Innovation and Industry Lissoni et al 2

University autonomy, the professor privilege and academic patenting: Italy, 1996-2007 Francesco Lissoni 1,2 , Michele Pezzoni 2,3 , Bianca Potì 4 , Sandra Romagnosi 5 1 GREThA – Université Bordeaux IV - France CRIOS – Università "L. Bocconi", Milano – Italy 2 3 4 5 DEMS, University of Milano-Bicocca CERIS-Consiglio Nazionale delle Ricerche, Rome - Italy National Agency for the Evaluation of Universities and Research Institutes (ANVUR), Rome – Italy Corresponding author: F.Lissoni, GREThA - UMR CNRS 5113, Université Montesquieu - Bordeaux IV, avenue Léon Duguit, 33608 Pessac cedex – FRANCE ( francesco.lissoni@u-bordeaux4.fr; tel. +33 (0)5 56 84 86 04) Abstract Using data on patent applications at European Patent Office, we search for trends in academic patenting in Italy, 1996-2007. During this time, Italian university underwent a radical reform process, which granted them autonomy, and were confronted with a change in IP legislation, which introduced the professor privilege. We find that, although the absolute number of academic patents has increased, (i) their weight on total patenting by domestic inventors has not, while (ii) the share of academic patents owned by universities has more than tripled. By means of a set of probit regressions, we show that the conditional probability to observe an academic patent has declined over time. We also find that the rise of university ownership is explained, significantly albeit not exclusively, by the increased autonomy of Italian universities, which has allowed them to introduce explicit IP regulations concerning their staff's inventions. The latter has effectively neutralised the introduction of the professor privilege. Keywords: academic patenting, university autonomy, professor privilege JEL codes: I23, O31, O34 Acknowledgements: Inventor data come from APE-INV, the project on "Academic Patenting in Europe" sponsored by the European Science Foundation (http://www.academicpatenting.eu). Participants to the "NameGame" APE-INV workshop series have all contributed to establish a robust methodology for name disambiguation, which has been essential for the creation of the database. Several Italian colleagues have provided us with data they collected over the years: Cinzia Daraio and Andrea Bonaccorsi (Aquameth data), Andrea Piccaluga (NetVal data), Rosa Grimaldi, Riccardo Fini, and Maurizio Sobrero (universities' IP regulations). Paolo Colchonero Freri provided valuable research assistance. 1. Introduction After being granted autonomy in 1989, Italian universities have generally moved towards a more active management of their sources of revenues and financial assets (Geuna and Rossi, 2013). Among other things, they have taken an active interest in the commercialization of research results, by introducing intellectual property (IP) regulations aiming at securing control of the relevant patents. These initiative clashed however, in 2001, with the introduction of piece of legislation the p ofesso p i ilege that nominally transferred all IP rights over academic research results from the universities to their faculty (Granieri, 2010; Baldini et al., 2010 and 2012). Italy is therefore an interesting case study, which allows assessing the relative impact on academic patenting of two contrasting forces: (i) a change in the governance and funding system of universities (the autonomy) and (ii) a change in the IP legislation (the professor privilege). As such, it contributes to the e e gi g lite atu e o Eu opea u i e sities’ ha gi g practices concerning IP management, as a result of broader policy changes in higher education and research (Della Malva et al., 2013; Mejer, 2012). It also contributes to the literature that has either reconsidered the professor privilege as an alternative to IP policies prevailing in US universities, or assessed its historical impact in the European countries which first introduced it, and then, in most cases, abolished it (Valentin and Jensen, 2007; Iversen et al., 2007; Lissoni et al., 2009; Greenbaum and Scott, 2010; Kenney and Patton, 2011; Geuna and Rossi, 2011; von Proff et al., 2012; Damsgaard and Thursby, 2013; Schön and Buenstorf, in this special issue). Our contribution is mainly empirical and exploit a new and original database in order to: - Build reliable estimates of academic patenting in Italy throughout the period 1996-2007; - Test for the existence of time trends concerning the weight of academic patenting over total patenting, and its ownership distribution between universities and other subjects; - Estimate the impact of both the autonomy of universities and the introduction of the professor privilege on the observed trends. We find that the weight of academic over total patenting has remained stable overtime, while the share of university ownership has increased significantly. We also find that the adoption of internal IP regulations by universities, an immediate consequence of their newly conquered autonomy, appears to have increased university patent ownership and effectively neutralized the professor privilege. We proceed as follows. In section 2 we put forward a number of propositions concerning the determinants of academic patenting and the distribution of its ownership between universities, their faculty, and industry. In section 3 we provide an historical account of Italian universities’ g adual i ple e tatio of autonomy in the 1990s and the abrupt introduction of the professor privilege. Section 4 describes in extreme synthesis our methodology and data. Section 5 contains our econometric exercise and the discussion of results. Section 6 concludes. 2. University autonomy, IP legislation and the ownership of academic patents We define "academic" any patent signed at least by one academic scientist, while working at his/her university, whether the patent is owned by the university, a public research organization (PRO), the scientist, a business company or any other organization, either exclusively or jointly with other assignees 1 (Lissoni, 2012; Dornbusch et al., 2013). This choice is now common in the literature and it reflects a fundamental theoretical issue, one which concerns: (i) the origin of academic inventions, (ii) the nature of IP legislation, and (iii) the governance model of universities in different countries. With respect to the origin of academic inventions, two broad categories can be put forward (Franzoni and Lissoni, 2009) . One category oi ides ith hat Je se a d Thu s all p oofs a d p otot pes fo sale , a el the inventive results of fundamental research, mostly public-funded, which necessitate further development in order to be commercialized. In the US, these inventions were the target of the Bayh-Dole Act of 1980, which assigned them exclusively to universities, as opposed to the funding agencies or the inventors, with the intent of pushing them to engage actively in their commercialization (Mowery, 2001). The second category consists of inventions arising from collaboration with industry, with three actors involved in negotiations over IP: the academic inventor, his/her business partner, and the university administration. Collaboration models range from consultancy to contract research, and to joint research activities. As for the resulting IP, we expect the inventor and the business partner to have a greater say when it comes to consultancy hi h i p i iple does ot i pi ge upo the u i e sit ’s esou ces), while administrations should have a greater say when moving in the direction of joint research activities (which may involve students, staff, and other resources of the institution). Overall, these inventions may be as important, technologically and commercially, than those in the first category (see Link et al., 2007, and Jensen et al. 2010, on consultancy; and Colyvas et al., 2002, on results of joint research). As for the IP legislation, historically this has intervened in two ways. One concerns the rules attached to public funding, as in the case of the Bayh-Dole Act (Mowery and Sampat, 2005). The other concerns the relationship between the academic inventor and his/her university. While general IP legislation assigns the p ope t of e plo ees’ i e tio s to thei e plo e , se e al Eu opea characterized by a special IP regime for academic inventions, k o ou t ies e e u til e e tl as p ofesso p i ilege (von Proff et al., 2012; Damsgaard and Thursby, 2013). Once common in all German-speaking and Scandinavian countries, and nowadays surviving only in Sweden and Italy, this regime prescribes that the university has o title o e the fa ult ’s i e tio s, unless decided otherwise by the inventor. Historically, this explains the high level of individual ownership of academic patents in Sweden, Denmark, and Germany (Lissoni et. al., 2008 and 2009; von Proff and Buenstorf in this issue). Fi all , the dist i utio of IP et ee fa ult , usi ess pa t e s, a d u i e sities is affe ted the latte ’s autonomy from governmental control, which determines their capability to steer and control their fa ult ’s activities, including inventive ones. A synthetic definition of autonomy is provided by Aghion et al. (2010), who consider two parameters: (i) whether a u i e sit ’s udget eeds to e app o ed the state; a d ii the pe e tage of the u i e sit ’s udget asso iated to o peti g grants, as opposed to block grants. A more complete conceptualization is provided by the European University Association (EUA), which easu es auto o looki g « […] at the a ilit of u i e sities to de ide o : • organisational structures and i stitutio al go e a e […] • financial issues, in particular the different forms of acquiring and allocating funding • staffing matters, in particular the responsibility for terms of employment [which may include IP matters] • academic matters, in particular the control over student admissions » (Estermann and Nokkala, 2009; p.7) 2 Historically, European universities have never enjoyed the same degree of autonomy of their US counterparts (Ben-David, 1977; Clark, 1993). This has limited their control over their finances (including IP assets and related revenues) and staff (including the freedom to set up clear rules over IP concerning their faculty's inventions, such as disclosure obligations and rewards). In the absence of such control, European universities have traditionally resisted being involved in IP management, and often took the shortcut of allowing scientists to take their own decisions, even in the absence of the professor privilege. When engaged in cooperative or contract research with third parties, the latter often signed blanket agreements leaving all IP rights i thei pa t e s’ ha ds. As a result, a large part of academic patents in Europe has for long gone unnoticed by official statistics, which classify the origin of the patent according to the applicants' identity, not the inventors'. It is only when recent studies have moved to re-classifying patents by inventor, and matched inventors' names to the names of university scientists, that the reality of academic patenting in Europe has emerged (review by Lissoni, 2012; more up-to-date information in the other articles of this issue). In all the countries considered by the literature, a significant percentage (from 3% to 8%) of corporate patents has been found to cover inventions by academic scientists. However, universities are the least important category of assignees of these same academic patents, with shares around 10% in most countries. Everywhere, they are superseded by business companies, whose shares, in the 1990s and early 2000s, ranged from 61% (in France) to 80% (in Sweden). The highest shares of university-owned academic patents were found in the Netherlands (26%) and in the United Kingdom (22%), whose universities enjoy the highest degree of autonomy in Europe (Estermann et al., 2009). In the US, which host a large number of private universities and whose public universities are not under federal control, the university share of academic patents can be estimated at 70%, and suggestions have been made that only inventions from consultancy escape the ad i ist atio s’ control (Thursby et al., 2009). From this discussion, we can formulate a simple research question: Q1 – Does the share of university ownership of academic patents increase with university autonomy? The o e ship dist i utio of IP a ha e o se ue es oth fo the u i e sities’ fi a es, a d their incentives to promote technology transfer activities. A detailed discussion of the first issue goes beyond the s ope of this pape . We li it ou sel es to poi t out that pate t o e ship a affe t a u i e sit ’s finances in two ways: a direct one, through the net results of IP management costs and IP licensing and trading revenues; and an indirect one, to the extent that exhibiting a strong patent portfolio improves either the evaluation of the university by funding agencies or its general reputation for technological excellence. The direct effect appears to benefit only a handful of large universities in the US (Bulut and Moschini, 2009; MacDonald, 2011), while the second one is highly disputed (Leydesdorff and Meyer, 2010; Thursby and Thursby, 2011) . As for the effects of ownership on incentives, we are mostly concerned with those leading to the introduction of inventions (for their development and commercialization, see Dechenaux et al., 2011; and Darmsgaard and Thursby, 2012). Our research question can be phrased as follows: Q2 - Does autonomy affect positively universities’ contribution to inventive activity in their country? The same two questions can be reformulated with reference to the introduction of the professor privilege. The expected profits may either push individual scientists to file a patent on their inventions, but also (and 3 with opposite effects) shield them from the pressure for commercialization eventually exerted by the university administration. Both scholars and legislators, however, have placed emphasis