Innovation and local externalities: what does geography of innovation say? Corinne Autant-Bernard, Nadine Massard* CREUSET, University Jean Monnet Saint-Etienne, 6 rue Basse des Rives, 42 000 Saint-Etienne, France.
Document de travail 2001
Abstract : Technological externalities are often set at the heart of the understanding of the role of proximity in the economic process, and particularly in the innovation process. This field refers to both endogenous growth theories and economic geography. Nevertheless, such an association between externalities and proximity is not obvious and deserves to be specified. The empirical literature on the geographic dimension of technological externalities has been significantly growing in the last ten years. Through a comparative approach of econometric studies, in the geography of innovation, we are analyzing here the different features of spillover modelisation. We underline the difficulties met by these econometric studies in co-modelising externalities and the local dimension. On this basis, we then discuss how to make comparisons between the results obtained in various contexts. Keywords : Technological externalities, Knowledge, Geographic proximity, local interactions, econometric studies. JEL Classification:
1. Introduction :
Most of the studies considering the local dimension of economic phenomena rely on the concept of spillovers. Despite their integration in different analyses, externalities and the idea of proximity are often associated. For instance, * Corresponding authors : email address:
[email protected],
[email protected] 1
the analyses in terms of networks or industrial districts are based on the presence of externalities to justify the role of proximity. In future, according to these approaches, the advantages linked to the local contexts would stem from positive spillovers. In another prospect, the models of technological competition1 have also connected externalities with proximity, in which local spillovers are used to account for the influence of vicinity in the strategies of adoption of technology. Despite a focus on pecuniary externalities, the local dimension of externalities is also present in the studies of "economic geography" (Krugman, 1991), allowing agglomeration economies to understand the concentration of industrial activities. Spillovers would then have a local character insofar as they stem from a close-by localization. In the theoretical context of endogenous growth, knowledge externalities are powerful factors of accumulation and, as such, set at the heart of the analysis of uneven regional patterns of growth (Grossman and Helpman, 1994, 1995). The accumulation process is showed as strongly dependent from the institutional elements modulating the public character of knowledge (as intellectual property rights for instance). However, the justifications of the geographic frontier of diffusion of externalities are hardly explained. Thus, it seems interesting to research an explanation for the connection between externalities and geographic proximity. Two main elements in the literature account for the presence of spillovers on the local scale: i) the interactive feature of the production process, especially concerning the innovation stage (which would require frequent exchanges generating spillovers), is favored by low geographic distance ; ii) the problems of appropriability which almost bring knowledge to the level of public wealth, propitious to generate spillovers (Arrow, 1962 ; Griliches, 1979). Thanks to the setting of local infrastructures which favors the appropriation of knowledge, geographic proximity then becomes a vector of externalities (Teece, 1992). As a consequence, spillovers would first have an impact on the innovation process, insofar as the latter requires interactions and as the question of knowledge is especially crucial in the innovation process. The notion of spillovers, ie of technological externalities, is then set at the heart of the understanding of the spatial dimension of innovation-based growth process. 1
See in particular Arthur (1994), Dalle (1995), An and Kiefer (1995), Cowan and Cowan (1997). 2
Paradoxically, despite the strong theoretical conclusions stemming from the hypotheses of the local dimension of externalities, no particular need of systematic empirical confrontation to data appears since the extreme end of the 80’s. Such a need emerged finally under the pressure of technological globalization and because of the role of the new Information and Communication Technology on the spatial diffusion of knowledge. Indeed, at this point, numerous voices raised to indicate that the association between externalities and proximity is not obvious. For instance, the rich literature on international spillovers2 attests that technological externalities do not take place at the local level only. Thus, the idea of a relation between technological externalities and proximity cannot continue being simply postulated. It appears necessary to try to assess empirically whether technological spillovers have a geographic dimension. This is the concern of the econometric studies on local externalities. These works, can be regrouped under the title "geography of innovation", similarly to "economic geography" from which they are inspired. They aim at testing empirically the assumption of the geographic dimension of technological externalities. To do so, the implemented methods are numerous. Most of them rely on the functions of knowledge production, but with very different ways to consider geographic spillovers. Such a variety reflects the difficulty to point out to both the phenomena of externalities and their geographic dimension. Associated to this, a first assessment of the results obtained after about ten years of studies in that field, suggest the necessity to improve the analytical base of empirical modelisation in order to better specify the relations between externalities and geographic proximity. The question of the local dimension of externalities is not autonomous. It can no longer be treated independently from a more precise comprehension of the process of knowledge diffusion allowing us to enter the "black box" of the ways and conditions of transmission of technological externalities. The paper is organized as follows. The second section analyses the main features of geographic spillover modelisation. It is particularly underlined that even if some studies do modelise the geographic dimension, they offer a rather poor representation of technological spillovers. Conversely, when the works really intend to modelise externalities, this is at the expense of an accurate measure of geographic proximity. The third section presents a comparison of the main results
2
See particularly Branstetter (1996), Coe and Helpman (1995) or Lichtenberg, Van Pottelsberghe (1996). 3
stemming from studies using data from different countries; various scores of ambiguities are underlined and we suggest ways to pass beyond them in specifying the reasons of the local dimension of spillovers. Final remarks are in section four.
2. The econometric methods to assess the geographic dimension of technological spillovers.
The econometric analyses on the local character of technological spillovers can be split up into four kinds of methodology. The most direct approach probably consists in researching indicators of technological externalities. This has been proposed by the studies based on patent citations (section 1). However, the use of patent citations as "paper trail" of spillovers is not without raising problems. Globally, it turns out to be difficult to find a relevant indicator of spillovers, insofar as -according to their definition- the latter cannot be directly measured. That is why, rather than researching a paper trail of innovations, it is possible to study spillovers through their effects. Within this prospect, some studies explain the concentration of innovating activities, grounding on the idea that if spillovers are localized, then some regions may benefit from the cumulative effects, fostering the production of innovations (section 2). We shall show why such a method does not seem to be thoroughly satisfactory. We shall particularly demonstrate that considering concentration as an innovation-explaining variable and not as the main goal of the study appears more relevant. This is what has been proposed by the analyses based on a function of knowledge production integrating an indicator of geographic proximity (section 3). This methodology first suggested for the study of the local dimension of spillovers, has been progressively sharpened. However, the representation of spillovers on which it relies on is not accurate enough. A last perspective has made it possible to have more visible spillovers thanks to a better modelisation of technological externalities and of local interactions (section 4).
