Geographic concentration of innovative activities in Germany

Structural Change and Economic Dynamics 20 (2009) 163–182

Contents lists available at ScienceDirect

Structural Change and Economic Dynamics journal homepage: www.elsevier.com/locate/sced

Geographic concentration of innovative activities in Germany Dirk Fornahl a,∗ , Thomas Brenner b a b

BAW Institute for Regional Economic Research, Wilhelm-Herbst-Str. 5, 28359 Bremen, Germany Philipps University Marburg, Chair of Economic Geography and Location Research, Germany

a r t i c l e

i n f o

Article history: Received November 2008 Received in revised form May 2009 Accepted May 2009 Available online 15 May 2009 JEL classification: R11 R12 O30 D83 Keywords: Spatial concentration Innovation Technology Concentration indices

a b s t r a c t The geographic concentration of industries has attracted much attention in recent economic and geographic literature. One mechanism employed to explain the emergence and comparative advantage of industrial agglomerations is based on the relationship between industrial agglomeration and local knowledge production and diffusion, and the resulting innovation activities. This paper analyses this relationship by identifying geographic concentrations of innovation activities and examining different causes for the emergence of these concentrations. The paper applies different concentration measures to patent data for German regions. We analyse 43 technological fields separately to identify which of these technologies tend to cluster in geographic space. The results are discussed in light of theoretical predictions of why specific technological fields concentrate while others do not. These explanations include the concentration of industrial activities, the role of dominant firms, dependence on scientific knowledge, and local interactions. © 2009 Elsevier B.V. All rights reserved.

1. Introduction The geographic concentration of industries has been repeatedly studied in the literature (Ellison and Glaeser, 1997; Braunerhjelm and Carlsson, 1999; Lafourcade and Mion, 2005; Brenner, 2005; Sternberg and Litzenberger, 2004; Alecke et al., 2006). Ellison and Glaeser (1997), for example, identified that in general, industries show different degrees of geographic clustering, but in the majority of industries an agglomeration of firms in a few regions can be observed. This paper takes up the core issue of geographic concentration but is focusing not on industrial but on innovative activities. Studies examining such spatial distributions of innovation activities are rare. Such a lack of research on

∗ Corresponding author. Tel.: +421 206 99 30; fax: +421 206 99 99. E-mail addresses: [email protected] (D. Fornahl), [email protected] (T. Brenner). 0954-349X/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.strueco.2009.05.001

the existence of concentrations of innovative activities is particularly interesting, since such activities are a central element linked to the development of industrial clusters which are seen as a core driver of regional development. In the literature it is argued that firms that are located in industrial agglomerations are more innovative than firms outside such regions (Audretsch, 1998). This would imply that innovation activities follow the concentration pattern of industries but are more pronounced. At the same time, it is argued that innovation activities cause firms’ success and growth (Smolny and Schneeweis, 1999). As a consequence, industrial activity should follow the spatial distribution of innovations. In addition, many arguments for the emergence of geographic concentrations of industries are based on mechanisms, such as knowledge spillovers, accumulation of technological knowledge, and cooperation, that function rather on a technological than an industrial level. We can conclude that innovation activities should also be geographically concentrated, probably causing geo-

164

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

graphic concentration of industrial activity, and at the same time being affected by these industrial activities. By identifying the causes of technological concentration, additional insights into the emergence and operation of industrial clusters can be generated. Additionally, explaining the existence of local knowledge pools helps to understand why firms locate within such pools, and to design policy measures that support the emergence of knowledge pools. The aim of this paper is to measure the spatial concentration of innovation activity, to provide and discuss several potential explanations for such concentrations of innovative activities, and to examine their relevance in a comparative empirical study. This is helpful in two ways. On the one hand, it allows judging whether it is possible to influence the spatial distribution of innovation activities directly or whether this distribution is caused by economic activities, so that these activities would have to be changed in order to influence the spatial distribution of innovation activities indirectly. On the other hand, it provides information about how the spatial distribution of innovation activities can be influenced through investigating various causes and their impact on this distribution. Our paper differs to previous studies such as Feldman (1994), Audretsch and Feldman (1996), Paci and Usai (2000), Zitt et al. (1999), and Caniëls (1999). Feldman (1994) analysed the geographic distribution of product innovations. She found that product innovations cluster in certain regions, and that regions have a tendency to focus on innovations in specific industries. Innovative clusters are identified by comparing the regional rate of innovative activities to the national rate of such activities in a certain industry. Feldman explains this clustering mainly by the impact of local conditions, focusing on certain inputs for the product innovation process, such as the technological infrastructure. Other authors, such as Paci and Usai (2000) and Zitt et al. (1999), use patents as proxies for innovative activities, but apply a similar calculation method. These papers use Gini or Herfindahl concentration indices to calculate the level of concentration, and most focus on the identification of concentration while saying little or nothing about the factors that lead to this tendency for concentration. In order to study the mechanisms behind the geographic concentration of innovation activities, we examine geographic concentration in 43 technological fields in Germany. The main idea of the paper is to utilise differences in the geographic concentration of technological fields in order to identify the characteristics of technologies which cause them to be more, or less, geographically concentrated. Having identified the characteristics which tend to lead to the spatial concentration of innovation activity, one can study the relationship between technological concentration and spatial concentration. In common with the literature in this area, we use patent applications as a proxy for innovation activity. The shortcomings of patent data are extensively discussed, e.g. by Grupp and Schmoch (1999) and Pavitt (1985). First, patents are only granted to ‘products’. Many services and immaterial goods are therefore not patentable. Further, as Schumpeter discussed, innovation is more than product innovation; it includes processes, organisational change,

marketing and the supply side (Leo et al., 2007). Second, not all innovations are filed as a patent, even when they are patentable. Firms sometimes decide against the temporary monopoly a patent provides in order to keep their inventions secret (Rammer, 2003). Hence, only a fraction of all patentable innovations are actually covered. Despite these shortcomings, patents still provide a rich source of data and information, and are a powerful indicator of technological output (Griliches, 1990; Audretsch, 1995). In order to analyse the geographic concentration of industrial activities, we will calculate four indices based on German patenting data for different technologies. These are the Gini coefficient, the Herfindahl index, the Ellison and Glaeser approach, and Moran’s I. These indices provide us with some quantitative measures of the geographic concentration of different innovation activities, and their spatial autocorrelation. The indices are used in two ways. First, we intend to give a general overview of the geographic concentration of innovation activity. We will discuss how the results vary using different indices, and which technological categories tend to cluster geographically. Second, the indices are used to test theoretical propositions concerning why specific technological fields concentrate in space while others do not. We examine the characteristics of the different technologies and how these characteristics relate to the technologies’ spatial concentration. The hypotheses we test are formulated on the basis of the different potential explanations for the geographic concentration of innovations. Our findings indicate that technologies have quite different tendencies to concentrate in geographic space. Of particular importance is the role played by relatively few (‘dominant’) firms within a technology, and the geographic concentration of industrial activities that are related to the technologies. The dependence on scientific knowledge is found to have some relevance, while the importance of local interactions only in some instances affects the technological concentration. The remainder of this paper proceeds as follows. In the next section the potential causes of spatial concentration in innovation are theoretically discussed, and a set of core hypotheses derived. In Section 3 we discuss the data set and the methods that are applied to the data. Section 4 presents the empirical findings on the tendency of different technologies to concentrate in geographic space. Furthermore, this section empirically tests the factors assumed to shape these different concentration tendencies. Section 5 concludes and discusses future research options. 2. Theoretical considerations In this section we discuss the different mechanisms that may lead to a geographic concentration of innovation activities. Three fundamentally different causes are distinguished. The analysis of each cause proceeds in two steps. First, the mechanism itself is discussed and its characteristics are analysed. Second, we deduce predictions for the different tendencies of technologies to concentrate in geographic space. These predictions depend on the relevance of the respective mechanism. Hence, the hypotheses are for-

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

mulated in the form of ‘if a certain mechanism is relevant, we expect a certain order of technologies with respect to their geographic concentration’. The predictions are empirically tested in Section 4. This section also discusses the implications of our results concerning the potential mechanisms that can lead to a geographic concentration of innovation activities. 2.1. Innovation activities dominated by a few firms First, it might be the case that, for certain technologies, only a few firms conduct all or almost all innovation. This does not hold for innovative activities in general, since around half of manufacturing firms declare that they carry out such activities in questionnaires (Evangelista and Smith, 1998; Beise et al., 1999). However, many studies revealed that patent activity is dominated by a few inventors who file for patents (see, e.g. Narin and Breitzmann, 1995). These inventors generate most of the patents, while the majority of other organisations patent sparsely, if at all. Although this holds for all technologies, the degree of domination might differ among technologies. If the patent activities of the dominating firms are geographically concentrated, we can expect a situation in which most patents originate in a few regions while all other regions do not add much to the patent activities in the particular technological field. Hence, if patent activities differ strongly between firms, and the dominating firms concentrate their patent activities in one or a few locations, a strong difference in patent activities between regions should be observed. If this is a main factor shaping the spatial distribution of patent activity, we should expect the following hypothesis to hold: Hypothesis 1. The more the patent activities in a technology are concentrated in a few firms, the more the patent activities are concentrated in geographic space. This hypothesis can be tested by examining the Herfindahl–Hirschman index (HHIfirms ) of the distribution of patents among firms. 2.2. Concentration of other economic variables As has been already pointed out in the introduction, the spatial distribution of industrial activities and innovation activities may be interrelated. This would imply that if industrial activities are geographically concentrated, innovation activities should also be geographically concentrated. Besides industrial activities, other economic and social factors may play a role in the occurrence of innovations. If these factors are unequally distributed across space, innovation can be expected to be similarly unequally distributed. There exists some empirical evidence to suggest that innovative activity within a region is also effected by university R&D (Feldman and Florida, 1994), science institutions (Blind and Grupp, 1999), business-service firms (Feldman and Florida, 1994), various kinds of human capital (Fröderer et al., 1998; Soete and Stephan, 2003), and by cooperation and networks (Pittaway et al., 2003).

