What determines technological hits?

Research Policy 33 (2004) 1565–1582

What determines technological hits? Geography versus firm competencies Myriam Mariania,b,c,∗ a

MERIT, University of Maastricht, Maastricht, The Netherlands b University of Camerino, Camerino, Italy c CESPRI, Bocconi University, Milano, Italy

Received 17 February 2004; received in revised form 16 July 2004; accepted 5 August 2004 Available online 11 November 2004

Abstract The centrality of firms vis-`a-vis regions underlines a general contrast between two models of producing innovations. This paper uses a new database composed of 4262 European chemical patents applied by 693 firms during 1987–1996 to compare the relative effect of firm and regional characteristics on the production of technological “hits” (highly cited patents). By using extensive controls, the main finding of the paper is that technological hits in the “traditional” chemical sectors are explained only by R&D intensity at the firm level and the scale of the research projects. Firm competencies, particularly technological specialisation, are still important in biotechnology. However, the distinct feature of the biotechnology model is that localised knowledge spillovers also matter. © 2004 Elsevier B.V. All rights reserved. Keywords: Firm competencies; Geographical spillovers; Patents; Patent citations

1. Introduction There is consensus in the literature about the importance of firm competencies in the production of innovations. The affiliation to an organisation with unique capabilities, internal communication systems, and specific routines is an effective mechanism for the ∗ Present address: CESPRI, Universit` a Commerciale Luigi Bocconi, Via Sarfatti, 25, 20136 Milano, Italy. Tel.: +39 0347 5180190. E-mail address: [email protected].

0048-7333/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.respol.2004.08.004

production and transmission of knowledge (e.g. Nelson and Winter, 1982; Dosi et al., 1988; Klepper, 2001). There is also evidence of the importance of competencies that pertain to territories and regions rather than specific firms. By facilitating the transfer of knowledge among close-by individuals and companies, the geographical cluster is another organisational form for innovation. Recent contributions confirm that knowledge spills over, and that the cost of acquiring knowledge is lower when individuals and firms are geographically close. This explains the tendency of innovative

1566

M. Mariani / Research Policy 33 (2004) 1565–1582

activities to cluster (Jaffe, 1986; Jaffe et al., 1993; Saxenian, 1994; Audretsch and Feldman, 1996; Swann et al., 1998; Cani¨els, 2000). While the literature has typically focussed on either the firm or the geographical cluster, little work has been done to compare them. This is the goal of the present paper. It explores empirically how much of the value of an innovation is affected by the affiliation of the inventors to the same organisation, as opposed to spillovers that arise when the inventors are geographically close to each other and to external sources of knowledge. The empirical investigation uses a detailed dataset constructed specifically for analysing this issue. The dataset is composed of a randomly selected sample of 4262 chemical EPO patents applied by 693 firms and invented in 208 European regions during 1987–1996. To test the effect of firm and regional characteristics on the probability of producing technological hits, the paper performs negative binomial regressions. It uses the number of citations received by the patents after the application date as a proxy for the value of the innovations.1 This is regressed on project and firm variables, scientific and technological characteristics of the regions in which the patents are invented, and a set of controls. The results highlight two models of innovation. I distinguish between biotechnology and the more traditional chemical sectors. Both sectors are researchintensive, but while biotechnology is a modern research sector that arose only a couple of decades ago, chemicals are the oldest high-tech industry in modern economies. Moreover, while the latter hinges upon large established companies, the former is associated with smaller firms that are often agglomerated geographically. I found that R&D intensity at the firm level and the scale of the research projects are important for developing valuable innovations in the chemical sectors. External knowledge spillovers and the technological characteristics of the area in which the firms are located play no role. By contrast, the drivers of valuable innovations in biotechnology are the technological characteristics of the regions and the 1 I use data on patent applications and not necessarily granted. Patent applications can be cited after their publication irrespective of whether they have been granted or not (see also Maurseth and Verspagen, 2002). As a matter of fact, 38.4% of the non-granted patents in my sample received one or more citations.

technological specialisation of the firms. These results are consistent with earlier work in the literature on the importance of knowledge spillovers and agglomeration economies in research-intensive sectors in general, and biotechnology in particular (Audretsch and Feldman, 1996; Klepper, 1996; Zucker et al., 1998a, 1998b). Section 2 reviews previous studies on firm competencies, geographical spillovers and the use of patent citations. Section 3 presents insights from the data, while Section 4 estimates an empirical model in which the probability of inventing valuable innovations in biotechnology and in traditional chemicals is a function of firm and regional characteristics. Section 5 summarises and concludes the paper.

2. Firm, geography and the value of innovations 2.1. Firm competencies and the geographical cluster The organisation and performance of innovative activities is influenced by decisions made by the firm as well as by factors external to it. A large part of the literature describes the firm as the natural mechanism to foster, select and coordinate R&D projects and activities (Nelson and Winter, 1982; Dosi et al., 1988; Patel and Pavitt, 1997). This is because the firm relies on specific competencies, learning processes, and communication systems that reduce the cost of coordinating different individuals and parts of the organisation (Nelson, 1995). Klepper and Sleeper (2002) describe the parental origins of firms’ distinctive capabilities for spinoff companies, and argue that these capabilities are difficult to reproduce without transferring the human capital employed by the companies. Teece et al. (1997) point out that the effectiveness of a firm’s learning processes, the capabilities to coordinate and integrate internal activities, and the ability to modify strategies and competencies when the outside conditions change are important factors in explaining firms’ competitive advantage (Levinthal and March, 1993; Christensen, 1997). A rising body of the literature, however, emphasises another means to coordinate individuals and activities: the geographical cluster. This literature underlines the importance of local infrastructures for innovation and growth (Marshall, 1920; Porter, 1998; Swann