4
2.1. The research of spillover trails. The most direct way to point out externalities consists in finding a measure. In this prospect, patent citations are used to test the geographic dimension of technological externalities (Jaffe, Trajtenberg and Henderson, 1993; Almeida and Kogut, 1997a, 1997b; Maurseth and Verspagen, 1999; Verspagen and Schoenmakers, 2000)3. The system of patents provides indications about the knowledge preceding an innovation4, and particularly about the previous patents linked to the new invention. Those citations are used to define clearly the specific contribution of the new patent. The citations contained in a patent can thus be considered as a stock of previous knowledge. Firms and Universities use this stock of knowledge so as to innovate. The public feature of this knowledge leads to the consideration that patent citations are a means of identifying technological spillovers, and especially their geographic dimension. Indeed, if spillovers are localized, a firm will be more up to use the stock of local knowledge. A link should thus be discovered between the localization of the new patents and which of the patents they refer to. However, for this link to be relevant, the fact that technological activities can be geographically concentrated beforehand must be taken into account. That is why, Jaffe, Trajtenberg and Henderson (1993), followed by Almeida and Kogut (1997a and 1997b), have built a control sample of patents5, designed to allow the comparison between the probability of a patent to be situated at a given place and its probability to be situated at the same place as the patent to which it refers.
3
Other studies such as Jaffe and Trajtenberg’s (1996), Henderson and Cockburn’s (1998) rely on citations (of patents or publications) to measure spillovers. However, they do not aim at explaining the local dimension of these phenomena. For this reason, they are not presented in this paper. 4 Those indications are actually divided into two groups : the references to the scientific literature and the references to other patents. But it seems that only references to previous patents have been used to test the geographic dimension of knowledge spillovers. 5 It is built by associating, with each patent (pi) referring to another (pc), a patent (pj) filed in the same technological field and at the same period, but which does not cite the same originating patent pc. Each pair of patents (pi, pj) allows to compare the localization of the control patent (pj) with that of the originating patent cited (pc) in the tested patent (pi). The local dimension of spillovers is then pointed out if the frequency obtained for the citations is significantly stronger than the frequency obtained for the control sample. 5
From the samples of citations and of control patents, the aim is then to test two hypotheses : the null hypothesis H0 : Pcit = Pcon and the alternate hypothesis H1 : Pcit > Pcon with Pcit being the probability of the citing patent to be localized at the same place as the originating patent and Pcon being the probability of the control patent to be localized at the same place as the originating patent. The observations of Jaffe, Trajtenberg and Henderson on the one hand and that of Almeida and Kogut on the other, lead to reject H0 and then imply that citations are localized. Thus, numerous patents are often filed in the same geographic area (countries, States or metropolitan areas) as the patent they refer to. Almeida and Kogut (1997b) have accounted for this phenomenon thanks to the presence of labor market local networks. The innovation courses would tend to be localized since knowledge would be transmitted from one firm to another thanks to the professional mobility of engineers. Now, this mobility would be geographically bounded due to the existence of labor market networks. Actually, Almeida and Kogut have noted that the regions of the most localized citations are also those of the most important professional mobility. Moreover, the probability to cite a patent is all the stronger for a firm as it has hired the engineer at the origin of this patent. The localized character of spillovers would thus stem from the tacit dimension of knowledge. Ideas would not be directly available to everyone, but on the contrary, they would (at least partly) be incorporated into men. Spillovers would then be diffused by means of localized labor transfers. More recently, the scientists of the MERIT have used patent citations in the European context6. Maurseth and Verspagen (1999) analyze the influence of geographic proximity on patent citations between European regions. Verspagen and Schoenmakers (2000) make a similar study but at the firm level instead of regions. This provides a larger number of observations and a finest geographic scale. The results confirm those obtained for the American case. They highlight the positive impact of the geographic proximity on the probability to cite a patent. However, the local dimension is likely to be overestimated. These two latter studies do not use control-sample, and patent-citations inside firms are introduced.
6
Maastricht Economic Research Institute on Innovation and Technology. 6
More generally, extrapolating from the results obtained for patent citations to the spillover ways of transmission is tricky. It supposes patent citations to be a good indicator of spillovers. Now, the relevance of the link between citations and spillovers can be challenged. Some citations may not point out spillovers and conversely, patent citations do not necessarily reflect every spillover. Actually, every spillover is not included in the patent citations; and this for two reasons. First, patents are partial indicators of innovation. They do not reflect the whole innovation. In particular, they lead to lay the stress on some sectors of activities insofar as there are more patented innovations in some fields than in others. Furthermore, they particularly point out the results of applied R&D, while fundamental R&D leads more rarely to patent filings; now, this is probably fundamental research which generates the most spillovers. Moreover, even for patented innovations, some spillovers cannot be identified by means of citations. The knowledge used to realize an innovation is not only the information contained in the former patents. The previous knowledge is only partly revealed by patent citations. For instance, another part stems from the information contained in scientific newspapers or works. More globally, there is a contradiction between the selected indicator (patents) and the observed phenomenon (spillovers). Indeed, Almeida and Kogut have connected the local character of spillovers to the tacit dimension of knowledge. Now, patents are, on the contrary, the typical reflection of codified knowledge. Moreover, citations do not always allude to spillovers. Indeed, every patent citation is not from the inventor. A third person entrusted with studying the patent makes some citations. This is generally a specialist of the technological field in which the patent has been filed. But the citations determined by a third person do not necessarily imply actual relations between the new patents and the patents to which they refer. At the invention time, the inventor was not necessarily aware of the cited patent existence, and even if he was, he did not necessarily use it for his invention. In this case, the citation does not reflect a spillover. A recent study made by Jaffe, Trajtenberg and Fogaty (2000) is specifying this gap between patent citations and spillovers. Thanks to a sample of two thousand inventors, they estimate that around one half of patent citations does not correspond to real knowledge flows. Then, the use of patent citations as paper trails of spillovers does not turn out to be thoroughly satisfactory. This explains that other studies have 7
headed towards alternative methods. The latter give up the idea of spillover trails and opt for an analysis of externalities through their effects.
2.2. The study of an effect of local spillovers: the geographic concentration of innovation. If spillovers have a local dimension, then one of their effects will be to favour the regrouping of the innovating activities in some places; proximity being necessary to benefit from the spillovers generated by other firms or by public organizations. Thus, if the distance has an impact on spillovers, then the geographic concentration of innovations should be observed. The study of the concentration of innovating activities then appears as a means to test spillovers geographic dimension. This approach has been adopted in several studies, based either on a measure of a concentration index (Audretsch and Feldman, 1994, 1996) or on spatial econometric tools (Caniëls, 1997). The first ones starts from the following statement: there is a connection between the concentration of production and that of innovation, but nevertheless the concentration of innovation varies depending on industries, independently of the production localization. The aim is consequently to account for this part of the concentration of innovation which does not stem from the concentration of production. In this prospect, the geographic concentration of innovating activities is assessed thanks to Gini's coefficients, weighted by the part of the economic activities set up in each geographic area (cf. Krugman, 1991c). The main tested equation is the following:
(1)
Ginii1 = γ1 Giniva1 + γ2 Indrdi + γ3 Skilledi + γ4 Unirei + ε.