165

For most of these factors a causal relation in the opposite direction also seems plausible, i.e. the geographic distribution of innovations may affect the distribution of these factors. This holds for example for the accumulation of human capital as well as for the cooperation, network or industrial activities (see Faggian and McCann, 2006 for the circular relation between innovations and human capital). This potential circularity has an influence on the interpretation of the results in Section 4. However, industrial activity in a region seems to be the best predictor of the respective innovation activity. Hence, we first focus on the relationship between the spatial concentrations of industrial and innovation activities. If the spatial distribution of innovation activities is shaped by the spatial distribution of the respective industrial activities, the following hypothesis should hold: Hypothesis 2. The higher the related industrial activity is concentrated in space, the more the innovation activities are geographically concentrated. This hypothesis can be tested by examining the HHIfirms of the spatial distribution of industrial employment. The importance of other local factors depends very much on the kind of technology employed in the different sectors, or the innovation drivers in these sectors. According to Pavitt’s taxonomy (Pavitt, 1984) innovation in some industries is predominantly science-based, while in others innovation is supplier-driven or scale-intensive. According to Pavitt, innovation processes in science-based industries utilise universities and research centres as knowledge sources. If this is the case, then we would expect the location of universities and research centres to affect the location and concentration of innovation activities in an industry (Pavitt, 1984; Marsili and Verspagen, 2001). In contrast, the other Pavitt categories lead to a dependence of the innovation process on firm-internal capabilities, suppliers or customers. The spatial distribution of these factors follows either the distribution of the core industry, the distribution of other industries, or is approximately uniform. Studying these factors requires information that is not provided by patent data. Therefore, we will restrict our analysis on the degree to which an industry is science-based. We assign the category ‘science-based’ as one dimension characterising different industries or their sectoral system of innovation (Malerba, 2002). Hence, we consider industries that are more or less science-based. The university R&D and public research that is relevant for a specific sectoral system of innovation can be quite concentrated in geographic space. In most cases, research conducted in a specific location tends to be focused on one or a few related areas, and is not equally relevant to all industries. As a consequence, relevant R&D undertaken by universities and public research institutes tends to be spatially concentrated. We conclude that innovation processes that are related to public research tend to be geographically concentrated. If the spatial distribution of innovation activities is strongly influenced by the spatial distribution of the respective public research, the following hypothesis should hold:

166

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

Hypothesis 3. The more a technology is science-based, the more the innovation activities are concentrated in geographic space.

3. Empirical method

This degree to which a technology is science-based can be measured by the share of non-patent citations on each patent, i.e. by the share of citations to scientific publications as opposed to citations to earlier patents.

The analysis builds on data extracted from the European Patent Organisation’s (EPO) Worldwide Statistical Patent Database version October 2006, the so-called ‘PATSTAT October 2006’ database. Using this source, we focus on inventions with a certain economic importance—or at least the applicants believe them to be of importance because it is more complicated and more costly to file patents at the EPO than at the German Patent Office (Frietsch et al., 2008).1 For the purpose of this paper we selected all patents filed at the European Patent Office for which at least one applicant was located in Germany. Using this source, our dataset contains 186,145 patent applications for the 10-year period from 1994 to 2003. Since the patents are assigned to regions by using post codes, we are unable to include data prior to 1994. This is because the system of postal codes changed in 1994—due to the German re-unification some years before. Assigning post codes of the former-GDR to regions is difficult. Our data set ends in 2003 because this is the year in which the employed database ended. We have assigned the patents to regions in Germany. The unit of analysis we chose is the administrative district, called ‘Raumordnungsregionen’, of which there are 97 in Germany.2 We assigned the post codes of the inventors’ locations rather than the organisations they are employed by. This is because the headquarters of large firms tend to be located far away from the place where the innovation took place (Paci and Usai, 2000). Inventors who are not located in Germany were only included in order to calculate the regional weights for the different patents. In the next step, all patents were assigned to different technological fields. This involved the identification of the International Patent Classification (IPC) for all relevant patents. Based on these IPCs, we assigned the patents to 43 technological fields with the help of a concordance (Schmoch et al., 2003). Since the pure counting of the extracted data would distort the whole picture of the analysis, the relevant patents were weighted by the region of the inventors and by the technology sector. For example, a patent which has two inventors, each living in a different region, and which belongs to three technological sectors, would count as 1/6 in each region and technological sector. Summing up all the patents, for which we can assign region and technology weighted by the two factors (43 technological sectors and 97 regions plus outside Germany), leads to the total number of patents included in the analysis which

2.3. Local interactions and collaborations Intended and un-intended information and knowledge exchange takes place within a region (Saxenian, 1994; Verspagen, 1999). Inter-organisational co-operations are more likely to occur within such a regional context. For instance, the spatial proximity between firms within a region enables more face-to-face contacts and, over time, the emergence of trust-based inter-firm relationships which lower transaction costs. Furthermore, heuristics and motivational factors lead to a bias in search processes: individuals and organisations start to search for solutions to technical problems, or to co-operate with partners inside their regions (Broekel and Binder, 2007). Hence, spillovers and local knowledge transfer allow firms to benefit from the research and innovation activities of other firms in the region (Acs et al., 1994; Anselin et al., 1997). These regional exchange processes, related to innovation activities, may be particularly important in technologies where joint research, the frequent exchange of knowledge, and/or rapid feedback processes are important. Spatial proximity eases such exchanges. Hence, we would expect that technologies characterised by collaborative innovation will tend to be more concentrated in space. The interaction of the above mechanisms may generate a regional self-reinforcing process that increases still further the tendency for a geographic concentration of innovation. This favours regions that already enjoy comparative advantages, while regions that do not initially enjoy these advantages fall still further behind. This is similar to the feedbacks discussed by the literature on local industrial clusters (Brenner, 2004). Hence, the geographic concentration of innovation processes might be caused by dynamics within the creation of innovations itself. It is argued in the literature that innovations involve a cumulative mechanism (Lundvall and Borras, 1999). People who generate an innovation are often in a good position to use their technological advancement to create further innovations (Ernst et al., 2000). Although it might take some time until the self-reinforcing processes lead to the establishment of regional concentrations, the technological fields in this paper – none of which are peculiarly young – should show signs of such concentration, if local interactions play a role. Therefore, if the spatial distribution of innovation activities is strongly affected by local interactions and collaborations, the following hypothesis should hold: Hypothesis 4. The more agents interact locally in the generation of innovations, the more the patent activities are concentrated in geographic space.

3.1. Empirical data

1 With such an approach we under-estimate the patent activities of smaller firms which mostly file at the German Patent Office. But as long as there are no systematic differences in the technologies in which smaller and larger firms are patenting, no bias should disturb the analysis. 2 We also conducted the analysis for German labour market regions, of which 270 exist. The results are comparable to those based on the ‘Raumordnungsregionen’. Furthermore, the analysis should neither be affected by a specific border effect nor by a peculiar effect with regard to the situation in East Germany after 1994 because we focus on a comparison between technologies which should be equally affected by such effects.

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

167

Table 1 Distribution of identification as most active patenting region. Region

Frequency of being region with highest share

Examples of technologies

Munich

11

Stuttgart Düsseldorf Rhine-Main Industrial Region Middle Franconia

8 6 4 2

Signal transmission, telecommunication; office machinery and computers Motor vehicles; energy machinery Special purpose machinery; fabricated metal products Pharmaceuticals; rubber and plastic products Leather articles; industrial process control equipment

Rhineland-Palatinate Black Forest-Baar-Heuberg Arnsberg Berlin Bielefeld

2 2 1 1 1

Basic chemicals; pesticides, agro-chemical products Weapons and ammunition; watches, clocks Lightening equipment Electric distribution, control, wire, cable Wood products

Brunswick Cologne Münster Schleswig-Holstein South South East Upper Bavaria

1 1 1 1 1

Paints, varnishes Petroleum products, nuclear fuel Agricultural and forestry machinery Tobacco products Wearing apparel

amounts to around 173,670. Thus, we are able to include 93.3% of the patents in the analysis. The small share of data that we loose is caused by the fact that, for some patents, the post codes was not detectable from the data. An overview of the central characteristics of the 43 technologies is provided in Table A.1 in Appendix A. The patents are very unevenly distributed among the technologies. There are some technologies such as ‘Motor vehicles’, ‘Basic chemicals’ or ‘Special purpose machinery’ with around 18,488, 15,212 and 12,781 patents respectively, while ‘Watches, clocks’, ‘Tobacco products’ and ‘Paints, varnishes’ only have 139, 116 and 45 patents. In related industries, different tendencies exist to patent new technologies. Other means of knowledge protection might be used in different technological regimes (Brouwer and Kleinknecht, 1999). Furthermore, the patenting activities differ between countries and, thus, a nation-specific technological profile emerges for Germany (Amiti, 1999). The list of regions with the highest patenting share in each of the 43 technologies (see Table 1) shows that inno-

vative activities are dominated in Germany by a few regions (15 different regions in total are at the top of the ranking of the 43 technologies). Some of these regions focus their activities in neighbouring technological fields such as Rhine-Main and Rhineland-Palatinate in the chemical and pharmaceutical area. Nevertheless, patenting activities are at the same time also widely spread in geographic space, at least to some extent. For eleven technologies all 97 regions have at least one patent. Even in the case of ‘Tobacco products’ around 42% of the regions have at least one patent. A Lorenz curve is presented in Fig. 1 showing the relationship between the share of regions and the share of patents aggregated over all technological fields. 10% of the regions are responsible for around 25% of the patents. Half of the regions account for nearly 80% of the patents. In some specific technological fields, activities can be much more concentrated. Thus, there exists a tendency of innovative activities to concentrate in geographic space, but this tendency seems to differ between technology fields and, in general, there are

Fig. 1. Lorenz curve for the regional distribution of the overall patent activities.

168

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

many different regions which are active in innovation in Germany. 3.2. Concentration indices

The second Gini coefficient that we use weights the number of inhabitants in each region. popsharer is the share of inhabitants in region r. We do this because the 97 regions strongly differ in their number of inhabitants.

 n × (ai + ci ) × hi n−1 n

In this section of the paper we explain the four different approaches that are used to calculate the degree of concentration.