M. Mariani / Research Policy 33 (2004) 1565–1582

et al., 1998) and argues that localised technological spillovers foster knowledge exchange and increase the returns from the investment in R&D (see, for example, Jaffe, 1986; Jaffe et al., 1993). This is particularly important when knowledge, and particularly “new knowledge which tends to be informal and uncodified” (Pavitt, 1987), is involved in the production of other knowledge, and when it relies on practice and learning-by-doing. By locating close to each other, people can access information, monitor other people’s behaviour, and foster communication among individuals, therefore reducing the complexity and uncertainty of the innovation process. Moreover, since knowledge tends to be cumulative also at the geographical level (Cantwell and Iammarino, 2001, 2003), scientific and technological progress is faster in regions that have accumulated high levels of innovative activities over time. The empirical evidence confirms the clustering of innovative activities and the geographical dimension of knowledge spillovers, and it estimates its effect on regional economic growth (Verspagen, 1997; Cani¨els, 2000). It also confirms that there are sectoral differences in spatial clustering with skilled and R&D intensive industries that benefit more from co-location and knowledge spillovers (Audretsch and Feldman, 1996; Breschi, 1999). Klepper (1996), for example, demonstrates that innovative activity in the early phases of an industry life cycle benefits more from locally bounded knowledge spillovers compared to the mature or declining stages.2 The biotechnology sector, a young science-based industry composed of a large number of small firms, is a case in point (Orsenigo, 1989). Powell et al. (1996) show that the locus of innovation in the biotechnology industry is a network of different organisations, rather then the individual firms. By using data on the formal agreements set up by 225 biotechnology firms they map the network structure of the industry, and argue that firms collaborate to expand their competencies. They describe two important biotech discoveries in the mid1990s that are co-authored by more than 30 researchers located in different organisations. Zucker et al. (1998a, 1998b) point out that the growth and location of intellectual human capital has been the main determinant 2 Wlash (1984) shows that knowledge-led factors are particularly important for innovation in the early phases of an industry.

1567

of the growth and location of the US biotechnology industry. By the same token, Zucker et al. (1998a, 1998b) demonstrate that geographical proximity to university research, which materialises in working relationships between the firm and the top scientists in the academia, positively affects firm R&D productivity in biotechnology. Such spillovers, however, do not occur unintentionally, but depend on specific complementary actions of the economic agents (Arora et al., 2001). As this brief assessment of the literature shows, the importance of the firm and the geographical cluster in the innovation process have typically been studied separately. The contribution of this paper is to analyse them together. Based on earlier work, the expectation is that different models of innovation exist in different industries, where firm competencies and regional characteristics play a different role in explaining the performance of R&D activities. As a research-intensive and prolifically patenting industry, the chemical sector is a suitable case study to investigate this issue. Moreover, as the industry is heterogeneous, ranging from bulk chemicals to biotechnology, it offers the possibility to explore the existence of different innovation models in sectors with different characteristics or in different stages of the industry life cycle. Specifically, while firm competencies are expected to be crucial for developing innovations both in the science-driven biotechnology sector and in the more mature and scale-intensive chemical industry, it is interesting to test whether biotechnology relies more than the latter on external sources of scientific and technological knowledge. 2.2. The value of innovations and patent citations This paper uses the number of citations received by patent applications in the 5 years after the application date as a proxy for the value or more generally for the importance of the innovations. The use of patent citations is now fairly standard in the literature (for a survey, see Hall et al., 2001). Citations made to previous patents are used as indicators of knowledge spillovers from the cited to the citing patent (Jaffe et al., 1993). Citations received by a patent are a proxy for the importance of the innovation. Several contributions demonstrate that there is a positive relationship between patent indicators and the actual ex-post value of the innovations as given by traditional accounting evaluation (Hall et al., in press). A classical contribu-

1568

M. Mariani / Research Policy 33 (2004) 1565–1582

tion is Trajtenberg (1990), who shows that there is a non-linear and close association between patent counts weighted by forward citations and the social value of innovations in the Computer Tomography Scanner industry. Harhoff et al. (1999) demonstrate that the number of backward citations to other patents and to non-patent literature, and the number of citations received after the publication of a patent are positively correlated with the value of the innovation. This holds also for patents applied in many countries, and for patents that incur in opposition and annulment procedures (Harhoff and Reitzig, 2004). Griliches et al. (1987) use data on patent renewal rates and fees to estimate the private value of patent rights. Lanjouw and Schankerman (in press) use multiple indicators to construct a composite measure of the quality of patents, and show that forward citations and claims are the most informative indicators with about 30% of the variation being related to quality. Forward citations are also the most important indicator in terms of reduction of the variance of the expected quality of patents. Patent indicators and patent citations have limitations as well (Griliches, 1990). For example citations cannot be made to or by innovations that are not patented, thus underestimating the actual importance of some of them. Second, there is the “truncation” problem: more recent patents are less cited. Third, not only is the number of citations received by any patent truncated in time, but patents applied in different years and technological classes differ in their propensity to be cited, suggesting that changes in the number of citations per patent might stem from factors other than the actual changes in the technological impact of the innovations. The econometric investigation will not ignore these problems, and, as a solution, it will scale each patent’s citations by the average number of citations of a cohort of patents with similar characteristics (as in Hall et al., 2001).

3. Insights from the data 3.1. Distribution of patents and citations across firms, regions and sectors The empirical investigation uses a detailed database at the level of the individual patent (European Patent Office, 1998), and links it to other sources of data on

firm and location characteristics. The dataset is composed of a random sample of 4262 chemical patents applied by 693 firms in 1987–1996.3 Appendix A describes the data collection and lists the variables gathered for each patent on: the project that led to the patent and the firm that applied for it; the European region in which the inventors were located while developing the innovation; a set of other controls for the characteristics of the innovation. The variable I want to explain in my regressions is the value of the individual patents (the “technological hits”). This is proxied by the number of citations received by the patents in the 5 years after the application date up to the year 2000 excluding self-citations, i.e. citations form patents applied by the same applicant. I label this variable CITS. I exclude self-citations because they may not have the same role in measuring the value of the innovation as compared to independent cites, as Hall et al. (in press) have suggested. For example, small companies might exploit technological trajectories in specialised niches. In this case, self-citations would be an indicator of internal spillovers and highlight the existence of cumulative processes of knowledge creation within the same firm, more than value per se. Large firms might cite themselves simply because they have large patent portfolios to cite, which could be quite independent of the actual value of the cited patents.4 To provide additional insights about CITS, Fig. 1 shows its distribution with and without self-citations.5 Consistently with other contributions (Scherer et al., 2000; Scherer and Harhoff, 2000) the distribution 3 From an initial sample of 10,000 chemical patents, I dropped 211 patents applied by public organisations including universities, 134 patents applied by individual inventors and 5393 patents for which none of the inventors was located in Europe. This produced a final sample of 4262 patents. For this sample, the share of patents with single applicant (93.6%), single inventor (16.4%), and single supplementary IPC class (19.8%), the year of application, and the nationality of the applicants and inventors are representative of the population as a whole. 4 At any rate, I performed the regressions in Section 4 by using CITS with and without self-citations, with no relevant changes in the overall results. 5 For the descriptive purposes of this section, I controlled for the truncation problem by using only the 3080 patents in my sample filed between 1987 and 1993. In the econometric analysis in Section 4 I will employ adequate controls for the truncation problem, and I will therefore use the whole sample of 4262 patents applied in 1987–1996.