Thus, the concentration of innovation (Ginii) would not only be explained by the concentration of production (Giniva), but also by elements fostering the production and transmission of knowledge: industrial R&D (Indrd), the share of skilled labor (Skilled) and academic research (Unire). The econometric results point out that all the coefficients are positive and significant. Thus, although the effect connected with the localization of production is neutralized, the factors at the root of new knowledge favour the concentration of innovation. Audretsch and
8
Feldman have deduced that the localization of production accounts less for the concentration of innovation than technological spillovers. But the observation of the polarization of innovative activities, thanks to concentration index only, accounts for the concentration inside regional boundaries. It does not tell us if this polarization produces spillovers on boundering areas. From this observation, Caniëls (1997) is using spatial econometric tools in order to estimate the potential inter-regional spillovers of innovation. She measures the spatial autocorrelation thanks to Moran’s statistic7. This method indicates if the spatial structure of innovative activities is random or not. The study of European regions shows a positive autocorrelation. It means that innovative activities are concentrated in few areas, and if an area has a high (respectively low) level of innovation, other strongly (weakly) innovative areas surround it. For Caniëls (1997), this spatial concentration results from the presence of local knowledge spillovers. However, it is not obvious that the observation of the degree of the concentration of innovation does reflect the existence of spillovers. Actually, explaining innovation concentration by the existence of knowledge spillovers supposes to associate the agglomeration of R&D activities with spillovers, i.e. to assume implicitly that technological spillovers between geographically close firms are at the root of agglomeration economies. Now, the study of Podolny and Shepard (1998) tends to demonstrate that knowledge spillovers, whilst having a geographic dimension, do not generate agglomeration economies. Therefore, the existence of local spillovers could not be deduced from the only observation of geographic concentration. Furthermore, those studies on innovation concentration appear essentially static. They aim at explaining the localization of innovating activities at a given time and connect localization with phenomena of local technological externalities. But one does not know whether concentration has a positive impact on the capability of a firm to innovate. Now, this is interesting to know whether positive or negative is the effect generated by knowledge spillovers on innovation. Technological change is actually one of the most powerful vector of growth. Thus, explaining the factors of innovation growth brings elements to understand the global economic dynamics. Then, studies focusing on the performances of a firm 7
This approach has recently been applied to several countries. See for instance the studies of Baumont and Legallo (2000) in the French case and of Paci and Usai (2000) in the Italian context. 9
in terms of innovation seem to be less static and are part of a wider approach. This is the case with the methods based on a function of knowledge production. The latter emphasizes either geographic coincidence, or local interactions.
2. 3. The measure of geographic coincidence. The aim is to test the relation between the capability of a firm to innovate and their proximity with respect to other firms or universities. In order to do so, an index of geographic coincidence is integrated in a function of knowledge production. Jaffe (1989) proposes the first index. The question is then to measure the correlation between the innovation output of a geographic area and the proximity of firms and universities in this geographic area. To do so, Jaffe is using a coefficient of correlation between the localization of public and private R&D laboratories : (2)
Ci = Σs Uis.TPis / [Σs Uis2]1/2 . [Σs TPis2]1/2
with TPis being the number of private R&D researchers in a metropolitan area situated in a State i ; and U being academic research spending. The correlation between academic research spending and the number of R&D workers is then calculated in accordance with the geographic areas. A State having a strong correlation coefficient is a State in which private R&D and public research are localized close by (i.e. in the same metropolitan areas). If spillovers have a geographic dimension, such a State should have a stronger capability to innovate. So as to test this hypothesis, the index is integrated as an explanatory variable in a function of knowledge production (Griliches, 1979) :
(3)
log (Pik) = ß1k log (Iik) + ß2k log (Uik) + ß3k [log (Uik) . log (Cik)] + eik
with P being private patents ; I private R&D spending ; e, a random perturbation ; i the observation scale and k the technological area (i.e. the industrial sector). The interpretation is the following : if innovation increases when geographic coincidence is high, then the R&D efforts of the ones are very likely to have a local impact on the others’ innovation. But Jaffe’s results are not very significant: if the positive influence of the American academic research on innovation is really pointed out, the role of 10
proximity is not demonstrated. In return, more significant results are obtained by changing the innovation output indicator. With Jaffe (1989), innovation is measured by means of patents, but this indicator is very partial. That is why Acs, Audretsch and Feldman (1991) have proposed another means to measure the innovation output : the actual introduction of innovations on the market (listed by the US Small Business Administration, for 1982). This change of indicator brings clearer results as for the impact of proximity on the capability to innovate, ß3 becoming significant. Moreover, it leads to a wider consideration of the local factors fostering innovation. Actually, the observation is situated far more downstream from the innovation production process : the reasoning turns on marketed innovations, i.e. innovations really introduced on the market, whereas patents concern a stage preceding the innovation development. Consequently, local spillovers widely go beyond the only relations between firms and universities8. In particular, the marketing stage requires the intervention of more various components. Feldman (1994) inserts then new factors : her model connects the level of output in the innovation of firms with the local factors capable to explain it. Thus, it is shown that a firm innovates all the more as it is in the vicinity of some specific factors. According to Feldman, those local factors, producing spillovers, make up the technological infrastructure, which contends: academic research, industrial R&D, the presence of related firms and the networks of service industries9. However, the increase in the number of factors is realized at the expense of an accurate measure of the space dimension. Feldman has not proposed a measure of the geographic proximity between innovating firms and the service industries or the bound firms which is comparable to the index of geographic coincidence between universities and innovating firms proposed by Jaffe (1989). The local dimension is simply measured by the localization inside one State. Now, Jaffe has demonstrated that the State is not a relevant geographic scale to point out spillovers. Complementary studies will make it possible to extend the analysis in this direction.