Gweight =

(3)

i=1

with 3.2.1. Herfindahl indices 

 n   p˛r HHI˛ = ˛−1

hi = (1)

3.2.2. Gini coefficients We apply two different Gini coefficients. The first is an unweighted Gini that is used, for example, by Wen (2004). pr is the share of patent activities in region r, n is the number of regions (= 97) and p is the mean of shares. Gini coefficients are insensitive to the number of regions (Davies, 1979). However, we assigned a value of 0.00000001 to cases where no patent exists in a particular region and technological field. This is done to provide a common basis for comparison. The Gini coefficients range between 1 − 1/n (for maximum concentration) and 0 (for equally distributed patents). We normalise by multiplying the Gini coefficient by n/(n − 1) so that the maximum value is 1. 1  n × |pj − pr | n−1 2n2 p¯ n

G=

n

r=1 j=1

(2)

popsharej −

j=1

r=1

We use a generalised Herfindahl–Hirschman index (HHI˛ ) with pr as the share of patent activities in region r, n is the number of regions (= 97) and ˛ is the exponent (where ˛ = 2 would resemble the normal Herfindahl index). This index decreases when the number of regions increases and is therefore sensitive to differences in n (Davies, 1979). As we previously mentioned, not all technological fields are represented in all regions. This could make a comparison based on HHI␣ problematic. In order to deal with this problem we included all 97 regions in the analysis, assigning a value of 0.000000001 to any region that is not active in a particular technology (the smallest share actually identified in active regions was around 0.0001). The Herfindahl index ranges between 1 (if all patents in a technological field were conducted in one single region) and 1/n (= 1/97 ≈ 0.01; if the patents in the field are equally distributed over the regions). The HHI␣ is sensitive to units of analysis with a high share of activities; in our case to regions with a high share of patenting activities. This effect is stronger, the larger is ˛. For ˛ > 2 the weight of these very active regions is higher than for the normal HHI␣ (with ˛ = 2), while the effect is smaller for ˛ < 2. In order to examine whether this characteristic plays a role, we calculate the HHI␣ with ˛ = 1.5 and with ˛ = 2.5. An advantage of the HHI␣ is that it is very sensitive to small differences in the distribution of activities over regions. A slight problem of using HHI␣ is that the estimated values can be relatively small, even if concentration is high.

i 

ai =

i 

i−1 

popsharej

(3a)

pj

(3b)

pj

(3c)

j=1

popsharej −

j=1

ci =

i−1 

i  j=1

popsharej −

j=1

i−1  j=1

3.2.3. Ellison–Glaeser index The Ellison–Glaeser index was originally developed to measure the extent to which the spatial distribution of industry-specific employment deviates from a random distribution (Ellison and Glaeser, 1997). The index is zero if the probability to find a firm in each region is proportional to the size of the region (measured in terms of total employment). Since employment is not equally distributed amongst firms, they also included the industry plant size distribution in their calculation. Their index is given by



n

=

r=1

(pr − xr )2 − 1 −



1−

n

x2 r=1 r



n

1−

x2 r=1 r

m

 m

z2 j=1 j

z2

j=1 j .

(4)

This equation is redefined and applied to the case of patents in the following manner. The share of patents in the regions pr is the core variable (originally the share of employment, si , in Ellison and Glaeser, 1997). These shares were originally compared to the regions’ share of total employment, xr . Instead, we make two alternative comparisons: 1. According to the share of the population xr that lives in region r. The resulting Ellison–Glaeser index implicitly assumes that each region should generate, in each technological field, a number of patents that reflects its population size. We call this the ‘weighted by population’ version. 2. According to the share of total number of patents xr that are generated in region r. In this case, the resulting Ellison–Glaeser index implicitly assumes that more innovative regions should generate more innovations in every technological field. We call this the ‘weighted by patents’ version. This version is in line with Ellison and Glaeser’s own approach since they relate the values of one variable (employment) in a specific field (industry) to this variable in total. Other definitions could be used. These would lie somehow between the two options that we have developed. For

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

169

Table 2 Correlation between selected concentration indices. Indices

HerfindahlIndex 1.5

HerfindahlIndex 2.5

Gini weighted by inhabitants

Gini unweighted

Herfindahl-Index 1.5 Herfindahl-Index 2.5 Gini unweighted Gini weighted by inhabitants Ellison–Glaeser (weighted by patents) Ellison–Glaeser (weighted by inhabitants) Moran’s I

1

0.9664*** 1

0.9195*** 0.8477*** 1

0.9494*** 0.8591*** 0.9596*** 1

Ellison–Glaeser (patents) −0.0253 −0.0574 0.1107 0.0862 1

Ellison–Glaeser (inhabitants)

Moran’s I

0.1693 0.2077 0.1683 0.1508

−0.2704* −0.1583 −0.2694* −0.3362**

0.7663***

−0.095

1

−0.0949

1

Note: Spearman rank correlation test used for the analysis; Spearman’s rho presented in the cells. * t < = 0.1. ** t < = 0.05. *** t < = 0.01.

example, one could argue that not all inhabitants in a region have the same propensity to innovate. Perhaps only the number of employed people should be used, or perhaps only the number of people employed in industries that are related to the studied technology should be considered. We propose that the two versions defined above represent the extreme cases that need to be considered. In practice, it turns out that the estimates of these two cases are nearly identical. Consequently, further options do not need to be considered. In the case of patents, j is the distribution of patent numbers across organisations. Hence j, as in Ellison and Glaeser (1997), is an index for each organisation, with m being the number of organisations active in patenting in the technology under consideration. zj is the share of all patents in a technology that come from organisation j. The values of this variable are also calculated using our patent data. In comparison to the Herfindahl index and the Gini coefficient, the Ellison–Glaeser index has the following specific characteristics. First, and most important for the empirical analysis, the Ellison–Glaeser index compares the empirical spatial distribution with a theoretical distribution that results from a random distribution of the given firms into the given regions according to their size. Hence, the distribution of patent activity among firms is included in the calculation of the ‘zero case’ as well as the different sizes of the regions. This implies that two causes for an unequal distribution among regions are included in the ‘zero case’: differences in the patent activity of firms and differences in regions. Hence, the spatial concentration that results from these two causes is not measured. Note that the ‘zero case’ is not that with the lowest concentration. The Ellison–Glaeser index takes both positive and negative values, ranging from −1 to +1. There might be more or less geographic concentration than expected according to the Ellison–Glaeser rationale.

3.2.4. Moran’s I Moran’s I was developed by Moran (1950) and is widely used in statistics that test the presence of spatial depen-

dence. It is an autocorrelation measure that is defined as I=

N·

N N

[wij · (xi − x¯ ) · (xj − x¯ )]

i=1

j=1

i=1

j=1

N N

wij ·

N

i=1

(xi − x¯ )2

(5)

where N is the number of spatial units and xi is the variable of interest at the spatial unit i (with the average given by x¯ ). wij is a matrix of spatial weights that characterises geographic proximity between the spatial units i and j. The Moran’s I measures to what extent the variable x takes similar values for nearby spatial units. Hence, the characteristics of the Moran’s I differ strongly from the characteristics of the other three measures of geographic concentration. Estimates using the other measures increase the more activities take place in the same few regions. By contrast, Moran’s I is the larger the more activities are concentrated in a few neighbouring, or nearby, regions. Hence, a high Moran’s I implies that activities are spread across a number of regions, with these regions being located next to each other. 4. Results 4.1. Distribution of technological activities The results of the calculations of the concentration indices are presented in Table A.2 of Appendix A. In the first step we examine the extent to which the results of the seven indices overlap or diverge. A rank correlation analysis of the seven indices indicates there exist some strong and significant correlations between some of the indices on the one hand, but also some significant differences on the other hand (Table 2). The first observation is that the two Ellison–Glaeser indices are correlated with a Spearman’s rho of around 0.75 but are not correlated with the other indices. This is probably due to the fact that the two Ellison–Glaeser indices are the only ones which already correct for the distribution of innovative activities across organisations. The second observation is that the two Herfindahl–Hirschman and Gini indices are correlated with at least 0.85. Neither the weighting by inhabitants nor different ˛’s used

170

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

Table 3 Ranking of technological fields. Ranking

Based on both Gini and Hirschman–Herfindahl indices

Based on both Ellison–Glaeser indices

Based on Moran’s I

1 2 3 4 5 39 40 41 42 43

Tobacco products Soaps, detergents, toilet preparations Pesticides, agro-chemical products Electronic components Watches, clocks Other transport equipment Furniture, consumer goods Special purpose machinery Rubber and plastic products Non-metallic mineral products

Soaps, detergents, toilet preparations Tobacco products Watches, clocks Weapons and ammunition Man-made fibres Pesticides, agro-chemical products Electric motors, generators, transformers Signal transmission, telecommunications Other electrical equipment Industrial process control equipment

Electric motors, generators, transformers Wearing apparel Domestic appliances Fabricated metal products Basis metals Wood products Accumulators, battery Paints, varnishes Pharmaceuticals Man-made fibres

in the Herfindahl–Hirschman estimates lead to a strong divergence of the indices. Finally, we see that Moran’s I is significant but negatively correlated with most of the Herfindahl–Hirschman and Gini coefficients. In other words, technological fields that are geographically concentrated according to the Hirshman–Herfindahl and Gini indices do not show signs of spatial autocorrelation, indicating that for these technologies neighbouring regions are not affected by high innovative activities in the core regions. For the examination of the rankings of the different technologies the indices in these three groups are aggregated and the results for the groups are presented. The analysis in Section 4.2 takes into account all seven indices separately. In order to give an impression of which technologies are most spatially concentrated, the indices are aggregated within the three groups, and ranked in Table 3. This aggregation is performed by calculating the average of the rank positions for the individual indices. We find that ‘Tobacco products’, ‘Soap, detergents, toilet preparations’, and ‘Pesticides, agro-chemical products’ are the technological fields with the strongest local concentration, as measured by HHI/Gini. A Ellison–Glaeser estimate with a threshold of 0.04 is used to identify those technological fields that tend to concentrate in space. Only two technological fields are identified: ‘Tobacco products’, and ‘Soaps, detergents, toilet preparations’. This is the case regardless of which version of the Ellison–Glaeser indices is used. Thus, hypotheses of a random distribution of activities are rejected for some technologies. For all technological fields there exists a positive spatial autocorrelation (Moran’s I > 0). The highest values can be identified for ‘Electric motors, generators, transformers’, ‘Wearing apparel’, and ‘Domestic appliances’. The three groups of indices cover different aspects of geographic concentration. There is some overlap but there are considerable differences. Notably, Moran’s I differs from the other indices. This is not unexpected given that Moran’s I analyses spatial autocorrelation and not just the general geographic distribution. The field ‘Electric motors, generators, transformers’ appears to have a very strong spatial autocorrelation–neighbouring regions show similar activities in this technological field. This field has a very low estimated ranking using the Ellison–Glaeser indices. This indicates that most of the concentration in this field can be explained by the distribution of activities over the firms and the inhabitants/patent numbers in the regions. The opposite holds for ‘Man-made fibres’. It

has low spatial autocorrelation and a higher geographic concentration than expected compared to a random distribution of activities. HHI/Gini and the Ellison–Glaeser indices show more overlap with ‘Tobacco products’, ‘Soap, detergents, toilet preparations’ and ‘Watches, clocks’ being most concentrated. But again the concentration of ‘Pesticides, agro-chemical products’ that is expected from HHI/Gini seems to be explainable by taking into account the distribution of innovation activities across firms and the distribution of inhabitants and patents across regions. 4.2. Hypotheses check We set up a number of hypotheses to test the different causes of spatial concentration in innovation and patent activities. The findings discussed in Section 4.1 indicate that patent activities are indeed concentrated in geographic space. However, this might be caused by one, a few, or all of the mechanisms proposed in Section 2. Therefore, the hypotheses should be seen as presenting alternative explanations which may also be jointly involved. Our task in this section is to test each of the individual explanations. Each hypothesis is dealt with in turn. Before this can be done, however, we have to define and discuss the variables and the statistical method that is used. 4.2.1. Explanatory variables In order to test the hypotheses we make use of empirical data on a number of explanatory variables. Tables A.4 and A.5 in Appendix A give an overview on these factors. All values were calculated based on our patent database described above or on data from the German ‘Institut für Arbeitsmarkt- und Berufsforschung (IAB)’. Again, if a patent was assigned to several technological fields, it was counted for each technological field only fractional, in such a way that the total count of each patent sums to 1. We test Hypothesis 4 in two ways which are complementary to each other. These two approaches differ with respect to their spatial connotation. We argue that one way of measuring whether collaboration plays an important role for the innovation processes in a technology is to calculate the share of patents with more than one inventor. The fact that agents profit from interacting in the innovation activities is reflected in this coinventorships. If patents have multiple inventors, the innovation can be assumed to be the result of the cooperation of actors, within the same firm or organisation or even beyond

