University Effects on Regional Innovation∗ Robin Cowan† BETA, UNIVERSITY OF STRASBOURG, 61 avenue de la Forˆ et Noire, 67085 Strasbourg, France UNU-MERIT, MAASTRICHT UNIVERSITY Keizer Karelplein 19, 6211 TC Maastricht, The Netherlands
Natalia Zinovyeva INSTITUTE OF PUBLIC GOODS AND POLICIES SPANISH NATIONAL RESEARCH COUNCIL (IPP-CSIC) Calle Albasanz 26-28, Madrid, 28037, Spain
April 2, 2012
Abstract This paper analyzes empirically whether expansion of a university system affects local industry innovation. We examine how the opening of new university schools in Italy during 1985-2000 affected regional innovation. We find that creation of new schools increased regional innovation activity already within five years. On average, an opening of a new school has led to a seven percent change in the number of patents filed by regional firms. The evidence suggests that the effect is mainly generated by high quality scientific research brought to the region with new schools.
Keywords: University research, regional innovation, publications, industry-university interaction
∗
We would like to acknowledge the inputs of members of the KEINS project, and particularly Francesco Lissoni and Bulat Sanditov, for their gracious openness and valuable help with the data. We are also grateful to Francesco Quatraro who provided us with the historic data on Italian regional R&D collected from various issues of ISTAT. We also acknowledge the helpful comments of Bronwyn Hall, Jacques Mairesse, Joel Baum and all the participants of XXXII Symposium of Economic Analysis in Granada and DIME Conference “Knowledge Based Entrepreneurship: Innovation, Networks, and System” in Milan. This research was supported by the DYREC Chaire d’ Excellence of Robin Cowan, funded by the French ANR, and grants from ESF COST and APE-INV projects. † Corresponding author at: UNU-MERIT, Keizer Karelplein 19, 6211 TC Maastricht, The Netherlands. Tel.: +31(0)433884408. Email addresses:
[email protected] (R. Cowan),
[email protected] (N. Zinovyeva).
1
Introduction
Between 1960 and 2000, there was a large expansion in universities in the industrialized countries. Early expansion was to deal with the baby boom coming of university age; later expansion was driven by the desire to increase the proportion of the population receiving tertiary education.1 Naturally, a rise in student numbers tended to be accompanied by a rise in the size of the professoriate and an increase in the sizes and numbers of universities. The clearest effect was just that: an increase in the general education level of the labor force. University expansion coincided with spectacular rise of innovation activity in industrialized world. In1963 US Patent Office granted around 45 thousand patents; by the end of nineties the yearly number of granted patents approached 160 thousand (Hall et al., 2001). How to maintain this competitiveness and get more innovation out of a knowledge system has become a hotly debated issue. Following the line taken in the literature on innovation systems, it is often suggested that stimulating academic research and close interactions between academia, industry and government are necessary to promote knowledge flows and innovation. These policy suggestions are often based on the idea that universities have within them some of the keys to increasing innovative activity.2 The fact that the increase of innovation activity during past decades coincides with the increase in the size of the university sector might suggest that the innovation performance of an economy is determined in part by the supply of universities in the innovation system. This hypothesis motivates our analysis. There have been many studies on the relationship between universities and industrial innovation, particularly at the regional level (see section 2.1). The vast majority of these studies analyze cross-sectional data, focussing on either the presence or size of universities and the relationship with local innovation activity. Generally, they document a strong relationship between university research activity and industrial innovation. But there are well-known difficulties in drawing conclusions from cross-sectional analysis about phenomena that take place over time, so while the results are suggestive, one must be cautious in drawing the “obvious” policy conclusions from them, particularly in terms of whether opening new universities is a good idea. Additionally, endogeneity problems are rife in this kind of work: some of the effects of university-industry interaction are driven by supply of knowledge, some by demand for it; external factors may drive both 1
According to the Global Education Digest 2009 by UNESCO Institute of Statistics, the share of students in North America and Western Europe that enroll in tertiary education during five years after the end of secondary education increased by 41 percentage points from 30% in 1970 to 71% in 2007. 2 An OECD 2007 report “Higher Education and Regions: Globally Competitive, Locally Engaged” estimates that only 10% of UK firms currently interact with universities with most university-industry links focusing on big business and a few hi-tech fields. Report concludes that “the potential of higher education institutions to contribute to the economic, social and cultural development of their regions is far from being fully realized”.
1
public and private research output simultaneously; the location of universities and firms is often endogenously determined (Mairesse and Mohnen, 2010). All imply that identifying empirically the effect of universities would ideally rely on exogenous shocks to university supply. Such shocks are rare in real life and most studies rely on strong assumptions to claim the existence of the supply-side effects. There was, however, a period of several years in the 1980s and 1990s in which Italy opened many new university schools in different regions of the country.3 University expansion was centralized and, as was acknowledged later by policy makers, the distribution of new schools across regions was largely independent of the properties of the regional economy. In fact no significant correlation can be observed between the number of new schools in a region and regional characteristics including population, share of graduates in the labor force, private and public investment in research and development, and value added produced by different economic sectors. We use this episode to ask directly whether expanding university activity by opening new universities has an identifiable effect on local industrial innovation. This is the first issue we address in this paper. The second issue has to do with the nature of the relationship between universities and industrial innovation. There have been several studies on the “channels” of interaction between university and industry (see section 2.2). By and large, these studies are based on firm surveys, asking firms about their external sources of knowledge or information. As one might expect, firms use many different channels for accessing university expertise: academic papers or patents, conferences, seminars, consulting, and so on. But one could frame the question in a slightly different way. What measures of university activity help explain their effects on local innovation? Scientific publications are thought to represent advances in basic knowledge. Patents represent advances in applied knowledge. Both of these activities indicate human capital capable of producing novel knowledge, basic and applied respectively. We construct measures of these activities using data from Thompson ISI and the European Patent Office. Additionally though, universities might possess other competences harder to quantify or describe, for example skills or accumulated knowledge that can be applied to issues other than creating novelty. These too could be of value in industrial innovation activities. In the latter part of the paper we perform an accounting exercise in an attempt to assess whether the human capital associated with creating new basic knowledge, creating new applied knowledge, or something different is what drives the university effect on industrial innovation. For two reasons we focus on the short-term effects of academic research. First, it is 3
In the Italian system teaching is organized into schools (facolt` a) and research is organized into departments (dipartimenti). Departments and schools may or may not coincide. To simplify presentation, we refer only to “schools”, and our measure of the date of opening of a new school is the year at which the first class was registered within a newly formed school. This should not be read to imply that university expansion affected only teaching. A new school in most cases implied creation of a new department. This conflation of schools and departments, teaching and research units, is not an issue for our analysis, as both measure university presence in the region.
2
likely that regional collaboration networks grow fastest in the first few years after opening of new university schools. Second, considering the short-run effect of universities allows us to identify the direct influence of academic research on innovation activity and to exclude other channels. In particular, it permits us to avoid the issue of how graduates contribute to innovation.4 So by focusing on the short term effects, we can identify direct knowledge spillover effect from university faculties to local industries. Our results suggest that there is indeed a significant effect of the creation of new university schools on regional research and innovation activity. Industrial patenting increases following the introduction of a new school to a region: on average, one new school has led to about a seven percent increase in the number of patents filed by regional firms five years later. But the quality of patents produced as a consequence of university supply shock is not different from the rest of regional patents. Given that the level of development of a region affects its absorptive capacity, one might expect that more developed regions with more intensive R&D activity benefit more from interactions with universities. However, contrary to this hypothesis, we find that less developed regions benefit more from university-industry interactions. Regarding the second issue, we find that the number of patents explains essentially none of the effect of universities on innovation. Publications corrected for quality explain most of the effect of universities on local industrial innovation. This suggests that in order to increase regional innovation local the intermediate policy goal should be to increase the amount of high quality academic research carried out in the region. The rest of the paper is organized as follows. Section 2 reviews the existent empirical findings concerning the role of academic research in innovation systems. Section 3 describes the data. Section 4.1 introduces the empirical model and comments on the main identification assumptions. The results of the empirical analysis are provided in sections 4.2 and 4.3. Finally, section 5 concludes.
2
Background literature
2.1
Identifying the effect of university R&D
There exists a large literature analyzing the relationship between academic research and industrial innovation activity. That university effects on industrial innovation might be localized stems from the nature of knowledge. While to a great extent the business of universities is to produce codified knowledge, tacit knowledge remains central in the 4
The effects of an increased quantity and quality of graduates in a region are likely to be very diffuse and hard to identify. However, they do not emerge within five years after a school opens: the official duration of most degrees in Italy (in analyzed period) is five years, fewer than 20% of graduates complete education on time and, on average, students take two more years to graduate after the end of the official program (Bagues et al. 2008).
