, suggesting a diminished productivityin basic research after the first patent for scientists involvedin the field of chemistry. This finding can be explained in thelight of the epistemology differences between hard science andengineering in defining the research scope as stressed in theprevious section. In Hypothesis 1, we predicted a positive effectof patenting on basic research, although stronger in magnitudefor engineers than for chemists. Our results do suggest that theeffect is field-dependent; however, we find no significant effectfor engineering and a negative effect for chemistry, which seemsto indicate that there might have been a crowding-out of basicresearch, associated to patenting in the latter group.

C. Scientific Productivity: Quality of Scientific Production

The final question of our analysis concerns the effect ofpatenting on the quality of a scientist’s research output. To an-swer this question, we first had to overcome the problem of

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 25

TABLE IVZINB ML REGRESSION ESTIMATES WITH INDIVIDUALLY CLUSTERED ROBUST STANDARD ERRORS

finding an appropriate measure of scientific quality. Several ap-proaches have been proposed by scholars: the most common oneis to measure the quality of a researcher’s output by countingthe total number of citations received [1], [10], [21]. However,this method is not immune from drawbacks, since it can be dra-matically affected by the specific characteristics of the scientificfield considered such as different publications rates, differentcross-citing practices, different citation trajectories along time,and so on.

Azoulay et al. [8] tried to overcome such drawbacks by con-structing two alternative metrics: the first is based on the propor-tion of publications in which the researcher appears in the first

and last position of the authors’ list. The second16 is based onthe average journal IF of the articles published in a given year.We followed the latter approach, although in a slightly differentway.

Given the different distribution of the average IF among jour-nals of different scientific fields,17 in addition to running sep-arate regressions according to the researcher’s main scientific

16Also adopted by [12].17The average IF for journals in chemistry (in our sample) is 2.37 (with

average IpIQ index of 3.53) versus 1.17 of engineering (IpIQ: 2.26) and 2.68 ofphysics (IpIQ: 3.64).

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

26 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

TABLE VOLS-FE REGRESSION ESTIMATES

field (as in the previous sections), we also standardized theIF score assigned to each publication of scientist i in year tas follows: stdifac = [(IFit − mean(IF))/Std. Dev.(IF)], wherethe mean and the standard deviation of IF are calculated withrespect to the scientific field of the journal in which the articleappeared. Thus, our dependent variable stdifacit is the averageof the standardized journal impact factor index (stdifac) for thearticles published in year t by scientist i.

This dependent variable is clearly affected by incidental trun-cation, since the IF is only observable when the researcher hasat least one publication in year t (i.e., if the dummy variabledumpublit is equal to 1). We treat truncation by means of aHeckman selection equation [29] (based on dumpublit as de-pendent variable) and a truncated regression (based on stdifacit

as dependent variable) that are estimated simultaneously alongwith the variance of the error component u1 of the outcomeequation σ (the variance of the error component in the selectionu2 is set to 1) and the correlation ρ between u1 and u2 .18 Resultsare reported in Table VI.19

Concerning the estimates on all observations (column 1) wefind that, after the first patent, scientists tend to publish injournals with higher (standardized) IF, as compared to thosewho never patent. When looking at each specific scientific field(columns 2–3), however, this finding only holds true for theengineers who show a positive and significant coefficient as-sociated to postpatit-1 , which is counterbalanced by the neg-ative and significant coefficient on the cumulative number ofgranted patents (cumpatit-1), suggesting that this increase inperformance comes at a decreasing marginal rate, and wouldeventually be neutralized and overwhelmed after about the thirdgranted patent. This evidence is therefore consistent with bothHypothesis 1 of a stronger beneficial effect for engineers and

18See [29] and [5] for details.19We included individual dummy variables (estimates not reported) in the

outcome equation to control for potential sources of unobserved heterogeneity(gender is omitted to avoid multicollinearity, given its nontime varying nature).

TABLE VIHECKMAN—ML REGRESSION ESTIMATES (DUMMY VARIABLE MODEL)

Hypothesis 2 of decreasing marginal returns of repeated patent-ing. For chemists, we found no significant effect of patenting onthe quality of scientific production.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 27

TABLE VIISUMMARY OF POSTPATENT EFFECTS: SIGN AND SIGNIFICANCE

V. COMMENTS ON RESULTS AND CONCLUSION

A summary of the effects estimated through the diverse mod-els is presented in Table VII. Our results suggest that whereaspatenting does not substantially alter a scientist’s publicationtrack on the overall sample, some specific patterns emergewithin different scientific fields.