on the first of these two effectsEvidence related to our research questions is still in its infancy, and too US-centric for lending itself to a generalization. As for Q2, Aghion et al. (2010), Bonaccorsi and Daraio (2007), and Estermann et al. (2009) suggest that autonomy is associated to higher efficiency and productivity. However, these studies focus on u i e sities’ general tasks (teaching, research, and technology transfer), and make no use of specific datasets on patents. Concerning Q1, some evidence exist on the effects of the abolition of professor privilege, which is however limited by lack of data. The latter is explained by the persistence, until recently, of a methodological problem, best described as a trade-off between accuracy, scope, and longitudinal depth. Identifying academic patents on the basis of i e to s e ui es at hi g the i e to s’ a es to the names of academic scientists. While the information on inventors come with patent data, for which very long time series are available, information on academic scientists is generally obtained through ad hoc requests to universities or governmental institutions, which almost always are satisfied for one year only. Most existing studies are therefore limited to either cross-sectional evidence for a single country (Meyer, 2003; Saragossi and van Pottelsberghe, 2003; Balconi et al., 2004), cross-country evidence (Lissoni et al., 2008) or university case studies. In one particular case, that of Germany, researchers have made use of a surrogate indicator of the academic p o e a e of the pate t, su h as the p ese e of the a ade i title of P ofesso i the i e to ’s a e, which allows to build time series (Schmiemann and Durvy, 2003; Gering and Schmoch in OECD, 2003; Czarnitzki et al., 2007; 2009a,b; 2011a,b; 2012) . However, Von Proff et al. (2012), who follow the same approach, point out its many limitations. A more innovative alternative has been recently proposed by Dornbusch et al. (2013), who verify the academic affiliations of inventors by means of bibliometric sources. In this case, we miss all patents by academic scientists whose publications cannot be retrieved (possibly due to deficiencies in bibliometric sources for remote years). Our own solution to this problem is illustrated in section 4. 3. Italian universities: autonomy and IP legislation The Italian university system has been for long characterized by a combination of academic corporatism and governmental bureaucracy, and a weak role for the university administration. Until the 1990s, academics were pure civil servants, paid directly by the State, which also regulated their careers and duties. Universities could not actively dispose of their revenues, personnel, and curricula (Giglioli, 1979). In 1989 a major reform (L168/1989) established new principles concerning the distribution of authority and coordination in the system, followed by three further pieces of legislation (L.341/1990; L.537/1993-art.5; and D.M.9/2/1996) that introduced autonomy also with respect to educational offer and financial management. Block grant funding was introduced, with a major fund (FFO, "Fondo di Finanziamento Ordinario") coming to replace direct transfers from the state to professors for wages and earmarked transfers to universities for all other expenses. A new system of research funding was also introduced, with more room for competitive, peer-reviewed allocations. Finally, universities were given permit to raise their own revenues by accessing financial credit, commercializing their research results, increasing student fees, and getting support from local authorities (Moscati and Vaira, 2008; Geuna and Rossi, 2013). As a result, 4 the sha e of Italia u i e sities’ e te al sou es of fu di g o e total e e ues has o ed f o ea l zero in 1994 to around 30% in 2010, while the weight of FFO has declined steadily (figure A3 in Additional Material). These trends, however, are the result of fo es that ha e little to do ith u i e sities’ o e ializatio efforts. First, they are due to a steady decline of block grant funding; after increasing, at constant prices, throughout the 1990s, this has steadily declined since 2000 (starting 2008, it has declined also at current prices). Second, while the reform laws mentioned explicitly technology transfer as part of the u i e sities’ mission, no clear indication nor incentive scheme was introduced (Bonaccorsi and Daraio, 2007)1. Last, in 2001, the professor privilege was introduced. At the beginning of our period of interest, patent matters in Italy were regulated by a rather old "law on inventions" (RD1127/1939), which did not include any specific provision for university. Academic inventions were presumed to belong to the inventor's employer, but it was not clear whether the latter was the university or the State, nor did any legal norm existed to compel disclosure. In this vacuum, academic inventors either retained tacitly the property of inventions or negotiated it with sponsor companies and funding agencies (Balconi et al., 2004; Baldini et al., 2006). With the advent of autonomy, universities came to be ega ded as the a ade i i e to s’ employers. Hence, several of them introduced explicit IP regulations, starting 1995. By 2008, over 70% of Italian universities had adopted one (Baldini et al., 2010 and 2012; see figure A4 in the Additional Material). At the same time, the universities began transforming the organization of technology transfer activities, from a discretionary function in the hands of Rectors to one performed by specific offices (TTOs), with a dedicated staff and funded by the University internal resources (Potì and Romagnosi, 2010; figure A4 in the Additional Material). This process, however, did not go hand in hand with the adoption of IP statutes, so that we do not observe any correlation between the diffusion of IP regulations and that of TTOs2. When policy-makers finally turned their attention to IP matters, they did so in quite an extemporaneous fashion, by inserting a 10-lines article in the annual Budget Law (L383/2001, art.7). The article transferred the exclusive ownership of IP rights over academic research results to the professorship. The novelty had not been anticipated by any consultation with universities or enquiry on their current IP management practices, and was motivated (indeed loosely) by arguing that incentives to file patents and commercialise research results would have worked better if assigned to individual scientists, rather than universities (see Granieri, 2010; esp. chapters 1 and 6). 1 The only exceptions were: - the introduction of a legal notion of "spinoff company" in 1997; - the introduction of the "professor privilege" concerning IPR matters in 2001 (see section 3.2); - a short-lived provision of subsidies for the creation of technology transfer offices, from 2005 to 2007. 2 We explain this oddity in two ways. First, establishing a TTO absorbs financial resources, while approving an IP regulation is inexpensive (but politically complex, as it affects the relationship between faculty and administration). In addition, TTO activities may well go beyond or not include IP management. 5 Baldini et al. (2010, 2012) illustrate at length the universities' negative reaction to this legislative change. Only few institutions complied immediately with it, by adapting their IP statutes, while others explicitly amended them in the direction of circumventing the new law and keeping IP for the university3. I hat follo , e e plo e ho this te sio et ee u i e sities’ auto o a d IP legislatio has affe ted the object of our research questions, namely the distribution of IP ownership over academic inventions, and the contribution of universities to total patenting in Italy. 4. Data 4.1 Methodology and sample The main database used in this paper consists of patent applications filed at EPO, the European Patent Office, with priority dates comprised between 1996 and 2007 and at least one inventor with an Italian address. Academic inventors and their patents are identified by means of a 3-step procedure. STEP 1: Disambiguation of inventors' names STEP 2: Name matching between disambiguated inventors and academic personnel, the latter's names made available, in 2000, 2005 and 2009 by the Italian Ministry of Education. This step produced 10118 "professor-patent" pairs obtained by attributing to each professor the patents signed by the matched inventors. STEP 3: Validation of "professor-patent" pairs, on the basis of automatic criteria, manual checking, telephone and email surveys, and two regression exercises. After completing these three steps, we were left with: - a dataset of Italian patents and inventors, containing all patents by inventors with an Italian address in the period of interest (42784 inventors for 51054 patents) - three datasets of Italian academic patents, containing espe ti el a lo e a da uppe ou d , a i te ediate , ou d esti ate of the phe o e o of i te est. Full details of step 1 are provided by Pezzoni et al. (2013). The Additional Material attached to this paper summarizes step 1 and provides full details of steps 2 and 3. In what follows we limit ourselves to illustrate the o te ts a d the diffe e es et ee the lo e The lo e ou d dataset o tai s ou d , i te 99 a ade i i e to s a d ediate , a d uppe 9 pate ts. The ou d datasets. e e ide tified afte validating the results of steps 1 and 2 by: - making use of any information available, either on the patent, the web, or past surveys of Italian academic inventors (Balconi et al., 2004; Lissoni et al., 2006 and 2008) - an e-mail survey, which obtained a 37.5% response rate. 3 Notwithstanding this diffused criticism, the norm on the professor privilege was maintained in the new Code of Industrial Property, introduced in 2005, although with some amendments that lifted the professor privilege in case of formal collaborations between university and industry. 6 The lo e ou d dataset is likely to be affected by time-related bias. This is because key information we used for validation is patent ownership. This means that university—owned academic patents are likely to be over-represented in the dataset, relative to academic patents owned by firms and other subjects. To the extent that university ownership increases over time (see below), the bias becomes larger in recent years. For this reason, we also run two distinct probit regressions that exploited available information on two samples of professors, one relative to those unreached by the survey (due to unavailability of any contact information), the other to non-respondents. Coefficients from the regressions were then used to predict which patents, respectively among those of unreachable and non-respondent professors, could be treated as academic. By adding predicted academic patents out of the unreachable cases to the lower bound dataset, e o tai ed the i te ediate dataset (2399 academic inventors and 3093 academic patents). By further adding the predicted academic patents out of non-response ases e o tai ed the uppe ou d dataset (2602 academic inventors and 3535 academic patents). We presume the latter to be the least affected by any time-related bias. 4.2 Descriptive analysis of trends Figure 1 shows that the number of academic patents in Italy has increased over time, no matter whether we consider our lower bound, intermediate, and upper bound estimates. Data for years after 2007 are not reported, as they are truncated due to publication delays; also data for 2007 have to be treated with caution, due to some right truncation problems. Figure 1 – Nr of academic patents, 1996-2007; upper, intermediate & lower bound estimates 400 Upper bound est. Intermediate est. Lower bound est. 350 300 250 200 150 100 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Figure 2, however, provides a different picture. It reports the same data, but as percentages over the total number of patents by Italian inventors, joint with estimated time trends, in the form of linear regressions (based on years 1996-2006, that is excluding observations for 2007). According to the type of estimate 7 considered, the 1996-2006 average share of academic patents is between 4.5% and 7%, two figures which are compatible with previous findings (Lissoni et al., 2008). We notice that lower bound estimates suggest the existence of a positive and significant trend, which is absent when considering the intermediate and upper bound estimates. This is in line with our expectation to observe a positive time-related bias when using lower bound estimates. Figure 2 Share of academic patents over all patents by Italian inventors, 1996-2006; upper, intermediate & lower bound estimates (% values) 8,0 Upper bound est. Intermediate est. Lower bound est. 7,5 y = -0,0266x + 7,022*** R² = 0,0553 7,0 6,5 6,0 y = -0,0525x + 6,2943*** R² = 0,1391 5,5 5,0 y = 0,0944**x + 4,5503*** R² = 0,3873 4,5 4,0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Table 1 – Distribution and ownership of academic patents by university (top ten vs. others), 1996-2007; upper bound estimates Ownership (% share by type of owner) nr % Gov't & Foreign univ University Company Individual PROs & PROs patents patents Milano 331 7.7 14.6 72.8 4.3 5.2 3.2 Politecnico Milano 290 6.7 25.9 67.5 3.4 2.2 0.9 Bologna 288 6.7 17.0 66.4 7.5 5.0 4.1 Roma "Sapienza" 241 5.6 27.0 58.9 4.8 7.0 2.2 Firenze 169 3.9 18.4 61.1 12.4 5.4 2.7 Napoli "Federico II" 169 3.9 11.8 64.7 11.2 4.3 8.0 Padova 168 3.9 10.1 70.4 10.1 6.7 2.8 Pisa 164 3.8 11.0 72.7 10.5 3.5 2.3 Catania 158 3.7 7.8 83.1 3.0 5.4 0.6 Torino 156 3.6 15.2 69.0 9.9 2.9 2.9 Total top 10 universities Other universities with 50 patents (1) Other universities with >1 patent 2134 49.4 18.4 74.1 7.7 5.2 3.2 1488 34.4 22.0 71.2 7.9 8.1 3.0 699 16.2 13.4 52.6 10.4 11.0 4.4 (1) Ferrara, Pavia, Modena & Reggio, Roma "Tor Vergata", Politecnico Torino, Genova, Parma, Perugia, Milano-Bicocca, Siena, Palermo, Bari, Udine, Trieste, Brescia, Salerno, Cagliari 