8
This is also certainly one of the reasons why the results of Acs, Audretsch and Feldman (1991) are more significant than those of Jaffe (1989). 9 The empirical observations allow not only to validate this idea, but also to point out a self - reinforcement of this infrastructure, the presence of one of those factors in a geographic area being strongly related to the presence of other factors. This result tends to confirm the analyses in terms of path dependence (Arthur, 1994). 11
Acs, Anselin and Varga (1997) improve Jaffe’s study in two ways. The first one, which we will deal with in this section, consists in suggesting new measures of geographic coincidence, at the States level. The second part of their study relies on another spatial level of analysis, looking at metropolitan areas instead of States. The method used in this case moves from the idea of geographic coincidence introduced by Jaffe, and insists on local interactions. For this reason, this second part of their study will be presented in the next section. Grounding on the approaches dealing with spatial econometrics, Acs, Anselin, and Varga (1997) add three other indicators of geographic coincidence to the index calculated by Jaffe: - a measure of gravity10 : the aim is to measure the distance between the counties where private R&D is localized and the counties where universities are localized. The lower is the average distance for the whole of the State counties, the stronger is geographic coincidence, and then the higher should be the innovation output of this State11. - two covering indicators : the aim is to take into account the effects -on the R&D of an industrial county- of public research realized in an area located 50 or 75 miles around the county12. The major interest of those new measures of geographic coincidence relies on their possible application to other variables than academic research. They could particularly be used so as to account for the impact of geographic proximity between innovating firms or between innovating firms and service industries. Globally, the use of an indicator of geographic coincidence provides with a direct measure of the role of geographic proximity and offers a consistent framework to analyze geographic spillovers. In return, the indications provided by such a method are to be considered cautiously. Indeed, just like for the measures of innovation concentration, nothing is said concerning the relations between firms or between firms and universities. The only question is geographic proximity. This analysis does not lead to a real observation of spillovers since it does not wonder whether there are actually local interactions. The risk is to assimilate too quickly 10
Spatial economics have been using the measure of gravity for a long time (see for instance Isard, 1960). 11 The index of geographic coincidence is given by : GRAVi = Σj Uj / (dij)², with dij the distance between county i and county j. 12 In this case, the index of geographic coincidence is calculated by : COVi = Σj δij .Uj, with δij= 1 if dij < 50 (or 75) miles, and δij= 0 otherwise. 12
proximity and externalities whereas the agglomeration of research activities does not necessarily mean that they are linked together. Zucker, Darby and Armstrong (1994) have particularly demonstrated that if interactions are taken into account, this is not geographic proximity that matters, but agents’ local relations. A fourth research direction has thus headed this way.
2. 4. The modelisation of spillovers and the consideration of local interactions. Should the activity of research generate spillovers, then the research effort of a firm or of a university is partly profitable to others. From this principle, several studies have observed the geographic character of knowledge externalities. Their interest is to give a real measure of spillovers by studying whether the R&D activities of close firms or universities have an impact on the capability of a firm to innovate. In order to do so, the elasticity of the innovative output of each firm (or group of firms) relatively to the R&D spending of other firms or universities situated in the same geographic area (or in an area in the vicinity) is estimated. Within this prospect, Acs, Anselin and Varga (1997) have followed Jaffe’s method (1989), but reasoning on the scale of metropolitan areas and not on the scale of States, as previously mentioned. This has allowed them to substitute another measure of space dimension for the indicator of geographic coincidence. Then, they have taken into account (in the equation of knowledge production) public and private research realized in and around the metropolitan area. The capability to innovate of a metropolitan area would thus be linked both to the R&D internal effort, but also to the repercussions generated by the research spending of the counties in the vicinity13. The results are significant, which would indicate that R&D "spills over". Thus, by reasoning on the scale of metropolitan areas rather than on the scale of States, and by using space variables, Acs, Anselin and Varga have proposed an alternative up to ? geographic coincidence. Instead of evaluating the concomitance between public research and private R&D, Acs, Anselin and Varga have measured the effect of each one on the innovation output. Jaffe’s method studies the effect of the concentration of geographic research within one State on the innovation output 13
They have tested the following equation : (1) K = + R + R75 + U + U50
13
of this State. In return, the method of Acs, Anselin and Varga goes beyond the mere effect of agglomeration. This method measures spillovers more directly, i.e. the internal impacts (variation in the innovation output of a metropolitan area) connected with external behaviors (level of research spending of the peripheral counties). Thus, the advantage of their indicator is to modelize spillovers rather than deducing them from the innovation output concentration, as it is the case with Jaffe (1989). In the same spirit, Antonelli (1994) has measured the effect of local technological externalities on the Italian productivity of firms, thanks to a Griliches’s (1979) function of production. He has linked the production growth rate to the growth rate of the capital stock, the labor stock, and the R&D spending carried out within and outside the firm. These technological spillovers are supposed to affect above all the regional scale. Consequently, the R&D spending external to the firm are calculated by considering only the firms localized in the same region. Furthermore, firms can only receive these spillovers if they know how to combine them with their internal knowledge. Antonelli has then supposed that the capability to receive technological spillovers depends on the R&D internal effort. In this prospect, he has integrated an interaction variable measuring the joined effect of the R&D spending of each firm and the intensity of the R&D spending14of close firms. The results confirm this assumption. The R&D direct spending and the R&D regional spillovers considered separately have an insignificant impact on productivity growth. In return, the interaction variable has an important positive effect on productivity growth, from which Antonelli has deduced that when there is an interaction between R&D external activities and the internal R&D, the firms benefit from the spillovers closely produced.