When we split the sample into subfield groups and run sepa-rate analyses for engineers and chemists, we find that the formercategory experiences an increase in the quantity of publicationsafter the invention. In addition, engineers increase the qualityof their publications after patenting, although at a decreasingmarginal rate, which tells us that this benefit tends to vanish forserial inventors.

On the contrary, the chemists of materials experience a de-cline of performance after patenting especially in terms of theamount of very basic publications, whereas, for what concernsthe quality of their publications, no significant effect seems toemerge.

In terms of magnitude, the marginal effects vary depending onthe estimation strategy adopted (according to the different para-metric model specified or the choice of weighted/unweighteddependent variable), but in most of cases remain far fromnegligible.

The results of our empirical exercise can be partially com-pared to the results of two other investigations, one fromAzoulay et al. [7] and the other from Breschi et al. [10]. Theformer is run on a large sample of American life scientists, thelatter on a sample of Italian tenured scientists in various fields.Our contribution shares with the previous investigations themethod used to account for endogeneity (IPTW - regressions),and adopts similar metrics (level of basicness). However, it addsto the previous investigations because it takes into account adatabase of individuals that include Ph.D.s, research assistants,and postdocs as well as a different scientific field (materialsciences), because it adopts a more robust set of econometrictechniques (excess zeros are accounted for20), a set of differentmetrics for quality, and because it addresses the issue of sub-field specialties both theoretically and methodologically. Boththe previous-mentioned analyses point at a beneficial effect of

20Excess-zeros treatment is an issue in every publication-based analysis,for the well-known skewness of publication distributions, often called “Lotka-Law” [37].

patenting on productivity, with no statistically significant effecton IF [7] and hint at a positive effect exerted on the quantity ofpublications [10], consistent with our overall results (althoughwe do not find the same positive effect when considering verybasic publications).

However, when the subfield dimension is included, our re-sults indicate that scientists contributing to the engineering sideof material science research were experiencing improved perfor-mances when working on industrial applications, whereas thiswas not the case for those that were contributing to materialsscience as chemists. We suggest that such dual effects might de-pend on two factors: first, on epistemological issues that provideengineering with more freedom to search for solutions to indus-trial problems and exploit new ideas beyond the rigid domainsof the scientific discipline; second, on the higher effectivenessof both patents and secrecy in protecting inventions of chemi-cal products, as opposed to other nonchemical materials, whichraises the benefits of secrecy and delayed publication. In chem-istry of materials, both productivity and basicness are likely todecrease, either because the feedback from industrial work ispoor or of little interest, or because the scientist wants to pro-tect the value of her patent even at the expense of her scientificrecords.

The latter issue is especially relevant in our case, because thescientists who were sampled work in the Italian academic sys-tem characterized by fixed wages, compensation that is largelyindependent from the effort and time devoted to academic re-search, and little monitoring of scientific merit. In similar sys-tems, which are widespread in most Continental European coun-tries (e.g., France, Germany, and Spain), incentives to disclosurein open science could be comparatively too small if challengedby the wish to attract industry funding or additional sources ofcompensation. Nonetheless, even in systems such as the UnitedStates, opportunistic rent-seeking behavior of scientists has beenlargely reported [38].

An alternative explanation of the findings of decreases in thebasicness experienced by chemists is that the scientist opts forpublishing articles that relate to patented materials to a low-ranking journal, in order to establish priority over a discovery,while at the same time avoiding massive diffusion until thepatent is secured. Such a practice may cause decline in theobserved journal level of the articles, which nonetheless doesnot correspond to a decline in the scientific impact.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

28 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

TABLE VIIISUMMARY OF POST-PATENT EFFECTS: SIGN AND SIGNIFICANCE (WITH ROBUST IPTW ESTIMATION)

In principle, it is plausible that similar results of different re-turns from industrial inspiration to different subfield disciplineswould be found in other subject domains, such as mathematicsand computing, or biology and biotechnology. Although our re-sults are hardly generalizable, if confirmed by further investiga-tions, they should stimulate discussion on a number of relevantissues.