M. Mariani / Research Policy 33 (2004) 1565–1582

1569

Fig. 1. Distribution of CITS—number of citations received by the patents in the 5 years after the application date, with and without self-citations. Source: Elaboration from EPO data. Sample: 3080 patents, 1987–1993. Note: The number of patent citations excluding self-citations ranges between 0 and 13 with mean 0.74 and standard deviation 1.34. The number of patent citations including self-citations ranges between 0 and 19 with mean 1.20 and standard deviation 1.91.

of patent citations is skewed. The inclusion of selfcitations increases the average number of CITS from 0.74 to 1.20, but it does not create major differences in the shape of the distribution. Fig. 2 compares the distribution of CITS in biotechnology and traditional chemicals. Patents in traditional chemicals receive a lower number of citations than in biotechnology. The share of patents with 0 citation is 63.2% in traditional chemicals compared to 53.9% in biotechnology. The average number of CITS is 0.69 in traditional chemicals; it is 1.03 in biotechnology. This difference is statistically significant.6

6 I grouped the patents in organic chemistry, materials, pharmaceuticals and polymers together in the “traditional chemical” sector because, compared to biotechnology, the descriptive statistics and the econometric estimates show that the model of innovation leading to high value innovations relies on similar firm and regional characteristics.

The distribution of patents is also concentrated in terms of applicants. The top five companies in terms of number of applications — Hoechst, Basf, Bayer, Ciba Geigy and Rhone-Poulenc — own more than one third of the patents. A long tail of companies applies for one or two patents. Table 1 lists the firms that applied for the top 20 biotechnology and traditional chemical patents in terms of the citations that they receive. Interestingly, in biotechnology, 12 companies in the top 20 are fairly small companies that develop a limited number of patents only in biotechnology. The other eight firms operate also in traditional chemicals. By contrast, all the top 20 firms in traditional chemicals are large and established companies. This confirms that the leading innovators in biotechnology are small and specialised companies, while innovation in traditional chemicals is dominated by large established corporations. The 3080 patents are invented in 208 NUTS2 and NUTS3 regions. Appendix B explains the criteria to

1570

M. Mariani / Research Policy 33 (2004) 1565–1582

Fig. 2. Distribution of CITS—number of citations received by the patents in the 5 years after the application date (excluding self-citations): biotechnology and traditional chemicals. Source: Elaboration from EPO data. Sample: 3080 patents (373 in biotechnology and 2707 in traditional chemicals), 1987–1993.

Table 1 Firms that applied for the top 20 patents in terms of patent citations Biotechnology

Traditional chemicals

Applicant

Number of citations

Applicant

Number of citations

B.A.T. Industries PLC Silica Apparatebau GmbH Imcera Group Inc. Ciba Geigy AG Hoechst AG Plant Genetic Systems N.V. Zeneca Group PLC E.I. DuPont de Nemours Max-Planck-Gesellschaft Transegene S.A. Akzo Nobel N.V. Biomerieux Alliance S.A. Got-a-Gene AB ISIS Pharmaceuticals Inc. Merial Rhone-Poulenc S.A. Royal Gist-Brocades N.V. Sclavo S.p.A. Solvay S.A. Amersham International PLC

11 10 9 8 8 7 7 6 6 6 5 5 5 5 5 5 5 5 5 4

Zeneca Group PLC Targor GmbH Bayer AG Glaxo Wellcome PLC Guerbet S.A. Basf AG C.H. Boehringer Sohn Rhone-Poulenc S.A. Dr. Zambeletti S.p.A. Hafslund Nycomed A/S Solvay S.A. Warner-Lambert Co. Ciba Geigy AG Monsanto Co. Polifarma S.p.A. Schering AG The Procter and Gamble Co. Vectorpharma International S.p.A. Colgate Palmolive Co. Hoechst AG

13 11 10 10 10 9 9 9 8 8 8 8 7 7 7 7 7 7 6 6

Source: Elaboration from EPO data. Sample: 3080 patents, 1987–1993.

M. Mariani / Research Policy 33 (2004) 1565–1582

assign the inventors’ addresses to NUTS. The geographical distribution of patents is skewed, with the top 25% regions hosting 63% of patents in biotechnology and 84% in traditional chemicals. However, biotechnology patents are less geographical concentrated than traditional chemical patents: the Herfindhal index is 0.020 for biotechnology, and 0.036 for traditional chemicals. Figs. 3 and 4 map the distribution of biotechnology and traditional chemical patents across the European regions. The symbols in the maps are drawn according to the distribution of patents across the regions. The size of the symbols increases by moving from the regions in the bottom quartile of the distribution to those in the second, third, fourth quartile, and in the top 10% class. The figures show that the bulk of biotechnology and traditional chemical patents are invented in roughly the same countries and regions: five of the top 10 biotechnology regions are also top 10 in traditional chemicals (Darmstadt-Hessen, Switzerland, Ile de France, Berkshire-Bucks-Oxfordshire, Surrey-East and West Sussex), and two thirds of the regions in the top quartile in biotechnology are also top quartile in traditional chemicals. 3.2. External versus internal spillovers in biotechnology and traditional chemicals This section provides preliminary evidence in support of the hypothesis that there are different models of innovation in biotechnology and traditional chemicals. I use the citations made by the 4262 patents in the sample to earlier patents as indicators of knowledge spillovers from the cited to the citing patent (Jaffe et al., 1993). The 7304 cited patents are classified geographically by the zip-code contained in the addresses of the inventors, and they are labelled as: local citation if at least one inventor of the cited and citing patents are located in the same NUTS region (378 patents); national citations if the inventors in the cited and citing patents are located in the same country, but in different regions (660 patents); international citations if the inventors are located in different countries (4095 patents). The fourth category is self-citations (2171 patents). Table 2 shows the share of local, national, international and self-citations in biotechnology and traditional chemicals. These are computed as the ratio S =

c / i il i ci , where i denotes the patent, cil the num-

1571

Table 2 Citations made to previous patentsa

Local citations National citations International citations Self-citations

Biotechnology

Traditional chemicals

Total

7.6 7.2 69.7 15.4

4.8 9.3 53.9 31.9

5.2 9.0 56.1 29.7

Source: Elaboration from EPO data. Sample: 4262 citing patents; 7304 cited patents. a Share of local, national, international, and self-citations in biotechnology and traditional chemicals.