8 The second column of table 1 shows that academic patenting appears to be quite concentrated by university, with four universities holding higher-than-5% individual shares of all academic patents, followed by 6 other institutions with higher-than-3% shares (for a C10 index almost equal to 50%; 60 universities have at least one patent). This concentration and ranking are stable over time (data available on request). Columns from third to last of table 1 provide information on the ownership of academic patents, by university, with double counting of patents owned by subjects belonging to different categories (but no double counting of patents owned by more than one subjects, if all from the same category)4. In all cases, business companies own the largest share, with universities a distant second. We also notice quite a remarkable heterogeneity by university. Figure 3 provides information on aggregate time trends concerning academic patent ownership (upper bound estimates only). Two very visible trends emerge, a negative one for company ownership, and a positive one for university ownership. This is in line with our expectation of an increasing control exerted by universities on IP over their scientists' inventions, as a result of their increasing autonomy. When using data from intermediate and lower bound estimates we get similar results in terms of trends, albeit not in levels (see figure A5 in additional material). Figure 3 – Ownership of academic patents 1996-2007: % of academic patents by type of owner (1) (2) 35,0 80,0 y = -1,8014x + 79,092 R² = 0,8296 30,0 70,0 Universities 25,0 60,0 PROs & Gov't Individuals Foreign univ. & PROs 20,0 50,0 Companies y = 2,1115x + 1,4022 R² = 0,9169 15,0 40,0 30,0 10,0 20,0 5,0 10,0 0,0 0,0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 (1) % of patent with company ownership on right axis; all others on left axis (2) upper bound estimates 4 Ownership information dates back not to the filing or priority date of the patent, but to information contained in the 2010 edition of PatStat. This suggests that some change of property may have occurred in the meanwhile (Sterzi, 2013). Consultation of alternative sources suggests them to be around 5%. 9 When disaggregating data on ownership by technological field we find that academic patenting is concentrated in science-based technologies such as Scientific Instruments, Pharmaceuticals & Biotechnology, Chemicals & Materials, and Electrical Engineering/Electronics (see table A6 in Additional Material). University ownership is the highest in the first two classes and the lowest in the other two (data available on request). These results are in line with previous studies and intuitively explained by differences with the origin of inventions (with consultancy and contract research being most important in Chemicals and Electronics) and the strategic value of patents (with Chemical and Electronic patents being valued by companies, but not universities, as defensive assets; and patents in Scientific Instruments and Pharma & Biotech which universities are more likely to see as a potential source of royalties). 5. Econometric analysis and discussion 5.1 Specification In what follows we run two probit regressions, where the dependent variables are, respectively, the probability to observe an academic patent, and the probability to observe university ownership, conditional on the patent to be academic. We run the two regressions both separately and as related steps in a Heckman selection model (ch. 19 in Wooldridge, 2010). Accordingly, we will refer to them as STEP1 and STEP2, both when run independently and when run jointly. Our main exercises will make use of the upper bound dataset of academic patents, with regressions based on intermediate and lower bound datasets used as robustness checks (section 3 of the Additional Material). Observations in regression STEP1 are EPO patent applications signed by at least one Italian inventor, with priority dates 1996-2007 for a total of 51504 patents. The dependent variable is a binary one, =1 for academic patents (around 7% of observations). Regressors include:  Year dummies, which capture any trend left after controlling for all other determinants of academic patenting, as well as the effects of the introduction of the professor privilege, in 2001 (reference year).  Technology dummies5  Other characteristics of patents, namely: the total number of inventors listed on the patent (N_INV), the share of backward citations to non-patent literature (SHARE_NPL), and the total number of backward citations (TOT_CIT). We expect a positive sign in all cases. For what concerns N_INV, this is a pure statistical effect, discussed by Lissoni et al. (2013). Non-patent literature citations are a common indicator of science-intensiveness of the patent, which makes its academic origin more likely. And the total number of citations is an indicator of patent quality, which some literature suggest being higher for academic patents (survey by Lissoni and Montobbio, 2013).  Average financial conditions of universities, namely: FFO_RATIO_REGION and SCIENCE_RATIO_REGION, which easu e, fo the u i e sities i the i e to ’s egio , the eight on total revenues of, respectively, block grants (FFO) and funds for scientific projects. We test here the hypothesis that universities that are less dependent from block grants contribute more to academic patenting. This is 5 As several patents fall in more than one technological field, we keep all dummies in the regression, with no reference case. 10 reasonable only to the extent that a low FFO_RATIO is due to a high share of revenues from collaboration with industry, rather than other sources of external funding (e.g. support from local authorities or student fees). In the light of the discussion conducted in section 3 we are then pretty cautious about the possibility to observe a significant effect. As for SCIENCE_RATIO_REGION, we expect it to be a sign of high scientific standing, positively correlated to academic patenting. Unfortunately these data were made by universities only starting from 2000 and are subject to problems of comparability and consistency.  IP and technology transfer policies of universities in the inventor's region, as measured by the diffusion of IP statutes and TTOs (respectively, FIRST_STATUTE_REGION and TTO_REGION), both expected to affect positively the dependent variable. We control for the number of universities active in the region in each year, as reported by ministerial sources (NR_UNIVERSITIES_REGION). Due to missing values for a few years, considering these variables reduces slightly the number of observations.  Regional R&D structure, as measured by the Business R&D intensity of the local economy (BERD/GDP) and the regional innovation system's dependence on public R&D (RD_SHARE_PAUNI). We expect RD_SHARE_PAUNI to be positively correlated to academic patenting, as it indicates how much the local inventive activity depends upon the academics' contributions. BERD/GDP is also expected to have a positive sign, to the extent that it signals the importance, in the local innovation system, of scienceintensive industries, which are the natural candidates for collaboration with university.  Regional dummies: they control for heterogeneity across regions besides the R&D structure and the diffusion of IP statutes and TTOs. For reasons of space, all descriptive statistics are relegated to section 2 of the Additional Material. All variables concerning the average financial conditions of universities, the IP and technology transfer policies of universities, and regional R&D structures are inserted with 1-year lags, following classic findings on R&D-patent lag structure (Hall et al., 1986; Griliches , 1990). Regressions with no lags or 2-year lags produce very similar results. In case of multiple inventors from different regions for the same patent, we use the cross-regional average values for continuous variables, and multiple values for dummies. Observations in regression STEP2 are a subset of those in STEP1, as they consist only of academic patents (3443 observations). The dependent variable is again a binary one, =1 if the patent assignee is a university or, in case of multiple assignees, if at least one of them is a university (17.7% of total observations). We run complementary regressions in which the dependent variable takes value 1 in case of (exclusive) business ownership or individual ownership. When commenting them, we will refer to them as STEP2-individual and STEP2-company regressions as opposed to STEP2-university regressions. The explanatory variables of STEP2 regressions include:  Year dummies, technological dummies and other characteristics of the patent and the R&D system of the u i e sit ’s egio (as in STEP1).  University-level variables, namely  FIRST_STATUTE and TTO, both of them being dummy variables. They take value 0 over the years, respectively, before the adoption of an IP statute by the university and the opening of the TTO, and value 1 afterwards. When several inventors from different universities are listed on the same patent, 11 we take the highest value. We expect the estimated coefficient for FIRST_STATUTE to take a positive value in STEP2-university regression and a negative one in STEP2-individual and STEP2-company. The same for TTO, although with some reservations, due to the quality of the data and the fact that the presence of a TTO may not be as indicative of the university having an explicit IP policy.  FFO_RATIO and SCIENCE_RATIO, which are analogous to FFO_RATIO_REGION and SCIENCE_RATIO_REGION, but for individual universities. We expect the former to affect negatively (positively) the dependent variable in STEP2-university (STEP2-company) regression. The opposite holds for the latter. None of them should affect the STEP2-individual regression.  University dummies, but only for the universities with at least 50 patents (dummies for universities with fewer patents result in completely determined) As in STEP1, in case of multiple inventors from different universities for the same patent, we consider the cross-region averages, for all regions listed on the patent, and multiple dummies. 5.2 Results Table 2 presents the results of regression STEP1 for three specifications: year dummies only (column 1); all variables, with the exception of FFO_RATIO_REGION and SCIENCE_RATIO_REGION (column 2); all variables (column 3), at the cost of eliminating observations for years 1996-2000, due to missing values. Results from column (1) can be directly compared to figure 2, as the sign of coefficients reflects differences between the share of academic patents in 2001 and other years. Moving to column (2), we notice that the sign and significance of coefficients of year dummies change. In particular, coefficients for years before 2000 become all positive and (with one exception only) significant, while all others become negative (and significant in two cases). This suggests the existence of a negative trend, which we interpret as follows: given the relationship between academic patenting and its determinants, changes in the latter should have led to an increase in the share of academic patents, which failed instead to materialize (we do not consider here 2007, whose negative sign and large absolute value are explained by right truncation). Among the most significant determinants of the probability of a patent to be academic, with similar values of the coefficients in specifications (2) and (3), we have: the technology dummies, the characteristics of the patent, and the structure of the regional R&D (sign and magnitude of coefficients are all in line with the descriptive analysis). On the contrary, none of the variables related to the universities' characteristics seem to matter. 12 Table 2 – STEP1 probit regression (dep. variable: probability of a patent to be academic; upper bound data) Year 1996 Year 1997 Year 1998 Year 1999 Year 2000 Year 2002 Year 2003 Year 2004 Year 2005 Year 2006 Year 2007 (1) 0.042 (0.045) 0.025 (0.044) -0.028 (0.044) 0.036 (0.042) -0.020 (0.042) 0.016 (0.041) -0.048 (0.041) 0.015 (0.040) -0.0093 (0.040) 0.027 (0.039) -0.088** (0.042) (2) 0.16*** (0.056) 0.12** (0.054) 0.024 (0.052) 0.11** (0.050) -0.012 (0.049) -0.028 (0.050) -0.12** (0.053) -0.096* (0.055) -0.100 (0.065) -0.050 (0.073) -0.18** (0.080) 0.046 (0.032) 0.31*** (0.028) 0.095*** (0.028) 0.57*** (0.031) -0.31*** (0.031) -0.40*** (0.036) -0.45*** (0.043) 0.15*** (0.0058) 0.41*** (0.026) 0.0091*** (0.0011) -0.0014 (0.083) 0.10 (0.066) 0.016 (0.012) 0.41** (0.17) 1.09*** (0.29) -1.49*** (0.030) N 51,054 0.00072 -3.27*** (0.24) Y 50,875 0.24 Electrical Eng.; Electronics Scientific instruments; Measurement Chemicals; Materials Pharmaceuticals; Biotechnology Industrial Processes Mechanical Eng.; Machines; Transport Consumer goods; Civil Eng. N_INV (nr of inventors) SHARE_NPL (% of citations to non-patent literature) TOT_CIT (tot nr of backward citations) TTO_REGION (regional diffusion TTOs) STATUTE_REGION (regional diffusion IP statutes) NR_UNIVERSITIES_REGION BERD/GDP (regional BERD/GDP) RD_SHARE_PAUNI (% of R&D by public administration & universities, in region) FFO_RATIO_REGION (block grant as % of univ.'s revenues, regional avg) SCIENCE_RATIO_REGION (public research funds % of univ.'s revenues, regional avg) Constant Regional dummies Observations Pseudo R2 (3) 0.047 (0.070) -0.068 (0.074) -0.046 (0.079) -0.055 (0.088) -0.0097 (0.097) -0.12 (0.11) -0.039 (0.042) 0.34*** (0.037) 0.080** (0.036) 0.53*** (0.039) -0.31*** (0.041) -0.50*** (0.048) -0.58*** (0.060) 0.15*** (0.0071) 0.45*** (0.032) 0.013*** (0.0020) 0.068 (0.11) 0.14 (0.093) 0.019 (0.015) 0.22 (0.24) 0.92** (0.41) 0.23 (0.27) 0.63 (0.56) -3.36*** (0.34) Y 32,317 0.26 Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1 13 Figure 4 Academic patenting in selected regions and technologies: predicted probabilities, 1995-2006 (§) 0,45 0,4 0,35 0,3 0,25 0,2 0,15 0,1 0,05 