with U50 being the public research outlays realized around the metropolitan area, in a perimeter of 50 miles ; and R75, private research outlays realized in a perimeter of 75 miles around the studied metropolitan area. 14 The general equation is the following : OUTPUTj = a + b INVj + c LABOURj + d PROD + e (RDI / OUTPUTj) and PROD = (RDE / OUTPUTe) . RDj with RDI being the internal R&D of the firm and RDE the R&D external to the firm. PROD measures the interaction between the level of each firm’s R&D spending (j) and the intensity of R&D spending in other firms (e), which represents the opportunities of each firm to have access to technological spillovers. 14
Nevertheless, the notion of "interaction" is ambiguous in this context. What is tested is the relation between the R&D importance in the other firms of the region and the amount of the R&D spending of the firm. Thus, it appears that the higher is the R&D spending of a firm relatively to the R&D importance in the region of the firm, the more the R&D effort of the firm generates a productivity gain. However, this does not necessarily indicate an interaction between the R&D of the firm and the R&D realized outside15. Such an interaction can only be pointed out through the direct testing of its shapes: interpersonal relations, cooperative agreements in terms of R&D, co-publications, purchase of patents or licenses, and
so on. This is what Zucker, Darby and Armstrong have suggested
(1994). Through the case of biotechnologies in California, Zucker, Darby and Armstrong (1994) have intended to show that geographic proximity is not enough to benefit from knowledge spillovers stemming from universities. According to them, a firm can only benefit from the academic researches if it is in direct relation with the scientists at the root of the discoveries. As a demonstration, they have wondered which scientists are the most productive, in order to study their relations with (affiliated or linked) firms and to measure the effect of this relation on the firm. With this aim in view, they have built a function connecting the performances of the firm on the one hand and innovation inputs –including one or several affiliated or linked scientists16- on the other. If actual interactions are at the root of spillovers, then the firms with most performances should be those which interact with researchers. This assumption is confirmed by the results: the distinction between the researchers in relation with firms and the merely academic ones shows that the latter generate no positive effect for firms. This is consequently the firm's relation with universities which allows it to benefit from public research activities, and not the mere fact of being localized in the vicinity. Thus, one can wonder whether, despite everything, geographic proximity does have a role, in which it would foster interactions. However, the work of 15
Antonelli (1994) does not directly interpret the interaction variable (PROD) as an indicator of effective relations between the research activities of the firm and those of its surrounding. He is rather in the perspective of Cohen and Levinthal (1989), with the idea that the internal level of research determines the capacity to absorb externalities. 16 The researcher is said to be "affiliated" if he has published as a member of the firm. He is said to be "linked" if he has published in collaboration with researchers of a firm situated in the same geographic area. 15
Zucker, Darby and Armstrong does not bring any answer, since this method postulates more than it demonstrates the existence of geographic spillovers. Actually, so as to point out the local dimension, the space in which spillovers are to take place is "a priori" bounded. The only phenomenon observed corresponds to the R&D efforts of the firms or universities situated in the vicinity. Thus, the existence of a correlation means that the innovation of a firm is affected by the R&D local intensity. But this does not mean that the local dimension has a particular role. Actually, the R&D of firms or universities localized in other regions would perhaps quite as much affect the innovation of the firm. Henceforth, Zucker, Darby and Armstrong have not really demonstrated the existence of geographic spillovers, insofar as they have not compared the impacts of the relations with local scientists and the effects of the relations with geographically distant scientists. Now, this is possible that the geographic area to which the academic researchers belong has no role and that the relations with geographically distant researchers are as profitable as the relations with local researchers. From then on, proximity would have no role, except if the relations with the academic researchers situated in the vicinity are more numerous, which remains undemonstrated. In the same way, if -in his indicator of external research- Antonelli had measured the R&D of the whole country firms and not the R&D of the only firms situated in the vicinity, the local specificity might not have appeared. For the same reasons, Acs, Anselin and Varga have tested the existence of technological spillovers but not their local dimension. Their method makes it possible to show that spillovers take place in a given geographic area : the metropolitan area benefits from the research of the close counties. But it does not reveal whether spillovers take place only on this scale. Thus, such an approach would only be valid to demonstrate the existence of geographic spillovers, in the framework of a comparative work between different levels of geographic decomposition. This is the way followed by Autant-Bernard (2000 and 2001) to study the French case. The local dimension of externalities is studied by taking into account not only the research activities conducted within a geographic area but also the research carried out nearby and finally the research conducted in a more distant neighborhood. If knowledge spillovers are geographically limited, then the level of
16
local innovation must be even more affected by neighboring research than by research carried out at a distance. The main equation is:
(4)
log(Ig) = α1 + β 1 log(Rg) + β 2 log(Rv(g)) + β 3 log(Rv’(g)) + β 4 log(VAg) + u1
with I the output of innovation (measured by patents). R measures the level of research inputs, inside area g, in the bordering areas (v(g)) and in the bordering areas of the bordering areas (v’(g)). α1 is a constant term and u1 a random perturbation. VA is an indicator of the value added, in order to account for the economic size of the area g. It then comes down to observing the relation between the production of innovations of a geographic area,17 and the research effort carried out locally and on the periphery by defining concentric circles around the area g. Externalities are studied by testing the possible effect of local research and of neighboring areas’ research on the innovation output of area g. This way, both the presence of spillovers and their local dimension are tested, and it is possible to introduce private as well as public research, and to distinguish the sources and channels of knowledge spillovers18.Then, confronting several geographic levels, it appears that knowledge spillovers do not only take place locally. The local phenomena are combined with more distant spillovers, based in particular on technological proximity and interpersonal relations.
3. The geographic dimension of technological spillovers: How to go beyond the ambiguities of the first results?
Measuring the geographic dimension of technological spillovers raises some problems. The two main problems are, obviously, the spillovers measure itself and the consideration of the space dimension. Depending on the selected 17
The basic areas considered are French administrative Departments. Most of these units include one metropolitan agglomeration and its surroundings. 18 Several variants of the general model presented in (4) have been studied. Autant-Bernard (2001a) analyses the private R&D spillovers, Autant-Bernard (2001b) includes public research and tests its consequences on innovation, and the relations between public and private research, Autant-Bernard (2000) introduces an analysis by technological fields that allows to confront intra and inter-sectoral spillovers and Autant-Bernard and Massard (2000) test the influence of co-authoring on these knowledge flows. 17
method, those difficulties are more or less well overcome. The methods integrating an index of geographic coincidence are probably the most likely to reflect the impact of proximity. Actually, the impact of proximity is directly measured on the innovation output. Nevertheless, spillovers are better considered by the studies modelising technological externalities. However and as mentioned before, the study of the geographic dimension is far more reduced in this approach. As indicated by Antonelli (1994), the question is to test the existence of technological spillovers, in a geographically bounded area. Thus, it seems that what is gained by modelising spillovers is lost while modelising the space aspect. Researches should then be developed so as to improve modelisation. Despite the variety of the implemented methods and the difficulties met by each of them, it seems that the whole of the econometric studies allows to point out to a number of observations as for the impact of proximity on the innovation activity. We will first lay stress on the ambiguities we can’t fail to observe when trying to make a comparison between the results obtained in various contexts. The main research channels emerging out nowadays and which are likely to help us to go beyond these ambiguities will then be presented.