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

organisational boundaries. Hypothesis 4 is based on the argument that the importance of local interaction implies a spatial concentration of innovation activities. Using coinventorship means that we implicitly assume that the importance of cooperation leads to more local interaction, either because the innovators co-locate to make interaction easier or because co-located innovators are more productive. The analysis of the share of patents with multiple inventors shows that in the chemical and pharmaceutical fields (including, e.g. ‘Basic chemicals’, ‘Pharmaceuticals’ or ‘Soaps, detergents, toilet preparations’) over 80% of the patents are done co-operatively, while in the more strongly consumer oriented fields (‘Watches, clocks’, ‘Leather articles’ or ‘Furniture, consumer goods’) there is a stronger tendency for single inventor patents. Our second variable for measuring the importance of local collaboration is based on the argument that if local interaction plays a role for the innovation process, we should see a high number of patents with inventors from the same region. In order to calculate the share of inventors who are from the same region we only consider patents with exactly two inventors. Through this we avoid a distortion between the technologies which might results from different average numbers of inventors per patent. The share of inventors from the same region shows the highest value for ‘Paints, varnishes’ and the lowest one for ‘Pesticides, agro-chemical products’. In the former case, around 63% of the patents with two inventors are with partners from the same region, while in the latter case this value amount only to 34%. The extent to which a technology is science-based is measured by the share of non-patent citations in the patent files here. It is argued that if technologies build strongly on scientific research, this research should be cited in the patent applications. ‘Pharmaceuticals’ has the highest average share of non-patent citations compared to all citations (non-patent plus patent citations to previous work), while ‘Leather articles’ has the lowest value. Although nonpatent citations cover many different kinds of material, especially scientific articles, books or papers that are cited on a patent from a technological field play an important role. The result strongly mirrors the scientific orientation of the technologies or is at least the best available proxy for this scientific orientation. While ‘Pharmaceuticals’ (in part including biotechnology as well) is science-based with many patents originating from research institutes or universities, consumer goods (including ‘Tobacco products’ or ‘Wearing apparel’) are less science-based and the share of freelance inventors is quite high in this field. In order to measure whether the innovation activities is equally spread among firms or dominated by a few firms, we use a Herfindahl index. We employ the Herfindahl index of the distribution of patent activity among organisations. In the theoretical part, we argued that the concentration of innovation activities should depend on the distribution of patent activities among firms. We stated that this does exactly hold only if the firms innovate in one place. The data that we have and use for the calculation of the above Herfindahl index declares for each patent an identification number of the patenting organisation. Small firms have one identification number and one location, so that our

171

assumption is satisfied. The same holds for some larger firms that have different locations but a unique identification number for each location. However, there are also some larger firms that do not satisfy our assumption because they have several locations in which they do research but only one identification number. Nevertheless, the best that we can do to be as near to our assumption as possible is calculating the Herfindahl index on the basis of identification numbers of organisations. This leads to the correct consideration of most firms with respect to our Hypothesis 1. In the case of a few large firms the assumption behind Hypothesis 1 is not given which has to be kept in mind in the interpretation of the results. This distribution varies strongly between technologies with a minimum of 0.002 (‘Rubber and plastic products’) and a maximum of 0.138 (‘Soaps, detergents, toilet preparations’). Hence, in some technological fields (including ‘Tobacco products’ or ‘Pesticides, agro-chemical products’) the patenting activities are strongly affected by single dominant organisations. Finally, the concentration of the respective industrial activity is measured by a Herfindahl index. We use employment data for all 3-digit industries and the 97 regions used here. In this data all employees are assigned to regions and industries according to the plant assignment in which they work. The employment numbers for each industry are weighted by the relevance of the industry for each technological field (Schmoch et al., 2003). By this we obtain a Herfindahl index of industry employment related to each technological field. The data shows that in the field of ‘Watches, Clocks’ the employment related to the technological fields is strongly concentrated in a few regions with a HHI of 0.199. For ‘Food, Beverages’ the employment is more equally distributed between the different regions (0.031). 4.2.2. Regression analysis In order to test which of the above defined variables are related to the geographic concentration of the innovation activity in the technological fields, a regression analysis is conducted. Instead of relying on conventional OLS regression which is focused on the explanation of the mean of the dependent variable (e.g. the mean level of performance), the presented paper aims to investigate different parts of the distribution separately by applying a quantile regression method. This method allows for regressing against every percentile of the distribution of the dependent variable separately (Buchinsky, 1998).3 Our intuition is that, especially in the tails of the performance distribution (for those technologies with a very high and a very low level of concentration), the coefficients estimated by a quantile regression will deviate from those of a conventional OLS approach. This holds especially since the two HHI and the two Ellison–Glaeser indices are not normally distributed. We calculated the 0.25, 0.5 (median) and 0.75 quantiles. The coefficients of the quantile regressions, reported in Table 5 , can be interpreted as the partial derivative of the conditional quantile of y with respect to particular regressors, ıQ (yit |xit )/ıx (Coad and Rao, 2006). 3 For more information on the advantages of quantile regressions, as well as for details on the calculations, see Koenker and Basset (1978), Koenker and Hallock (2001), and Buchinsky (1998).

172

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

Table 5 Quantile regression results. Share of patents with multiple inventors Model 1 HHI15

Constant

R2

0.1538* 0.1094*** 0.0498

0.121 0.0073 −0.0043

0.3089 0.4437 0.5653

0.6325*** 0.8556*** 1.4359***

0.2249* 0.1426** 0.0365

0.0175 0.0153 −0.0221

0.2606 0.391 0.5655

0.0572 0.2955 0.68

0.4587 0.3923 0.3567

Share of patents where all inventors are from one region

Share of non-patent citations in relation to all citations

−0.015 0.101 0.2792

– – –

0.4111*** 0.5356*** 0.8683***

0.0219 0.0662 0.0796*

– – –

HHI Firms

HHI Industries

0.25 0.5 0.75

0.0158 0.0081 0.1568

HHI25

0.25 0.5 0.75

0.0183 0.0083 0.0142

Gini W

0.25 0.5 0.75

0.2844*** 0.078 −0.0477

0.0575 0.0157 −0.3525

– – –

2.1086*** 2.2404*** 2.1746***

2.1654*** 1.5581*** 0.9282**

0.0856

−0.0322

–

2.3492***

1.7461***

0.3098

0.618

0.2434*** 0.1338 0.1543

0.084 0.551 −0.1894

– – –

1.3653*** 1.434*** 1.2726*

1.3463*** 1.0337*** 0.8281*

0.3408*** 0.4690*** 0.6357

0.4996 0.4554 0.3922

0.1312

0.0038

–

1.4229***

1.1541***

0.499***

0.6746

0.0242 0.0125 0.0299

0.4041 0.1686 0.0737

OLS Gini not w

0.25 0.5 0.75 OLS

E–G pat

0.25 0.5 0.75

0.0015 −0.0072 −0.0183

−0.5583* −0.0165 −0.0305

– – –

−0.7598*** −0.4648*** 0.3931***

0.2235*** 0.1519*** −0.0049

E–G EW

0.25 0.5 0.75

0.0093 −0.004 0.0009

−0.0271 0.0275 0.0527***

– – –

−0.5370*** −0.3843*** 0.2112***

0.222** 0.1604*** 0.0670***

0.0065 −0.0087 −0.0245***

0.2911 0.0879 0.1283

Moran

0.25 0.5 0.75

−0.2064*** −0.1724 −0.2523***

−0.0120 0.2245 −0.0124

– – –

0.1550 0.2090 0.4228**

0.3266*** 0.0605 −0.2045

0.4399*** 0.3485** 0.5596***

0.2655 0.1558 0.2406

−0.2244***

0.114

–

0.4325*

0.0959

0.4264***

0.3343

0.0173 0.0126 0.0096

0.3185 0.4522 0.5638

0.0275 −0.0072 −0.0251

0.2771 0.3991 0.4587

OLS Model 2 HHI15

0.25 0.5 0.75

– – –

−0.0132 0.006 0.0156

0.0348 0.0248 0.0182

0.3950*** 0.5156*** 0.8409***

0.1484** 0.1062*** 0.0475

HHI25

0.25 0.5 0.75

– – –

−0.0261 0.556 0.0985*

0.0399 0.027 0.0138

0.6558*** 0.8219*** 0.4892***

0.2097 0.1442** 0.0211

Gini W

0.25 0.5 0.75

– – –

0.0896 0.1075 −0.3936

0.3863* 0.2255 0.0368

2.3005*** 2.4810*** 2.2580***

1.9799*** 1.6156*** 1.1405*

0.175 0.2460* 0.6311*

–

−0.071

0.1198

2.4305***

1.7807***

0.3643**

0.6164

– – –

0.0614 0.2562 −0.1634

0.3517*** 0.1864 0.209

1.3532*** 1.7224*** 1.6667***

1.2703*** 1.3728** 0.7536*

0.4556*** 0.5241*** 0.6809***

0.4395 0.4269 0.4036

–

−0.0563

0.1927**

1.5429***

1.2083***

0.5817***

0.6684

0.0276 −0.0284** −0.0022

−0.8512*** −0.4332*** 0.3065***

0.2383*** 0.1473*** 0.0227

0.0076 0.0066 0.024

0.4138 0.1693 0.0719

−0.4672*** −0.3880*** 0.2126***

0.2104* 0.1654*** 0.0671***

−0.0015 −0.0194 −0.0241***

0.3031 0.0903 0.1281

OLS Gini not w

0.25 0.5 0.75 OLS

0.4273 0.3937 0.3552

E–G pat

0.25 0.5 0.75

– – –

−0.0282 −0.0075 −0.0409

E–G EW

0.25 0.5 0.75

– – –

−0.0069 0.0405 0.0525***

0.021 0.0097 0.0012

Moran

0.25 0.5 0.75

– – –

0.2257 0.2346 0.3864**

−0.3111*** −0.2504 −0.3182***

0.0349 −0.0328 0.5824*

0.3740** 0.2691 −0.0587

0.2344** 0.2475* 0.2183**

0.2117 0.1513 0.1438

–

0.2134*

−0.2286***

0.2062

0.0086

0.2818***

0.2303

OLS

Note: Coefficients reported in cells; shaded cells are at least significant with p < = 0.1. * p < = 0.1. ** p < = 0.05. *** p < = 0.01.