3
diffusion process (see for example Cowan et. al 2000). While codified knowledge can be diffused very widely, and now very rapidly, tacit knowledge, by its nature, cannot. Jaffe et al. (1993) showed that diffusion of the knowledge contained in patents, which are by definition highly codified, has a strong geographical pattern — diffusion is very much local, and access to the knowledge spreads geographically over time. Breschi and Lissoni (2009) revisited this issue and showed that in fact it is social rather than geographic distance over which the diffusion takes place. That is, inventors learn about the existence of a patent (and presumably the knowledge it contains) through their direct social contacts. Since most social contacts are local, we can expect (geographically) localized knowledge diffusion. As early as the 1980s it was suggested that technology clusters such as those in Massachusetts and California would be impossible without the technology transfer from universities in these areas (Saxenian, 1985; Dorfman, 1983). It was not long though, before several case studies questioned the generality of the role of university as an accelerator of regional innovation (Feldman, 1994a; Rogers and Larsen, 1984) and suggested that various characteristics of regional technological infrastructure (business services, technologically related firms, etc.) are necessary for development of university research outputs. To understand the magnitudes of possible effects of university research on industrial innovation, Jaffe (1989) provided a more aggregate econometric analysis. He used data for 8 years for 29 US states to test whether there is an impact of university R&D on industrial patenting, and found a significant positive effect. Several later studies confirmed Jaffe’s finding, using firms’ product and process innovations instead of patents as a measure of innovative activity (Acs et al., 1992; Feldman, 1994b; Feldman and Florida, 1994). The challenge that runs throughout the empirical literature on technology transfer is the problem of identifying causation in a system rife with endogeneity. A positive association between academic research and industrial innovation may not necessarily imply that universities increase local innovation activity. It is quite possible that increases in university outputs are be “caused by” increase in industrial R&D and associated with easy access to industrial inputs such as equipment or materials. A more active industrial R&D sector may facilitate and stimulate university knowledge production. Thus the causation may work in the “opposite” direction, and if there is inertia in the variables (industrial R&D and university papers and patents) as seems very likely to be the case, then extracting the causal direction is difficult statistically. Increases in university research production might also be reinforced by the self-selection of academics able to benefit from interaction with industry into highly innovative industrial districts. This will introduce a bias into the types of activities of universities, changing the types of outputs depending on the nature of local innovation activities. This would be consistent with results showing a positive corelation between professors’ scientific output (as measured by published papers), and their applied output as measured by patents (Carayol and Matt, 2004; Stephan et al., 4
2007). Again this causes problems for statistical understanding of causation. In order to address those endogeneity problems, Jaffe (1989) estimated a system of three equations: the first equation characterizing the effect on industrial and university R&D on patenting, and two equations describing the determinants of, respectively, industrial R&D and university R&D. To identify the model, Jaffe assumed that industry R&D does not depend on the number of private and public institutions and that university R&D does not depend on manufacturing value added, once, respectively, university and industrial R&D are taken into account. Thus the consistency of Jaffe’s findings depends on the validity of these assumptions. Econometric analysis of university effects on industrial innovation at the regional level relies on the assumption that knowledge diffusion and spillovers are geographically localized. Botazzi and Peri (2003) analyzed the effect of total regional R&D on innovation and found that in Europe the effect of R&D is very localized and exist only within a distance of 300 km. Andersson et al. (2009) provide evidence suggesting that spillovers from university investment might be even more localized. They analyze the effects of changes in the Swedish university system and find that roughly half of the productivity gains from aggregate university investments are manifest within 5-8 km of the community in which they are made. Some authors claimed that the evidence of firms’ disproportionate location in areas close to universities already suggests that the potential positive interactions between industry and university are likely to be quite localized (Abramovsky et al., 2007; Audretsch and Stephan, 1996). Still, there is also evidence suggesting that university R&D might be related to patenting activity much further away following the collaboration networks of university professors (Ponds et al., 2010). Many studies at the firm level have confirmed that those firms that collaborate with university have higher propensity to innovate (Loof and Brostrom, 2008; Zucker et al., 1998), especially in those technological areas that require frontier scientific knowledge (Hall et al., 2003). Once again, even if the empirical evidence suggests that industrial innovation and university research tend to cluster in the same locations and university-industry collaborations are more frequent in highly innovative firms, it is difficult to claim empirically that the intensity of university research influences the innovativeness of industrial sector, and not vice versa. Typically, one needs to rely on observable characteristics, which could be used as controls, and to assume that they are sufficient to exclude any correlated effect. To be policy relevant, it is also important that the observed university effect it not actually a crowding out effect of private R&D, i.e. that university supply does not lead to a substitution of private funding for public funding (Mairesse and Mohnen, 2010).
5
2.2
Transfer of knowledge and expertise
University research is conducted under a very different system of incentives than is that in the private sector (Stephan, 1996; Dasgupta and David, 1994). This could easily lead to the accumulation of very different types of human capital and knowledge in the two sectors, and thus raises the possibility of synergies between them in innovation activity. The most obvious competence associated with university researchers is that aimed at expanding the knowledge frontier. This has been the traditional research function of universities, and is seen as perhaps the main activity of the professoriate. Publication is traditionally the activity most highly valued in the university setting. More recently, though, academic scientists have been encouraged to produce applied knowledge. Here again, since the aim has been to express this knowledge in terms of patents, novelty is paramount. The types of human capital needed to produce these types of novelty are likely to be well-proxied by publication and patenting measures, and indeed, the literature on the channels of knowledge communication between industry and university finds that publications and patents, particularly the former, tend to be important channels of communication. Academic patents are often discussed, especially by policy makers, as one of the main channels of knowledge and technology transfer from university. In part, this belief motivated the U.S. Bayh-Dole Act (1980), which gave permission for US universities to patent technology developed with federal funds. The underlying rationale was that this should speed up technology transfer by bringing new commercialization opportunities to the market.5 In Europe, many universities have also recently adopted technology transfer policies. But at the same time, many academics expressed concerns about potential detrimental effects of incentives to patent on the type and the quality of the research output produced (Lundvall, 1992; Henderson et al., 1998). Contrary to the apparent belief of policy makers, the empirical evidence tends to suggest that academic patenting per se is not a key channel of technology transfer (Agrawal and Henderson, 2002; Arundel and Geuna, 2004; Cohen et al., 2002; D’Este and Patel, 2007). It is often argued that the transfer of university knowledge could be also spread through the more traditional academic channels such as scientific publications, seminars or face-toface interactions. In fact, publications (as well as co-publishing with industry) are often found to be among the most important channels of knowledge transfer from academia as perceived by survey respondents (Cohen et al., 2002; Cassiman et al., 2008; Bekkers and Freitas, 2008). While publication is necessarily strongly correlated with the ability to produce frontier knowledge, other channels discussed in that literature may be much less so. Conferences, workshops, consulting and so on may be avenues for transmitting something 5
Market failure theory suggests that due to the public good nature of knowledge private companies have little incentive to invest in developing an invention that is not protected by a patent.
6
other than cutting-edge knowledge. Academics may have general analytic or synthetic skills that can be used for things other than publishable research; they have (in principle at least) pedagogic skills gathering, synthesizing, codifying and delivering existing knowledge. All of these could be transferred or facilitate transfer of useful knowledge. One challenge arising from this literature stems from the suggestion that subjectiveness of survey responses can lead to very different opinions about channels’ importance, depending on who is asked to evaluate it. For instance, Bekkers and Freitas (2008) compare perceptions of the importance of academic patents (among other things) as a channel of technology transfer among academics and private sector R&D workers and report that the private sector considers them to be twice as important as does academia. In this paper we avoid this problem and do not focus on channels of transfer, rather we use an accounting procedure to address the issue of different types of knowledge flow.
2.3
This study in the literature and hypotheses
The extent to which knowledge flows from university affect regional innovation and which type of academic expertise matters in this respect, are important empirical questions. Still, as discussed above, the empirical analysis of these issues face several methodological problems related to the identification of the effects of university research. The presence of those problems often opens a window for criticism of the recent empirical findings on university effects on industry. In what follows we perform a statistical analysis to examine whether a rapid growth of the university system in Italy had an effect on local industrial innovation activity. University schools creation in the 1980’s and in the 1990’s was part of a plan to expand educational supply, adopted by the Italian government in the beginning of the 1980s.6 The plan sought to unload over-crowded universities and to improve graduation rates.7 We exploit the fact that, as we show in section 4.1, the rationale of this rapid university expansion and consequently the timing of new university openings was independent of the demands of local innovation systems. This allows us to avoid many of the endogeneity problems typically attendant on this type of study. 6
Some new university openings were already approved in late 1970’s. However, the substantial reform came with law n. 382 11/7/1980, which provided that any variation in the existing university supply should be included in a development plan (piani triennali), to be approved by the Minister of Education every three years (Law n. 590 14/8/1982). Some autonomy has been introduced starting from 1995. For more details see Bratti et al. (2008). 7 In most cases the new schools were opened within previously existing universities, but often located in different towns. With time some of these schools became independent universities (for example, what is now the University of Eastern Piedmont was founded in 1998 on the basis of schools of the University of Turin located in Vercelli, Alessandria and Novara). In a few cases the new units appeared as a result of the split of overcrowded universities in big megalopolises (the University of Rome III was founded in 1992 simply by taking part of the staff from University of Rome La Sapienza). There are very few examples of opening completely new universities from scratch (one such is the University of Teramo, founded in 1993).