First, if the impact of academic patenting is strongly field andsubfield dependent, every unique and generalized policy of tech-nology transfer and industrial liaison of academic institutions isinappropriate. Because patenting may provoke opposite effectsdepending on the subfield of research, more appropriate policiesshould probably be discussed within departments or schools.

Second, where industry funding is potentially rich and de-mands secrecy of results or a strong level of protection that con-flicts with open and widespread circulation of results, incentivesto publish in open science need to be empowered accordinglywithin academic institutions, for instance, by enhancing the re-wards attributed to top-level publications. Claims for a moreflexible and more context-adapted approach is increasingly ad-vocated by both scholars and practitioners [32]. This should bethe case particularly for those systems where wages are largelyfixed and monitoring of scientific excellence is poor. In suchsystems, incentives to earn additional sources of compensationand/or laboratory funding from industry need to be adequatelycounterbalanced by career and funding mechanisms based onscientific excellence.

APPENDIX

IPTW COMPUTATION AND “ROBUSTNESS” CHECKS

Given the logit estimates of Table III, define pit to be thepredicted probability for subject i of being treated for each yeart. Our regime of treating specification assumes that the status of“being an academic patenter” for each researcher lasts until theend of the “follow-up,” then pit equals one for each year afterthe first patent. The denominator21 of the weights are computedas follows.

1) Calculate the probability of each subject i to receive theobserved treatment at time t

IPTW∗it = p∗itPOSTPATit + (1− pit)∗(1−POSTPATit).

21The numerator is computed similarly using the predicted probabilities ac-cording to the model specifications in Table III, columns 4–6.

2) Estimate each subject’s probability of complete treatmenthistory up to each year t

IPTWit =t−1∏k=0

IPTW∗it−k .

We also reestimated the predicted probabilities pit by meansof a nonparametric approach that does not require any specifica-tion of the density structure of the model (e.g., logit distribution)nor any specific link function for connecting the regressors withthe conditional expectation of the dependent variable (e.g., addi-tive linear, with quadratics, with interactions, and so on). Severalkernel estimators for categorical data have been proposed in lit-erature [3]; in this paper, we follow the method proposed byLi and Racine [36], which is designed for estimating a densityfunction defined over both discrete (xd) and continuous (xc)variables using the following joint kernel density estimator:

f(xd, xc) =1

nhx

n∑i=1

L(Xdi = xd)W

(XC

i − xC

hx

)

where L(Xdi = xd) is a categorical data kernel function,

W [(XCi − xC )/hxC ] is a continuous data kernel function, and

hx is the bandwidth for the continuous variable chosen viacross-validation methods (see [36] for further details).

Given the computationally intensive nature of these proce-dures whose complexity increases exponentially with the sam-ple size, we estimated the nonparametric version of the IPTWonly for the subset of the engineers and the chemists and not forthe whole sample.

We then reran all the ZINB-ML, OLS-FE, and Heckman-FE models with robust IPTW estimated nonparametrically. Thefindings are similar to the previous ones22 and are summarizedin Table VIII.

REFERENCES

[1] A. Agrawal and R. Henderson, “Putting patents in context: Exploringknowledge transfer from MIT,” Manage. Sci., vol. 48, no. 1, pp. 44–60,2002.

[2] G. Ahuja, “Collaboration networks, structural holes, and innovation: Alongitudinal study,” Adm. Sci. Q., vol. 45, no. 3, pp. 425–457, 2000.

[3] J. Aitchison and C. G. G. Aitken, “Multivariate binary discrimination bykernel method,” Biometrika, vol. 63, no. 3, pp. 413–420, 1976.

22Detailed results and estimation routines are available upon request.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

CALDERINI et al.: UNEQUAL BENEFITS OF ACADEMIC PATENTING FOR SCIENCE AND ENGINEERING RESEARCH 29

[4] T. J. Allen, Managing the Flow of Technology: Technology Transfer andthe Dissemination of Technological Information Within the R&D Organi-zation. Cambridge, MA: MIT Press, 1977.

[5] T. Amemiya, Advanced Econometrics. Cambridge, MA: Harvard Univ.Press, 1985.

[6] J. H. Austin, Chase Chance and Creativity. The Lucky Art of Novelty.New York: Columbia Univ. Press, 1978.

[7] P. Azoulay, W. Ding, and T. Stuart, “The impact of academic patentingon the rate, quality, and direction of (Public) research output,” NBERWorking Paper 11917, Jan. 2006.