ber of local/national/international/self-citations made by each patent to previous patents, and ci the total number of citations made by each patent. The results show that the overall share of local citations is small (7.4%), suggesting that geographical proximity is of limited importance for a patent to be cited. However, this proximity is relatively more important in biotechnology (7.6% of local citations) than in traditional chemicals (4.8% of local citations). Second, the share of international citations is high in both sectors though, again, it is higher in biotechnology (69.7% versus 53.9% in traditional chemicals). Finally, self-citations are more frequent in traditional chemicals than in biotechnology (31.9% versus 15.4%). These results suggest that the process of innovation in traditional chemicals relies more strongly than biotechnology on knowledge internal to the firm. As noted earlier, this could partially be because firms in traditional chemicals have large patent portfolios to cite. Biotechnology relies more strongly on external knowledge. This is true both of international and local sources of knowledge.

4. Firms versus regions in biotechnology and traditional chemicals: empirical analysis 4.1. Variables used in the regressions This section tests the combined effect of firm and regional characteristics on the probability of producing technological hits in biotechnology and in traditional chemicals. The analysis uses the whole 1987–1996 sample of 4262 patents: 525 in biotechnology and 3737

1572

M. Mariani / Research Policy 33 (2004) 1565–1582

Fig. 3. Geographical distribution of biotechnology patents in Europe. Source: Elaboration from EPO data. Sample: 373 biotechnology patents invented in Europe.

M. Mariani / Research Policy 33 (2004) 1565–1582

1573

Fig. 4. Geographical distribution of traditional chemical patents in Europe. Source: Elaboration from EPO data. Sample: 2707 traditional chemical patents invented in Europe.

1574

M. Mariani / Research Policy 33 (2004) 1565–1582

in traditional chemicals. The dependent variable is CITS discussed earlier. Because of the count nature of this variable, I perform negative binomial regressions. I use four sets of variables as regressors (see Appendix A for the definition of the variables). The first set is composed of firm characteristics. SALES is a proxy for the size of the firms, which is expected to positively affect the probability of inventing technological hits. Large firms can benefit from economies of scale and scope in research, and from the advantages in the financial markets. They can also avoid duplication in research, and they can internalise knowledge spillovers by employing a large number of specialised and complementary researchers in coordinated activities, although the efforts and costs to coordinate the spillovers internally might partly offset the advantages. Once controlling for the size of the firm, R&D intensity (R&D/SALES) is expected to have a positive impact on CITS: R&D intensive firms are likely to hire many researchers and to start a large number of different projects, therefore increasing the probability of developing technological hits. A firm’s R&D investment also enhances the capacity to absorb knowledge coming from external sources. To explore the effect of firms’ technological specialisation I constructed two variables: TECHSPEC and TECHCOMP. As discussed in Appendix A, with the help of an expert pharmacologist I classified all the patents in the sample in five major technological sectors: biotechnology, materials, organic chemistry, pharmaceuticals and polymers. TECHSPEC is the number of EPO patents in the same technological sector filed by the firm during the 5 years before the date of each patent in the sample. TECHCOMP is the same as the previous variable but it uses the number of EPO patents in the other four technological sectors. A dummy for the missing values of SALES and R&D is included in the regressions, as well as a dummy for the large non-chemical companies. The second set of variables account for project characteristics. The number of inventors listed in the patent (INVENTORS) is a proxy for the scale of the research project, and it is expected to positively affect the probability of developing big innovations both in biotechnology and in traditional chemicals. MAPPL, a dummy that indicates if the patent is the output of a collaboration among different institutions, is a proxy for the breadth of the project that goes beyond the firm boundaries. I hypothesise that a joint research effort increases

the probability of developing technological hits.7 The third project-level indicator is DLOC, a dummy for the co-location of the inventors in the same NUTS region. Its inclusion is based on the idea that the more a research project is complex and interdisciplinary, the higher is the probability that it is mastered internationally by pulling together a wide range of competencies from different firms’ units, organisations and locations. This would produce a positive correlation between DLOC and CITS. This dummy also controls for the arbitrary geographical assignment of patents when the inventors are located in different NUTS (see Appendix B). The third set of variables includes regional characteristics. The idea is that in technology-intensive regions, where innovative activities agglomerate, it is easier to find the specialised and complementary competencies needed in complex R&D projects. Moreover, both in traditional chemicals and in biotechnology the proximity to university research and the collaboration with the academia has shown to be of primary importance for the productivity of firms’ R&D activities (see, for example, Rosenberg, 1998; Zucker et al., 1998a, 1998b; Arundel and Geuna, in press). This paper uses the number of higher education laboratories (REGHLABS) as a proxy for the location of scientific institutions in the region, and the average number of patents invented in each region in all sectors (REGPATS) as an indicator of the regional technological capabilities.8 The area of the regions (AREA), the population density (POP) and the per capita GDP (GDP) are exogenous controls. Consistently with the idea of agglomeration economies, the population density and the per capita GDP are expected to positively correlate with CITS, while AREA is expected to have a negative effect. Finally, I employed other controls. The first one is a citation benchmark value (CITSSEC) that controls for the truncation problem and for differences 7 Since there are strategic reasons to apply for individual patents even if the innovations are the output of a joint-project, MAPPL might underestimate the actual number of collaborations. 8 Compared to the number of chemical patents invented in the regions and to the number of chemical laboratories, these indicators are not endogenous with respect to the decision of the firms to locate in a region: i.e. the firms themselves do not determine the technological characteristics of the regions. The REGIO database (1999) does not provide information on the regional characteristics of Switzerland, Finland, Norway and Sweden.