7
The focus on Italy is especially relevant since it is one of the countries that has to catch up with other European countries in the level of innovation activity in firms. According to the Community Innovation Survey (CIS) 2008, Italy is still lagging behind the majority of European countries in the percentage of firms that innovate. Finally, we explore which type of academic knowledge and expertise is effectively transferred to industry. In particular, we are interested in understanding the extent to which industrial innovation is affected by the activities and human capital associated with professors’ publishing as opposed to those associated with the more applied activities associated with patenting. The answer to this question can help universities and policy makers to judge about the alignment of existent incentive structures in academia to the necessities of local industries.
3
Data and variables
The analysis is performed using Italian data at the regional level. The database includes characteristics of the university system, indicators of industrial and academic innovation activity and economic indicators observed for 20 Italian regions between 1984 and 2000. Our main indicator of the university presence in the region is the number of university schools in science, medicine and engineering. We consider the date students were first enrolled in the degree program of the school as the date of the creation of this school. Information about the number of first-year students at the school level was obtained from different issues of the Italian National Statistical Bureau bulletins (L’universit`a in cifre and Lo Stato dell’Universit`a ).8 According to this definition, 65 schools in science, medicine and engineering were opened for enrollment between 1985 and 2000. (Figure 1 describes the dynamics of university expansion across time and Figure 2 shows the geographical distribution of new schools.) Out of the total of 65 new schools, 29 schools were in civil and industrial engineering, 12 in sciences, 11 in agriculture and veterinary, and 13 in medicine, pharmaceutics and chemistry. The average Italian region has nine schools and every fifth region in a given year opened a new school (Table 1). The number of schools might seem too aggregate an indicator of university presence, since schools can vary considerably by size. However, we have sufficient evidence to believe that the variation in the number of schools was exogenous to regional innovation activity and the dynamics of other factors affecting innovation. By contrast, the number of professors hired by these new schools is likely to have been determined by the demand for education. This demand could be correlated with the innovation activity in the region. 8
The data are available from 1984 from the printed annual editions of L’universit`a in cifre and Lo Stato dell’Universit` a (available in most university libraries in Italy). Data for years from 1988 on were accessed at http://ionio.cineca.it, in November 2007.
8
12
Number of new schools
10 8 6 4
2000
1999
1998
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
0
1985
2
Year
Figure 1: Annual number of new university schools (excluding schools in humanities and social sciences), 1985-2000.
(4,8] (2,4] [0,2]
Figure 2: Location of schools created during 1985-2000 across Italian regions. The colours indicate numbers of schools opened: dark grey indicates that between 5 and 8 schools were opened in the region over the entire period; light grey, 3 or 4; pale grey, between 0 and 2.
9
Table 1: Descriptive statistics (1)
(2)
(3)
All regions Mean
(4) North
Std. Dev.
Mean
Std. Dev.
(5)
(6) Center
Mean
Std. Dev.
(7)
(8) South
Mean Std. Dev.
Schools: 9 6 9 7 10 4 8 5 - Engineering 2 2 2 2 2 2 3 2 - Sciences 2 1 2 1 2 1 2 1 - Medicine, Chemistry and Pharmacy 3 2 3 3 4 1 2 2 - Veterinary and Agriculture 2 1 1 1 2 1 2 1 New schools opened in 1985-2000: 0.20 0.55 0.20 0.53 0.16 0.46 0,24 0,62 - Engineering 0.09 0.35 0,08 0,30 0.09 0,33 0,11 0,41 - Sciences 0.04 0.19 0,04 0,19 0,04 0,19 0,04 0,19 - Medicine, Chemistry and Pharmacy 0.04 0.21 0,05 0,25 0,01 0,11 0,05 0,23 - Veterinary and Agriculture 0.03 0.18 0,03 0,17 0,03 0,16 0,04 0,21 Publications 1020 1111 1308 1238 1260 1257 521 527 Citations per publications 17 5 17 5 18 3 15 6 Patents: 142 241 284 327 80 73 22 25 - Academic patents 7 12 13 17 8 9 2 3 - Industrial patents 134 230 272 311 72 65 20 22 Citations per patent: 0.67 0.46 0,72 0,37 0,67 0,43 0,60 0,56 - Academic patents 0.78 1.29 0,97 1,24 1,16 1,66 0,63 0,62 - Industrial patents 0.66 0.47 0,72 0,36 0,60 0,36 0,34 0,78 Non-patent literature (NPL) citations 0.66 0.88 0,41 0,33 1,02 1,12 0,70 1,03 per patent - Academic patents 2.3 3.34 1,99 2,10 3,67 4,62 1,66 3,04 - Industrial patents 0.48 0.69 0,33 0,29 0,68 1,06 0,48 0,63 Private R&D investment, mln euros 229 421 439 582 147 205 48 59 Public Non-University R&D invest90 188 76 71 199 341 27 29 ment, mln euros Public University R&D investment, 150 128 154 125 186 140 119 117 mln euros Population, mln 2.4 1.8 2,7 2,3 2,1 1,4 2,1 1,6 Population of 19-olds in total popula15.7 4.2 13,3 3,4 15,0 3,7 18,9 3,3 tion, % University graduates in the labour 7.6 2.2 6,9 2,1 8,3 2,5 7,9 1,8 force, % VA per capita, thousand euros: 14.8 5.5 18,1 5,5 14,9 4,6 11,0 3,3 - Industrial VA in total VA , % 21.9 7.5 25,7 7,7 23,1 6,7 16,7 4,3 - Services VA in total VA , % 67.2 6.7 65,2 7,3 67,4 7,3 69,4 4,7 - Agriculture VA in total VA, % 4.3 2.0 3,1 1,2 3,6 1,3 6,1 1,6 - Construction VA in total VA, % 6.6 2.1 6,1 1,9 5,9 1,7 7,8 2,1 Notes: (*) Total number of regions is 20. Regions classified as “Northern” are Piedmont (PIE), Aosta Valley (AOS), Lombardy (LOM), Friuli-Venezia-Giulia (FVG), Trentino-Alto Adige (TAA), Veneto (VEN), Emilia-Romagna (EMR), Liguria (LIG); regions classified as “Central” are Tuscany (TUS), Umbria (UMB), Marche (MAR), Lazio (LAZ), Sardinia (SAR); regions classified as “Southern” are Abruzzo (ABR), Basilicata (BAS), Calabria (CAL), Campania (CAM), Molise (MOL), Apulia (APU), Sicily (SIC). Mean values for the period 1984-2000.
10
So the number of professors is unlikely to be an exogenous shock. Similarly to the case of school size, any indicator of university-industry collaborations (such as the number or size of university technology-transfer offices) would suffer from endogeneity.9 An additional problem of indicators of formal collaborations is that they unavoidably miss the important component of informal collaborations (Link et al. 2007).10 All these arguments justify our focus on the number of university schools as an indicator of university presence in the region. We measure regional innovation productivity by the number of patents registered in the European Patent Office, using the location of inventors to determine the region where the patent is produced.11,12 In order to disentangle the knowledge spillover effect of universities from the direct effect of university R&D investment on patenting and crowding out effect, we split patents into two groups: those that are produced with university participation (or academic patents), and the rest of patents (industrial patents). Note that until recently it has been difficult to attach patenting activity to university research. In fact, in contrast to the US case, up to the present, in Italy universities did not generally retain the property rights on inventions done by their researchers. Often “IPRs over inventions derived from sponsored research programmes were left to the sponsors”(see Balconi et al., 2004). The recent KEINS EP-INV database on academic patenting (Lissoni et al., 2006) matches the names of the inventors of the patents with a list of university professors. Thanks to this methodology, the KEINS database includes not only any patent owned by universities, but also all patents that involve university scientists, whether the patents are owned by firms, public research organizations, universities, or the scientists themselves. In the following we use the KEINS database to identify industrial and academic patents. We observe that an average region produced annually 142 patents, seven of which were produced with the participation of academic inventors. We measure the quality of innovation by the average number of citations received by these patents before 2004. Naturally, series on patent citations are subject to a truncation bias since the number of citations any patent receives grows with time, and our data include citations received only until 2004.13 We correct for truncation bias following the method developed by Jaffe and Trajtenberg (1996) in the version where the diffusion process is assumed to have the same shape in all technological sectors.14 Figure 3 shows the 9
In Italy, there were almost no universities that adopted a patent or technology transfer policy until 1996 (see Baldini et al. 2006), so this is not an option for our analysis. 10 However, it is likely that the intensity of informal collaboration is correlated with formal collaboration activities due to potential complementarities (Cohen et al. 2002, Grimpe and Hussinger 2008). 11 Specifically, the database includes all patent applications that passed a preliminary examination in the EPO. The assigned date of the patent is the priority date, which is the date of the first filing world-wide. 12 Patents with inventors from two different regions are counted twice. 13 More precisely, citation variables count the number of citations received by regional patents from all Italian patents until 2004. 14 Results of this paper are not affected if instead we use a 5-year window for patent citations.