3.1. About some ambiguities in the results obtained.
The first studies carried out in the geography of innovation basically concerned the US and focused on the 1980s’ data. Within this context, the stidies quite generally conclude to the localized feature of technological externalities or, in any case, to the strong sensitivity of geographic proximity in the scienceindustry relations within both the states and the American metropolitan areas. We can’t help noticing that the extension of these studies to more recent periods and to other countries has considerably contributed to moderate this first result. (AutantBernard, 2000 for France or M.Beise and H.Stahl for Germany). Three types of explanations to the rather growing ambiguities of this general result can be put forward : - first a methodological explanation: in the first studies one generally confines oneself to the study of only one pre-definite geographic level (American state or metropolitan area). The increase in the number of studies and especially the development of approaches allowing the comparison of the results obtained on many geographic levels finally show that if the local dimension exists, it is far 18
from being exclusive. The enterprises in a same area are indeed simultaneously influenced by local and much more global knowledge flows. Indeed, this local dimension of spillovers is not always obvious (Jaffe, 1989). This point is confirmed by Autant-Bernard (2000) through another way. Using panel data, it can be accounted for spatial heterogeneity. Then, the impact of geographic proximity is less obvious than for only one year of observation. - one explanation linked to the influence of institutional contexts : in the first studies, only one country is observed: the US. The studies then carried out on other countries sometimes convey different results. Doubtless, here comes to light the influence of institutional contexts which underlie, in each nation, different methods of production and knowledge diffusion (Feldman and Massard, 2001). The interaction modes between public and private research for example can strongly vary from a country to another as well as the precise conditions of the appropriateness of the results in research. - Finally, an explanation through the possible role of Information Technology (IT) : the most recent and moderate studies about the existence of localized externalities focus on the 1990s. We may see here the effect of the development of IT which would explain the differences in the results obtained on the 80s and 90s periods.
3.2. Explanation of the conditions and ways of transmission of knowledge externalities. Many of the ambiguities observed can be explained by a vision, far too dichotomic, of the problem in which it would only be a question of concluding on the localized or globalized feature of externalities. In this way, the deepening ways likely to go beyond these ambiguities consist in getting away from a dichotomic approach in explaining the ways and conditions of transmission of externalities. One can discover here and now that these ways and conditions of transmission considerably depend on the origin of transmitted externalities, of modes taken by the individual interactions and at last of the determinants of absorptive capacity.
1) The role of proximity varies depending on the origin of spillovers. In particular the origin is linked to the nature of knowledge transmitted: public vs private (since studies consensually demonstrated that spillovers not only 19
emanate from public research but also from private one); fundamental vs applied or tacit vs codified. In that sense, one of the main results of the studies presented above is to split the relations initially established between public research, fundamental research and global externalities on the one hand and, private research, applied research and local externalities on the other hand. For example the interpretations of Jaffe, Trajtenberg and Henderson (1993) or of Audretsch and Feldman (1996) about this question are opposed to Mansfield’s (1995). Methodological problems must not be neglected. Indeed, the methods implemented in geographic spillover econometrics do not easily allow to reveal the type of knowledge. Nevertheless, these oppositions show the existence of much more complex processes which need to be explained in relation with the institutional context which found them. The origin of spillovers is also linked to the intra or inter sectorial feature of the diffusion of knowledge. People are expecting from such an analysis, new elements in the debate on the specialized or diversified nature of the agglomerations of innovative activities. However, the results on this question are rather contradictory. The studies of Jaffe (1986), Trajtenberg and Henderson (1993), and Audretsch and Feldman (1995) tend to demonstrate that a significant part of spillovers stem from the firms which are not in the direct technological vicinity of the considered firm. According to Jaffe (1989) however, the academic research influence is clearer on the scale of sectors than on the local scale, which would indicate that spillovers are specific (to a sector) and that academic research does not produce a diffuse effect. An idea, emerging in recent studies (AutantBernard, 2000), is that R&D spillovers require a geographic proximity above all when the firms are technologically remote, ie when spillovers are stemming from a sector of activity different from that of the firm in question. However, further analysis, based on the confrontation of different geographical level of observations, seems necessary in order to confirm this. 2) The role of proximity varies depending on the ways of transmission of externalities. The best way to go further in the discussion about the nature of knowledge might be to orient the analysis on the ways taken by knowledge flows. Thus, Almeida's and Kogut's study (1997b) points out the ways of transmission of spillovers. They would be diffused by means of labor transfers. Thus, knowledge would be incorporated into men, which tends to confirm the hypothesis of tacit 20
knowledge, whereas on the contrary codified knowledge can be transmitted thanks to non-human means. In the same way, the results of Zucker, Darby and Armstrong (1994) indicate that firms only benefit from public research if they work in collaboration with an academic researcher. The mere fact of being localized in the vicinity of a university is not sufficient. This statement attests again that the knowledge locally transmitted is incorporated into men, and that it cannot be conveyed by codified mediums. Nevertheless, even if those elements seem to imply that the role of proximity is associated with the tacit dimension of knowledge, a much too simplifying vision should be avoided. As externalities are more important in the case of labor transfers or of actual interactions between individuals, the locally transmitted knowledge would then be essentially tacit. But one can also think that proximity fosters the transfer of codified knowledge. Thus, the study of Jaffe, Trajtenberg and Henderson (1993) indicates for instance that codified knowledge, incorporated into patents, also generates geographically-bounded externalities. It can then be assumed that, so as to exploit this knowledge, one must possess tacit knowledge. In this case, tacit knowledge and codified knowledge would be complementary (Lung, Rallet, Torre, 1997), and transmitted both locally and remotely. Then, the link between proximity and knowledge is perhaps to be researched more in the way of combining tacit knowledge with codified knowledge, than in the type of knowledge itself. In that sense we should be very interested in studies trying to analyze the possible impact of the development of Information and Communication Technologies (such as internet) on the geographic diffusion of knowledge externalities. Specially in order to clarify the interaction between face to face contact and contact by codified medium such as internet: are they substitutes or complementary ways of inter-individual relations? In the former case we could expect a tendency to the dispersion of externalities in the geographic space. Conversely, in the latter, forces of agglomeration could be reinforced.
3) Finally the role of geographic proximity appears through the absorptive capacity necessary to benefit from externalities.
21
Dosi (1988) and Cohen and Levinthal (1989) show that the capacity to catch externalities depends on previous knowledge and on its diversity. The empirical studies carried out on this question confirm these ideas (see Cohen and Levinthal, 1989; Cockburn and Henderson, 1998; Varga, 1998; Maurseth and Verspagen, 1999). However, seldom are the studies analyzing the link between absorptive capacity and the local dimension of spillovers. Yet, the level of research, as well as its diversity, may influence not only the level of externalities which can be caught, but also their geographical origin. In any case, this is what emerges from the analysis of the French case (Autant-Bernard, 2000). The fact to dispose of a high and varied level of internal scopes of activities looks determining in the capacity of deriving profit from the distant sources of knowledge. Conversely, the very little active areas, as regards research, or the much specialized ones seem to be more in a position to derive profit from the neighboring sources of externalities. The capacity of absorption would then play a more important role on the capacity to benefit from distant sources of externalities than on the level of captured externalities. Yet, this hypothesis still requires confirmation.