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Lombardy -Pharma Campania -Pharma Lazio -Pharma Lombardy -Electronics Campania -Electronics Lazio -Electronics (§) Predicted probability in year t, for technology k and region z, estimated at the following values  continuous variables N_INV, SHARE_NPL, and TOT_CIT: mean values 1996-2007, for technology k  continuous variables: TTO_REGION, STATUTE_REGION, NR_UNIVERSITIES, BERD_GDP and RD_SHARE_PAUNI: mean values 1996-2007, for region z  dummy variables set at one for year t, technology k, and region z; zero otherwise As an illustration of the marginal effects associated to the estimated coefficients, figure 4 reports the predicted probability of a patent to be academic for three large Italian regions, respectively in the North, Centre and South of the country, in two of the most important fields of academic patenting (Electronics and Pharma-Biotech), based on estimates from column (2) of table 2. The decline of academic contribution to patenting is quite visible. The figure also suggests that regional differences are quite large (when compared to the size of the time trend), and inversely correlated to the industrial and R&D strength of the region (which are the highest in Lombardy and the lowest in Campania). Table 3 presents the results of STEP2 regressions for university ownership (columns 1 and 2), individual ownership (columns 3 and 4), and company ownership (columns 5 and 6). Odd columns refer to specifications for the complete period of observations (1996-2007), while even columns include variables on universities' financial conditions, at the cost of excluding years in which they are not available (19962000). Estimated coefficient for year dummies in column (1) confirm the existence of a positive trend in university ownership, with three significantly negative coefficients before 2001 and two positive and significant coefficients in the following period. However, several post-2001 coefficients are not significant and even take a negative sign. This suggests that our regressors may explain away part of the trend. 14 Table 3 – Heckman probit regressions (STEP1, unreported; STEP2: prob. of an academic patent to be owned by university/individual/company) – upper bound estimate data Year 1996 Year 1997 Year 1998 Year 1999 Year 2000 Year 2002 Year 2003 Year 2004 Year 2005 Year 2006 Year 2007 Electrical Eng.; Electronics Scientific instruments; Measurement Chemicals; Materials Pharmaceuticals; Biotechnology Industrial Processes Mechanical Eng.; Machines; Transport Consumer goods; Civil Eng. N_INV (nr of inventors) SHARE_NPL (% of citations to non-patent literature) TOT_CIT (tot nr of backward citations) BERD/GDP (regional BERD/GDP) RD_SHARE_PAUNI (% of R&D by public administration & universities, in region) FIRST_STATUTE (IP regulation in place) TTO (TTO in place) FFO_RATIO (block grant as % of revenues) SCIENCE_RATIO (research as % revenues) Constant University ownership Individual ownership (1) (2) (3) (4) -0.20 -0.034 (0.17) (0.18) -0.48*** -0.18 (0.18) (0.17) -0.41** 0.034 (0.17) (0.16) -0.29** -0.13 (0.15) (0.16) -0.011 -0.14 (0.13) (0.16) 0.16 0.13 0.24 0.14 (0.13) (0.14) (0.15) (0.19) -0.048 -0.024 0.21 0.034 (0.14) (0.15) (0.15) (0.20) 0.087 0.16 -0.00094 -0.30 (0.13) (0.15) (0.16) (0.22) -0.047 0.071 0.047 -0.15 (0.14) (0.16) (0.16) (0.22) 0.28** 0.35** -0.23 -0.42* (0.14) (0.16) (0.18) (0.24) 0.36** 0.39** -0.20 -0.31 (0.15) (0.18) (0.19) (0.26) -0.17* -0.0032 -0.61*** -0.28* (0.093) (0.10) (0.12) (0.16) 0.29*** 0.45*** -0.074 -0.071 (0.078) (0.082) (0.095) (0.13) -0.038 -0.029 -0.43*** -0.34*** (0.072) (0.082) (0.090) (0.12) 0.13 0.40*** -0.20* -0.096 (0.094) (0.093) (0.11) (0.15) 0.25** 0.17 0.14 0.19 (0.099) (0.12) (0.12) (0.15) -0.090 -0.19 0.091 0.22 (0.14) (0.15) (0.15) (0.21) -0.017 -0.0070 0.41*** 0.42 (0.17) (0.21) (0.16) (0.26) 0.0089 0.075*** -0.19*** -0.19*** (0.018) (0.024) (0.028) (0.032) 0.62*** 0.84*** 0.12 0.12 (0.091) (0.089) (0.11) (0.16) -0.00081 0.0062 -0.00023 -0.0057 (0.0033) (0.0040) (0.0035) (0.0062) -0.079 -0.31 -0.58* -0.13 (0.26) (0.30) (0.30) (0.43) 0.42 0.035 0.023 0.79 (0.40) (0.46) (0.47) (0.67) 0.35*** 0.23** -0.058 -0.14 (0.083) (0.100) (0.094) (0.13) -0.087 -0.14 0.032 0.13 (0.085) (0.094) (0.097) (0.13) 0.014 -1.03* (0.37) (0.53) -0.42 -0.61 (0.59) (0.86) -1.90*** -2.77*** 0.32 0.56 (0.52) (0.60) (0.68) (0.97) Y Y Y Y 50,793 32,041 50,793 32,041 0.065 0.62** -0.31* -0.46** 47437 30187 47437 30187 0.16 0.17 0.13 0.14 University dummies(§) Observations Rho Censored observations Pseudo R2 Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1 (§) Only for universities with >50 patents (all other universities as reference case) ; (#)The nr. of observations is slightly less than that reported in section 5.1, due to missing values Firm ownership (5) (6) -0.019 (0.14) 0.29** (0.13) 0.23* (0.13) 0.30** (0.12) 0.28** (0.12) 0.095 -0.093 (0.12) (0.15) 0.063 -0.076 (0.12) (0.15) 0.066 -0.17 (0.12) (0.16) 0.12 -0.13 (0.12) (0.17) 0.033 -0.24 (0.12) (0.17) 0.056 -0.28 (0.13) (0.19) 0.45*** 0.23** (0.084) (0.11) -0.17** -0.30*** (0.071) (0.10) 0.25*** 0.22** (0.065) (0.089) -0.052 -0.24** (0.088) (0.12) -0.036 -0.18 (0.092) (0.13) 0.24** 0.12 (0.12) (0.17) 0.018 0.031 (0.15) (0.23) 0.050** 0.072** (0.023) (0.036) -0.54*** -0.68*** (0.079) (0.11) 0.00036 0.0015 (0.0028) (0.0047) 0.22 -0.087 (0.23) (0.32) -0.98*** -1.25** (0.35) (0.49) -0.22*** -0.22** (0.072) (0.10) 0.0080 0.027 (0.075) (0.099) -0.48 (0.40) -0.19 (0.63) 1.29** 2.27*** (0.51) (0.82) Y Y 50,793 32,041 -0.17 -0.17 47437 30187 0.09 0.11 15 Technology dummies confirm that university ownership tends to be quite high in Scientific Instruments and, to less extent, Pharma & Biotech. Among the characteristics of the patent, the share of citations to non-patent literature is the only one to exhibit a significant and positive sign. More importantly, a positive determinant of university ownership is the adoption of an IP regulation, whose coefficient is positive and significant in both specifications (1) and (2). Notice that university dummies control for fixed effects, so that our result can be interpreted in a causal way, as indicative of a change in the university' strategic attitude towards patenting, made possible by the newly gained autonomy. As for the presence of TTOs, this looks irrelevant. Neither the R&D structure of the university's region, nor the university's financial conditions (FFO_RATIO and SCIENCE_RATIO) seem to bear any effect on the probability of university ownership6. Figure 5 University-ownership of academic patents in Pharma-Biotech, before/after the adoption of an IPR statute: predicted probabilities for selected universities, 1995-2006 (§) 0,6 Milano (adopted) Roma Sapienza (adopted) Roma Sapienza (not adopted) 0,5 Padova (adopted) Padova (not adopted) 0,4 0,3 0,2 0,1 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 (§) Predicted probability in year t, at university k, for the following values of regressors:  continuous variables N_INV, SHARE_NPL, and TOT_CIT: mean values 1996-2007, for Pharma-Biotech  continuous variables BERD_GDP and RD_SHARE_PAUNI: local mean values 1996-2007 (Milan: 0.78; 0.32 / Rome: 0.53; 0.62 / Padova: 0.46; 0.49)  dummy variables FIRST_STATUTE and TTO set at one for university k in the adoption year and the following; zero in the adoption year and previous  other dummy variables set at one for year t, Pharma-Biotech, and university k; zero otherwise Figure 5 reports the predicted probabilities of university ownership for three universities in the top ten list of table 1, for Pharma-Biotech academic patents, over the period 1996-2007 (during which two universities 6 We tried also to insert quadratic terms, to no avail 16 adopted an IPR statute). The positive trend is quite visible and we notice that its overall magnitude is considerable: in the case of the university of Milan (which adopted its IPR statute before 1996), the probability of university ownership doubles in the period considered. We can also appreciate the impact of the adoption of an IPR statute, which explains entirely the difference between the universities of Padova and Milan, and the increased difference, after 2000, between these universities and that of Rome "La Sapienza". Notice that neither the dummies for Milan and Padova are significant, while that for Rome "La Sapienza" is positive and significant. Results for company ownership (columns 5 and 6 of table 3) are the mirror image of those for university ownership. The signs of the estimated coefficients are always the opposite, with the only exceptions of year dummies 1996, 2002, 2006, and 2007 in column 5 (which are anyway not significant) and N_INV (see below). We also notice that the coefficient of FIRST_STATUTE, as expected, is negative and significant. These results suggest that universities have increased their share of academic patents by bargaining more actively with the same business companies to whom, in the past, they would have relinquished all IP. Moving to individual ownership (columns 3 and 4 of table 3) we notice that no increase occurred after the introduction of the professor privilege (no year dummy is significant, with the exception of 2006 in specification (4), which is anyway negative). We also notice that:  individually-owned academic patents are more likely to be found not in science-based fields, but in Consumer Goods (specification 3);  the estimated coefficient of N_INV is negative and significant, while it is positive both for company and university ownership.  the estimates of the Heckman Rho are positive (and significant in specification (2)) in the university ownership regressions, while they are negative and significant in the case of individual ownership (and negative but never significant for company ownership). These results suggest that the nature of inventions protected by individually-owned patents is different from that of both company- and (especially) university-owned ones. The latter are likely to derive from a scientific research programme, typically pursued by a team and with some relationship to academic disciplines, while the former look more like the results of extemporaneous individual initiatives, and possi l so e ga age i e tio s p odu ed u i e sit s ie tists outside the eal of thei p ofessio . 6. Discussion and conclusions This paper has proposed the very first longitudinal analysis of academic patenting in Italy, and one of the first worldwide. We find that, from 1996 to 2006, the share of academic patenting over total patenting at the EPO has declined, conditional on the typical characteristics of academic patents and on the evolution over time of the Italian R&D system. This suggests that, ceteris paribus, Italian universities have met increasing difficulties to contribute to inventive activities, at least those subject to patenting. We do not have a ready explanation for this trend, but we suspect this is not due to lack of funding (the share of R&D spent by higher education and public laboratories did not decline in the ten years considered). It may be possibly due to lack of demand of collaboration with industry, the Italian one being less and less oriented 17 towards R&D-intensive activities. For sure, autonomy alone did not prove sufficient to push the overall academic system towards a greater technology transfer effort. We also find that the strength of academic patenting if positively affected both by the R&D intensity of the local (regional) economy, and by the local share of public R&D. The latter is the highest in less advanced, less-R&D intensive regions. This suggests that the origin of Italian academic patents may be very heterogeneous: some may stem from academics' collaboration with industry, others from purely academic research, which in the Southern regions is the main (or only) source of inventions. This may imply a high heterogeneity also in terms of quality and commercialization potential. The most noticeable time trend concerns the ownership distribution of academic patents, with universities reclaiming an ever-increasing share of academic patents. University ownership is explained by the characteristics of the patents, the local share of public R&D, and the introduction, in most universities, of specific IP regulations. The latter was an institutional innovation made possible by the newly acquired autonomy, often adopted in the absence of clear ministerial directives, and in contrast with the introduction of the professor privilege in 2001. The examination of marginal effects suggests that university ownership depends first and foremost on the nature of research funds (public vs. private), followed by universities' strategies (as measured by the adoption of an IP statutes and university dummies). The introduction of dedicated Technology Transfer Offices (TTOs) seems not to have exerted any positive influence. However, our TTO data hide a great heterogeneity in terms of size and skills, which we are not yet able to measure. As for policy conclusions we observe that the introduction of the professor privilege has neither encouraged academic patenting, nor favoured individual ownership. In fact, it has been effectively neutralized by universities, through the introduction of IP statutes. This deposes against the transformative potential of the privilege, in a context in which universities exploit their autonomy to increase their control over their faculty and resources. We suggest that debating over the professor privilege may be less relevant than debating the use made by universities of their increasing autonomy, when it comes to IP matters. We are not yet in a position to evaluate the observed trends in terms of financial returns to universities, and impact on innovation levels in the country. That requires further data collection, which is under way. Existing evidence for the subperiod 1996-2001, however, suggest that, in the case of Italy, university-owned patents are less cited than company-owned academic patents and non-academic ones (Lissoni and Montobbio, 2013). This may imply that Italian universities, in the 1990s, were doing a bad job with picking up the right inventions, or with managing effectively the patents they owned. If this was found to hold also in recent years, we should conclude against encouraging universities to expand their patent portfolios. The same research, however, provides opposite evidence for Dutch universities, which have enjoyed autonomy and accumulated experience in handling IP for a longer time than their Italian counterparts. Similar conclusions may be drawn from the study on Flemish universities by Callaert et al., in this issue. It is then possible that, nowadays, Italian universities have improved their selection and management skill concerning patentable inventions. Our main research objective for the immediate future consists in testing this hypothesis. 