11
evolution of the number of patents, patent citations and corrected citations in industrial and academic sectors. Patents, both industrial and academic, and citations to them grew steadily over the period.15
Figure 3: Evolution of academic (left panel) and industrial (right panel) innovation outcomes in Italy, 1985-2000
Information on professors’ publication records are obtained from ISI Web of Science. We use all publications from 1984 to 2000 having at least one coauthor with an Italian affiliation. We observe citations received by these publications up to 2009. An average Italian region has produced more than a thousand publications a year; each of them has received around 17 citations. The extent to which technological innovations rely on scientific knowledge is measured by the propensity of patents to cite non-patent literature (NPL). Not surprisingly, we observe that patents with inventors from academia draw more on scientific knowledge than pure industrial patents (column 1, Table 1). We use information from the Italian National Statistical Bureau on several regional characteristics including private and public spending on research and development, valueadded produced by different economic sectors (industry, services, agriculture and construction), population, population of the age 19, and the proportion of university graduates in the labor force. We apply the depreciation coefficient used by Gordon (1990) (19.3%) to the time-series of R&D investment in order to construct an indicator of the stock of R&D facilities available in a region. Note that the intensity of innovation activity is very heterogeneous across Italian regions (columns 3-8, Table 1). Between 1984 and 2000, regions in the North of Italy were investing almost ten times more in R&D than Southern regions. These differences are also reflected in the number of patent applications done by inventors from these regions. The gap in the innovation activity across these regions are not due to the size effect: there are no important population differences across the regions. There are also no substantial differences in the university presence or in the educational level of the labor 15
Other ways to treat the truncation bias in citations are discussed in the Appendix.
12
force. Largely the differences in the innovation activity could be attributed to a relatively low income level in the South and to the differences in the industrial structure: in the North manufacturing has a larger weight in the economy than in the South, whereas in the South service sector and agriculture are relatively more important.16
4
Empirical Analysis
We start by observing simple correlations between the number of new schools opened in a region and the variation in various indicators of research and innovation activity (Table 2). First, we analyze whether the opening of new schoolsan is associated with the increase in the academic research activity. We find that university expansion is associated with the growth of university R&D stock in the region observed about two-five years later (column 1). Similarly, higher university presence is associated with the rise of the number of scientific publications already within the next three years (column 2). We do not observe any clear (significant at standard level of 5%) relationship between university expansion and the growth of academic patenting within first five years (column 3). Second, we analyze the relationship between university expansion and the growth of industrial innovation activity. We find that there is a positive correlation between university expansion and the growth of industrial patenting a few years later (column 4), even though new university units are not associated with the growth of private R&D and income per capita (columns 5 and 6).17 Table 2: Correlation between new school opening and variations in the indicators of private and public research activity
Number of years after school opening:
(1)
(2)
(3)
(4)
(5)
(6)
University R&D stock
Publications
Academic Patents
Industrial Patents
Private R&D stock
VA per capita
-0.036 -0.058 0.021 -0.056 0.006 0.141** 0.096 0.142**
0.042 0.018 -0.020 -0.021 -0.034 -0.018 0.006 0.016
0.099 0.001 -0.061 -0.074 -0.112 -0.039 0.046 -0.036
-2 n/a -0.074 -1 -0.173 0.005 0 0.184 0.031 1 0.189 0.155** 2 0.267** 0.226*** 3 0.287*** 0.111* 4 0.250*** 0.016 5 0.249*** 0.071 Notes: * p-value <0.100, ** p-value <0.050, *** p-value <0.010.
0.025 -0.103 0.130* -0.132* -0.072 0.032 0.059 0.011
Though suggestive, the results in Table 2 should be considered with caution. In principle, the creation of new schools might be not independent of regional innovation 16
Evangelista et al. (2001) also observe that, according to the 1995 wave of the Community Innovation Survey, there are very few science-based firms in the South. Regionally disaggregated data is not available in later CIS waves. 17 Note that the difference between the effect of new schools on industrial patenting observed in years 3, 4, and 5 is not statistically significant. Throughout the paper, we adopt the conventional level of significance – p-value less than 5%. The estimates with p-value between 5% and 10% are considered as marginally significant.
13
activity. In addition, the above correlations could be confounded by characteristics of the economy which vary simultaneously with university expansion and regional innovation. In the following we make explicit the assumptions of our identification strategy and analyze the above finding in more detail.
4.1
Empirical model and identification strategy
The main problem we seek to address is the possibility of circular causation between university research and industrial innovation. In order to address it, we analyze and build upon the standard reduced-form relationship between industrial innovation output, Pi,t , and the number of schools in the region, Ui,t , Pi,t = α + βUi,t + Xi,t γ + ct + ci + i,t
(1)
The simplest way to avoid potential simultaneity problem in model (1) is to consider the right-hand-side variables — including university presence — with a time lag. The time lag between university presence and innovation activity could be also justified since the effects of institutional changes could take time to realize (see the above findings in Table 2). Below we consider the right-hand-side variables with a 5-year lag. Nevertheless, given the presence of strong autocorrelation in the data, lagging independent variables is likely to be insufficient to avoid endogeneity. In order to capture some differences across regions, we include extensive list of observable regional characteristics among controls, Xi,t . The size of the region — both in terms of population, especially younger population, and in terms of economic production — may reflect inherent local demand for higher education as well as the propensity to innovate. Therefore we control for population, the proportion of 19 year olds, and aggregate value-added. Industrial composition affects the propensity of a region to patent. This in turn may affect the value of certain types of high-skilled human capital and so the ability of a region to lobby the central government for more university resources. Thus, we control for proportions of regional value-added in industry, construction, services and agriculture. An additional control for industrial structure is the share of graduates in the local labor force. Public non-university R&D and private R&D both affect industrial patenting, and may reflect a general attitude towards the value of knowledge production and training, thus again affecting the ability to lobby for more university resources. We include both controls in our estimations. University expansion was stronger in the 1990’s than in the 1980’s coinciding with the rapid growth of innovation activity (see Figures 1 and 3). In other words, the timing of university expansion generally might be not independent of the time trend affecting innovation activity, ct . In order to account for time effects influencing all regions simultaneously, we introduce a set of year dummies among controls. 14
Notwithstanding the inclusion of the above controls, one might suspect that regions differ on other perhaps non-observable dimensions; and these differences might explain both the university presence in the region and the development of innovation activity. In other words, the unobserved regional effects, ci , might be correlated with the university presence. Given the panel structure of our data, we can account for ci in two ways: using a fixed effect estimator (or, equivalently, including the regional dummies among controls) or using a difference estimator. Below we report results obtained using the firstdifference estimator. We note, though, that fixed effect estimation produces results that are statistically similar to the ones presented here. We prefer the difference estimator to the fixed effect specification since the former does not require strictly exogenous regressors (that is, it does not require that industrial innovation has no impact on future right-handside variables included in Xi,t , such as value added and R&D) (Wooldridge, 2002). With regard to the count nature of our dependent variable, in the following we adopt the traditional approach of modeling regional innovation activity using a log-log relationship between university presence and regional patents (Jaffe, 1989; Feldman and Florida, 1994). We prefer this model to a negative binomial specification for two reasons. First, given that the mean number of regional patents is quite high (134 patents, see Table 1 for more details), the negative binomial distribution is essentially normal and the loglog model provides a good approximation. Second, the linear model allows capturing region-specific effects without imposing the strong exogeneity assumption. Performing a negative binomial estimation with predetermined regressors requires a GMM estimation for which our sample size is not sufficiently large (Blundell et al. 2002). Ultimately, our identifying assumption is that the error term, i,t , is uncorrelated with university presence in the region once regional time-invariant effects, time effects and observable time-varying characteristics are taken into account. In other words, we assume that during the analyzed period no variations in regional characteristics (apart from the ones included in Xi,t ) affected both the variation in the number of university schools and the variation in the regional innovation 5-years later. Is this assumption justified? To answer this question we need to understand the factors that influenced university expansion. As was acknowledged by policy makers ex post, the distribution of new units across the regions was largely independent of regional labor market demands. The openings seemed to be associated with an indiscriminate allocation of funds across the regions. In this regard, the Observatory for the evaluation of the university system in the Ministry of Education and Research (MURST) after analyzing the expansion of university system in the beginning of 90’s concludes: The rules by which new institutions were created does not seem to have followed any logic tailoring university development to territorial specificities. It seems not to have made reference to a demand for university education 15
(that is, responding to the potential scope of use of the new initiatives), nor does it seem to have made reference to the demand for graduates (the formative needs of the country) or to existing infrastructure. In substance, no rigorous evaluations of the initiatives were done, either in absolute terms, or concerning compatibility with the rest of the system. The criterion actually favored was geographical re-equilibrium, which aimed to bring the offer of university education and subjects near to the demand, ignoring not only the “real” size of this demand (which sometimes turned out to be less than the minimum requirements for the initiative to be efficient and effective), but also the importance of the transportation system, the receptive capacity of the population of students and students’ financial support in determining access to university establishments. So, [. . . ] at least to a large extent, the prevalent logic was the one of incremental expansion and distribution “by drops of rain”, without evaluating other initiatives that were suppressed [. . . ]. (p.3, Verifica dei piani di sviluppo dell’universita 1986-90 e 1991-93, Osservatorio per la valutazione del sistema universitario, MURST, 1997; authors’ translation). This evaluation of Italian Ministry of Education and Research supports our identifying assumption. In Table 3 we also analyze the contemporaneous correlation between observable regional characteristics included in Xi,t and university expansion. Consistently with characterization done by the Italian Ministry, observable regional characteristics in the analyzed period seem to be poorly correlated with university expansion. Correlation between the number of new schools and the contemporaneous dynamics in observed regional characteristics, including the growth of regional patents, is very poor as well.