M. Mariani / Research Policy 33 (2004) 1565–1582

1575

Table 3 Descriptive statistics: traditional chemicals and biotechnology (biotechnology in italics) Variable

Mean

Standard deviation Firm and project characteristics 1.16 1.54

Minimum

Maximum

0 0

13 11

CITS

0.58 0.84

SALES

17,574 10,889

14,895 13,575

4 4

87,542 79,643

R&D/SALES

0.06 0.09

0.10 0.11

0.002 0.002

3.12 0.85

TECHSPEC

307 44

351 59

0 0

1,230 252

TECHCOMP

1,073 506

1,174 971

0 0

6,264 4,502

INVENTORS

3.19 3.15

1.85 1.84

1 1

16 14

DLOC

0.62 0.59

0.49 0.49

0 0

1 1

MAPPL

0.06 0.10

0.23 0.31

0 0

1 1

REGHLABS

46 45

Regional characteristics 71 69

0 0

461 461

REGPATS

610 467

609 605

0.8 0

2,263 2,263

GDP

16.9 17.1

4.4 4.3

7.4 7.6

27.5 27.5

POP

0.70 0.67

0.87 0.92

0.01 0.005

5.93 5.73

AREA

6,207 6,499

3,981 5,574

97 97

35,291 55,401

SCIENCE

1.16 3.14

1.8 2.7

0 0

24 17

CITSSEC

1.44 1.70

0.58 0.67

0.13 0.07

4.02 3.49

Other controls

Number of observations: 4262 patents (525 in biotechnology and 3737 in traditional chemicals).

in the citation intensities over time and across sectors that are unrelated with the value of the patents (Hall et al., 2001). Differences in EPO practices over time and across technological fields, the natural increase in the number of citing patents, and the fact that patents applied in different years suffer in different degrees from truncation make it difficult to compare patents according to the total number of citations received.

CITSSEC solves the problem by weighting the number of citations of each patent by the average citation intensity of a group of patents with similar characteristics. Moreover, CITSSEC is a detailed control for technology-specific citation intensities as it uses a three-digit IPC classification that includes more than 150 chemical and chemical-related classes. I also control for the extent to which a patent is related to basic

1576

M. Mariani / Research Policy 33 (2004) 1565–1582

research. This is the meaning of SCIENCE defined as the number of citations made by each patent to the nonpatent literature like scientific journals, books and proceedings. Since more “scientific” work might be used in a large number of different applications, SCIENCE is expected to positively correlate with the probability of receiving forward citations. All the regressions also include: dummies for the country of the inventors and the country of the applicants to capture the effect of regional characteristics that are independent of the variation across countries; dummies for the year in which the patent was applied for; and dummies for the five technological sectors in which the patents are classified to control for technology-specific effects that are independent of the changes identified by CITSSEC. Table 3 provides the descriptive statistics of the variables used in the regressions. 4.2. Empirical results Table 4 shows the results of the econometric estimates. Since the database is constructed at the patent level, there is intra-group error correlation for patents developed by the same firm or in the same region, leading to heteroskedasticity in the regressions. Robust estimators are therefore included in order to produce robust standard errors. All the variables are in logs.9 In traditional chemicals the probability of producing valuable patents depends on firms’ internal R&D effort. The elasticity of CITS with respect to R&D/SALES is positive (0.151) and statistically significant. The probability of developing technological hits is higher also when a research project involves a large network of inventors and different organisations: the coefficient of INVENTORS and MAPPL are positive and statistically significant (0.200 and 0.278, respectively). By contrast, SALES, TECHCOMP and DLOC do not add anything to the expected value of traditional chemical innovations. Only the technological specialisation of companies is significantly and negatively correlated 9

I checked for over-dispersion in the data, which seems highly likely given the pattern of the data in Fig. 2. Both the value of α (significantly different from zero in all regressions) and the LR-χ2 -test of the null hypothesis that α = 0 suggest that there is over-dispersion in the data. Therefore, the negative binomial model fits better than the Poisson.

with CITS but, rather than the effect of the technological diversification of the companies, the negative correlation of TECHSPEC is suggestive of the high propensity to patent in the chemical sector. Traditional chemicals is indeed populated by large firms that have the financial strength to apply for patent protection not only for important innovations, but also for less valuable ones. As the number of patent applications in the specific sector raises their average quality decreases, leading to a lower expected number of CITS.10 The technological characteristics of the regions in which the inventors are located do not raise the probability of developing technological hits in traditional chemicals. This result holds both for the proximity to higher education laboratories and for the location in technologically-intensive regions: the coefficient of REGHLABS and REGPATS are not statistically significant. Since there is a large number of controls in these regressions, this suggests that apart from economies in R&D internal to the firm, the model of innovation that leads to high expected value patents in traditional chemicals is dominated by large firms that invest heavily in internal R&D activities and large scale projects, with no role for the spillovers from near-by research laboratories or from the general technological environment in which the research is conducted.11 Firm characteristics still matter in biotechnology. Specifically, technological specialisation matters: the estimated coefficients of TECHSPEC and TECHCOMP are 0.264 and −0.196, and they are statistically significant. The expected value of innovations also rises when firms carry out R&D projects in collaboration with other institutions. The coefficient of MAPPL is positive (0.539) and statistically significant. Moreover, consistently with the more basic nature of biotechnology research compared to traditional chemicals, SCIENCE is positive (0.302) and statistically significant. But the real distinct feature of biotechnology vis-`a-vis traditional chemicals is the importance of 10 Since traditional chemical companies also tend to cite themselves more than companies in biotechnology, the negative effect of TECHSPEC disappears when self-citations are included in CITS. 11 The regional characteristics are also jointly insignificant. In Table 4 I tested the unrestricted model via a likelihood ratio test against a restricted model with REGHLABS = REGPATS = GDP = POP = AREA = 0. The null hypothesis cannot be rejected.

M. Mariani / Research Policy 33 (2004) 1565–1582

1577

Table 4 Estimates of negative binomial regressions Traditional chemicals SALES R&D/SALES TECHSPEC TECHCOMP INVENTORS DLOC MAPPL

−0.002 (0.027)* 0.134 (0.043)*** – – 0.183 (0.062)*** – 0.271 (0.133)**

REGHLABS REGPATS GDP POP AREA

0.054 (0.041) −0.025 (0.058) −0.220 (0.236) 0.036 (0.089) 0.006 (0.101)

SCIENCE CITSSEC

−0.055 (0.056) 0.653 (0.224)***

Number of observations Log-likelihood Pseudo-R2 α

3518 −3460.4 0.0358 1.54 (0.12)

Biotechnology Firm and project characteristics −0.035 (0.056) −0.018 (0.116) – – 0.125 (0.119) – 0.504 (0.229)** Regional characteristics −0.061 (0.081) 0.303 (0.151)** −0.474 (0.526) −0.176 (0.185) −0.300 (0.210) Other controls 0.293 (0.116)** 1.447 (0.405)*** 497 −568.5 0.0797 1.04 (0.17)