18 References AGHION, P., DEWATRIPONT, M., HOXBY, C., MAS-COLELL, A. & SAPIR, A. 2010. The governance and performance of universities: evidence from Europe and the US. Economic Policy, 25, 7-59. BALCONI, M., BRESCHI, S. & LISSONI, F. 2004. Networks of inventors and the role of academia: an exploration of Italian patent data. Research Policy, 33, 127-145. BALDINI, N., FINI, R. & GRIMALDI, R. 2012. The Transition Towards Entrepreneurial Universities: An Assessment of Academic Entrepreneurship in Italy. Available at SSRN: http://ssrn.com/abstract=1979450 or http://dx.doi.org/10.2139/ssrn.1979450. BALDINI, N., FINI, R., GRIMALDI, R. & SOBRERO, M. 2010. The Institutionalisation of University Patenting Activity in Italy: Diffusion and Evolution of Organisational Practices. SSRN eLibrary. BALDINI, N., GRIMALDI, R. & SOBRERO, M. 2006. Institutional changes and the commercialization of academic knowledge: A study of Italian unive sities’ pate ti g a ti ities et ee 9 a d . Research Policy, 35, 518-532. BEN-DAVID, J. 1977. Centres of learning: Britain, France, Germany, United States, New York NY, McGrawHill. BONACCORSI, A. & DARAIO, C. (eds.) 2007. Universities And Strategic Knowledge Creation. Specialization and Performance in Europe: Edward Elgar. BULUT, H. & MO“CHINI, G. 9. U“ u i e sities’ et etu s f o pate ti g a d li e si g: a ua tile regression analysis. Economics of Innovation and New Technology, 18, 123-137. CLARK, B. 1993. The Research foundations of graduate education: Germany, Britain, France, United States, Japan, Berkeley CA, University of California Press. COLYVAS, J., CROW, M., GELIJNS, A., MAZZOLENI, R., NELSON, R. R., ROSENBERG, N. & SAMPAT, B. N. 2002. How Do University Inventions Get into Practice? Management Science, 48, 61-72. CZARNITZKI, D., GLANZEL, W. & HUSSINGER, K. 2007. Patent and publication activities of German professors: an empirical assessment of their co-activity. Research Evaluation, 16, 311-319. CZARNITZKI, D., GLÄNZEL, W. & HUSSINGER, K. 2009a. Heterogeneity of patenting activity and its implications for scientific research. Research Policy, 38, 26-34. CZARNITZKI, D., GRIMPE, C. & TOOLE, A. A. 2011a. Delay and Secrecy: Does Industry Sponsorship Jeopardize Disclosure of Academic Research? SSRN eLibrary. CZARNITZKI, D., HUSSINGER, K. & SCHNEIDER, C. 2009b. Why Challenge the Ivory Tower? New Evidence on the Basicness of Academic Patents. Kyklos, 62, 488-499. CZARNITZKI, D., HUSSINGER, K. & SCHNEIDER, C. 2011b. Commercializing academic research: the quality of faculty patenting. Industrial and Corporate Change, 20, 1403-1437. CZARNITZKI, D., HUSSINGER, K. & SCHNEIDER, C. 2012. The nexus between science and industry: evidence from faculty inventions. The Journal of Technology Transfer, 37, 755-776. DAMSGAARD, E. F. & THURSBY, M. C. 2013. University entrepreneurship and professor privilege. Industrial and Corporate Change, 22, 183-218. DELLA MALVA, A., LISSONI, F. & LLERENA, P. 2013. Institutional change and academic patenting: French universities and the Innovation Act of 1999. Journal of Evolutionary Economics, 23, 211-239. DORNBUSCH, F., SCHMOCH, U., SCHULZE, N. & BETHKE, N. 2013. Identification of university-based patents: A new large-scale approach. Research Evaluation, 22, 52-63. ESTERMANN, T. & NOKKALA, T. 2009. University Autonomy in Europe - I. Brussels. FRANZONI, C. & LISSONI, F. 2009 Academic entrepreneurs: critical issues and lessons for Europe. In: VARGA, A. (ed.) Universities, Knowledge Transfer And Regional Development: Geography, Entrepreneurship and Policy. Cheltenham: Edward Elgar. GEUNA, A. & ROSSI, F. 2011. Changes to university IPR regulations in Europe and the impact on academic patenting. Research Policy, 40, 1068-1076. GEUNA, A. & ROSSI, F. 2013. L'università e il sistema economico, Bologna, Il Mulino. GIGLIOLI, P. P. 1979. Baroni e burocrati Bologna, Il Mulino. GRANIERI, M. 2010. La gestione della proprietà intellettuale nella ricerca universitaria. Invenzioni accademiche e trasferimento tecnologico, Bologna, Il Mulino. 19 GREENBAUM, D. & SCOTT, C. 2010. Hochschullehrerprivileg—A Mode I a atio of the P ofesso ’s Privilege to Promote University to Industry Technology Transfer. Science Technology & Society, 15, 55-76. GRILICHES, Z. 1990. Patent Statistics as Economic Indicators: A Survey. Journal of Economic Literature, 28, 37. HALL, B. H., GRILICHES, Z. & HAUSMAN, J. A. 1986. Patents and R&D: Is There A Lag? National Bureau of Economic Research Working Paper Series, No. 1454. IVERSEN, E., GULBRANDSEN, M. & KLITKOU, A. 2007. A baseline for the impact of academic patenting legislation in Norway. Scientometrics, 70, 393-414. JENSEN, R., THURSBY, J. & THURSBY, M. C. 2010. University-industry spillovers, government funding, and industrial consulting. National Bureau of Economic Research. JENSEN, R. & THURSBY, M. 2001. Proofs and Prototypes for Sale: The Licensing of University Inventions. American Economic Review, 91, 240-259. KENNEY, M. & PATTON, D. 2011. Does inventor ownership encourage university research-derived entrepreneurship? A six university comparison. Research Policy, 40, 1100-1112. LEYDESDORFF, L. & MEYER, M. 2010. The decline of university patenting and the end of the Bayh–Dole effect. Scientometrics, 83, 355-362. 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, 641-655. LISSONI, F. 2012. Academic patenting in Europe: An overview of recent research and new perspectives. World Patent Information, 34, 197-205. LISSONI, F., LLERENA, P., MCKELVEY, M. & SANDITOV, B. 2008. Academic patenting in Europe: new evidence from the KEINS database. Research Evaluation, 17, 87-102. LISSONI, F., LOTZ, P., SCHOVSBO, J. & TRECCANI, A. 2009. Academic patenting and the professor's privilege: evidence on Denmark from the KEINS database. Science and Public Policy, 36, 595-607. LISSONI, F. & MONTOBBIO, F. 2013. The ownership academic patents and their impact. Evidence from five European countries. mimeo. LISSONI, F., SANDITOV, B. & TARASCONI, G. 2006. The Keins Database on Academic Inventors: Methodology and Contents. CESPRI Working Paper Milano: Università Bocconi. MACDONALD, S. 2011. Seducing the goose. A review of patenting by UK universities. Intellectual Property Quarterly, 4, 323-344. MEJER, M. 2012. Essays on Patent Systems and Academic Patenting. PhD, Université Libre de Bruxelles. MOSCATI, R. & VAIRA, M. (eds.) L'università di fronte al cambiamento. Realizzazioni, problemi, prospettive, Bologna: Il Mulino. MOWERY, D. & SAMPAT, B. 2005. The Bayh-Dole Act of 1980 and University-Industry Technology Transfer: A Model for Other OECD Governments? In: LINK, A. N. & SCHERER, F. M. (eds.) Essays in Honor of Edwin Mansfield. Springer US. MOWERY, D. C., NELSON, R. R., SAMPAT, B. N. & ZIEDONIS, A. A. 2001. The growth of patenting and licensing by U.S. universities: an assessment of the effects of the Bayh–Dole act of 1980. Research Policy, 30, 99-119. PEZZONI, M., LISSONI, F. & TARASCONI, G. 2012. How To Kill Inventors: Testing The Massacrator© Algorithm For Inventor Disambiguation. Cahier du GREThA nr. 29. Groupe de Recherche en Economie Théorique et Appliquée - Université Bordeaux IV. POTI', B. & ROMAGNOSI, S. 2010. Effectiveness of the new tools for University research valorisation: TTOs characterization in IP management and their performance. International Workshop on Organizational Models of Technology Transfer Offices and the technology transfer manager profession. Alghero, September 10. RAFFO, J. & LHUILLERY, “. 9. Ho to pla the Na es Ga e : Pate t et ie al o pa i g diffe e t heuristics. Research Policy, 38, 1617-1627. SARAGOSSI, S. & VAN POTTELSBERGHE DE LA POTTERIE, B. 2003. What Patent Data Reveal about Universities: The Case of Belgium. The Journal of Technology Transfer, 28, 47-51. SCHMIEMANN, M. & DURVY, J.-N. 2003. New Approaches to Technology Transfer from Publicly Funded Research. The Journal of Technology Transfer, 28, 9-15. 20 STERZI, V. 2013. Patent quality and ownership: An analysis of UK faculty patenting. Research Policy, 42, 564576. THURSBY, J., FULLER, A. W. & THURSBY, M. 2009. US faculty patenting: Inside and outside the university. Research Policy, 38, 14-25. THURSBY, J. G. & THURSBY, M. C. 2011. Faculty participation in licensing: Implications for research. Research Policy, 40, 20-29. VALENTIN, F. & JENSEN, R. L. 2007. Effects on academia-industry collaboration of extending university property rights. The Journal of Technology Transfer, 32, 251-276. VON PROFF, S., BUENSTORF, G. & HUMMEL, M. 2012. University Patenting in Germany before and after 2002: What Role Did the Professors' Privilege Play? Industry and Innovation, 19, 23-44. WOOLDRIDGE, J. M. 2010. Econometric Analysis of Cross Section and Panel Data (2nd ed.), Cambridge MA, MIT Press. 21 University autonomy, the professor privilege and academic patenting: Italy, 1996-2007 (Additional Material) Francesco Lissoni 1,2 , Michele Pezzoni 2,3 , Bianca Potì 4 , Sandra Romagnosi 5 1 GREThA – Université Bordeaux IV - France CRIOS – Università "L. Bocconi", Milano – Italy 2 3 4 5 DEMS, University of Milano-Bicocca CERIS-Consiglio Nazionale delle Ricerche, Rome - Italy National Agency for the Evaluation of Universities and Research Institutes (ANVUR), Rome – Italy Corresponding author: F.Lissoni, GREThA - UMR CNRS 5113, Université Montesquieu - Bordeaux IV, avenue Léon Duguit, 33608 Pessac cedex – FRANCE ( francesco.lissoni@u-bordeaux4.fr; tel. +33 (0)5 56 84 86 04) Abstract This document contains additional material for Lissoni et al. (2012), as submitted for publication to Industry & Innovation, after revision. It is organized in four sections. Section 1 describes the methodology followed for creating the dataset of Italian academic inventors and patents used in the analysis. Section 2 contains detailed descriptive statistics for the data used in the regression analysis. Section 3 contains the results of robustness checks. Section 4 contains additional tables and figures to which Lissoni et al. (2012) refers. 22 1. Academic inventors and patents database: methodology and data 1.1 Overview The database originates from two different sources: the complete list of inventors with an Italian address, as reported on patent applications at the European Patent Office, 1978-2010 (from the October 2011 edition of PatStat); and several lists of assistant, associate, and full professors active in Italia universities in 2001, 2005, and 2010. The procedure for the database construction consists of three steps: 1. i e to s’ disa iguatio ; 2. identification of academic inventors and patents, through inventor-professor name matching; 3. validation of the resulting professor-patent pairs, through data manual inspection and a survey of matched professor; followed by: 3bis: analysis of professors who resulted either unreachable or did not respond to the validation survey; and identification, by means of two distinct econometric exercises (one for the unreachable cases, the other for the non-responses), of further professor-patent pairs to be retained as valid The outcome of this procedure consists of three alternative datasets of academic patents:  Lower-bound estimates: it contains results from steps 1 and 2; only patents in professor-patent pairs validated manually or through the survey are retained and considered academic; it contains  4743 pairs, for a total of 2199 academic inventors and 2679 academic patents. Intermediate estimates: it contains results from step 1, step 2, and the econometric exercise of step 3bis concerning unreachable cases; on the basis of the latter we estimate the probability that a professor-patent pair is valid (the patent is indeed an academic one). Non-respondents to the validation survey are treated as non-academic. It contains 5204 pairs, for a total of 2399 academic  inventors and 3093 academic patents. Upper bound estimates it contains results from step 1, step 2, and both the econometric exercises of step 3bis (one for unreachable cases, the other to non-responses). It contains 5733 pairs, for a total of 2602 academic inventors and 3535 academic patents. 