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

Yasar et al. (2006) stated that the derivative may be seen as the marginal change in y at the th conditional quantile due to marginal change in a particular regressor. In other words, for each determinant the point estimates may be interpreted as the impact of a one-unit change of the determinant on the level of concentration. For those concentration indices for which the assumption of a normal distribution cannot be rejected (both Gini indices and Moran’s I), we additionally applied conventional robust OLS regressions (conditional mean effect). To avoid problems of multicollinearity (see Table A.3 in Appendix A) two different models are tested. In the first model we include the ‘share of patents with multiple inventors’ and in the second one the ‘share of non-patent citations in relation to all citations’. The results of the OLS and the quantile regressions for the two tested models and the seven different spatial concentration measures as dependent variables are given in Table 5.4 The discussion of the four hypotheses is based on the results in that table. 4.2.3. Hypothesis 1 The first hypothesis states that, if the distribution of innovation activities among firms is an important determinant for the spatial distribution of innovation activities, spatial concentration should positively depend on the Herfindahl index for the distribution of patents among firms. Examples for such dominance of one or a few firms are: • 50% of all patent applications in the technology ‘Soaps, detergents, toilet preparations’ come from Düssseldorf with the dominant firm Henkel. • 30.6% of all patent applications in the technology ‘Tobacco production’ are from Schleswig-Holstein South with the dominant firm Reemtsma (which is located in Hamburg, but many inventors are residing in Schleswig-Holstein South). • 28.6% of all patent applications in the technology ‘Electronic components’ originate from Munich with the dominant firms Siemens and Infineon. • 25.1% of all patent applications in the technology ‘Signal transmission, telecommunication’ are generated in Munich with the dominant firm Siemens. • 24.4% of all patent applications in the technology ‘Motor vehicles’ come from Stuttgart with the dominant firms Bosch, Porsche and Daimler. These are just some examples for the fact that one or a few firms might cause the concentration of innovation activities in one technological field within one region. Our regression results confirm this impression for measuring geographic concentration of innovation activities with the Herfindahl index (both variants) and the Gini coefficient (both variants). This means that the unequal distribution of innovation activities among firms contributes to the geographic concentration of innovation activities.

4 We also applied robust OLS regressions with logarithmised and transformed dependent variables as well as Tobit, truncated and robust OLS regressions with the original dependent variables. There are some differences with regards to the significance of the variables, but the signs of the coefficients and, hence, the directions of causes remain the same.

173

Slightly different results emerge for the relationship between HHI Firms and the geographic concentration measures represented by the Ellison–Glaeser indices and the Moran’s I. The Ellison–Glaeser index explicitly takes into account the distribution of patents among the firms. It measures geographic concentration beyond the amount that is expected according to this distribution. We do find significant effects of HHI Firms, which was unexpected since HHI Firms is already included in the Ellison–Glaeser index, but we find significantly negative ones for the lower two quantiles and a positive one for the upper quantile. This can be explained as follows: If HHI Firms is very large, meaning that patent activity is concentrated in a few firms, the denominator in the Ellison–Glaeser index becomes small. The idea in the Ellison–Glaeser index is that the potential for deviations from the predicted distribution in the numerator is also smaller. If this is not the case because the deviations have other statistical properties than assumed in the Ellison–Glaeser index, deviations lead to higher absolute values of the Ellison–Glaeser index for high values of HHI Firms. A reason for the different statistical properties might be the fact that we use organisations with different identifications for the calculation of HHI Firms as discussed above. The implication would be that the most positive and the most negative values of the Ellison–Glaeser index result for the technologies with the highest values of HHI Firms. This is what we observe. Therefore, we conclude that the arguments above are correct and the results are not caused by economic mechanisms but by the definition of the Ellison–Glaeser index. Moran’s I measures the relationship between the innovation activities in neighbouring regions. The domination of the patent activities by a few firms does not imply a relationship of the innovation activities in neighbouring regions per se. However, if a few firms dominate the patent activities, these firms are usually large. This might increase the likelihood that not all of their workers live in the same region in which the firm is located, but probably in a neighbouring region. In an analysis of inventor residences, as done here, this might lead to a positive relationship of patent activities in neighbouring regions. Our regression results confirm this especially for the 0.75 quantile, while for other quantiles no significant relationship is found. Hence, especially those technological fields which are highly concentrated in neighbouring regions, are those in which a strong concentration of innovative activities in a few firms has a positive effect. To sum up, our empirical results are well in line with the argument that part of the geographic concentration of patent activities is caused by the fact that in some technological fields a few firms do almost all patent activity. The mixed effect for the Ellison–Glaeser indices is due to the specific characteristics of these indices. Only those technologically fields which are highly concentrated in neighbouring regions, are positively affected by the concentration of innovative activities in a few firms. 4.2.4. Hypothesis 2 Hypothesis 2 states that the geographic concentration of innovation activities is strongly related to the geo-

174

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

graphic concentration of the employment in the respective industries. This implies that the innovation activities are concentrated in the same regions as the respective industrial activities. We do not compare the locations of innovation and industrial activities here. Instead, we check whether technological fields that are strongly concentrated in space correspond to industries that also show such a strong geographic concentration. For example, innovations in the technological field ‘Weapons and ammunition’ are strongly concentrated in space and the same holds for the industries ‘Manufacture of weapons and ammunition’ (Sternberg and Litzenberger, 2004). This example supports Hypothesis 2, but there are also counter-examples: ‘Non-metallic mineral products’ is weakly concentrated in space, while nearly all mining related industries are strongly concentrated (Sternberg and Litzenberger, 2004; Alecke et al., 2006). We obtain a much clearer result from our regression analysis (see Table 5). The Herfindahl index for the related industry employment (‘HHI Industries’) is positively related to the geographic concentration. Significant confirmations for such a relationship are obtained for all measures and nearly all quantiles. Hence, the concentration of industrial activities is related to the concentration of innovative activities as assumed by Hypothesis 2. At the same time we find that often for the 0.75 quantile no significant results can be identified and for nearly all indices the coefficients decrease from the 0.25 to the 0.75 quantile. Thus, the higher the geographic concentration in a technological field, the less impact can be ascribed to the industrial concentration. This also holds for Moran’s I for which only the 0.25 quantile is significant. Only those technological fields with a low spatial autocorrelation, are positively affected by the geographic concentration of the related industries (HHI Industries). This might be interpreted as follows. The spatial concentration of industries has, of course, an impact on the spatial concentration of the related innovation activity. However, the spatial concentration of innovations that results from this transfer from industries to innovation activity is far below the original spatial concentration in the industries (the coefficients for HHI Industries are far below 1). Therefore a strong spatial concentration of the innovation activities can only be reached if additional causes, maybe within the innovation activity itself, are given. Hence, for those technologies with a low or medium concentration level the concentration of the related industries is a good explanatory variable, while a high concentration has to be explained by other additional variables, e.g. by local co-invention as our results suggest. To sum up, Hypothesis 2 is by-and-large confirmed by our regression results. There is a positive relation between the spatial distribution of industrial employment and innovation activities. But this effect is the smaller, the higher the concentration of technological activities in geographic space. 4.2.5. Hypothesis 3 Hypothesis 3 states that science-based technologies should be more geographically concentrated than other technologies. This is partly confirmed by our regression analysis. A significantly positive relationship is found for

the two Gini coefficients. For the unweighted Gini coefficient the OLS regression delivers a positive and significant coefficient. The quantile regressions indicate that this is especially based on the 0.25 quantiles. A significantly negative relationship is obtained for Moran’s I and the Ellison–Glaeser index weighted by patents. No significant results are identified for the two Herfindahl indices and the Ellison–Glaeser index weighted by inhabitants. Hence, whether technologies are science-based seems to matter for the geographic concentration of innovation activities, but only for those technological fields that are less geographically concentrated. The more concentrated technological fields are not affected significantly by the geographic proximity to science. The less geographically concentrated industries are more concentrated if their innovation activities rely on scientific inputs. The interpretation is similar to what we found for the spatial concentration of the related industries. An importance of scientific research seems to restrict the spatial spread of innovation activity to some extent as proposed in Hypothesis 3. However, this impact leads only to a quite weak concentration in space. All spatial concentration beyond that point (beyond the lowest quantile) can only be explained by other factors and mechanisms. The coefficients for the Ellison–Glaeser index weighted by patents are all very small and just one of them is significant. This might be a statistical artefact. Hence, we will not discuss this finding in depth. It is also not supported by the result for the Ellison–Glaeser index weighted by inhabitants. The negative relationship between the science-base of a technology and the Moran’s I of its patent activity can be interpreted as follows: Innovation activities in science-based technologies take place nearby the locations of public research, meaning mainly in big cities and agglomeration centres. These regions are spread across Germany with quite some space between them. In contrast, innovation activities in technologies that are less sciencebased might also spread to other regions and, hence, might locate in a number of regions that are neighbouring each other. A higher spatial correlation is found for these technologies. To sum up, Hypothesis 3 is only confirmed partly. Among the less geographically concentrated technologies science-based technologies show a higher geographic concentration in their innovation activity. 4.2.6. Hypothesis 4 The fourth hypothesis is based on the argument that technologies with a strong importance of (local) collaboration for the innovation process should be more concentrated in space. Two variables that measure the importance of (local) collaboration are used in our regression: the share of patents with multiple inventors and the share of patents with all inventors located in the same region. Hypothesis 4 is partly confirmed by our analysis. The two Gini coefficients show a significant positive relationship to the share of patents with multiple inventors for the 0.25 quantiles. We might conclude that, at least, technologies that are less concentrated in space are positively affected by the number of inventors. We can again