16
17
-0.00
0.69***
-0.02
0.85***
0.02
Notes: * p-value <0.100, ** p-value <0.050, *** p-value <0.010.
(28) Log Academic Patents
0.05
-0.01
0.02
∆Log Academic Patents
(27)
0.77***
-0.03
0.91***
-0.03
0.20***
-0.07
(26) Log Publications
0.03
-0.54***
0.06
-0.77***
0.03
0.02
-0.18***
0.64***
-0.01
0.00
0.07
-0.06
0.01
0.04
-0.10*
0.14**
-0.04
0.05
-0.16***
0.04
-0.06
-0.06
0.12**
-0.08
-0.31***
-0.13**
0.02
0.50***
-0.25***
0.13**
0.07
0.09*
0.07
-0.07
0.05
0.20***
-0.14**
0.22***
1
(5)
-0.01
-0.03
0.31***
0.77***
-0.13**
0.89***
-0.15**
0.27***
-0.04
0.02
-0.02
-0.03
0.00
-0.03
0.11*
0.05
-0.49***
0.02
0.73***
0.08
1
(4)
∆Proportion of VA in -0.09 -0.01 services (18) Proportion of VA in ser- -0.02 -0.16** vices (19) ∆Proportion of indus- 0.11* -0.04 trial VA (20) Proportion of industrial -0.00 0.31*** VA (21) ∆Proportion of agricul- 0.03 0.00 tural VA (22) Proportion of agricul- 0.09 -0.04 tural VA (23) ∆Proportion of VA in -0.04 0.07 construction (24) Proportion of VA in con- 0.00 -0.56*** struction (25) ∆Log Publications 0.06 -0.20***
(17)
-0.07
-0.07
0.79***
0.07
-0.06
(13)
(16) VA per capita
0.68***
-0.09
-0.03
-0.12**
-0.08
-0.01
0.42***
0.02
-0.06
∆Log Public nonuniversity R&D stock (14) Log Public nonuniversity R&D stock (15) ∆VA per capita
-0.02
0.04
0.07
-0.01
0.04
0.83***
0.02
-0.03
0.01
0.17***
1
0.10*
-0.02
Log Population
(6)
0.05
0.66***
0.08
∆Log Population
(5)
-0.02
1
-0.02
(3)
-0.03
Log Industrial Patents
(4)
0.04
0.02
(2)
-0.05
∆Log Industrial Patents
(3)
1
(1)
∆Population 19-olds in total population (8) Population 19-olds in total population (9) ∆Graduates in the labour force (10) Graduates in the labour force (11) ∆Log Private R&D stock (12) Log Private R&D stock
Log Schools
(2)
(7)
∆Log Schools
(1)
-0.05
0.00
-0.06
0.06
0.07
-0.01
0.01
0.07
-0.08
-0.08
0.03
-0.06
0.04
0.17***
-0.04
-0.11*
1
(7)
0.79***
0.04
0.79***
-0.19***
0.03
0.02
0.00
0.05
-0.51*** -0.17***
0.08
-0.18***
0.01
0.31***
-0.06
-0.14**
-0.00
0.08
0.06
0.81***
-0.05
0.78***
-0.10*
0.30***
0.04
0.07
-0.07
1
(6)
1
(9)
-0.35***
-0.01
-0.20***
0.07
0.60***
-0.14**
0.66***
-0.08
-0.31***
-0.02
-0.04
0.14**
-0.84***
-0.20***
-0.20***
0.35***
-0.37***
0.23***
0.03
-0.12**
0.05
-0.01
-0.05
-0.08
-0.06
-0.06
0.02
-0.05
0.01
0.12**
0.06
0.01
0.05
-0.12**
0.05
0.05
-0.41*** 0.18***
-0.08
1
(8)
0.37***
0.02
0.47***
-0.11**
-0.51***
0.01
-0.24***
0.04
-0.29***
0.09
0.55***
-0.10*
0.42***
-0.18**
0.42***
-0.28***
0.32***
-0.20***
1
(10)
-0.13**
0.03
-0.15**
0.01
0.17***
0.09*
0.21***
-0.02
0.01
-0.14**
-0.13**
0.08
-0.23***
0.04
-0.21***
0.14**
-0.18***
1
(11)
0.82***
0.03
0.78***
-0.20***
-0.70***
0.17***
-0.58***
0.04
0.47***
-0.08
-0.13**
-0.06
0.43***
0.21***
0.86***
-0.12**
1
(12)
-0.15**
-0.02
-0.06
0.12**
0.07
-0.07
0.15**
-0.09
0.11**
0.06
-0.19***
0.05
-0.41***
0.06
-0.20***
1
(13)
0.17***
0.35***
-0.01
-0.27***
-0.26***
0.26***
1
(15)
0.10*
0.17**
0.01
0.15**
-0.15**
1
(16)
0.78***
0.04
0.85***
-0.19***
-0.59***
0.12**
0.22***
-0.02
0.08
-0.04
-0.15**
0.29***
0.42***
0.01
0.22***
-0.09
-0.45***
0.15**
-0.34*** -0.29*** -0.68***
0.03
0.22***
-0.02
0.05
-0.07
0.29***
0.13**
1
(14)
Table 3: Cross-correlation table
-0.07
-0.03
-0.08
0.09
0.1
-0.26***
0.01
-0.53***
-0.02
-0.61***
-0.01
1
(17)
-0.17***
0.00
-0.10*
0.02
0.06
-0.08
-0.12**
0.00
-0.88***
0.06
1
(18)
-0.06
0.02
-0.02
0.09
-0.02
-0.23***
0.07
-0.12**
-0.07
1
(19)
0.46***
0.01
0.35***
-0.09
-0.47***
0.15***
-0.29***
-0.01
1
(20)
0.03
-0.02
0.02
-0.08
-0.05
-0.12**
0.09
1
(21)
-0.48***
0.00
-0.22***
0.12**
0.50***
-0.23***
1
(22)
0.17***
0.06
0.15**
-0.20***
-0.09
1
(23)
1
(25)
0.02
0.06
1
(26)
1
(27)
-0.65*** -0.18*** 0.74*** 0.27***
-0.04
-0.72*** -0.19***
0.14**
1
(24)
1
(28)
Similarly, the opening of new schools does not seem to be related to the demand for certain types of professions. In Figure 4 we plot the degree of the fit of educational supply to local demand for skilled labor in a region-discipline (measured as the number of new graduates incorporated in the labor force over the number of new graduates from local universities) versus the number of new schools in each region-discipline.18 An economically driven policy might aim to locate schools in regions that were importing skilled labor, leading to a positive correlation between university expansion and our measure of the fit of educational supply. Visually, no positive correlation is apparent, and indeed, the correlation between educational fit and the number of new schools in a corresponding discipline and region is on aggregate -0.055. Again, this is consistent with the MURST analogy between school creation and drops of rain.