4. Conclusion
After a little more than ten years’ research in the geography of innovation, we have wished to go back over the major results obtained concerning the geographic dimension of externalities. In describing in a detailed way the econometric methods we have used, we have showed the difficulties of a simultaneous measure of the externality phenomenon itself (as an external stock of research) and of its geographic dimension. Despite both these difficulties and the diversity of these approaches (not always allowing the comparison of the results), we have noticed a growing tendency to ambiguity. We have showed then that the exceeding of these ambiguities is possible, provided that the studies free themselves from a too much dichotomic conception of the process in questioning oneself more precisely about the ways and conditions of transmission of knowledge externalities in space. Relying both on the features of innovation interactivity and on the determinants of appropriateness, these conditions are strongly influenced by the institutional context. More advanced studies, taking into account the origin of transmitted knowledge, the modes of interpersonal interactions enabling transmission and the 22
determinants of absorptive capacity of knowledge, still today, prove to be necessary to a good understanding of the spatial determinants of the diffusion of externalities.
REFERENCES ACS Zoltan J., AUDRETSCH David B. and FELDMAN Maryann P. (1991), " Real Effects of Academic Research : Comment ", The American Economic Review, vol. 82, no. 1, March, p. 363-367. ACS Zoltan, ANSELIN Luc, VARGA Attila (1997), " Local Geographic Spillovers Between University Research and High Technology Innovations ", Journal of Urban Economics, no. 42, p. 422-448. ALMEIDA Paul and KOGUT Bruce (1997a), " The Exploration of Technological Diversity and the Geographic Localization of Innovation ", Small Business Economics, no. 9, p. 21-31. ALMEIDA Paul and KOGUT Bruce (1997b), " The Localization of Ideas and the Mobility of Engineers in Regional Networks ", Working Paper, June, p. 45. ANSELIN Luc (1988), Spatial econometrics: methods and models, Kluwer Academic Publishers, Netherlands, p. 284. ANTONELLI Cristiano (1994), " Technological Districts Localized Spillovers and Productivity Growth. The Italian Evidence on Technological Externalities in the Core Regions ", International Review of Applied Economics, p. 18-30. ARROW Kenneth J. (1962), "Economic welfare and the allocation of resources of invention", in NELSON Richard R. (ed.), The rate and direction of inventive activity: economic and social factors, Princeton University Press, Princeton, p. 609626. ARTHUR W. Brian (1994), Increasing Returns and Path Dependence in the Economy, The University of Michigan Press, coll. Economics, Cognition, and Society, p. 201. AUDRETSCH David B. and FELDMAN Maryann P. (1994), " R&D Spillovers and the Geography of Innovation and Production ", Discussion Paper, FS IV, no. 2, Berlin, p. 31. AUDRETSCH David B. and FELDMAN Maryann P. (1996 a), " Innovative Clusters and the Industry Life Cycle ", Review of Industrial Organization, no. 11, p. 253-273. AUDRETSCH David B. and FELDMAN Maryann P. (1996 b), " R&D Spillovers and the Geography of Innovation and Production ", The American Economic Review, vol. 86, no. 3, jun, p. 630-640. AUDRETSCH David B. and FELDMAN Maryann P. (1999), " Innovation in Cities: Science-Based Diversity, Specialization, Localized Competition ", European Economic Review, n°43, p. 409-429. AUDRETSCH David B. and VIVARELLI Marco (1994), " Small Firms and R&D Spillovers : Evidence from Italy ", Revue d’Economie Industrielle, no. 67, p. 225237. AUTANT-BERNARD C. (2000), Geographie de l'innovation et externalités locales de connaissances. Une étude sur données françaises, Thèse pour le doctorat en sciences économiques, Université Jean Monnet St-Etienne. AUTANT-BERNARD C. (2001a), " The geography of knowledge spillovers and technological proximity", Economics of Innovation and New Technology, vol.10. AUTANT-BERNARD C. (2001b), "Science and knowledge flows: evidence from the french case", Research Policy. AUTANT-BERNARD C. and MASSARD N. (2000), "Scientific interactions, geographic spillovers and innovation. An empirical study on the French case", 40th European Regional Science Association Congress, Barcelone, 29 août-1er sept.
23
BAUMONT C. et LE GALLO J. (2000), "Geographic spillovers and growth. A spatial econometric analysis for european regions", Xème journée du SESAME, Dijon, septembre. BEISE M. et STAHL H. (1999), "Public research and industrial innovations in Germany", Research Policy, n°28, p. 397-422. BLIND K. et GRUPP H. (1999), "Interdependencies between the science and technology infrastructure and innovation activities in German regions : empirical findings and policy consequences", Research Policy, n°28, p.451-468. BRANSTETTER L. (1996), " Are Knowledge Spillovers International or Intranational in Scope? Microeconometric evidence from the US and Japan ", Working Paper no. 5800, NBER, october. CABALLERO R. J. and JAFFE Adam B. (1993), " How High are the Giants’Shoulders : An Empirical Assessment of Knowledge Spillovers and Creative Destruction in a Model of Economic Growth ", p. 11-74. CANIELS M.C.J. (1997), "The geographic distribution of patents and value added accross european regions", Working paper, MERIT, Août, p.10. CARRINCAZEAUX Christophe (2001), " The Role of Geographical Proximity in the Organization of Industrial R&D", in FELDMAN Maryann and MASSARD Nadine (ed.), Institutions and systems in the geography of innovation, Kluwer Academic Publishers. COCKBURN I. and HENDERSON R., 1998, "Absorptive capacity, coauthoring behaviour, and the organization of research in drug discovery ", The Journal of Industrial Economics, vol. XLVI, n. 2, Jun, p. 157-182. COE D., HELPMAN E. (1995), " International R/D Spillovers ", European Economic Review, vol. 39, May. COHEN Wesley and LEVINTHAL David A. (1989), Innovation and learning: the two faces of R&D", The Economic Journal, n°99, september, p. 569-596. DOSI Giovanni (1988), " Sources, Procedures, and Microeconomic Effects of Innovation ", Journal of Economic Litterature, vol. XXVI, september, p. 1120-1171. DOSI Giovanni et alii (1988), Technical Change and