1.2 STEP1 – Inventor disambiguation A longstanding problem for scholars using patent data for micro-econometrics, is the correct reclassification of patents by inventor. The reclassification effort usually consists in applying to raw patent 23 data a computer algorithm for assessing whether two inventors, uniquely identified by their name and address, are the same person, with some level of uncertainty. The computer science literature refers to this e e ise as disa iguatio o, o e ge e all , as e tit esolutio . Many scholars have made efforts and invested considerable resources in improving the quality of their patent data, most often on an individual basis and with a limited sharing of the results of their disambiguation exercises. In order to avoid duplicating efforts and to work in a non cumulative fashion, the European Science Foundation financed, in 2009-13, the APE-INV (Academic Patenting in Europe) database harmonization project. One of the main tasks of the project consisted in producing an inventor database, based on EPO data retrieved from the PatStat database, to be distributed freely and used immediately for the identification of academic inventors, as we do in the present paper. Two authors of the present paper contributed actively to the creation of the database, by producing the algorithm (Massacrator© 2.0) used in the disambiguation step (for full description: Pezzoni et al., 2013). The algorithm proceeds in three steps: (i) parsing & cleaning; (ii) matching; (iii) filtering7. In the parsing & cleaning step, the algo ith e o es f o the i e to s’ a es all ha a te s i luded i an ad hoc list of punctuations and non-ASCII characters. Addresses are also parsed into street, city, zip code, state and country. In the matching step inventors with identical names, but different addresses, or similar names are matched on the basis of the 2-gram distance between tokens composing the name, and an arbitrary distance threshold. The latter is as loose as possible in order to avoid false negatives during the matching phase, at the cost of introducing a large amount of false positives (for the definition of false negatives and positives see again Pezzoni et al., 2013). These are then filtered in the third step. In the filtering step, the algorithm calculates, for each pair of matched inventors, a "similarity score", based upon a set of 17 criteria. By comparing the score obtained by each pair to a threshold value, the algorithm then selects the valid matches (positive matches), and which to discard as non-valid matches (negative matches). The criteria considered are 17 grouped in 6 families: network, geographical, applicant features, te h olog , pate ts’ itatio s, a d othe s Ta le A . Thei sele tio , as ell as the ali atio of the threshold value, is the result of a Monte Carlo simulation, aimed at finding an efficient balance between precision (percentage of false positives over total positives) and recall (percentage of false negatives over total positives). Before the parsing phase of the algorithm, the number of distinct Italian inventors listed on EPO patents since 1978 were 83836, while after applying the entire disambiguation process they decrease up to 68157, about 19% less. 7 We o o the pa si g, at hi g, filte i g te i olog f o Raffo a d Luhille 9. 24 Ta le A1: List a d des riptio of riteria applied i the filteri g stage of the i Name of the criterion and family of criteria Network family 1 2 Common co-inventor 3 degrees of separation Geographical family 3 4 5 6 City Province Region State 7 Street Applicant family Applicant Small Applicant Group 8 9 10 Technology classes 11 12 13 IPC 12 IPC 6 IPC 4 Citation family 14 15 Citations ASE 16 Others Rare surname 17 Priority date differs for less than 3 years e tors’ disa iguatio algorith . Description Two matched inventors who turn out to be socially close are likely to be the same person. Social distance is measured in terms of co-inventorship chains, as in Breschi and Lissoni, 2004 The two matched inventors I and J have a common co-inventor. At least one of I’s o-inventors and one of J’s o-inventors are co-inventors Two matched inventors who turn out to be close in space are likely to be the same person. The two matched i e to s’ add esses e hi it the sa e it The two matched i e to s’ add esses e hi it the sa e p o i e NUT“ The two matched i e to s’ add esses e hi it the sa e p o i e NUT“ The two matched i e to s’ add esses e hi it the sa e states it applies o l to federal states). The two matched i e to s’ add esses e hi it the sa e st eet a d u e This family exploits the characteristics of the patent applicant. Both matched inventors have signed >1 patents filed by the same applicant. As with Applicant, when the applicant has less than 50 inventors affiliated. Both matched inventors have signed >1 patents filed by applicants belonging to the same group Two matched inventors who turn out to be technologically close are likely to be the same person. Technological distance is measured by considering IPC (International Patent Classification; http://www.wipo.int/classifications/ipc/en/, last visited, 30/4/2013) of their patents. IPC is a 12 digit hierarchical classification, with cut points at 1, 4, and 6 digits. The two matched i e to s’ pate ts sha e at least o e IPC ode at the digit le el The two matched i e to s’ pate ts sha e at least o e IPC ode at the digit le el The two matched i e to s’ pate ts sha e at least o e IPC ode at the digit le el Two matched inventors who turn out to be technologically close are likely to be the same person. Technological distance is measured by considering citation links between patents At least o e of i e to I’s pate ts ites >1 patents by J, or vice versa ASE stands for Approximated Structural Equivalence (Huang and Walsh, 2010). It occurs when two patents by the matched inventors stand in the same position within the network of citations Miscellaneous At least one a o g the at hed i e to s’ su a es e hi its a f e ue lo e tha the a e age alue i the i e to ’s ou t . Two matched inventors whose patents are close in time are more likely to be the sa e pe so . Dista e i ti e is easu ed the pate ts’ priority date (Martinez, . We fi st al ulate the i i u te po al dista e et ee the t o i e to s’ sets of patents, which turns out to be highly skewed. We set a threshold value of 3 years as a filtering criterion. Source: Pezzoni et al. 2013. 1.3 STEP2 – Professor-inventor matching We extract from the dataset of disambiguated inventors all inventors with at least one Italian address (as reported on their patent applications) and proceed to match their names to those of assistant, associate, and full professors active in various years in Italian universities, in scientific, medical, and engineering disciplines. 25 The p ofesso s’ data e e olle ted o e the ea s CE“PRI a d the KITE“, t o esea h e t es of Bocconi University, Milan. Several research projects supported the data collection effort, the most important ones being KEINS and APE-INV, from which as many databases of the same name were derived. The KEINS database methodology consisted of 2 steps (Lissoni et al., 2006): 1. Name matching between disambiguated inventors and academic personnel, the latter's names made available, in 2000 and 2005 by the Italian Ministry of Education. This step produced a number of professor-patent pairs obtained by attributing to each professor the patents signed by the matched inventors. 2. Filtering of professor-patent pairs, on the basis of automatic criteria, manual checking, and a telephone survey (with around 80% response rate). The APE-INV project i p o es upo oth the disa iguatio a d the p ofesso s’ at hi g algo ith s used by KEINS (as described in section 1.2 above) in order to extend the database over time , while at the same time allowing for variations both in surnames and names in order to increase the overall recall rate8. For Italy, we extracted the patents with at least one Italian inventor, and re-matched all of them to professors from the 2000 and 2005 lists and a new ministerial list updated to 2009. For the filtering stage, we exploited any available information either on the patents' assignees or already contained in the KEINS database, and then run an e-mail survey of remaining professor-patent pairs. The available information for all 39393 professors in cohorts 2000, 2005 and 2009 include their name, surname, date of birth, discipline, rank, and the date of nomination to the rank. As some professors are present in one or more cohorts, but not in others, we can classify them in four main groups:  listed in all cohorts ;  listed in 2005 and 2009 (that is, those who get a tenured position after 2000);  listed only in 2009 (tenured position acquired after 2005)  listed in 2000 and 2005 (presumably retired or transferred outside Italy or the academic before 2009). As shown in table A2, these classes make up 91.5% of all observations. Three residuals classes contain the individuals listed only in 2000, or in 2005, or (quite oddly) only in 2000 and 2009. 8 As explained by Lissoni et al., 2008, the KEINS project had to the objective of proving that pre-existing estimates of the extent of academic patenting in Europe were downward bias. To this end, it had to avoid any risk of overestimation. Thus, it made use of algorithms geared towards maximizing the precision rate, that is minimizing Type I errors. 26 Table A2: Number of professors, by data cohort Years of observation 2000; 2005; 2009 2005; 2009 2009 2000; 2005 2000 2005 2000; 2009 Total Professors 21422 7311 4105 3182 3180 174 19 39393 % 54.4% 18.6% 10.4% 8.1% 8.1% 0.4% 0.0% 100% The inventor data used for STEP1 come from 51881 patent applications filed at the EPO from 1996 to 2007, that is 59% of the patents by Italian inventors treated by STEP1 (see figure A1). The reason for leaving out all patents filed before 1996 is that they are likely to include a large number of academic patents whose inventors were already retired in 2000, and thus escape our identification effort, with the consequent risk of underestimation of the phenomenon of interest. Patents filed after 2007 are left out due to right truncation problem9. Figure A1: Patents of Italian inventors applied at EPO, by priority date 6000 5000 4000 3000 2000 1000 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 0 We produced professor-inventor matches as illustrated in figure A2: we name-matched professors active in 2001 to inventors of patents with priority dates 1996-2001, and professors active in 2005 and 2009 to inventors of patents with priority dates from 2001-2005 and 2006-2010, respectively. This matching st ateg e ploits the th ee oho ts of p ofesso s’ data i o de to o tai lo gitudi al dataset of a ade i patents. When having just one cohort of professors (year Y), we end up underestimating the number of 9 The October 2011 version of PatStat does not include patent applications filed in the same year, very few patents filed in 2010, and much less than 100% of patents filed in 2008 and 2009; it also underestimate the number of patents filed in 2007, but not as much as to make data for that year useless. 27 academic patents for any year Y-t , especially for large values of t. This is because many professors active in year Y-t may have retired or changed job before Y, which makes it impossible to identify their patents. With three cohorts at hand (Y1, Y2, Y3; from less to more recent), we can match professors from cohort Yi (i=1,2,3) to inventors of patents filed in years comprised between Yi and Yi -t, with t reasonably low (5 years, in our case). Due to the short interval, we can safely presume that most professors in cohort Yi were active throughout the period, with very few entering the university after Yi or leaving it before Yi -t. Figure A2 - Identification of academic inventors and patents: data collection methodology As for the name matching procedure used at this stage, this is similar to the one already employed in STEP1 for the disambiguation of inventors. After matching the professor and the inventor, for the validation process, we turned our attention to the resulting professor-patent pairs, for two reasons. First, the same inventor may have done several patents, and even when we discover that he/she is now an academic, we cannot exclude that some patents were produced before the start of his/her academic career, or after its end. Second, n>1 i e to s a e at hed to the sa e p ofesso , all of the fo e ’s a es ei g suffi ie tl si ila to the latte ’s; this is e eali g eithe of a t pe II e o i the i e to disa iguatio phase (the n inventors should have been identified as the same person, and they were not; in jargon, they are false negatives); or of a type I error in the professor-inventor matching phase (the n inventors are not same person, and they should not have been all matched to the same professor; in jargon, the professorinventor match is false positive). 1.4 STEP3 – Validation by information- he ki g: lo er ou d esti ate of a ade i pate ti g STEP2 left us with 10118 professor-patent pairs (for a total of 3775 professors and 6484 patents) to be validated either by checking the information contained in the patent, or by surveying the inventors. 28 Validation through information- he ki g oi ided o fi ostl fo ussed o the appli a ts’ ide tit : he e e the latte ith the p ofesso ’s u i e sit , the p ofesso -patent pair was retained as valid and the patent ed as a ade i . Othe ite ia used fo the appli a t’s add ess o tai ed alidatio o sisted i he ki g hethe the i e to ’s o o ds su h as dipa ti e to , fa oltà , o istituto , hi h a e indicative of an academic affiliation. On the contrary, some professor-patent pairs were considered invalid filte ed out due to i o siste ies et ee the p ofesso ’s age a d the p io it date of the pate t, o the p ofesso ’s dis ipli e a d the pate t’s te h ological classification. When necessary, and if available, we collected additional information relative to the inventor and the professor on the web. Finally, we used information from the surveys already conducted by Balconi et al. (2004) in 2002 and the KEINS project in 2006 (Lissoni et al., 2006). Overall, this validation effort allowed us to filter out 763 professor-patent pairs and to confirm as valid 2501 pairs (see table A3). For the remaining 6854 pairs we set up an e-mail survey, which consisted in presenting each matched professor with the list of matched patents and the request to confirm/deny them as his/hers. We managed to retrieve a valid email address for 5424 professor-patent pairs, corresponding to 1756 professors and 4215 patents10. On the contrary, we were not able to find any contact information for 1430 pairs (750 professors; 1190 patents). Most of these u ea ha le ases o e p ofesso s p ese t o l i cohorts 2000 and (possibly) 2005, for which the institutional email address at the university of affiliation did not exist anymore, the professor having retired or left the academy. The response rate for reachable cases was 37.5%, which means that we obtained information for 2036 professor-patent pairs: 993 pairs were positive (the professor confirmed the patent to be his/hers), while the remaining 1043 were negative (the professor denied to be the inventor of the matched patent). However, for a large number of non-responses we could exploit information coming from respondents, as these were asked not only to identify their won patents but also to identify who, among their co-inventors, also was an academic. In this was we managed to validate further 1249 professor-patent pairs, and filter out 278, leaving us with only 1861 professor-patent pairs without response. Summing up, after the automatic and manual check and the email survey, we were able to either validate or filter out 6827 professor-patent pairs (2378 professors and 5105 patents) , that is about 67% of total professor-patent pairs obtained by STEP2 (62% of professors; 79% of patents)11. 