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

argue that an importance of co-patenting leads to a certain degree of spatial concentration. However, the spatial concentration that results from this effect is not very strong and higher concentrations can only be explained by other mechanisms. Moran’s I indicates a strongly significant, negative relationship with the share of patents with multiple inventors. If many actors are involved in innovation activities, this innovation activity does not show spatial correlation, but might be concentrated in the core region only or might link geographically distant inventors. We can conclude that collaborations do not spread locally and, hence, do not involve neighbouring regions. Additionally our results indicate that for those technological fields which are highly concentrated, the share of local inventors has a positive and significant effect for the 0.75 quantiles; although the coefficients are relatively small. This result holds for one Herfindahl index (HHI25) and the Ellison–Glaeser index weighted by inhabitants. Here we see the opposite effect to what we find for many of the other variables. Many of the other variables, especially the spatial concentration of the relevant industries, the relation to science and the importance of co-patenting, cause spatial concentration to a certain degree but not beyond. In contrast, the importance of local interaction seems only to have a visible effect if a certain degree of spatial concentration is exceeded. Either this impact only works if spatial concentration is high, or other impacts are stronger for smaller concentrations, so that it is statistically not detected. Moran’s I is positively related to the share of patents with all inventors in the same region for the 0.75 quantile. It implies that a strong importance of local interaction in the innovation activity leads to correlations between innovation activities in neighbouring regions. It is interesting to compare this result to the effect of the rate of co-patenting on spatial autocorrelation. We found above that a high share of co-patenting reduces the relationship between neighbouring regions. Here we find that a high share of local co-patenting increases the relationship between neighbouring regions. Hence, only if space matters for the interaction in innovation activities, innovations activities are connected between neighbouring regions. Otherwise we even see the opposite. On the other side the Ellison–Glaeser weighted by patents has negative coefficients and the one for the 0.25 quantile is significant. Those technologies which are active in the patent intensive regions have relatively low values of this Ellison–Glaeser index. It seems as if these are the same technologies which are also linked to local interaction. Hence, local interaction takes place in those technologies which are active in patent intensive regions since they offer a lot of opportunities for such interaction. To sum up, we find some evidence for Hypothesis 4. We interpret the negative dependence for the Ellison–Glaeser index as a specific finding with the above interpretation. The remaining results can be separated into two parts. First, for spatial autocorrelation (Moran’s I) we find the interesting result that the importance of local interaction increases the connection between neighbouring regions, while the importance of collaboration in principle decreases this

175

connection. Second, the other results imply that the importance of collaboration leads to some degree of spatial concentration in innovation activities. The importance of local interaction increases the spatial concentration if a certain degree of such concentration is exceeded. 5. Conclusions The aims of this paper are two-fold. First, to analyse which innovative activities tend to concentrate in geographic space. Second, to examine the characteristics of innovative activities which lead to different concentrations. In doing this, we have sought to develop a new perspective for analysing industry concentration (one that is complementary to the more common analysis of industrial employment), and to provide an empirical application. The main factors influencing the geographic concentration of innovative activities are the effect of single dominant firms, the distribution of industrial activities, the relatedness of technologies to the scientific base, and knowledge transfer based on local interactions. We have used patent applications as a proxy for innovative activities in 43 technological fields and 97 regions in Germany. Our findings indicate that some regions play a central role in certain technological fields, resulting in a high share of innovative activities in these fields. The last phenomenon can at least partly be explained by the local industrial structure, including single firms located in these regions. Furthermore, innovative activity in Germany is dominated by 11 regions, with Munich, Stuttgart, Düsseldorf and, Rhine-Main at the top of the list. Still, we find considerable variation across the regions, and innovations are wide-spread across geographic space. The concentration of the technological fields has been measured using Hirschman–Herfindahl indices, Gini coefficients, Ellison–Glaeser indices and Moran’s I. We find considerable overlap exists, and there is high positive correlation between some of the indices employed. At the same time, there is a high divergence between groups of indices. These findings of internal coherence but external divergence are also reflected in the regression results. They are to be understood in the following way. The two Herfindahl–Hirschman and Gini indices are highly correlated. Neither the weighting by inhabitants nor the different ˛ for the Herfindahl–Hirschman index can lead to a strong divergence of the indices. Since the Gini coefficients are normally distributed, they are more applicable to standard regression analysis. Since it makes no difference whether the inhabitants are included in the coefficient or not, the unweighted Gini coefficient should be used in future research. The HHI with ˛ = 2 or higher produces some particular results for technologies that are strongly concentrated. If these are the focus of the research, then this index should also be included. The Ellison–Glaeser indices differ from the other indices in that they are the only ones which correct for the distribution of innovative activities across organisations. Furthermore, the Ellison–Glaeser indices take into account the size of the region, as measured by the number of inhabitants or patents. Whether this correction should be included in the index or during the regression analysis

176

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

depends on the research question. From the regression analysis performed in this paper, we have the impression that the correction included in the Ellison–Glaeser index does not work equally for the different parts of the distribution. Hence, if a regression analysis is conducted with the distribution of activities over organisations being one of the explanatory variables, it is probably better to use an uncorrected concentration measure. Having said this, any regression analysis needs to take into account the fact that the dependent variable is unlikely to be normally distributed. If an index is to be used to describe the geographic distribution of innovative activities, then the Ellison–Glaeser index can be employed. The advantage is that it corrects for an influence that is very strong, and is based on a mechanism that is not connected to spatial economic forces, such as clustering. Moran’s I is negatively correlated with the Herfindahl–Hirschman index and with both of the Gini coefficients. This is because Moran’s I is a measure of spatial autocorrelation. Moran’s I (or an alternative measure of spatial autocorrelation) should be included in future research because it provides additional insight to the discussion of spatial concentration. This is supported by the regression results. We find that technological activities strongly differ in their tendency to concentrate in geographic space. Furthermore, the ranking strongly depends on the concentration index that is used. Notably, the field ‘Electric motors, generators, transformers’ was found to have very strong spatial autocorrelation while it has a very low Ellison–Glaeser ranking. This indicates that most of the concentration in the core German regions can be explained by the distribution of firm activities, and by the inhabitants/patent numbers in those regions. Still, we found that some overlap exists in certain technological fields. We additionally tested each of the individual factors which could influence the spatial concentration of innovative activities. There is stronger supporting evidence for some of these factors than for others. The most strongly supported factor is the existence of a small number of dominant firms, who are very active in their respective fields but are located in just a few locations. In addition, the geographic concentration of industrial activities related to technologies is identified as empirically significant. However, the concentration of related industrial activity only leads to a certain degree of spatial concentration of the innovative activity. Concentration beyond this point is explained by other factors. Dependence on scientific knowledge was only found to have some relevance for some indices. Once again, this factor can only explain a certain level of spatial concentration of innovation. Interestingly, this variable has a negative effect on Moran’s I, indicating that public scientific research limits the location of innovative activity to certain regions–this tends to be large cities and agglomeration centres, which are geographically scattered. We found only limited evidence for the importance of interactions in the innovation process. Yet the limited evidence which does exist provides some interesting insights. The importance of collaboration independent of space increases the concentration of innovation activ-

ity somewhat, though only up to a quite low level. We find that local interaction leads to an additional increase in concentration provided a certain level of concentration is attained. Again, Moran’s I provides some different insights. Technologies with a highly collaborative innovation process appear to be characterised by low spatial autocorrelation, while technologies with strong local copatenting are characterised by high autocorrelation. Hence, collaboration per se has the opposite effect of local collaboration. We conclude that innovative activities have different tendencies to concentrate in geographic space, and that there exist underlying mechanisms which affect the propensity of the innovations to concentrate. Some policy implications can be drawn from these initial findings. First, if the policy objective is to generate local technological ‘hotspots’ then it is important to select the right technology because not all technology fields have a tendency to concentrate. Second, the correct mechanisms must be put in place to support concentration. Our empirical findings indicate that not all mechanisms reinforce concentration processes. Of the mechanisms which do appear to be significant, the role of scientific institutes is perhaps one of the most interesting ones for policy makers. Our findings indicate that science-based innovation activities in Germany are strongly concentrated in large cities and agglomeration centres where there are public research organisations in the relevant fields. Local collaboration between inventors is also found to play an important role in some technologies. For these technologies we find above-average spatial autocorrelation. This implies that, in sectors where local proximity is important for collaboration, innovations mainly take place in neighbouring regions. Policy makers should take this into account when supporting industrial collaboration. Looking forward, the work presented in this paper can be extended in four interesting directions. First, the framework can be adapted to different countries in order to make international comparisons. Second, following the 1999 Ellison and Glaeser paper, local factors (such as local universities) can be included to explain the geographic distribution of innovative activities. This would provide an opportunity to distinguish between concentrations that are due to certain local factors and those that are due to self-reinforcing processes. Third, an analysis of the change in concentration tendencies over time is possible. This would enable one to investigate the stability of the identified tendencies. Finally, the level of analysis can be altered in order to identify regional technological clusters. Each of these directions offer rich opportunities for new research.

Acknowledgements We are grateful to Despoina Filiou, Bart Verspagen, Claudia Werker and Paul Windrum, for helpful comments and discussions. Especially we would like to thank Anne Otto for her help with the quantile regressions. We also would like to thank two anonymous referees for their suggestions to improve and refine the article. All errors are ours.