[8] P. Azoulay, W. Ding, and T. Stuart, “The determinants of faculty patentingbehavior: Demographics or opportunities?,” J. Econ. Behav. Org., vol. 63,pp. 599–623, 2007.

[9] B. Barber and R. C. Fox, “The case of the floppy-eared rabbit. An instanceof serendipity gained and serendipity lost,” Am. J. Sociol., vol. 64, no. 2,pp. 128–136, 1958.

[10] S. Breschi, F. Lissoni, and F. Montobbio, “University patenting andscientific productivity. A quantitative study of Italian academic inven-tors,” CESPRI Working Paper 189, Bocconi University, Milan, Italy, Nov.2006.

[11] S. Breschi, F. Lissoni, and F. Montobbio, “The scientific productivity ofacademic inventors: New evidence from Italian data,” Econ. InnovationNew Tech., vol. 16, no. 2, pp. 101–118, 2007.

[12] M. Calderini, C. Franzoni, and A. Vezzulli, “If star scientists do notpatent. The effect of productivity, basicness and impact on the decision topatent in the academic world,” Res. Policy, vol. 36, no. 3, pp. 303–319,2007.

[13] N. Carayol, “Academic incentives, research organization and patenting ata large French university,” Econ. Innovation New Tech., vol. 16, no. 2,pp. 119–138, 2007.

[14] W. M. Cohen, R. R. Nelson, and J. P. Walsh, “Protecting their intellec-tual assets: Appropriability conditions and why U.S. manufacturing firmspatent (or not),” NBER Working Paper 7552, 2000.

[15] W. M. Cohen, R. R. Nelson, and J. P. Walsh, “Links and impacts: Theinfluence of public research on industrial R&D,” Manage. Sci., vol. 48,no. 1, pp. 1–23, 2002.

[16] J. S. Coleman, “Social capital in the creation of human capital,” Am. J.Sociol., vol. 94, pp. 95–120, 1988.

[17] P. Dasgupta and P. A. David, “Towards a new economy of science,” Res.Policy, vol. 23, pp. 487–521, 1994.

[18] A. A. del Campo, A. Sparks, R. C. Hill, and R. T. Keller, “The transfer andcommercialization of university-developed medical imaging technology:Opportunities and problems,” IEEE Trans. Eng. Manag., vol. 46, no. 3,pp. 289–298, Aug. 1999.

[19] A. M. Diamond, “What is a citation worth?,” J. Human Res., vol. 21,no. 2, pp. 200–215, 1986.

[20] H. Etzkowitz, “Entrepreneurial scientists and entrepreneurial universitiesin American academic science,” Minerva, vol. 21, pp. 198–233, 1983.

[21] K. R. Fabrizio and A. D. Minin, “Commercializing the laboratory: Facultypatenting and the open science environment,” Res. Policy, vol. 37, no. 5,pp. 914–931, 2008.

[22] C. Franzoni and G. Scellato, “Paper in the drawer. Estimating the deter-minants of the patent-publication lags in Europe and the USA,” WorkingPaper SSRN 1 084 062, 2007.

[23] C. Freeman, “Networks of innovators: A synthesis of research issues,”Res. Policy, vol. 20, no. 5, pp. 499–514, 1991.

[24] E. Forti, C. Franzoni, and M. Sobrero, “The effect of patenting on thenetworks and connections of academic scientists,” presented at the Acad.Manage. Conf., Philadelphia, PA, Aug. 2–4, 2007.

[25] A. Geuna and L. Nesta, “University patenting and its effects on academicresearch,” SEWPS Working Paper 99, 2003.

[26] M. Granovetter, “The strenght of weak ties: A network theory revised,”Sociol. Theory, vol. 1, pp. 201–233, 1983.

[27] W. O. Hagstrom, The Scientific Community. New York/London: BasicBooks, 1965.

[28] B. H. Hall, “Exploring the patent explosion,” J. Tech. Transfer, vol. 30,no. 2_2, pp. 35–48, 2005.

[29] J. J. Heckman, “Sample selection bias as a specification error,” Econo-metrica, vol. 47, pp. 153–161, 1979.

[30] R. Henderson, A. B. Jaffe, and M. Trajtenberg, “Universities as a sourceof commercial technology: A detailed analysis of university patenting,1965–1988,” Rev. Econ. Statist., vol. 80, no. 1, pp. 119–127, 1998.