Economic Theory, Pinter Publishers, London and New York, p. 646. FELDMAN Maryann P. (1994), The Geography of Innovation, Economics of Science, Technology and Innovation, vol. 2, Kluner Academic Publishers, Dordrecht, Boston, London, p.155. FELDMAN Maryann P. (1999), " The New Economics of Innovation, Spillovers and Agglomeration : a Review of Empirical Studies ", Economics of Innovation and new technology, vol.8, p. 5-25. FELDMAN Maryann P. and FLORIDA Richard (1994), " The Geographic Sources of Innovation : Technological Infrastructure and Product Innovation in the United States " Annals of the Association of American Geographers, vol. 84, no. 2, p. 210229. FELDMAN Maryann and MASSARD Nadine (ed.), Institutions and systems in the geography of innovation, Kluwer Academic Publishers. FORAY Dominique and FREEMAN Christopher (1992), dir., Technologie et richesse des nations, Economica, Paris, p. 513. GLAESER Edward L., KALLAL Hedi D., SCHEINKMAN Jose A. and SHLEIFER Andrei (1992), " Growth in Cities ", Journal of Political Economy, vol. 100, no.6, p. 11261152. GRILICHES Zvi (1979), "Issues in assessing the contribution of research and development to prductivity growth ", The Bell Journal of Economics, vol. 10, no.1, p. 92-116. GROSSMAN Gene and HELPMAN Elhanan (1994), " Endogeneous Innovation in the Theory of Growth ", Journal of Economic Perspectives, no. 8, p. 23-44. GROSSMAN Gene M. and HELPMAN Elhanan (1995), Innovation and Growth in the Global Economy, The M.I.T. Press, Cambridge, Massachusetts and London, England, 5th edition (1st edition 1991), p. 359. HARRISON Bennett, KELLEY Maryellen R. and GANT John (1996), " Innovative Firm Behaviour and Local Milieu : Exploring the Intersection of Agglomeration, Firm Effects, and Technical Change ", Economic Geography, vol. 72, no. 3, July, p. 233258. 24
ISARD Walter (1960), Methods of regional analysis: an introduction to regional science, The MIT Press, Cambridge, Massachusetts, and London, England, 7th edition p. 794. JAFFE Adam B. (1986), " Technological Opportunity and Spillovers of R&D : Evidence From Firm’s Patents, Profits and Market Value ", The American Economic Review, vol. 76, no. 5, December, p. 984-1001. JAFFE Adam B. (1989), " Real Effects of Academic Research ", The American Economic Review, vol. 79, no. 5, December, p. 957-970. JAFFE Adam B., TRAJTENBERG Manuel (1996), "Flows of knowledge from universities and federal labs: modeling the flow of patent citations aver time across institutional and geographic boundaries", NBER Working paper Series, Working paper n°5712, Cambridge, p.29. JAFFE Adam B., TRAJTENBERG Manuel and HENDERSON Rebecca (1993), " Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations ", The Quaterly Journal of Economics, August, p. 577-598. JAFFE Adam B., TRAJTENBERG Manuel and FOGARTY Michael S. (2000), "The meaning of patent citations: report of the NBER/case-Western reserve survey of patentees", NBER Working paper Series, Working paper n°7631, avril, Cambridge. KENNEY Martin and FLORIDA Richard (1994), " The Organization and Geography of Japanese R&D : Results From a Survey of Japanese Electronics and Biotechnology Firms ", Research Policy, no. 23, p. 305-323. KRUGMAN Paul (1991a), " Increasing Returns and Economic Geography ", Journal of Political Economy, vol. 99, no. 3, p. 483-499. KRUGMAN Paul (1991b), "History of industrial location: the case of manufacturing belt", AEA Papers and Procedings, vol.81, n°2, Mai, p.80-83. KRUGMAN Paul (1991c), Geography and Trade, Leuven University Press, Leuven, Belgium, The MIT Press, Cambridge, Massachusetts, London, England, 4th edition, p. 136. LICHTENBERG F. and VAN POTTELSBERGHE de la POTERIE (1996), " International R/D Spillovers : a re-examination ", Working Paper no. 5668, NBER, July. LUNG Yannick, RALLET Alain and TORRE André (1996), " Innovative Activity and Geographical Proximity ", European Regional Science Association, 36th European Congress, ETH Zurich, Switzerland, 26-30 August, p. 20. MANSFIELD Edwin (1995), " Academic Research Underlying Industrial Innovations : Sources, Caracteristics, and Financing ", The Review of Economics and Statistics, vol. LXXVII, no.1, February, p. 55-65. MANSFIELD Edwin (1998), " Academic Research and Industrial Innovation : An Update of Empirical Findings ", Research Policy, vol. 26, no. 7-8, April, p. 773-776. MARSHALL Alfred (1906), Principes d’économie politique, Gordon & Breach, Paris, Londres, New York, edition 1971, tome 2, p. 576. MAURSETH Botolf et VERSPAGEN Bart (1998), "Knowledge spillovers in Europeand its consequences for systems of innovation", ECIS Working Paper, n°98-001, octobre, p.21. MOHNEN P. (1991), " Survol de la littérature sur les externalités technologiques ", in DE BANDT J. and FORAY D. (eds.), L’évaluation économique de la recherche et du changement technique, Editions du CNRS, Paris, First Part, chap. 1. MOHNEN P. (1998), " International R/D Spillovers and Economic Growth ", Paper presented at the UNU Project Meeting on Information Technology and Economic Development, Jun, 12th and 13th . PACI Raffaele et USAI Stephano (2000), "Externalities, knowledge spillovers and the spatial distribution of innovation", CRENOS Working Paper, Mars,p. 28. PODOLNY Joel M. and SHEPARD Andrea (1998), " Firm Agglomeration and Technological Spillovers : Citation Patterns in the U.S. Semiconductor Industry ", Working Paper, July, p. 44. ROMER Paul (1990), "Endogenous Technological Change", Journal of Political Economy, vol. 98, no. 5, p. 71-102. TEECE, D.-J., (1992), "Competition, cooperation, and innovation. Organizational arrangements for regimes of rapid technological progress", Journal of Economic Behavior and Organization, 18, p. 1-25. 25
VARGA Attila (1998), "Local academic knowledge spillovers and the concentration of economic activity", Research Paper n°9803, West Virginia University, march, p. 28. VERSPAGEN Bart et SCHOENMAKERS Wilfred (2000), "The spatial dimension of knowledge spillovers in Europe : evidence from firm patenting data", Working paper, avril, p.18. ZUCKER Lynne G., DARBY Michael R. and ARMSTRONG Jeff (1994), " Intellectual Capital and the Firm : the Technology of Geographically Localized Knowledge Spillovers ", NBER Working Paper Series, Working Paper, no. 4946, NBER, Cambridge, p. 59.
26