Traditional chemicals

Biotechnology

0.019 (0.031) 0.151 (0.043)*** −0.093 (0.040)** 0.031 (0.037) 0.200 (0.066)*** 0.033 (0.072) 0.278 (0.131)**

0.037 (0.065) −0.079 (0.124) 0.264 (0.108)** −0.196 (0.074) *** 0.154 (0.129) 0.169 (0.168) 0.539 (0.232)**

0.052 (0.042) −0.021 (0.059) −0.231 (0.238) 0.041 (0.090) 0.010 (0.101)

−0.076 (0.081) 0.323 (0.154)** −0.373 (0.550) −0.186 (0.182) −0.274 (0.209)

−0.063 (0.056) 0.676 (0.223)*** 3494 −3439.5 0.0366 1.54 (0.11)

0.302 (0.111)*** 1.090 (0.404)*** 490 −559.9 0.0851 0.098 (0.17)

Dependent variable: number of citations received by the patent in the 5 years after the application date excluding self-citations (CITS). Sample: 4262 patents. Variables are in logs. Cluster-robust standard errors are in parentheses. All regressions include dummies for non-chemical companies, missing value for R&D and SALES, inventor country, applicant country, year of application and technological field. ∗ Significant at 0.1 level. ∗∗ Significant at 0.05 level. ∗∗∗ Significant at 0.01 level.

the regions. The technological environment in which the research is carried out influences the probability of developing biotechnology hits: after including extensive controls for firms, projects and regions, the net effect of REGPATS is positive (0.323) and statistically significant. Surprisingly, however, the expectation that the geographical proximity to university laboratories is correlated with the probability of inventing important innovations in biotechnology is not confirmed. This is probably due to the fact that the number of patents coming from a region is a proxy for the general scientific and technological environment in which the inventors are located, and measure the output of the research carried out by both public and private institutions in the area. These results, together with the maps drawn in Figs. 3 and 4 suggest that the regional effect is a “true” effect: although biotechnology and traditional chemical patents are produced to a good extent in the same regions, the proximity to external

sources of knowledge affects the probability of developing valuable innovations only in the former.12 Following Zucker et al. (1998a, 1998b) and Klepper and Sleeper (2002), an interpretation of these results could be that some regions are better at doing certain types of research activities because of the location of 12 In place of REGHLABS I alternately used the number of chemical laboratories, private laboratories and all laboratories located in the regions; in place of REGPATS I used the number of chemical patents invented in the regions. The results are consistent with those in Table 4: the number of chemical patents is positive and significant, while the number of laboratories of any type does not affect CITS. The results in Table 4 are also robust to the use of citations in all years after the application date and to the inclusion of self-citations in CITS. Finally, I checked the robustness of the results by using an Ordered Probit model, by controlling for heteroskedasticity with cluster-robust estimators for firms and regions alternatively, and with dummy variables for firms and regions. The significance of the estimated coefficients is sometimes smaller, but the results are consistent with those in Table 4.

1578

M. Mariani / Research Policy 33 (2004) 1565–1582

top universities or high-tech companies. They spawn a large number of R&D performing start-ups founded by the personnel employed in the top organisations, and produce a large number of patents together with high citation intensities in the areas. In this case spillovers are not produced by the technological environment as proxied by the number of higher education laboratories or the number of patents invented in the region, but come from the initial technological characteristics of the regions in which these firms are located. Still, this paper shows that this would happen in biotechnology and not in traditional chemicals. In short, the sources of spillovers could be generic knowledge spillovers, or they could stem from human capital mobility or the foundation of new firms. Whatever the sources, there are sectors in which such external factors are important, and others in which they are not.

5. Conclusions Firm competencies and regional characteristics are often discussed in the literature as sources of firms’ competitive advantages. By estimating how much of the value of an innovation depends on the characteristics of the organisation to which the inventors are affiliated, and how much it is affected by the characteristics of the location in which it is invented, this paper compares the firm and the geographical cluster as models for producing innovations. The results suggest that the effect of the regions in which the research is conducted differs in new research-intensive sectors like biotechnology vis-`a-vis more established research-intensive industries like traditional chemicals. Big innovations in biotechnology are likely to be produced by firms that are technologically specialised in the sector and that rely on knowledge spillovers from being located in a technological intensive region. By contrast, in the traditional chemical sectors, large established companies benefit from the internalisation of knowledge spillovers within the firm, and local knowledge spillovers do not add to the probability of inventing technological hits. There are some potentially unresolved issues on why geographical knowledge spillovers are important only in biotechnology. For example, because of its more practical and learning-by-doing nature, traditional chemical technologies might entail lower transferability through market-based mechanisms. Geo-

graphical spillovers should then play a prominent role because they allow for knowledge transmission via mobility of personnel, informal contacts, or the intermediating role of specialised suppliers. If so, my result about the insignificant effect of regions in traditional chemicals would be surprising. One explanation might be the different industrial organisation of the two sectors. In this respect, the smaller start-up firms in biotechnology imply greater geographical spillovers compared to the large established companies in traditional chemicals because of the lower internalisation of knowledge activities in smaller concerns. However, I control for company size in the regressions, and its insignificant impact suggests that this is not key for explaining differences between the two sectors. Another possibility is that patents have different connotations in biotechnology compared to traditional chemicals. For example, patents might be more important in biotechnology, as a source of knowledge, and hence patent citations could count more for important innovations than in chemicals. These are issues that require further thoughts. As discussed in Section 2.1, I interpreted my results as indicating that local knowledge spillovers may be more prominent in research-intensive sectors that are in the early stages of the industry life cycle. Many studies have examined the importance of the firm in developing innovations. Others have studied the geographical cluster, and found that geographical proximity is also important for innovation. This paper suggests that firm and regional forces lead to different models: one that relies on factors internal to the organisation, and the other founded also on the relationships among firms and institutions in a geographical cluster. The contribution of this paper is that it confirms the existence of such alternative models of innovation by using a novel and extensive database along with several controls.