10 The total number of professor-patent pairs for which we obtained an e-mail address is obtained by summing rows (b) and (c) in table A3. The corresponding figures for professors and patents differ from such sum, due to the presence of several professors with more than one patent and several patents co-invented by more than one professor. 11 The figure for professor-patent pairs is obtained by summing rows (a) and (b) in table A3. As explained in the previous footnote, the corresponding figures for professors and patents differ from such sum due to the presence of multi-patent professors and coinventorship. 29 We use these figures as a "lower bound" estimate of the number of academic patents in Italy for the period considered, based on the assumption that all unreachable and non-response cases are equivalent to negative responses. Under these assumptions, we count as academic 2679 patents, which correspond to 2199 professors with at least one patent and 4743 professor-patent pairs. However, this estimate is subject to time-related bias. In fact, unreachable and non-response cases are all related to patents owned by business firms, individuals and other non-university entities, so that "lower bound" data would return biased estimates of the ownership distribution. In addition, to the extent that unreachable cases include a high proportion of patents from the 1990s, we could also observe a bias with respect to the time distribution of academic patents namely, a negative bias for early years and positive bias of any estimated time trend. Table A3: Professor-patent pairs, results of filtering stage and subsequent estimates (a) (a1) (a2) (b) (b1) (b2) (b3) (b4) (c) (c1) (c2) (d) (d1) (d2) (i) (ii) (iii) (iv) (v) Automatic/Manual check of which: - confirmed - rejected Professor-patent pairs Nr % 3264 32.3% 2501 763 35.2% Professors nr 1540 Patents Nr 2145 1356 217 1479 693 1236 2015 412 262 472 87 899 968 1012 236 814 1669 298 516 472 1197 750 1190 247 523 420 821 E-mail survey (responses) of which: - confirmed - rejected - confirmed, via extra info (i) -rejected, via extra info (i) 3563 E-mail survey (no responses) of which: - confirmed (estimate  table 6) - rejected (estimate  table 6) 1861 E-mail survey (unreachable) of which: - confirmed (estimate  table 6) - rejected (estimate  table 6) 1430 Total (ii) of which: - confirmed (lower bound estimate) (iii) - confirmed (intermediate estimate) (ii) - confirmed (upper bound estimate) (v) 10118 100% 3775 6484 4743 5204 5733 46.9% 51.0% 56.7% 2199 2399 2602 2679 3093 3535 993 1043 1249 278 18.4 % 529 1332 14.1 % 461 969 For several non responses, information was available from responses by other professors (who provided information on co-inventors) For professor-patent pairs: Total= (a)+(b)+(c)+(d); for professors and patents, totals may differ from (a)+(b)+(c)+(d), due to the possibility of having more than one patent per professor and vice versa For professor-patent pairs: Total= (a1)+(b1)+(b3); for professors and patents, totals may differ (see note (ii) above) For professor-patent pairs: Total= (a1)+(b1)+(b3)+(c1); for professors and patents, totals may differ (see note (ii) above) For professor-patent pairs: Total= (a1)+(b1)+(b3)+(c1)+(d1); for professors and patents, totals may differ (see note (ii) above) 30 1.5 STEP 3bis: Estimation of positive matches among unreachables and non-responses As just discussed, surveying professors who are present only in early cohorts (say, 2001 or 2005) may be difficult: having they retired or left the academy, there may be no way to reach them. It is at this point that data from former research projects turns out again to be useful, as they include information from surveys run at a time when most professors from these cohorts could still be reached (in particular, the survey conducted in 2002 by Balconi et al., 2004; and the 2006 KEINS survey, by the KEINS project; see figure A2). Based on such information, we run two probit regression exercises whose estimated coefficients allow us to predict whether the professor-patent pairs corresponding to unreachable or non-response cases can be validated as academic or not. Observations in both regressions consist of professor-patent pairs comprised both in our survey (falling, respectively, among the non-response and unreachable cases) and in either the 2002 and/or the 2006 surveys (in which case they were among the respondent cases). In both regressions, errors are treated as clustered on the professors. The dependent variable is a binary one, which takes value one in case the patent was validated as an academic one, and zero if it was not. For example, consider the case of professor J, who we contacted both in 2006 and in our most recent survey in order to validate as his three patents j1, j2, and j3, the former two filed before 2005, the latter after then. Assume further that professor J could be reached and returned his/her response in 2006, but not later. We then include both patents j1 and j2 in our regression exercise, with the dependent variable taking value one or zero depending on the answer he/she provided. The choice of running separate regressions for unreachable and non-response cases is due to differences between the two groups. The unreachable group is by and large composed of professors from the early data cohorts, now retired, who were active at a time when the legal, cultural, and economic circumstances differed from those in which they younger colleagues (more numerous among non-respondents) act nowadays. This left us with 160 observations for the regression exercise concerning unreachable cases, and 850 for the non-responses (see table A4). Table A4: unreachable and non-response statistics Patents Pairs validated or rejected manually or by past research (B) (B)/(A) 1124 2820 160 (of which 121 confirmed) 11% 750 1190 850 (of which 655 confirmed) 25% Pairs to be validated (A) Professors unreachable 1430 Non-response 3388 In both regressions, the explanatory variables include:  The professor's Age in the patent's priority year and the professor's Year of birth; the former is meant to capture a life cycle effect (senior professors may be more likely to patent than junior 31 ones, who may have either fewer contacts with industry or a higher opportunity cost in terms of time subtracted to publication activities), while the latter captures a cohort effect (professors belonging to different generations may have different attitudes or expertise faced to patenting  opportunities). The p ofesso ’s dis ipli e. Due to the lo u e of o se atio s, e fou d it i possi le to control for disciplines, so we limited ourselves to introduce one dummy (ICAR discipline) which points to a disciplinary group that include both civil engineers and urbanists, and we expect to have a lower propensity to patent (reminder: the professor-inventor matching exercise already excluded  all social and human scientists). A dummy variable (Different name) taking value one in case of non-perfect homonymy between the professor and the matched inventor. The effect of this variable is expected to be negative, as the matches between professors and inventors with different names are more likely to turn out to  be false positives. A dummy variable (Different region taki g alue o e he the p ofesso ’s u i e sit appea s to e lo ated i a diffe e t egio tha the at hed i e to ’s it , as epo ted i the i e to ’s add ess. The effect of this variable is expected to be negative, as the matches between professors and inventors located far away in the geographical space are more likely to turn out to be false  positives. The number of non-patent literature citations (NPL citations) listed on the patent. The effect of this variable is expected to be positive, NPL citations being an indicator of the existence of academic  inputs to the invention, which makes less likely that the inventor is him/herself an academic. Two dummy variables ased upo the pate t’s O“T te h ologi al lassifi atio 12 . We distinguish between technological classes as follows:  Science-based: Electrical engineering & Electronics; Scientific & Measurement Instruments; Chemicals & Materials; Pharmaceuticals & Biotechnology  Non science-based: Industrial processes; Mechanical engineering, Machines & Transport; Consumer goods & Civil engineering For each group we create a dummy variable. The two dummies are not exclusive (none of the two ought to be omitted from the regression), due to the possibility of multiple classifications for the same patent. As suggested by the existing literature, academic patents are more likely to be found in science-based technologies, so we expect a positive sign for the coefficient of the relative dummy, and a negative one for the other 12 according to OST classification as described in Schmoch 2008, Concept of a technology classification for country comparisons, Final report to the World Intellectual Property Organization (WIPO), Fraunhofer Institute for Systems and Innovation Research, Karlsruhe 32  The p ofesso ’s te u e status. The du e uals if the pate t esults from the research activity of the professor before getting the tenure (Not active professor) Table A5: Regressions results for non-response and unreachable professor-patent pairs Non-response unreachable Age Year of birth ICAR discipline Different name Different region NPL citations Non science-based technology Science-based & Non science-based technology Not active professor Constant Observations (academic=1 in parentheses) -0.071** (0.029) -0.088*** (0.027) -1.72** (0.70) -0.59* (0.34) -1.14*** (0.24) 0.044* (0.024) -0.99*** (0.25) -0.68*** (0.24) -1.48*** (0.27) 177*** (54.4) 850 (655) -0.13*** (0.038) -0.21*** (0.048) -1.48* (0.77) GOODNESS-OF-FIT (PREDICTED PROBABLITIES Predicted academic =1 Predicted academic =0 % correctly classified % false positives % false negatives Probability threshold set at: 738 112 86.70% 2.29% 50.26% 0.85 115 34 93.13% 4.96% 12.82% 0.5 REGRESSION RESULTS -1.41*** (0.42) 0.21** (0.10) -1.70** (0.82) 412*** (95.4) 160 (121) Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1 In both regressions, we retain only the explanatory variables whose estimated coefficients turn out to be significant (that is, we insert the regressors according to a backward stepwise procedure). The upper box in table A5 reports the estimated coefficients, while the lower box report goodness-of-fit measures based upon counting of correct predictions. Concerning the latter, we obtained them by calculating in-sample predicted probabilities for patents to be academic or not on the basis of estimated coefficient, to be compared with ad hoc threshold values. The latter were chosen with the objective of minimising the number of false positives (type I errors), without increasing much the false negatives (type II errors). This resulted in setting a very high threshold value for the non-response regression (0.85) and a standard one for unreachable (0.5), which return 87% and 93% of correctly classified observations. 1.6 Intermediate and upper bound estimates of academic patenting We apply the estimated coefficients and the selected threshold values from the probit regression exercises to the overall samples of unreachable and non-response cases in order to predict how many patents from each group can be validated as academic. For unreachable cases, we obtain positive predictions for 461 out of 1430 professor-patent pairs (247 professors and 420 academic patents; as reported in table A3). For 33 non-response, we obtain positive predictions for 529 professor-patent pairs out of 1861 (298 professors and 472 academic patents; as reported in table A3). Table A6: Academic patent database: structure and contents Statistics based on 1996-2007 sample Italian inventors (A) STEP1: Inventor disambiguation Italian disambiguated inventors (B) [(B)-(A)]/(A) Italian patents (C) (C)/(B) Italian professors (D) STEP2: Professor-inventor matching professor-inventor matches (E) % academic inventors over tot. professors (E)/(D) % academic inventors over tot. inventors (E)/(B) STEP2 (cont.): Validation Academic inventors: lower bound estimate (AI1) % of prof-inv matches confirmed by filtering (AI1)/(E) % academic inventors over tot. academics (AI1)/(D) % academic inventors over tot. inventors (AI1)/(B) STEP3: Prediction of non-validated matches Academic inventors: intermediate estimate (AI2) % of prof-inv matches confirmed by filtering (AI2)/(E) % academic inventors over tot. academics (AI2)/(D) % academic inventors over tot. inventors (AI2)/(B) Academic inventors: upper bound estimate (AI3) % of prof-inv matches confirmed by filtering (AI3)/(E) % academic inventors over tot. academics (AI3)/(D) % academic inventors over tot. inventors (AI3)/(B) Individuals (inventor/professor) 51391 42784 -17% 51054 1.19 [patents/inventor] 39393 3775 9.6% 8.8% 2199 58% 5.6% 5.1% 2399 64% 6.1% 5.6% 2602 69% 6.6% 6.1% Predicted academic patents out of un ea ha le ases, a e added to a ade i pate ts i the lo e a ade i pate t dataset AI i ta le A i o de to p odu e a i te ou d ediate dataset AI i table A6): this includes 5204 validated patent-professor pairs (2199 inventors). Predicted academic patents out of non-responses a e fu the added i o de to p odu e the uppe ou d dataset, for a total of 5733 confirmed professor-patent pairs (2602 inventors; AI3 in table A6). 