[31] M. A. Hernan, B. Brumback, and J. M. Robins, “Marginal structuralmodels to estimate the joint causal effect of nonrandomized treatments,”J. Am. Statistical Assoc., vol. 96, pp. 454, 440–448, 2001.

[32] M. Jelinek and S. K. Markham, “Industry-university IP relations: Integrat-ing perspectives and policy solutions,” IEEE Trans. Eng. Manag., vol. 54,no. 2, pp. 257–267, May 2007.

[33] S. Lach and M. Schankerman, “Royalty sharing and technology licens-ing in universities,” J. Eur. Econ. Assoc., vol. 2, no. 2/3, pp. 252–264,2004.

[34] E. T. J. Layton, “Technology as knowledge,” Tech. Cult., vol. 15, no. 1,pp. 31–41, 1974.

[35] Y. S. Lee, “‘Technology transfer’ and the research university: A searchfor the boundaries of university—Industry collaboration,” Res. Policy,vol. 25, no. 6, pp. 843–863, 1996.

[36] Q. Li and J. S. Racine, “Nonparametric estimation of regression functionswith both categorical and continuous data,” J. Econometrics, vol. 119,no. 1, pp. 99–130, 2004.

[37] A. J. Lotka, “The frequency distribution of scientific productivity,” J.Wash. Acad. Sci., vol. 16, no. 12, pp. 317–324, 1926.

[38] G. Markman, P. Gianiodis, and P. Phan, “Full-time faculty or part-timeentrepreneurs,” IEEE Trans. Eng. Manag., vol. 55, no. 1, pp. 29–36, 2008.

[39] E. Mansfield, “Academic research underlying industrial innovations:Sources, characteristics, and financing,” Rev. Econ. Statist., vol. 77, no. 1,pp. 55–65, 1995.

[40] M. A. McFayden and A. A. Cannella, “Social capital and knowledgecreation: Diminishing returns of the number and strength of exchangerelationships,” Acad. Manage. J., vol. 47, no. 5, pp. 735–746, 2004.

[41] Milken Institute, “Mind to market: A global analysis of university biotech-nology transfer and commercialization,” Milken Inst., Res. Rep., Sep.2006.

[42] D. C. Mowery, R. R Nelson, B. N. Sampat, and A. A. Ziedonis, “Thegrowth of patenting and licensing by US universities: An assessment ofthe effects of the Bayh–Dole act of 1980,” Res. Policy, vol. 30, no. 2,pp. 99–119, 2001.

[43] D. C. Mowery and B. N. Sampat, “Patenting and licensing universityinventions: Lessons from the history of research corporation,” Ind. Corp.Change, vol. 10, pp. 317–355, 2001.

[44] D. C. Mowery and A. A. Ziedonis, “Academic patent quality and quantitybefore and after the Bayh–Dole act in the United States,” Res. Policy,vol. 31, pp. 399–418, 2002.

[45] F. Murray, “Innovation as co-evolution of scientific and technologicalnetworks: Exploring tissue engineering,” Res. Policy, vol. 31, no. 8/9,pp. 1389–1403, 2002.

[46] J. Nahapiet and S. Ghoshal, “Social capital, intellectual capital and theorganizational advantage,” Acad. Manage. Rev., vol. 23, no. 2, pp. 242–266, 1998.

[47] F. Narin and K. S. Hamilton, “Bibliometric performance measures,” Sci-entometrics, vol. 36, no. 3, pp. 293–310, 1996.

[48] F. Narin, G. Pinski, and H. H. Gee, “Structure of the biomedical literature,”J. Am. Soc. Inf. Sci., vol. 25, pp. 25–45, Jan./Feb. 1976.

[49] R. R. Nelson, “The market economy, and the scientific commons,” Res.Policy, vol. 33, pp. 455–471, 2004.

[50] OECD, “Turning science into business. Patenting and licensing at publicresearch organizations,” presented at the Org. Econ. Co-operation De-velop., Paris, France, 2003.

[51] ProTon, “The ProTon Europe 2005 annual survey report,” ProTon, Europe,Rep. Draft, Apr. 2007.

[52] J. M. Robins, M. A. Hernan, and B. Brumback, “Marginal structuralmodels and causal inference in epidemiology,” Epidemiology, vol. 11,pp. 550–560, 2000.