Number of new schools
3
2
1
0 0
1
2
3
4
Hired 1992 graduates / Local 1992 graduates, by discipline and region
Figure 4: Number of new schools open between 1984 and 2000 by regional demand for corresponding professions
Overall, the above evidence implies that exploiting the variation in the number of schools within regions across time allows consistent estimation of the effect of university presence on regional innovation.
4.2
Regional innovation activity
Quantity The estimation results for model (1) with innovation activity being measured by the (log) number of industrial patents are presented in Table 4. We find that opening a new university school significantly increases regional innovation activity. The coefficients here are elasticities, so an increase of one percent in the number of schools in a region 18
New school openings are for the period from 1985 to 2000. The mismatch ratio uses data from the triannual Italian National Statistical Bureau (ISTAT) representative survey of graduates, the 1995 edition, which surveys students graduating in 1992. It covers information concerning graduates’ university-towork transition, asking, inter alia, where and in what discipline they graduated, and where they work. The description of the data could be found in Bagues et al. (2008).
18
increases industrial patenting in that region by 0.68 percent. The mean number of schools per region is nine, and the mean number of patent applications per region per year is 134 (Table 1), so on average, one new university school brings about ten new patent applications by regional non-academic inventors five years later.19 Table 4: The effect of the university expansion on industrial patents
Log Schools Log Population Population of 19-olds in total population Share of graduates in the work force Log Private RD stock Log Public (non university) RD stock VA per capita Share of VA produced in industrial sector Share of VA produced in agricultural sector Share of VA produced in construction sector Constant
(1)
(2)
(3)
Log Patents
Citations per patent
NPL citations per patent
0.68*** (0.21) 0.15 (2.67) 0.08 (0.16) 0.01 (0.07) 0.19** (0.09) 0.31 (0.23) -0.04 (0.13) -0.02 (0.04) 0.08 (0.05) -0.12 (0.13) 0.23 (0.14)
-0.11 (0.32) 2.37 (5.74) 0.25 (0.23) -0.05 (0.17) 0.31** (0.13) -0.79* (0.40) 0.04 (0.12) 0.13* (0.07) 0.07 (0.10) 0.13** (0.05) 0.74** (0.30)
-0.27 (0.41) -2.41 (3.95) 0.15 (0.16) -0.06 (0.21) -1.07** (0.41) -0.88* (0.48) -0.08 (0.19) -0.05 (0.06) -0.23** (0.09) -0.32** (0.15) 0.52 (0.45)
Year dummies Yes Yes Yes Adjusted R-squared 0.058 0.020 0.091 Number of observations 220 220 220 Notes: First-difference model estimates. Independent variables are 5-year lagged. In parentheses standard errors clustered by region. * p-value <0.100, ** p-value <0.050, *** p-value <0.010.
Quality The number of patents might be an appropriate indicator to capture the quantity of innovation, but it might hide important changes in quality. An observed increase in the quantity of patents as a result of a new university school in a region does not guarantee that the overall value of regional innovation activity grows. Therefore it is important to analyze the effect of universities on the quality of the patents produced. We check for effects on average patent quality, measured as the average number of received citations by regional patents. Evidence exists suggesting that patent citations represent a valid way to capture patent importance. (See Jaffe et al. (2000) for example.) Specifically, we estimate equation (1) using the average number of citations to regional industrial patents as the dependent variable. Results are presented in column 2 of Table 4. Opening a new school has no significant effect on the number of citations per regional patent. 19
See Appendix for robustness checks.
19
One might also ask whether not just the quality, but also other patent characteristics have changed. For instance, one might be interested to know whether the presence of a university affects the rate with which industrial innovation draws information directly from scientific publications. We use non-patent-literature (NPL) citations done by industrial patents to capture this patent characteristic. This is a very noisy measure (not least because many citations are actually added by patent examiners (Akers, 2000)), but it can nevertheless reflect any substantial changes in inventors’ reliance on scientific publications and basic knowledge. Still, we do not observe any significant effect of new schools on the nature of industrial patenting as measured by NPL citations (column 3, Table 4). Interestingly, while private R&D tends to increase the quantity and the quality of produced industrial patents, it is negatively related to the degree to which industrial inventors rely on scientific publications. One interpretation of the negative effect is that private R&D and local academic research can serve as substitutes in the innovation process. Universities can serve as knowledge sources when industry underinvests in private R&D.20 Regional heterogeneity A natural question is whether these effects hold uniformly across regions, or whether regions with different economic development respond differently. In Table 5 we explore whether university spillovers differ by the type of region. The results suggests that the strongest spillover effect occurs in the center and in the South of Italy (columns 1 and 2). As we have mentioned before, Southern and central regions differ from the northern regions in the level of income, private R&D investment and the industrial structure. To test whether these characteristics of the economy are in fact conditioning the strength of the spillover effect, we analyze how the effect differs across regions with different loadings along these dimensions. Specifically, for each dimension, we split the regions according to the median values of each variable. Note that regions can move from one group to another across time. We observe that regions with low per capita income benefit from knowledge spillovers from universities, whereas, on average, no positive university effect could be observed for high income regions (columns 3 and 4). Regions that have relatively low levels of R&D and those with a less educated labor force benefit more from universities (column 5-8). In less industrialized regions universities have a significant effect on innovation. In more industrialized regions the effect of universities is potentially larger, but is very noisy and not statistically significant (columns 9 and 10).21 20
Again this must be treated cautiously since many citations are entered in a patent not by the inventors but by examiners. 21 The number of new schools is not sufficiently large to perform detailed heterogeneity analysis of the effect of different types of schools. However, according to our data, the strongest effects were generated by schools in medicine, chemistry and pharmacy, veterinary, and agriculture. Science and engineering schools have stronger effect only in relatively more industrialized regions. Results are available upon request.
20
Table 5: The effect of the university expansion on the number of regional industrial patents (1)
(2)
Geographic location
Log Schools
(3)
(4)
Value Added per Capita
(5)
(6)
R&D stock
(7)
(8)
Graduates in the labour force
(9)
(10)
Industrial VA/ Agricultural VA
Center and South
North
Low
High
Low
High
Low
High
Low
High
0.90*** (0.13)
-0.05 (0.36)
1.01*** (0.29)
-0.06 (0.18)
0.79** (0.32)
0.45 (0.35)
1.04* (0.57)
0.77** (0.28)
0.57** (0.25)
0.86 (0.63)
Number of observations 88 132 110 110 110 110 110 110 110 110 Notes: First-difference model estimates. Regional characteristics and year dummies are included in all regressions. Independent variables are 5-year lagged. In parentheses standard errors clustered by region. * p-value <0.100, ** p-value <0.050, *** p-value <0.010. Regions with high level of VA per capita (abbreviations are the same as in Table 1, in parentheses – the year when a region first reaches the median): ABR (1997), EMR (1991), FVG (1994), LAZ (1991), LIG (1995), LOM, MAR (1995), MOL (2000), PIE (1991), TAA, TUS (1993), UMB (1995), VEN (1995), AOS (1989). Regions with high level of private R&D: ABR (1992), CAM, EMR, FVG (1990), LAZ, LIG, LOM, PIE, SIC (1998), TUS, VEN. Regions with high ratio of industrial VA to agricultural VA: ABR (2000), EMR (1989), FVG (1988), LAZ (1986), LIG (1986), LOM, MAR (1991), PIE, TUS, UMB (1988), VEN, AOS.
To summarize, we have observed that an increase in the number of schools is followed by an increase in industrial innovation activity in the region: the number of industrial patents increases. The effect of new schools depends on economic characteristics of the region: poorer regions with relatively low human capital and low investment in R&D benefit most from university presence. The average characteristics of industrial patents seems not to change with university presence, at least in the very short run. This suggests that in the short run the new industrial patents induced by the creation of new schools are on average not different from the rest of industrial patents.