Appendix A. See Tables A.1–A.5. Table A.1 Overview of the technological fields. Technological field

Total number of patents

Mean number of patents

Std. Dev. patents

Maximum share of patents in a region

Minimum share of patents in a region

Maximum number of patents in a region

Region with highest share in patents

Number of regions with at least one patent

1 2

Food, beverages Tobacco products

920.01 116.22

9.48 1.20

12.46 4.81

7.4% 30.6%

0% 0%

68.15 35.54

92 41

3 4

Textiles Wearing apparel

699.04 174.42

7.21 1.80

14.29 3.22

11.6% 11.4%

0% 0%

81.07 19.90

5

Leather articles

187.01

1.93

3.77

14.1%

0%

26.36

6 7 9

Wood products Paper Petroleum products, nuclear fuel Basic chemical Pesticides, agro-chemical products Paints, varnishes Pharmaceuticals Soaps, detergents, toilet preparations Other chemicals Man-made fibres Rubber and plastict products Non-metallic mineral products Basis metals Fabricated metal products Energy machinery Non-specific purpose machinery Agricultural and forestry machinery Machine-tools Special purpose machinery Weapons and ammunition Domestic appliances Office machinery and computers

177.89 892.35 379.64

1.83 9.20 3.91

2.89 14.77 6.84

8.0% 10.6% 9.6%

0% 0% 0%

14.21 94.56 36.49

Munich Schleswig-Holstein South Düsseldorf South East Upper Bavaria Industrial Region Middle Franconia Bielefeld Munich Cologne

15,212.26 1,116.43

156.83 11.51

344.61 36.13

13.9% 20.2%

0.005% 0%

2,115.95 225.18

Rhineland-Palatinate Rhineland-Palatinate

97 89

45.20 11,355.15 1,167.38

0.47 117.06 12.03

0.92 213.70 60.01

8.8% 9.0% 50.0%

0% 0.001% 0%

4.00 1,022.95 583.68

Brunswick Rhine-Main Düsseldorf

43 97 77

1,415.18 152.22 7,113.71

14.59 1.57 73.34

26.52 3.55 80.22

11.6% 11.6% 5.2%

0% 0% 0%

164.22 17.60 372.52

Düsseldorf Rhine-Main Rhine-Main

93 63 96

4,422.90

45.60

49.84

6.7%

0.022%

295.79

Rhine-Main

97

2,733.14 7,380.36

28.18 76.09

50.77 115.13

12.2% 10.6%

0% 0.005%

334.39 778.75

Düsseldorf Düsseldorf

95 97

8,250.89 7,199.99

85.06 74.23

128.67 108.68

11.8% 12.6%

0% 0.011%

976.09 905.34

Stuttgart Stuttgart

96 97

1,645.60

16.96

27.59

10.1%

0%

166.77

Münster

95

5,033.13 12,780.67

51.89 131.76

88.80 168.05

14.9% 7.9%

0.004% 0.006%

751.26 1,011.34

Stuttgart Düsseldorf

97 97

541.83

5.59

13.35

19.5%

0%

105.80

77

4,183.89 7,956.24

43.13 82.02

58.46 191.01

6.2% 21.2%

0% 0%

257.97 1,686.36

Black Forest-Baar-Heuberg Munich Munich

10 11

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

87 64 63 73 92 86

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

No.

93 96 177

178

Table A.1 (Continued ) Technological field

Total number of patents

Mean number of patents

Std. Dev. patents

Maximum share of patents in a region

Minimum share of patents in a region

Maximum number of patents in a region

Region with highest share in patents

Number of regions with at least one patent

29

Electric motors, generators, transformers Electric distribution, control, wire, cable Accumulators, battery Lightening equipment Other electrical equipment Electronic components Signal transmission, telecommunications Television and radio receivers, audiovisual electronics Medical equipment Measuring instruments Industrial process and control equipment Optical instruments Watches, clocks

1,662.51

17.14

31.71

12.6%

0%

209.87

Stuttgart

91

4,000.17

41.24

60.32

8.0%

0%

320.98

Berlin

94

932.65 828.77 2,221.38

9.61 8.54 22.90

19.23 15.17 41.08

11.8% 12.6% 10.4%

0% 0% 0%

110.02 104.55 230.00

Stuttgart Arnsberg Munich

85 86 94

4,606.78 11,525.77

47.49 118.82

141.15 337.06

28.6% 25.1%

0.000% 0.003%

1,318.12 2,895.82

Munich Munich

96 97

2,026.62

20.89

47.32

17.4%

0.000%

352.09

Munich

89

6,505.29 8,214.12 1,815.81

67.06 84.68 18.72

93.75 131.19 35.89

8.9% 10.3% 13.2%

0.000% 0.015% 0.000%

581.14 848.71 239.04

95 97 93

2,689.94 139.00

27.73 1.43

47.50 3.38

12.7% 17.5%

0.000% 0.000%

341.72 24.28

Motor vehicles Other transport equipment Furniture, consumer goods Overall

18,487.67 2,061.22

190.59 21.25

486.81 30.29

24.2% 11.5%

0.004% 0.003%

4,480.43 237.82

Munich Stuttgart Industrial Region Middle Franconia Munich Black Forest-Baar-Heuberg Stuttgart Munich

2,699.39

27.83

35.74

7.1%

0.000%

191.17

Stuttgart

94

173,669.81

1790.41

2531.20

8.6%

0.010%

15,004.12

Stuttgart

97

30 31 32 33 34 35 36

37 38 39 40 41 42 43 44

96 53 97 97

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

No.

Table A.2 Concentration indices for the technological fields. Technological field

HHI 1.5

HHI 2.5

Gini weighted by inhabitants

Gini unweighted

E&G (patents)

E&G (inhabitants)

Moran’s I

1 2 3 4 5 6 7 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44

Food, beverages Tobacco products Textiles Wearing apparel Leather articles Wood products Paper Petroleum products, nuclear fuel Basic chemical Pesticides, agro-chemical products Paints, varnishes Pharmaceuticals Soaps, detergents, toilet preparations Other chemicals Man-made fibres Rubber and plastic products Non-metallic mineral products Basis metals Fabricated metal products Energy machinery Non-specific purpose machinery Agricultural and forestry machinery Machine-tools Special purpose machinery Weapons and ammunition Domestic appliances Office machinery and computers Electric motors, generators, transformers Electric distribution, control, wire, cable Accumulators, battery Lightening equipment Other electrical equipment Electronic components Signal transmission, telecommunications Television and radio receivers, audiovisual electronics Medical equipment Measuring instruments Industrial process control equipment Optical instruments Watches, clocks Motor vehicles Other transport equipment Furniture, consumer goods

0.024 0.140 0.041 0.036 0.040 0.031 0.031 0.035 0.048 0.092 0.046 0.038 0.190 0.036 0.055 0.020 0.020 0.033 0.027 0.027 0.025 0.030 0.030 0.023 0.053 0.025 0.047 0.036 0.028 0.042 0.034 0.035 0.068 0.064 0.048 0.025 0.028 0.037 0.032 0.055 0.052 0.025 0.023

0.032 0.201 0.059 0.050 0.058 0.040 0.042 0.047 0.070 0.125 0.055 0.050 0.319 0.051 0.069 0.025 0.025 0.054 0.041 0.041 0.041 0.045 0.052 0.031 0.083 0.033 0.085 0.054 0.036 0.060 0.051 0.051 0.132 0.118 0.076 0.036 0.042 0.058 0.049 0.079 0.102 0.038 0.031

0.375 0.880 0.572 0.571 0.626 0.604 0.483 0.530 0.583 0.766 0.720 0.530 0.784 0.516 0.716 0.280 0.300 0.416 0.379 0.402 0.313 0.534 0.423 0.368 0.718 0.414 0.533 0.535 0.439 0.596 0.548 0.490 0.637 0.574 0.596 0.384 0.396 0.537 0.460 0.721 0.552 0.356 0.388

0.606 0.909 0.742 0.732 0.751 0.709 0.687 0.723 0.752 0.867 0.812 0.747 0.888 0.706 0.831 0.541 0.533 0.644 0.615 0.611 0.572 0.641 0.624 0.586 0.782 0.628 0.715 0.688 0.664 0.741 0.698 0.690 0.762 0.748 0.751 0.608 0.623 0.689 0.661 0.813 0.703 0.580 0.588

0.001 0.071 0.004 0.001 0.001 0.013 −0.011 −0.006 −0.012 −0.045 0.009 0.007 0.102 0.000 0.013 0.003 0.005 0.011 0.006 −0.006 0.001 0.021 0.005 0.001 0.016 −0.033 −0.009 −0.047 −0.031 −0.015 −0.004 −0.060 −0.034 −0.069 0.000 0.003 −0.013 −0.066 0.000 0.016 −0.011 0.000 0.006

0.003 0.049 0.008 0.009 0.005 0.005 −0.001 −0.004 −0.007 −0.038 0.002 0.009 0.112 0.001 0.016 0.004 0.004 0.009 0.010 0.004 0.009 0.010 0.017 0.001 0.015 −0.029 0.013 −0.035 −0.031 0.003 −0.002 −0.046 −0.003 −0.034 0.018 0.007 −0.001 −0.050 0.011 0.024 0.016 0.002 0.010

0.311 0.346 0.334 0.475 0.374 0.303 0.319 0.364 0.337 0.354 0.276 0.269 0.357 0.341 0.268 0.392 0.402 0.424 0.424 0.403 0.413 0.360 0.422 0.403 0.363 0.425 0.331 0.536 0.350 0.294 0.423 0.351 0.386 0.330 0.332 0.307 0.379 0.422 0.305 0.404 0.365 0.338 0.373

Overall

0.025

0.037

0.324

0.587

nd

nd

nd

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

No.

179

180

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

Table A.3 Correlation matrix of explanatory variables. Share of patents with multiple inventors Share of patents with multiple inventors Share of patents where all inventors are from one region Share of non-patent citations in relation to all citations HHI Firms HHI Industry (weighted by relevance for technological class)

Share of patents where all inventors are from one region

1 −0.2269

Share of non-patent citations in relation to all citations

HHI Firms

HHI Industry (weighted by relevance for technological class)

1 ***

0.0283

0.4334*** 0.1663

−0.1838 −0.0984

0.6602

1 0.194 0.0135

1 0.2803*

1

Note: Correlation coefficients reported in cells. * p < = 0.1. ** p < = 0.05. *** p < = 0.01. Table A.4 Descriptive statistics of explanatory variables.

Share of patents with multiple inventors Share of patents where all inventors are from one region Share of non-patent citations in relation to all citations HHI Firms HHI Industry (weighted by relevance for technological class)

Obs.

Mean

Std. Dev.

Min

Max

43 43 43 43 43

0.633 0.528 0.129 0.035 0.055

0.147 0.066 0.082 0.035 0.027

0.298 0.336 0.027 0.002 0.031

0.941 0.633 0.442 0.138 0.199

Table A.5 Values of the considered factors for the various technological fields. Tech. No.