Acknowledgements I thank Ashish Arora, Alfonso Gambardella, Marco Giarratana, John Hagedoorn, Bronwyn Hall, Dietmar Harhoff, Steven Klepper, Pierre Mohnen, Pierre Regibeau and three anonymous referees for helpful comments. I also thank Bart Verspagen for providing me with patent citations, and Fabio Pammolli for providing me with firm-level data that I could not find

M. Mariani / Research Policy 33 (2004) 1565–1582

from other sources. Useful comments were made by the participants in the EPIP conference (European Policy for Intellectual Property, Munich, 2003) and in the Innogen Workshop (Innovation, Growth and Market Structure, London, 2003). I am grateful to Rossana Pammolli for developing the correspondence table between IPC classes and the five sectors used in this paper. Support from the European Commission TMR “Marie Curie Fellowship” (Grant #HPMF-CT2000-00694) and from the European Commission Key Action “Improving the Socio-Economic Knowledge Base” (Contract #HPSE-CT-2002-00146) is acknowledged. The usual disclaimers apply.

Appendix A. Data collection and list of variables For each of the 4262 patents I collected information on the innovation, such as the name and address of the applicants, the number and address of the inventors, the year of application, and the number of citations received by the patents after the application date. An expert pharmacologist read the description of the three-digit International Patent Classification codes and developed a correspondence table between the IPC codes and five major technological sectors: biotechnology, materials, organic chemistry, pharmaceuticals and polymers. Each sample patent was classified in one of these five technological sectors. The zip-code contained in the address of the inventors was used to assign each patent to the specific NUTS region in which it was invented at the most disaggregated NUTS3 and NUTS2 level (see Appendix B).

1579

Data about the characteristics of the regions such as the GDP, the population, the size, the total number of patents invented in the area are drawn from the EUROSTAT REGIO database (1999). From the European R&D database (Reed Elsevier Publisher, 1996) I downloaded a stock of about 20,000 R&D laboratories located in Europe as for December 1995, and classified them as private laboratories, higher education laboratories (i.e. universities), government laboratories, and chemical laboratories if they focus on chemical research. Each laboratory was also assigned to its NUTS region. By using the Who Owns Whom database (1995) I merged the parent and affiliate names of the 4262 applicants under the same name, and I collected company data from different sources. For patents with multiple applicants (7.6%) I collected data on the first applicant of the list. First, Aftalion (1991) lists the top 250 chemical companies worldwide in 1988, and provides firm-level information. This ensured that I covered the most important chemical firms in the sample. I complemented these data with sales and R&D information from Compustat (1999) and from an Internet search for some smaller concerns. In the end I was unable to find information on a tail of applicants covering 852 patents in the sample. These are fairly unknown firms with one or two patents in the sample, and their distribution across regions and technological classes is not biased in any particular direction. Finally, for each applicant I collected the number of EPO patents filed in the 5 years before the date of each patent. This information was used to develop a measure of firm competencies and firm technological specialisation in the period before the patent application.

List of variables

CITS SALES R&D/SALES TECHSPEC

Firm and project characteristics Dependent variable. Number of citations received by the patent in the 5 years after the application date, excluding self-citations Company sales in 1988 (millions of 1988 US$) Company R&D spending over sales in 1988 Number of EPO patents in the same technological sector of each patent in the sample (biotechnology, materials, organic chemistry, pharmaceuticals, and polymers) filed by the firm during the 5 years before the application date

1580

M. Mariani / Research Policy 33 (2004) 1565–1582

TECHCOMP

Number of EPO patents filed by the firm in the 5 years before the date of each patent in the other four technological sectors

INVENTORS DLOC MAPPL NOCHEM

Number of inventors that collaborate to develop the innovation Dummy. It takes the value 1 if all the inventors listed in the patent are located in the same region; 0 if at least one inventor is located in a different NUTS region Dummy. It takes the value 1 if there are multiple applicants; 0 otherwise Dummy for non-chemical companies

MISSING

Dummy for missing values on SALES and R&D

REGHLABS REGPATS GDP

Regional characteristics Number of higher education laboratories located in the region (stock in 1995) Number of patent applications in all sectors invented in the region (units, average 1987–1996) Regional per capita gross domestic product in millions of purchasing power parity corrected for inflation (average 1987–1996)

POP

Population density of the region (thousands, average 1987–1996)

AREA

Area of the region (km2 )

SCIENCE CITSSEC

Other controls Number of citations made by the patent to the past scientific literature Citation intensity of patents computed as the average number of citations received by patents applied in the same year and technological class of the patent application (three-digit IPC classes)

INVCY APPLCY YEAR SECTOR

Dummy for the country of the inventors (At, Be, Ch, De, Dk, Es, Fi, Fr, Gr, Ie, It, Lu, Nl, Se, Uk) Dummy for the country of the applicant firm (At, Be, Ch, De, Dk, Es, Fi, Fr, Gr, Ie, It, Lu, Nl, Se, Uk, Jp, Us, others) Dummy for the year of application (1987–1996) Dummy for the sector in which the patent is classified: biotechnology, materials, organic chemistry, pharmaceuticals and polymers

Appendix B. Regional classification used in the paper, and criteria for matching inventors’ addresses and NUTS regions The Nomenclature des Unit`es Territoriales Statistiques (NUTS) is a Eurostat classification that subdivides the European Union in groups of regions (NUTS1), regions (NUTS2) and provinces (NUTS3). In order to have homogeneity in the size of the regions I used the NUTS3 regions for Austria, Denmark, Spain, Finland, France, Italy and Sweden, and the NUTS2 classification for Belgium, Germany, Greece,

The Netherlands, and the UK. Luxemburg, Ireland and Switzerland were considered as a whole. To assign inventors’ addresses to NUTS regions I adopted the following criteria. When there is only one inventor in the patent or when all the inventors are located in only one region, the zip-code contained in their address is used to identify the NUTS region in which they developed the innovation. For patents invented by multiple inventors located in different regions (33.5% of the total sample), I assign the patent to the region in which the largest share of inventors is located. When the share is 50% of the inventors in one

M. Mariani / Research Policy 33 (2004) 1565–1582

region, and the other 50% in another one, I assign the patent to the first inventor of the list. Even if this implies some degree of arbitrariness, the problem is confined to 24% of the patents for which the assignment is based on the region with the highest share of inventors, and 9.5% for the 50–50 rule. I also control for these patents in the regressions by including a dummy for the co-localisation of the inventors in the same region, as opposed to having at least one inventor in a different region. In a few cases in which the inventors are located on the border between different regions they are classified as being located in only one region according to the rule described above. The list of NUTS regions used in this paper is obtainable from the author on request.