4.3
Type of knowledge transfer
In the previous section we found that at the regional level, university presence has a positive influence on the quantity of industrial innovation. Recall that in our analysis we define industrial innovation as innovation done without academic inventors. Therefore the observed increase in industrial patenting after new school creation captures a spillover effect of university on industrial sector. In this section we ask whether it is possible to identify the type of university expertise that generates this spillover effect on industry. Specifically, we are interested in understanding whether firms are benefiting from university capability to produce frontier scientific research, from a more applied inventive potential of academic researchers or from other types of human capital or endowments. Understanding the type of expertise that is effectively transferred from university to local industry may be important for the design of university incentive structures. University capability to produce frontier scientific research can be measured by the current publications record of professors employed in a university. A quality-adjusted measure would weight the importance of each publication by its impact in terms of received 21
citations. This is a common way to characterize the quality of academic researchers and there seems to be a trend now towards making evaluation and incentives for researchers formally related to their (quality-adjusted) publication records. We define the quality weighted measure of scientific publications as the total citations received between date of publication and 2009. Recently, in addition, there has been an emphasis on academic patenting: in a variety of ways academic researchers have been encouraged to patent their findings, possibly in collaboration with firms. Professors’ patents (and the number of citations received by these patents) is our measure of academics’ inventive potential. These are both topical measures, since today publications and especially academic patents are often used to measure university contribution to economic activity. In this section we ask to what extent these measures of academic research activity can explain the effects we have found in the previous section. We undertake the following accounting exercise. To the previous model, we add variables representing our measures of professors’ expertise, and ask how their inclusion affects the estimated coefficient of the number of schools. Specifically we include in model (1) publications and academic patents as well as their citations, Puni i,t : Pi,t = α + βUi,t + Xi,t γ + Puni i,t δ + ct + ci + i,t
(2)
Note that scientific articles and academic patents might take up to several years to be published or listed in EPO. Additionally, it might take some time for schools to hire the necessary staff. So to capture correctly the increase in the regional human capital due to the opening of new university schools one would need to consider the change in publications and academic patents over several years. Consequently, we allow the effect to be distributed in time and include all lags of our measures in Puni i,t . A positive β in equation (1) signals the existence of a causal relationship between the opening of a new school and regional innovation output. However, if the inclusion of Puni i,t in equation (2) reduces significantly the size of β relative to its value in equation (1), we can claim that the corresponding university research output proxies the type of human capital which is effectively translated into regional innovation. If, controlling for professors’ publications and academic patents, we still observe a significant residual effect of universities on industrial patenting, we might conclude that there is something beyond the academic human capital captured by publications and academic patents that generates positive effects on industrial innovation. Note that university publications and academic patents are potentially endogenous to industrial patenting. An active industrial R&D sector could generate spillover effects inducing university activity, perhaps directly as industry seeks partners, or through some less direct, spillover mechanism. On the other hand, for various cost efficiency reasons, firms might engage in collaboration with university crowding our their independent re22
Table 6: Explaining the effect of schools on industrial patents
Log Schools
Publications Publication citations Academic patents Academic patent citations
(1)
(2)
(3)
0.68*** (0.21)
0.70*** (0.24)
(4)
0.48** 0.41 (0.19) (0.29)
Yes Yes
(5)
(6)
(7)
0.69*** (0.24)
0.70*** (0.23)
0.70*** (0.24)
Yes
Yes Yes
Yes Yes Yes
Adjusted R-squared 0.058 0.093 0.075 0.104 0.050 0.038 0.027 Number of observations 220 220 220 220 220 220 220 Notes: First-difference model estimates. Regional characteristics and year dummies are included in all regressions. Independent variables are 5-year lagged. In parentheses standard errors clustered by region. All lags of publication and patent indicators are included. In parentheses standard errors clustered by region. * p-value <0.100, ** p-value <0.050, *** p-value <0.010. The variance inflation factor (VIF) of Log Schools in column 4 is 1.45.
search. Therefore we do not interpret the estimated direct effects of publications and academic patents here. The results are presented in Table 6. The first, very clear observation is that academic patenting (columns 5, 6, and 7) captures none of the effects of university presence on industrial innovation. Expertise in applied knowledge generation inside a university seems to have little effect on local industrial innovativeness. Second, expertise measured by a simple count of publications (column 2) also seems completely ineffective. However, if we include quality in the measure of expertise (column 3) the coefficient on Log Schools falls by roughly 30 percent. Including both quantity and quality (column 4) reduces the coefficient by about 40 percent, and it becomes statistically insignificant. This suggests that what matters from the point of view of industrial innovation is to have a local university producing a significant amount of high-quality research. Strict interpretation of statistical significancy would imply that high quality research is all that matters, however the size of the coefficient and its high standard error suggests some caution, and opens the door to the possibility that there is something more, not captured by standard measures of university output.
5
Conclusions
In this paper we focus on the economic effects of universities, and in particular on their effects on innovation. It is widely believed that the presence of a university in a region is beneficial for industrial innovation activity. We have taken advantage of certain unusual features of university expansion in Italy during the 1980s and 1990s in an attempt to identify the effect of university presence on regional innovation. According to ex post evaluation of the expansion programmes, university schools were created “like rain”, independently of underlying economic features of the regions. This experiment permits a 23
nice way out of standard endogeneity problems. Our first result indicates that there is indeed a significant effect of the creation of new university schools on the regional innovation activity. In response, industrial patenting activity in the region increases quite significantly even within five years of a new school opening. Thus the general impression seems to be correct: university activity is positively correlated with local innovation activity, and a policy tool to increase the latter is indeed to increase the former. The effect of new schools depends on the economic characteristics of the area. Poor regions with low levels of R&D and human capital investment are the ones that benefit most from an increase in university presence. This suggests that one role of universities is to fill gaps in missing R&D infrastructure. If this is the case, and if there are positive feedbacks in innovation dynamics, then opening a university in an innovation-poor region can be an effective part of a development strategy for that region. Our results suggest that in an initial period when the region is poorly endowed with “innovation assets”, the university presence can compensate. This, under the proviso that other necessary assets are present, could help to push a region onto a higher innovation path. How are these benefits created? Marshall might suggest that it arises simply from the agglomeration of agents pursuing related activities; Mike Lazaridis22 asserts that it arises through the production of highly trained graduates; supporters of the Bayh-Dole act assert that it comes from controlled technology transfer through academic patenting. Given the time frame we examine, namely effects within 5 years of a school opening, we exclude from the analysis any effects of universities driven by the new graduates who enter regional labor markets. To address the question of other channels through which this influence flows, we have performed an accounting exercise estimating how the gross effect of increasing the number of universities is affected when we add to the model the proxies of factors that might be intermediary in the process. The factors that we focus on include professors’ ability to produce scientific research and their ability to produce patentable inventions. We measure the former by the number of ISI publications and the latter by the number of patent applications done with the participation of academic inventors. We observe that the human capital associated with traditional university production, as measured by scientific publications and their citations, has a strong effect on innovation, whereas academic patents can not explain observed university effects. At the same time, we do not exclude the possibility that other types of academic human capital, apart from the ability to produce frontier research, could be relevant as well. Such competences as collecting, generalizing and classifying existent knowledge might be relevant for industrial inventors. Universities perform many activities, and contain many varieties of skill and knowl22
Founder and former CEO of Research in Motion, maker of the Blackberry.
24
edge. Almost certainly it is a mixture of skills and knowledge that serves the interests of the local economy. However, the results presented in this paper suggest that it is possible to tilt the skill set in directions that are not in fact useful. They suggest that forcing universities to change tack towards more applied research may not, in fact, provide for the needs of local industry. At the very least, introducing patent activity as a new measure of university performance may not capture what matters, and, in the face of Goodheart’s law, as a policy tool to make universities more relevant to industry may be self-defeating. The traditional tool that values high-quality scientific research may remain more to the point.