1 2 3 4 5 6 7 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Technological fields

Share of patents with multiple inventors

Share of patents where all inventors are from one region

Share of non-patent citations in relation to all citations

HHI Firms concentration index

HHI Industry (weighted by relevance for technological class)

Food, beverages Tobacco products Textiles Wearing apparel Leather articles Wood products Paper Petroleum products, nuclear fuel Basic chemical Pesticides, agro-chemical products Paints, varnishes Pharmaceuticals Soaps, detergents, toilet preparations Other chemicals Man-made fibres Rubber and plastic products Non-metallic mineral products Basis metals Fabricated metal products Energy machinery Non-specific purpose machinery Agricultural and forestry machinery Machine-tools Special purpose machinery Weapons and ammunition Domestic appliances Office machinery and computers Electric motors, generators, transformers Electric distribution, control, wire, cable

0.681 0.677 0.754 0.298 0.413 0.605 0.671 0.723

0.458 0.366 0.461 0.597 0.395 0.429 0.453 0.502

0.205 0.034 0.116 0.034 0.027 0.077 0.082 0.073

0.008 0.126 0.023 0.019 0.029 0.019 0.018 0.026

0.031 0.094 0.058 0.049 0.032 0.039 0.052 0.093

0.876 0.941

0.487 0.336

0.187 0.259

0.045 0.122

0.071 0.076

0.821 0.885 0.919

0.633 0.509 0.598

0.197 0.442 0.131

0.035 0.012 0.138

0.063 0.049 0.063

0.805 0.863 0.518 0.569 0.655 0.438 0.584 0.586

0.546 0.526 0.524 0.506 0.507 0.572 0.561 0.546

0.133 0.220 0.087 0.103 0.185 0.040 0.085 0.083

0.023 0.030 0.002 0.003 0.012 0.003 0.012 0.003

0.064 0.069 0.039 0.044 0.041 0.037 0.042 0.045

0.440

0.558

0.048

0.018

0.035

0.542 0.591 0.547 0.581 0.609

0.585 0.570 0.572 0.538 0.606

0.102 0.085 0.067 0.068 0.220

0.004 0.009 0.047 0.040 0.031

0.043 0.042 0.050 0.049 0.059

0.632

0.548

0.158

0.063

0.053

0.646

0.573

0.039

0.042

0.051

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

181

Table A.5 (Continued ) Tech. No.

Technological fields

Share of patents with multiple inventors

Share of patents where all inventors are from one region

Share of non-patent citations in relation to all citations

HHI Firms concentration index

HHI Industry (weighted by relevance for technological class)

31 32 33 34 35

Accumulators, battery Lightening equipment Other electrical equipment Electronic components Signal transmission, telecommunications Television and radio receivers, audiovisual electronics Medical equipment Measuring instruments Industrial process control equipment Optical instruments Watches, clocks Motor vehicles Other transport equipment Furniture, consumer goods

0.799 0.505 0.590 0.731 0.613

0.515 0.553 0.535 0.588 0.555

0.218 0.074 0.135 0.247 0.228

0.030 0.031 0.066 0.084 0.099

0.050 0.060 0.051 0.050 0.047

0.569

0.624

0.197

0.027

0.050

0.620 0.665 0.649

0.472 0.550 0.551

0.074 0.191 0.162

0.005 0.016 0.078

0.036 0.050 0.053

0.634 0.481 0.636 0.527 0.343

0.558 0.534 0.553 0.452 0.598

0.154 0.085 0.095 0.077 0.073

0.011 0.031 0.041 0.012 0.002

0.052 0.199 0.048 0.042 0.033

36 37 38 39 40 41 42 43 44

Note: In order to calculate this share only those patents were used on which 2 inventors were filed.

References Acs, Z.J., Audretsch, D.B., Feldman, M.P., 1994. R&D spillovers and recipient firm size. Review of Economics and Statistics 76, 336–340. Alecke, B., Alsleben, C., Scharr, F., Untiedt, G., 2006. Are there really high-tech clusters? The geographic concentration of German manufacturing industries and its determinants. Annals of Regional Sciences 40, 19–42. Amiti, M., 1999. Specialization patterns in Europe. Weltwirtschaftliches Archiv 135, 573–593. Anselin, L., Varga, A., Acs, Z., 1997. Local geographic spillovers between university research and high technology innovations. Journal of Urban Economics 42, 422–448. Audretsch, D., 1995. Innovation and Industry Evolution. MIT Press, Cambridge. Audretsch, D.B., 1998. Agglomeration and the location of innovative activity. Oxford Review of Economic Policy 14, 18–29. Audretsch, D.B., Feldman, M.P., 1996. Innovative clusters and the industry life cycle. Review of Industrial Organization 11, 253–273. Beise, M., Ebling, G., Janz, N., Licht, G., Niggemann, H., 1999. Innovationsaktivitäten im Verarbeitenden Gewerbe? Ergebnisse der Innovationserhebung 1997. In: Janz, N., Licht, G. (Eds.), Innovationsaktivitäten in der deutschen Wirtschaft. Nomos, Baden-Baden, pp. 9–98. Blind, K., Grupp, H., 1999. Interdependencies between the science and technology infrastructure and innovation activities in German Regions: empirical findings and policy consequences. Research Policy 28, 451–468. Braunerhjelm, P., Carlsson, B., 1999. Industry clusters in Ohio and Sweden 1975–1995. Small Business Economics 12, 279–293. Brenner, T., 2004. Local Industrial Clusters—Existence, Emergence, and Evolution. Routledge, London. Brenner, T., 2005. A Stochastic Theory of Geographic Concentration and the Empirical Evidence in Germany, Papers on Economics & Evolution #0523. Broekel, T., Binder, M., 2007. The regional dimension of knowledge transfers—a behavioral approach. Industry and Innovation 14, 151–175. Brouwer, E., Kleinknecht, A., 1999. Innovative output, and a firm’s propensity to patent an exploration of CIS Micro Data. Research Policy 28, 615–624. Buchinsky, M., 1998. Recent advances in quantile regression models: a practical guideline for empirical research. The Journal of Human Resources 33, 88–126. Caniëls, M., 1999. Regional Growth Differentials—The Impact of Locally bounded Knowledge Spillovers. Faculty of Economics and Business Administration, Maastricht University, Dissertation No. 99-42. Coad, A., Rao, R., 2006. Innovation and firm growth in high-tech-sectors: a quantile regression approach. LEM Working Paper Series 2006/18, Pisa.

Davies, S., 1979. Choosing between concentration indices: the isoconcentration curve. Economica 46, 67–75. Ellison, G., Glaeser, E.L., 1997. Geographic concentration in U.S. manufacturing industries: a dartboard approach. Journal of Political Economy 105, 889–927. Ernst, H., Leptien, C., Vitt, J., 2000. Inventors are not alike: the distribution of patenting output among industrial R&D personnel. IEEE Transactions on Engineering Management 47, 184–199. Evangelista, R., Smith, K., 1998. Measuring innovation in European Industry. International Journal of the Economics and Business 5, 311–333. Faggian, A., McCann, P., 2006. Human capital flows and regional knowledge assets: a simultaneous equation approach. Oxford Economic Papers 58, 475–500. Feldman, M.P., 1994. The Geography of Innovation. Kluwer Academic Publishers, Dordrecht/Boston/London. Feldman, M.P., Florida, R., 1994. The geographic sources of innovation: technological infrastructure and product innovation in the United States. Annals of the Association of American Geographers 84, 210–229. Frietsch, R., Köhler, F., Blind, K., 2008. Weltmarktpatente—Strukturen und deren Veränderungen. Studie zum deutschen Innovationssystem Nr. 7-2008, Expertenkommission für Forschung und Innovation, http://www.e-fi.de. Fröderer, K., Krey, K., Palme, K., 1998. Innovation und Mittelstand, eine Umfrage bei 1871 Unternehmen. In: Beiträge zur Gesellschafts- und Bildungspolitik. Institut der Deutschen Wirtschaft, Köln. Griliches, Z., 1990. Patent statistics as economic indicators: a survey. Journal of Economic Literature 27, 1661–1707. Grupp, H., Schmoch, U., 1999. Patent statistics as economic indicators. Research Policy 28, 377–396. Koenker, R., Basset, G., 1978. Regression quantiles. Econometrica 46, 33–50. Koenker, R., Hallock, K., 2001. Quantile regression. Journal of Economic Perspectives 15, 143–156. Lafourcade, M., Mion, G., 2005. Concentration, agglomeration and the size of plants. Working Paper No. 2005-42, Paris-Jourdan Sciences Economiques. Leo, H., Reinstaller, A., Unterlass, F., 2007. Motivating sectoral analysis of innovation performance. Europe Innova, Innovation Watch, Workpackage 4. Lundvall, B.A., Borras, S., 1999. The Globalising Learning Economy: Implications for Innovation Policy. Office for Official Publications of the European Communities, Luxembourg. Malerba, F., 2002. Sectoral systems of innovation and production. Research Policy 31, 247–264. Marsili, O., Verspagen, B., 2001. Technological regimes and innovation: looking for regularities in Dutch manufacturing. ECIS, Eindhoven University of Technology, Working Paper. Moran, P.A.P., 1950. Notes on continuous stochastic phenomena. Biometrika 37, 17–33.

182

D. Fornahl, T. Brenner / Structural Change and Economic Dynamics 20 (2009) 163–182

Narin, F., Breitzmann, A., 1995. Inventive productivity. Research Policy 24, 507–519. Paci, R., Usai, S., 2000. Technological enclaves and industrial districts: an analysis of the regional distribution of innovative activity in Europe. Regional Studies 34, 97–114. Pavitt, K., 1984. Sectoral patterns of technical change: towards a taxonomy and a theory. Research Policy 13, 343–373. Pavitt, K., 1985. Patent statistics as indicators of inventive activities: possibilities and problems. Scientometrics 7, 77–99. Pittaway, L., Robertson, M., Munir, K., Denyer, D., Neely, A., 2003. Networking and innovation: a systematic review of the evidence. International Journal of Management Reviews 5/6, 137–168. Rammer, C., 2003. Patente und Marken als Schutzmechanismen für Innovationen, Studie zum deutschen Innovationssystem. Nr. 11-2003, Zentrum für Europäische Wirtschaftsforschung, Mannheim. Saxenian, A.L., 1994. Regional Advantage Culture and Competition in Silicon Valley and Route 128. Harvard University Press, Cambridge (MA) and London. Schmoch, U., Laville, F., Patel, P., Frietsch, R., 2003. Linking Technology Areas to Industrial Sectors. Final Report to the European Commision, DG Reasearch.

Smolny, W., Schneeweis, T., 1999. Innovations growth and employment. Jahrbücher für Nationalökonomie und Statistik 218, 451–472. Soete, B., Stephan, A., 2003. Nachhaltiges wirtschaftliches Wachstum durch Innovationen: Die Rolle von kleinen und mittleren Unternehmen. DIW-Wochenbericht, 38/03. Sternberg, R., Litzenberger, T., 2004. Regional clusters in Germany—their geography and their relevance for entrepreneurial activities. European Planning Studies 12, 767–791. Verspagen, B., 1999. 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, pp. 236–256. Wen, M., 2004. Relocation and agglomeration of Chinese industry. Journal of Development Economics 73, 329–347. Yasar, M., Nelson, C., Rejesus, R., 2006. Productivity and exporting status of manufacturing firms: evidence from quantile regressions. Weltwirtschaftliches Archiv 142, 675–694. Zitt, M., Barré, R., Sigogneau, A., Laville, F., 1999. Territorial concentration and evolution of science and technology activities in the European Union: a descriptive analysis. Research Policy 28, 545–562.