34 2. Econometric analysis : descriptive statistics Table A7 the complete descriptive statistics for the STEP1 regression's dependent variable (with values for the dependent variables for both lower bound, intermediate, and upper bound estimates) and regressors. Table A7 - STEP1 regression: descriptive statistics Obs Dependent variable (Academic patent): 51054 upper_bound 51054 Intermediate 51054 lower_bound Regressors: 51054 Year 1996 51054 Year 1997 51054 Year 1998 51054 Year 1999 51054 Year 2000 51054 Year 2002 51054 Year 2003 51054 Year 2004 51054 Year 2005 51054 Year 2006 51054 Year 2007 51054 1.Electrical eng.; Electronics 51054 2.Instruments 51054 3.Chemicals; Materials 51054 4.Pharmaceuticals; Biotech. 51054 5.Industrial processes 51054 6.Mechanical eng.; Machines; Transport 51054 7.Consumer goods; Civil eng. 51054 N_INV 51054 SHARE_NPL 51054 TOT_CIT 50931 TTO_REGION 50875 STATUTE_REGION 50931 NR_UNIVERSITIES_REGION 50930 BERD/GDP 50927 RD_SHARE_PAUNI 32395 FFO_RATIO_REGION 32395 SCIENCE_RATIO_REGION Regional dummies (obs=51054): Mean Std. Dev. Abruzzo 0.019 0.137 Basilicata 0.003 0.054 Calabria 0.004 0.062 Campania 0.020 0.141 Emilia-Romagna 0.175 0.380 Friuli VG 0.036 0.186 Lazio 0.059 0.236 Liguria 0.028 0.164 Lombardia 0.354 0.478 Marche 0.024 0.153 Molise 0.001 0.029 Mean Std. Dev. Min Max 0.067 0.059 0.051 0.251 0.235 0.221 0 0 0 1 1 1 0.059 0.065 0.069 0.076 0.083 0.087 0.091 0.094 0.100 0.102 0.090 0.172 0.150 0.136 0.099 0.253 0.243 0.186 2.097 0.368 4.010 0.517 0.395 7.077 0.665 0.424 0.44 0.12 0.236 0.247 0.254 0.266 0.276 0.282 0.287 0.292 0.300 0.303 0.287 0.378 0.357 0.342 0.299 0.435 0.429 0.389 1.587 0.418 6.859 0.332 0.323 4.024 0.339 0.181 0.1 0.04 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0.033 0.12 0.02 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 49 1 217 1 1 12 1.48 1 0.87 0.28 Piemonte Puglia Sardegna Sicily Toscana Trentino AA Umbria Val d'Aosta Veneto Unknown region Mean 0.140 0.013 0.005 0.019 0.065 0.015 0.011 0.002 0.134 0.219 Std. Dev. 0.347 0.113 0.068 0.135 0.246 0.123 0.105 0.042 0.340 0.414 Table A8 reports the complete descriptive statistics for the STEP2 regression, only for upper bound estimate data (lower bound and intermediate data are available on request). Notice that in the STEP1 35 regression, the number of observations is the same from whatever estimate we draw values for the dependent variable. On the contrary, in STEP2 the type of estimate affects the number of observations, as it changes the counting of academic patents. Table A8 - STEP2 regression: descriptive statistics (for upper bound estimate of academic patenting) Obs Dependent variables (Patent ownership): University 3443 Individual 3443 Company 3443 Regressors: Year 1996 3443 Year 1997 3443 Year 1998 3443 Year 1999 3443 Year 2000 3443 Year 2002 3443 Year 2003 3443 Year 2004 3443 Year 2005 3443 Year 2006 3443 Year 2007 3443 1.Electrical eng.; Electronics 3443 2.Instruments 3443 3.Chemicals; Materials 3443 4.Pharmaceuticals; Biotech. 3443 5.Industrial processes 3443 6.Mechanical eng.; Machines; Transport 3443 7.Consumer goods; Civil eng. 3443 N_INV 3443 SHARE_NPL 3443 TOT_CIT 3443 BERD/GDP 3438 RD_SHARE_PAUNI 3438 STATUTE 3443 TTO 3361 FFO_RATIO 1954 SCIENCE_RATIO 1954 University dummies (obs=3343): Mean Std. Dev. Bari-Politecnico 0.017 0.131 Bologna 0.092 0.290 Catania 0.047 0.212 Ferrara 0.040 0.195 Firenze 0.056 0.230 Genova 0.032 0.176 Milano-Bicocca 0.026 0.160 Milano 0.105 0.307 Milano-Politecnico 0.085 0.279 Modena 0.039 0.193 Napoli "Federico II" 0.049 0.215 Padova 0.055 0.229 Mean Std. Dev. Min Max 0.177 0.087 0.731 0.382 0.282 0.443 0 0 0 1 1 1 0.065 0.069 0.066 0.082 0.080 0.090 0.083 0.098 0.099 0.108 0.076 0.207 0.256 0.275 0.380 0.110 0.069 0.036 3.680 0.600 7.255 0.602 0.484 0.643 0.476 0.467 0.132 0.246 0.253 0.248 0.275 0.272 0.286 0.276 0.297 0.299 0.311 0.266 0.405 0.437 0.446 0.486 0.313 0.253 0.186 2.331 0.396 10.570 0.309 0.203 0.479 0.499 0.123 0.063 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0.145 0 0 0.009 0.0002 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 49 1 200 1.48 1 1 1 0.9 0.43 Palermo Parma Pavia Perugia Pisa Roma "La Sapienza" Roma "Tor Vergata" Siena Torino Torino-Poilitecnico Udine Mean 0.025 0.035 0.048 0.028 0.054 0.078 0.037 0.029 0.048 0.034 0.017 Std. Dev. 0.155 0.184 0.214 0.165 0.227 0.267 0.189 0.169 0.213 0.181 0.128 36 3. Econometric analysis : robustness checks Table A9 replicates STEP1 regressions with values for the dependent variable coming respectively from lower bound and intermediate estimates of academic patenting. Results for patent characteristics are the same as those obtained with upper bound data. The same holds for BERD/GDP and RD_SHARE_PAUNI. Regressions based on lower bound estimates exhibit positive and significant signs for STATUTE_REGION (and negative, and in one case significant, for TTO_REGION). Also the coefficient for NR_UNIVERSITIES_REGION is positive and significant. The strength of results for STATUTE_REGION may derive from the bias of lower bound estimates, whose share of university-owned academic patents is artificially high. As we know (from previous sections) that such share grows when universities introduce IP regulations, we may conclude that the estimated coefficient for STATUTE_REGION in columns (1) and (2) is positively biased. The same applies to columns (3) and (4), albeit with lower significance, as intermediate estimates correct only in part for errors in lower bound estimate data. Table A10 replicates the Heckman probit regressions, with university ownership of academic patents as the dependent variable. In this case, the results are almost identical to those obtained with upper bound estimate data. 37 Table A9 – STEP1 Probit regression (dep. variable: probability of a patent to be academic; lower bound and intermediate estimate data Year 1996 Year 1997 Year 1998 Year 1999 Year 2000 Year 2002 Year 2003 Year 2004 Year 2005 Year 2006 Year 2007 Electrical Eng.; Electronics Scientific instruments; Measurement Chemicals; Materials Pharmaceuticals; Biotechnology Industrial Processes Mechanical Eng.; Machines; Transport Consumer goods; Civil Eng. N_INV (nr of inventors) SHARE_NPL (% of citations to non-patent literature) TOT_CIT (tot nr of backward citations) TTO_REGION (regional diffusion TTOs) STATUTE_REGION (regional diffusion IP statutes) NR_UNIVERSITIES_REGION BERD/GDP (regional BERD/GDP) RD_SHARE_PAUNI (% of R&D by public administration & universities, in region) FFO_RATIO_REGION (block grant as % of univ.'s revenues, regional avg) SCIENCE_RATIO_REGION (public research funds % of univ.'s revenues, regional avg) Constant Regional dummies Observations Pseudo R2 Lower bound (1) (2) 0.015 (0.063) -0.025 (0.061) -0.049 (0.058) 0.011 (0.055) -0.018 (0.053) -0.100* -0.059 (0.055) (0.076) -0.14** -0.12 (0.058) (0.080) -0.080 -0.059 (0.060) (0.085) -0.029 -0.041 (0.072) (0.095) 0.11 0.076 (0.080) (0.10) -0.037 -0.068 (0.088) (0.12) 0.046 -0.0014 (0.036) (0.045) 0.33*** 0.38*** (0.031) (0.039) 0.15*** 0.16*** (0.030) (0.038) 0.60*** 0.55*** (0.033) (0.042) -0.26*** -0.21*** (0.034) (0.043) -0.41*** -0.43*** (0.041) (0.052) -0.51*** -0.51*** (0.053) (0.067) 0.16*** 0.16*** (0.0061) (0.0074) 0.41*** 0.46*** (0.029) (0.035) 0.0088*** 0.010*** (0.0012) (0.0021) -0.23** -0.083 (0.092) (0.11) 0.17** 0.20** (0.073) (0.10) 0.038*** 0.026* (0.013) (0.016) 0.28 0.11 (0.18) (0.25) 0.88*** 0.64 (0.31) (0.43) 0.35 (0.30) 0.059 (0.62) -3.33*** -3.34*** (0.26) (0.36) Y Y 50,875 32,317 0.26 0.26 Intermediate bound (3) (4) 0.13** (0.058) 0.11** (0.055) 0.015 (0.054) 0.12** (0.051) 0.021 (0.050) -0.079 -0.017 (0.052) (0.073) -0.14*** -0.097 (0.055) (0.077) -0.093 -0.053 (0.057) (0.082) -0.079 -0.055 (0.068) (0.092) 0.019 0.026 (0.076) (0.10) -0.096 -0.075 (0.083) (0.12) 0.054 -0.016 (0.033) (0.044) 0.31*** 0.36*** (0.029) (0.038) 0.13*** 0.14*** (0.029) (0.037) 0.57*** 0.53*** (0.031) (0.041) -0.24*** -0.20*** (0.032) (0.041) -0.34*** -0.37*** (0.037) (0.048) -0.39*** -0.50*** (0.045) (0.062) 0.15*** 0.16*** (0.0059) (0.0073) 0.40*** 0.45*** (0.027) (0.033) 0.0090*** 0.012*** (0.0012) (0.0021) -0.12 -0.028 (0.087) (0.11) 0.11* 0.21** (0.068) (0.097) 0.025** 0.026* (0.012) (0.016) 0.39** 0.13 (0.17) (0.24) 1.06*** 0.69 (0.29) (0.42) 0.30 (0.29) 0.51 (0.59) -3.35*** -3.36*** (0.25) (0.35) Y Y 50,875 32,317 0.23 0.25 Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1 38 Table A10 – Heckman Probit regressions (STEP1, unreported; STEP2: probability of an academic patent to be owned by university) – lower bound and intermediate estimate data (1) Year 1996 Year 1997 Year 1998 Year 1999 Year 2000 Year 2002 Year 2003 Year 2004 Year 2005 Year 2006 Year 2007 Electrical Eng.; Electronics Scientific instruments; Measurement Chemicals; Materials Pharmaceuticals; Biotechnology Industrial Processes Mechanical Eng.; Machines; Transport Consumer goods; Civil Eng. N_INV (nr of inventors) SHARE_NPL (% of citations to non-patent literature) TOT_CIT (tot nr of backward citations) BERD/GDP (regional BERD/GDP) RD_SHARE_PAUNI (% of R&D by public administration & universities, in region) FIRST_STATUTE (IP regulation in place) TTO (TTO in place) -0.19 (0.19) -0.47** (0.19) -0.41** (0.18) -0.24 (0.16) -0.00072 (0.14) 0.27* (0.14) -0.00072 (0.15) 0.14 (0.14) -0.023 (0.15) 0.29** (0.15) 0.37** (0.16) -0.19* (0.10) 0.28*** (0.084) -0.098 (0.078) 0.056 (0.10) 0.16 (0.10) -0.13 (0.15) 0.056 (0.20) -0.0100 (0.021) 0.67*** (0.098) -0.00093 (0.0037) -0.17 (0.29) 0.26 (0.44) 0.30*** (0.089) -0.065 (0.091) FFO_RATIO (block grant as % of revenues) SCIENCE_RATIO (research as % revenues) Constant Observations Rho obs. censored obs. Pseudo R2 -1.58*** (0.59) 50,809 0.10 50809 48259 . (2) (3) (4) -0.21 (0.18) -0.49*** (0.18) -0.41** (0.18) -0.30** (0.15) -0.044 (0.14) 0.24 0.23* 0.18 (0.15) (0.13) (0.15) 0.062 -0.00080 0.019 (0.16) (0.14) (0.16) 0.28* 0.13 0.22 (0.16) (0.14) (0.16) 0.18 -0.0062 0.12 (0.17) (0.15) (0.17) 0.47*** 0.31** 0.41** (0.17) (0.14) (0.17) 0.51*** 0.35** 0.42** (0.20) (0.15) (0.20) -0.0092 -0.17* 0.025 (0.11) (0.097) (0.11) 0.43*** 0.29*** 0.43*** (0.087) (0.082) (0.089) -0.065 -0.080 -0.065 (0.092) (0.075) (0.092) 0.34*** 0.094 0.36*** (0.10) (0.098) (0.10) 0.13 0.18* 0.17 (0.12) (0.099) (0.12) -0.20 -0.16 -0.20 (0.15) (0.14) (0.15) 0.039 -0.054 0.040 (0.22) (0.17) (0.22) 0.056** -0.0035 0.055* (0.028) (0.020) (0.029) 0.89*** 0.62*** 0.84*** (0.097) (0.096) (0.097) 0.0080 -0.0012 0.0053 (0.0049) (0.0034) (0.0042) -0.28 -0.11 -0.30 (0.32) (0.27) (0.32) 0.072 0.29 -0.045 (0.49) (0.42) (0.50) 0.20* 0.35*** 0.23** (0.11) (0.086) (0.11) -0.14 -0.060 -0.14 (0.100) (0.088) (0.10) 0.40 0.14 (0.41) (0.41) -0.27 -0.49 (0.64) (0.65) -2.95*** -1.58*** -2.56*** (0.65) (0.57) (0.72) 32,120 50,805 32,094 0.62** 0.034 0.53** 32120 50805 32094 30637 47876 30495 . . . Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1 (§) only for universities with >50 patents (all other universities as reference case) (£) computed for STEP2 as stand-alone regression 39 4. Additional figures and tables Figure A3. Weight of block grants (FFO) and public funds for scientific project funds (SCIENCE) over universities' total revenues, 1995-2009 80% 0,16 70% 0,14 60% 0,12 50% 0,1 40% 0,08 30% 0,06 20% 0,04 FFO/Tot revenues (left axis) 10% Public funds for research/Tot revenues (right axis) 0% 0,02 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Sources: own elaborations on AQUAMETH and CNSVU data 40 Figure A4. Diffusion of technology transfer offices (TTOs) and IP statutes, all Italy (1995-2009) 90% 80% 70% 60% 50% IPR statute 40% TTO 30% 20% 10% 0% 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Sources: own elaborations on NETVAL survey and CNSVU 41 Figure A5 – Ownership of academic patents 1996-2007; lower bound estimates 35,0 80,0 30,0 70,0 y = -1,6487x + 72,618 R² = 0,6517 60,0 25,0 50,0 20,0 40,0 15,0 y = 2,359x + 3,6369 R² = 0,9211 30,0 10,0 20,0 5,0 10,0 0,0 0,0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Figure A6 Share of academic patents over all patents by domestic inventors, 1996-2007 – by technical field and subperiod; upper bound estimates (% values) 1.Electrical eng.; Electronics 2.Instruments 3.Chemicals; Materials 4.Pharmaceuticals; Biotechnology 5.Industrial processes 1996-2000 6.Mechanical eng.; Machines; Transport 2001-2005 2006-2007 7.Consumer goods; Civil eng. 0,0 5,0 10,0 15,0 20,0 25,0 30,0 42 References Balconi M., Breschi S., & Lissoni F. (2004). Networks of inventors and the role of academia: an exploration of Italian patent data, Research Policy, 33(1):127-145. Hua g H., Walsh J.P. . A e Na e-Mat hi g App oa h fo sea hi g Pate t I e to s , pape presented at the APE-INV 2nd Name-Game Workshop, Madrid 9-10 December (http://www.esf-apeinv.eu/index.php?page=10#madrid) Lissoni F. Pezzoni M. Poti B. and Romagnosi S (2012), University autonomy, IP legislation and academic patenting: Italy, 1996-2007, Cahier du GREThA 2012-26 (http://cahiersdugretha.ubordeaux4.fr/2012/2012-26.pdf) Lissoni F., Sanditov B., & Tarasconi G. (2006) The Keins database on academic inventors: methodology and contents. WP Cespri, 181 (http://www.francescolissoni.com). Martinez (2011) Patent families: When do different definitions really matter?, Scientometrics 86(1):39-63 Pezzoni M. Lissoni F. and Tarasconi G. (2013), How To Kill Inventors: Testing The Massacrator Algorithm For Inventor Disambiguation, Cahier du GREThA 2012-29, Université Montesquieu – Bordeaux IV (http://cahiersdugretha.u-bordeaux4.fr/2012/2012-29.pdf) Raffo J. and Lhuillery S. (2009). How to play the name game: Patent retrieval comparing different heuristics Research Policy, 38(10):1617-1627. 43