References Abramovsky, L., R. Harrison, and H. Simpson (2007). University Research and the Location of Business R&D. The Economic Journal 117 March, C114 – C141. Acs, Z.J., D.B. Audretsch, and M.P. Feldman (1992). Real Effects of Academic Research: Comment. The American Economic Review 82 (1), 363–367. Agrawal, A. and R. Henderson (2002). Putting Patents in Context: Exploring Knowledge Transfer from MIT. Management Science 48 (1), 44–60. Akers, N. (2000). The referencing of prior art documents in European patents and applications. World Patent Information 22 (4), 309–315. Andersson, R., J. M. Quigley, M. Wilhelmsson (2009). Urbanization, productivity, and innovation: Evidence from investment in higher education. Journal of Urban Economics 66 , 2–15. Arundel, A. and A. Geuna (2004). Proximity and the use of public science by innovative European firms. Economics of Innovation and New Technology 13 (6), 559–580. Audretsch D. and P. Stephan (1996). Company-scientist locational links: the case of biotechnology. American Economic Review 86 , 641–652. Bagues, M., M. Sylos Labini, N. Zinovyeva (2008). Differential Grading Standards and University Funding: Evidence from Italy CESifo Economic Studies 54 (2), 149–176. Balconi, M., S. Breschi, and F. Lissoni (2004). Networks of inventors and the role of academia: an exploration of Italian patent data. Research Policy 33 (1), 127–145. Baldini, N., R. Grimaldi, and M. Sobrero (2006). Institutional changes and the commercialization of academic knowledge: A study of Italian universities’ patenting activities between 1965 and 2002. Research Policy 35 (4), 518–532. Bekkers, R. and I.M. Bodas Freitas (2008). Analysing knowledge transfer channels between universities and industry: To what degree do sectors also matter? Research 25
Policy 37 , 1837–1853. Blundell, Richard, Rachel Griffith, and Frank Windmeijer (2002). Individual Effects and Dynamics in Count Data Models, Journal of Econometrics 108 (1), 113–131. Bottazzi, L. and G. Peri (2003). Innovation and spillovers in regions: Evidence from European patent data. European Economic Review 47 , 687–610. Bratti, M., D. Checchi, G. de Blasio (2008). Does the expansion of higher education increase the equality of educational opportunities? Evidence from Italy, Bank of Italy working paper 679. Breschi, M. S. and F. Lissoni (2009). Mobility of skilled workers and co-invention networks: an anatomy of localized knowledge flows Journal of Economic Geography 9 , 439–468. Carayol, N., and M. Matt (2004). Does research organization influence academic production? Laboratory level evidence from a large European university Research Policy 33 , 1081–1102. Cassiman, B., R. Veugelers, and P. Zuniga (2008). In search of performance effects of (in)direct industry science links. Industrial and Corporate Change 17 (4), 611–646. Cohen, W.M., R.R. Nelson, and J.P. Walsh (2002). Links and Impacts: The Influence of Public Research on Industrial R&D Management Science 48 (1), 1–23. Cowan, R., P. A. David, and D. Foray(2000). The Explicit Economics of Knowledge Codification and Tacitness. Industrial and Corporate Change 9 (2), 211–253. Dasgupta, P. and P. A. David (1994) Toward a new economics of science. Toward a new economics of science 23 (5), 487–521. Dorfman, N. (1983). Route 128: The Development of a Regional High Technology Economy. Research Policy 12 , 299-316. D’Este, P. and P. Patel (2007). Universityindustry linkages in the UK: What are the factors underlying the variety of interactions with industry? Research Policy 36 , 1295–1313. Evangelista, R., S. Iammarino, V. Mastrostefano, A. Silvani (2001). Measuring the regional dimension of innovation. Lessons from the Italian Innovation Survey. Technovation 21 , 733–745. Feldman, M.P. (1994a). The University and Economic Development: The Case of Johns Hopkins University and Baltimore. Economic Development Quarterly 8 (1), 67–76. Feldman, M.P. (1994b). Knowledge complementarity and innovation. Small Business Economics 6 (5), 363–372.
26
Feldman, M.P. and R. Florida (1994). The geographic sources of innovation: Technological infrastructure and product innovation in the united states. Annals of the association of American geographers 84 (2), 210–229. Grimpe, C. and K. Hussinger (2008). Formal and Informal Technology Transfer from Academia to Industry: Complementarity Effects and Innovation Performance. ZEW discussion paper No. 08-080. Hall, B. H., A. Jaffe, and M. Trajtenberg (2001). The NBER Patent Citations Data File: Lessons, Insights And Methodological Tools. NBER working paper series 8498. Hall, B.H., A. N. Link, and J. T. Scott (2003). Universities as Research Partners. The Review of Economics and Statistics 85 (2), 485–491. Henderson, R., A.B. Jaffe, and M. Trajtenberg (1998). Universities as a Source of Commercial Technology: A Detailed Analysis of University Patenting 1965-1988. Econometrica 52 , 909–938. Jaffe, A.B. (1989). Real effects of academic research. American Economic Review, 79, 957–970. Jaffe, A.B. and M. Trajtenberg (1996). Flows of Knowledge From Universities and Federal Labs: Modeling The Flow of Patent Citations Over Time and Across Institutional and Geographic Boundaries. Proceedings of the National Academy of Sciences, Vol.93, pp.12671-12677. Jaffe, A.B., M. Trajtenberg, M. Fogarty (2000). The Meaning of Patent Citations: Report on the NBER/Case-Western Reserve Survey of Patentees. NBER Working Paper , 7631. Jaffe, A.B., M. Trajtenberg, and R. Henderson (1993). Geographic localization of knowledge spillovers as evidenced by patent citations. Quarterly Journal of Economics 108 (3), 577–598. Link, A.N., D.S. Siegel, and B. Bozeman (2007). An Empirical Analysis of the Propensity of Academic to Engage in Informal University Technology Transfer. Industrial and Corporate Change 16 (4), 641-655. Lissoni, F., B. Sanditov, and G. Tarasconi (2006). Keins database on academic inventors: Methodology and contents. CESPRI Working Paper 181. Loof, H. and A. Brostrom (2008). Does knowledge diffusion between university and industry increase innovativeness? Journal of Technology Transfer 33, 73–90. Lundvall, B. (1992). National Systems of Innovation. Pinter Publishers, London. Mairesse, J. and P. Mohnen (2010). Using Innovation Surveys for Econometric Analysis, Handbook of the Economics and Innovation, Vol. 2, 1129-1155.
27
Ponds, R., F. van Oort, and K. Frenken (2010). Innovation, spillovers and universityindustry collaboration: an extended knowledge production function approach. Journal of Economic Geography 10 , 231-255. Rogers, E. and J. Larsen (1984). Silicon Valley Fever. Basic Books, New York. Saxenian, A. (1985). Silicon Valley and Route 128: Regional Prototypes or Historic Exceptions?, 91–105. Sage Publications. Stephan, P.E. (1996). The Economics of Science. Journal of Economic Literature 34, 1199–1235. Stephan, P.E., S. Gurmu, A. J. Sumell and G. Black (2007). Who’s patenting in the university? Evidence from the survey of doctorate recipients Economics of Innovation and New Technology 16(2), 71–99. Wooldridge, J.M. (2002). Econometric Analysis of Cross Section and Panel Data, Cambridge and London: MIT Press. Zucker, L.G., M.R. Darby, and J. Armstrong (1998). Geographically Localized Knowledge Spillovers or Markets? Economic Inquiry 36 , 65–86.
28
Appendix. Robustness and Specification Checks 1. Measurement of patent characteristics In our dataset we observe information on patents only until 2004. There exist several ways to tackle the truncation problem in citations for recent patents. Our preferred method is to extrapolate citations using the method proposed by Jaffe and Trajtenberg (1996). Alternatively, one can impose the same truncation bias on all observations and count only those citations that were received within a certain time window after patent filing. Another alternative would be to normalize citations for patents submitted in the same year. In order to account for different propensities to cite across fields of science and technology, one could also normalize citations within the same field. Table A-1 summarizes estimations for (i) extrapolated patent citations, (ii) patent citations received within 5 years after publication, and (iii) patent citations normalized by the year of patent publication and patent class group. None of these definitions reveals any effect of new schools on patent quality. Table A-1: Measurement of patent characteristics a)
Citations per patent Extrapolated 5-year window after citations publication
Normalized for patents of the same year and patent class
Log Schools
-0.11 (0.32)
-0.16 (0.26)
-0.09 (0.17)
R-squared Number of observations
0.109 220
0.147 220
0.107 220
Absolute number
Normalized for patents of the same year and patent class
Log Schools
-0.27 (0.41)
-0.10 (0.23)
R-squared Number of observations
0.174 220
0.216 220
b) NPL citations per patent
Notes: First-difference model estimates. Regional characteristics and year dummies are included in all regressions. Independent variables are 5-year lagged. In parentheses standard errors clustered by region. * p-value <0.100, ** p-value <0.050, *** p-value <0.010.
Given that potentially there might be differences across patent classes in the propensity to refer to scientific literature, we also redo estimations for non-patent literature citations 29
normalizing them for patents within the same type of patent class and year. Once again, we do not observe any significant effect of new schools on the propensity to refer to scientific literature. 2. Time clustering of new schools’ opening In the paper we hypothesize that the short time span (5-years) considered after the opening of new schools allows us to exclude the knowledge transfers occurring through graduates. Still, if there exists time clustering in the opening of new schools within a region, this might affect the interpretation of our results. To avoid this problem we have always clustered standard errors for observations belonging to the same region, allowing them to be correlated within region across time. We also perform estimations only on the subsample of observations, for which there were no prior school openings during at least 3 years. Results are presented in Table A-2. The estimates, if anything, are large on this subsample. We also perform estimations on the subsample of regions excluding outliers in terms of university expansion and innovation – Piedmont, Lombardy and Emilia-Romagna. 22 of 65 new schools were opened in these three regions and the average R&D stock here is four times bigger than the country average. Once again, our results are robust to the exclusion of these outliers. Table A-2: Time clustering of new schools’ opening
Log Schools
All observations
No openings for at least 3 years before considered period
Rest of observations
Excluding Lombardy, Lazio and Emilia Romagna
0.68*** (0.21)
1.13*** (0.44)
0.80 (0.59)
0.65*** (0.21)
R-squared 0.144 0.229 0.272 0.148 Number of observations 220 94 126 187 Notes: First-difference model estimates. Regional characteristics and year dummies are included in all regressions. Independent variables are 5-year lagged. In parentheses standard errors clustered by region. * p-value <0.100, ** p-value <0.050, *** p-value <0.010.
30