References Aftalion, F., 1991. A History of the International Chemical Industry. University of Pennsylvania Press, Philadelphia, PA. Arora, A., Fosfuri, A., Gambardella, A., 2001. Markets for Technology. MIT Press, Cambridge, MA. Arundel, A., Geuna, A., in press. Proximity and the use of public science by innovative European firms. Economics of Innovation and New Technology. Audretsch, D.B., Feldman, M.P., 1996. Knowledge spillovers and the geography of innovation and production. American Economic Review 86 (3), 630–640. Breschi, S., 2001. Spatial patterns of innovation: evidence from patent data. In: Gambardella, A., Malerba, F. (Eds.), The Organisation of Innovative Activity in Europe. Cambridge University Press, Cambridge, MA. Cani¨els, M.C.J., 2000. Knowledge Spillovers and Economic Growth: Regional Growth Differentials Across Europe. Edward Elgar, Cheltenham. Cantwell, J.A., Iammarino, S., 2001. EU regions and multinational corporations: change, stability and strengthening of technological comparative advantages. Industrial and Corporate Change 10 (4), 1007–1037. Cantwell, J.A., Iammarino, S., 2003. Multinational Corporations and European Regional Systems of Innovation. Routledge, London. Christensen, C.M., 1997. The Innovator’s Dilemma, When New Technologies Cause Great Firms to Fail. Harvard Business School Press, Boston, MA. Dosi, G., Freeman, C., Nelson, R., Silverberg, G., Soete, L., 1988. Technical Change and Economic Theory. F. Pinter, London. Griliches, Z., 1988. Patent statistics as economic indicators: a survey. Journal of Economic Literature 28, 1661–1707. Griliches, Z., Pakes, A., Hall, B., 1987. The value of patents as indicators of inventive activity. In: Dasguptam, P., Stoneman, P. (Eds.), Economic Policy and Technological Performance. Cambridge University Press, Cambridge, MA.

1581

Hall, B., Jaffe, A., Trajtenberg, M., 2001. The NBER patent citations data file: lessons, insights and methodological tools. NBER Working Paper No. 8498. Hall, B., Jaffe, A., Trajtenberg, M., in press. Market value and patent citations. Rand Journal of Economics. Harhoff, D., Reitzig, M., 2004. Determinants of opposition against EPO patent grants — the case of biotechnology and pharmaceuticals. International Journal of Industrial Organization 22, 443–480. Harhoff, D., Scherer, F., Vopel, K., 1999. Citations, family size, oppositions and the value of patent rights: evidence from Germany. Mimeo, February. Jaffe, A., 1986. Technological opportunity and spillovers of R&D: evidence from firms’ patents profits and market value. American Economic Review 76 (8), 984–1001. Jaffe, A., Trajtenberg, M., Henderson, R., 1993. Geographic localization of knowledge spillovers as evidenced by patent citations. Quarterly Journal of Economics 63 (3), 577–598. Klepper, S., 1996. Entry, exit, growth, and innovation over the product life cycle. American Economic Review 86 (4), 562–583. Klepper, S., 2001. Employee startups in high tech industries. Industrial and Corporate Change 10, 639–674. Klepper, S., Sleeper, S., 2002. Entry by spinoffs. Mimeo. Lanjouw, J.O., Schankerman, M., in press. Patent quality and research productivity: measuring innovation with multiple indicators. Economic Journal. Levinthal, D.A., March, J.G., 1993. The myopia of learning. Strategic Management Journal 14, 95–112. Marshall, A., 1920. Principles of Economics, 8th ed. MacMillan, London. Maurseth, P.B., Verspagen, B., 2002. Knowledge spillovers in Europe: a patent citations analysis. Scandinavian Journal of Economics 104 (4), 531–545. Nelson, R., 1995. Recent evolutionary theorizing about economic change. Journal of Economic Literature 33, 48–90. Nelson, R., Winter, S., 1982. An Evolutionary Theory of Economic Change. Harvard University Press, Cambridge, MA. Orsenigo, L., 1989. The Emergence of Biotechnology: Institutions and Markets in Industrial Innovations. F. Pinter, London. Patel, P., Pavitt, K., 1997. The technological competencies of the world’s largest firms: complex and path dependent, but not much variety. Research Policy 26, 141–156. Pavitt, K., 1987. On the Nature of Technology. University of Sussex, SPRU, Brighton. Porter, M.E., 1998. Clusters and the new economics of competition. Harvard Business Review, 77–90. Powell, W., Kaput, K., Smith-Doerr, L., 1996. Interorganisational collaboration and the locus of innovation: networks of learning in biotechnology. Administrative Science Quarterly 41 (1), 116–145. Rosenberg, N., 1998. Technological change in chemicals: the role of university–industry relations. In: Arora, A., Landau, R., Rosenberg, N. (Eds.), Chemicals and Long-term Economic Growth: Insights from the Chemical Industry. Wiley, New York. Saxenian, A., 1994. Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Harvard University Press, Boston, MA.

1582

M. Mariani / Research Policy 33 (2004) 1565–1582

Scherer, F., Harhoff, D., Kukies, J., 2000. Uncertainty and the size distribution of rewards from technological innovation. Journal of Evolutionary Economics 10, 175–200. Scherer, F., Harhoff, D., 2000. Technology policy for a world of skew-distributed outcomes. Research Policy 29, 559–566. Swarm, P., Prevezer, M., Stout, D. (Eds.), 1998. The Dynamics of Industrial Clustering. International Comparisons in Computing and Biotechnology. Oxford University Press, Oxford, UK. Teece, D.J., Pisano, G., Shuen, A., 1997. Dynamic capabilities and strategic management. Strategic Management Journal 18, 509–534. Trajtenberg, M., 1990. A penny for your quotes: patent citations and the value of innovation. Rand Journal of Economics 21 (1), 172–187.

Verspagen, B., 1997. European regional clubs: do they exist and where are they heading? On economic and technological differences between European regions. In: Adams, J., Pigliaru, F. (Eds.), Economic Growth and Change: National and Regional Patterns of Convergence and Divergence. Edward Elgar, Cheltenham. Wlash, V., 1984. Invention and innovation in the chemical industry: demand-pull or discovery-push? Research Policy 13, 211–234. Zucker, L., Darby, M., Armstrong, J., 1998a. Geographically localized knowledge: spillovers or markets? Economic Inquiry 36, 65–86. Zucker, L., Darby, M., Brewer, M., 1998b. Intellectual human capital and the birth of U.S. biotechnology enterprises. The American Economic Review 88 (1), 290–306.