[53] P. R. Rosembaum and D. B. Rubin, “The central role of the propensityscores in observational studies for causal effects,” Biometrika, vol. 70,pp. 41–55, 1983.

[54] N. Rosenberg, Inside the Black Box. Technology and Economics. NewYork: Cambridge Univ. Press, 1982.

[55] P. E. Stephan, S. Gurmu, A. J. Sumell, and G. Black, “Who’s patentingin the university? Evidence from a survey of doctorate recipients,” Econ.Innovation New Tech., vol. 16, no. 2, pp. 71–99, 2007.

[56] J. G. Thursby and S. Kemp, “Growth and productive efficiency of univer-sity intellectual property licensing,” Res. Policy, vol. 31, pp. 109–124,2002.

[57] J. G. Thursby, R. Jensen, and M. C. Thursby, “Objectives, characteristicsand outcomes of university licensing: A survey of major U.S. universities,”J. Tech. Transfer., vol. 26, pp. 59–72, 2001.

[58] B. V. Looy, M. Ramga, J. Callaert, K. Debackere, and E. Zimmermann,“Combining entrepreneurial and scientific performance in academia: To-wards a compounded and reciprocal Matthew-effect?,” Res. Policy,vol. 33, no. 3, pp. 425–441, 2004.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.

30 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 56, NO. 1, FEBRUARY 2009

[59] B. Verspagen, “University research, intellectual property rights and Euro-pean innovation systems,” ECIS Working Paper 06.05, 2006.

[60] W. G. Vincenti, What Engineers Know and How They Know it. AnalyticalStudies From Aeronautical History. Baltimore, MD: The John HopkinsUniv. Press, 1990.

[61] J. M. Wooldridge, Econometric Analysis of Cross Section and Panel Data.Cambridge, MA: MIT Press, 2002.

Mario Calderini received the B.Sc. and M.Sc. de-grees in mechanical engineering from the Politec-nico di Torino, Turin, Italy, and the Ph.D. degreein economics from the University of Manchester,Manchester, U.K.

He is currently a Full Professor of strategy and in-novation management in the Department of Produc-tion Systems and Management (DISPEA), Politec-nico di Torino. He was an expert in several workinggroups of the European Commissions and the UnitedNations. His current research interests include eco-

nomics and policy of innovation, regional development, and technology transfer.Prof. Calderini is a research associate of the Bureau of Research on Inno-

vation, Complexity and Knowledge (BRICK), Collegio Carlo Alberto. He iscurrently the President of Finpiemonte, the Piedmont Region Agency for Eco-nomic Development and Financial Holding Company and Delegate for PublicPolicies for of the Fondazione COTEC per l’Innovazione Tecnologica.

Chiara Franzoni received the Ph.D. degree in eco-nomics and management of technology from the Uni-versity of Bergamo, Bergamo, Italy.

She is currently with the Department of Produc-tion Systems and Management (DISPEA), Politec-nico di Torino, Turin, Italy, where she is engaged inresearch and teaching. Her work experience includesthe National Research Council of Italy and AndrewYoung School of Policy Studies (visiting 2005–06).Her current research interests include economics ofintellectual property rights and economics of science

and innovation.Dr. Franzoni is a research associate of the Bureau of Research on Innovation,

Complexity and Knowledge (BRICK), Collegio Carlo Alberto.

Andrea Vezzulli received the M.A. (Laurea) degreein political sciences and the Ph.D. degree in eco-nomics from the Universita degli Studi di Milano,Milan, Italy, in 2002 and 2006, respectively.

He was a Visiting Research Fellow at the Univer-sity of York, York, U.K., and at Fondazione GiovanniAgnelli, Turin, Italy. He is currently an Adjunct Lec-turer in econometrics and applied economics at theUniversity of Milan–Bicocca, Milan, Bocconi Uni-versity, Milan, and the University of Brescia, Brescia,Italy, and a Postdoctoral Fellow with the Knowledge,

Internationalization and Technology Studies (KITES), Bocconi University. Hiscurrent research interests include innovation and technology transfer strategy,applied microeconometrics, Bayesian and nonparametric estimation methods,and models for credit risk assessment.

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on January 22, 2010 at 10:40 from IEEE Xplore. Restrictions apply.