How sub-national conditions affect regional innovation systems: The case of the two Germanys




We compare two leading regional innovation systems (RIS) in East Germany with two RIS in West Germany of about the same size and degree of agglomeration. Our analyses show that differences in the performance between the regions cannot easily be related to the structural properties of the respective innovation networks because distinct challenges and macroeconomic conditions in the two parts of the country, as well as differences in integration of the regions into their neighbouring spatial environment, play an important role. We conclude that an analysis of RIS should account for the (sub-) national economic conditions as well as for the region's position in its spatial environment; merely focusing on region alone is not sufficient.


Comparamos dos sistemas de innovación regional (SIR) líderes en Alemania oriental con dos SIR en Alemania occidental de aproximadamente el mismo tamaño y grado de aglomeración. Nuestros análisis muestran que las diferencias entre regiones en cuant a su desempeño no pueden relacionarse fácilmente con las propiedades estructurales de las respectivas redes de innovación debido al importante papel que juegan tanto los diferentes retos y condiciones macroeconómicas de ambas partes del país, como las diferencias en integración de las regiones dentro del ambiente del espacio que las rodea. Nuestra conclusión es que cualquier análisis de SIR debería tener en cuenta no solo las condiciones económicas (sub) nacionales sino también la posición de la región dentro del ambiente del espacio que la rodea: no es suficiente con estudiar únicamente la región.

inline image

1 The embeddedness of regional innovation systems

The concept of national innovation systems (NIS) is a comprehensive approach to analysing innovation processes. It particularly stresses the division of innovative labour between actors and the embeddedness of innovation processes in an institutional and macroeconomic environment (Lundvall 1992, 2007; Nelson 1993; Edquist 1997). Following introduction of the NIS concept, many authors proposed focusing on the systemic view of innovation processes in regions, sectors, or technologies, leading to the regional innovation systems (RIS) (Cooke 2004; Asheim and Gertler 2006), sectoral innovation systems (Malerba 2002), and the technological systems (Carlsson 1994) approaches. The RIS approach has become especially popular in empirical research. However, it is still unclear to what degree the RIS approach is appropriate for empirical analyses or, particularly, as a basis for policy decisions. Lundvall (2007), for example, argues that although this concept, like others such as sectoral and technological systems, may lead to important insights, the NIS approach should be favoured because the nation-state is the most important level of policy making. In a recent paper, Shearmur (2010) voiced the criticism that many analyses of RIS treat regions as ‘islands’ with characteristics such as local networking, institutions, tacit knowledge, and clusters “that have little or no effect on another's” (Shearmur 2010, p. 5). He calls for a more continuous conceptualization of space in innovation studies that will account for the surrounding space and particularly for the fact that different kinds of innovation may require different intensities of face-to-face contact and different degrees of geographic proximity (for another such an argument, see also McCann 2007).

This study investigates the effects of internal interaction and the wider geographic context, as well as influences that may result from historical developments, on the performance of RIS. Specifically, we are interested in the relationship between structural properties of regional innovator networks and system performance, taking into account the importance of sub-national conditions. By comparing regions in East and West Germany, we can analyse how the different histories and resulting macroeconomic conditions of the two parts of the country have shaped regional innovation activity. During the time period we study, East and West Germany have had more or less identical formal institutions, but vary considerably as to their soft institutions, economic structures, and the behavior of their residents (Fritsch 2004; Kronthaler 2005). We are thus able to perform a kind of international comparison within one country.

We selected two of the leading RIS of East Germany, Dresden and Jena, and matched them with two well-performing RIS in West Germany that are of comparable size and degree of agglomeration, Aachen and Karlsruhe. We use patent statistics to construct networks of innovators in the four regions under study (Cantner and Graf 2006; Graf and Henning 2009). Our analysis shows important differences with regard to the structural properties of innovator networks between the two East German and the corresponding West German RIS that clearly indicate different modes of innovation that may be traced back to different historical development and macroeconomic conditions. We find that the two East German RIS show relatively low levels of innovation performance but considerably higher degrees of R&D co-operation than their more efficient West German counterparts, which is in contrast to what would be predicted by the systems of innovation approach (Cooke 2004; Asheim and Gertler 2006). This may indicate an effect of differences in the organization of innovation activities as well as of different macroeconomic conditions in the two parts of the country. Another factor that could contribute to explaining this phenomenon is that the two East German RIS (Dresden and Jena) are spatially isolated ‘hot spots’ in the East German innovation landscape, whereas the two West German RIS (Aachen and Karlsruhe) are embedded in relatively prosperous regional neighbourhoods. This suggests that the performance of RIS depends to a considerable degree on their wider spatial environment and that sub-national conditions have a strong effect on the performance of RIS that should not be neglected in either empirical analyses or in policy design. The results can be regarded as some indication of the importance of NIS to the performance of the embedded RIS and may shed some light on the question of which approach − NIS or RIS − is most appropriate. Moreover, since East Germany experienced a turbulent change from a socialist system to a market economy, our study also contributes to research on transformation processes.

We introduce the regional innovation systems in the next section (Section 2). Section 3 contains a brief overview of the historical background and innovation performance in East Germany compared to the western part of the country in general, and Section 4 provides information on characteristics of and performance in the four case study regions. The comparative analysis of the four regions is reported in Section 5. Section 6 discusses development of the regions in the subsequent period. Section 7 concludes.

2 Literature: Regional innovation systems and the network perspective

The systemic view of innovation processes (e.g., Lundvall 1992; Nelson 1993; Edquist 1997), emphasizes the importance of a division of innovative labour and knowledge transfer between innovative actors. The RIS approach particularly stresses that the conditions for innovation activities may differ substantially between regions and that knowledge flows tend to be regionally bounded (Jaffe et al. 1993; Cooke 2004; Asheim and Gertler 2006). The main argument for the spatial dependence of knowledge flows is that knowledge has tacit components that can be transferred only via personal relationships that may be facilitated by geographical proximity (Boschma 2005; Breschi and Lissoni 2009). These ideas are put to an empirical test by Fleming et al. (2007), who show that differences in regional innovative performance can be traced back to the connectedness of the respective inventor networks.

However, if there is ‘too much proximity’ (Boschma 2005) or, as Uzzi (1997) terms it, ‘overembeddedness’, we can expect adverse impacts on learning and innovation as dense local structures can pose a barrier to inflows of novel information created external to the system. Taking a perspective different from the view that focuses solely on the benefits of knowledge spillovers within local networks, several authors discuss the importance of interaction with actors external to the local system (Gertler 1997; Bathelt et al. 2004; Bathelt 2005). As such, the benefits of a dense local network arise as these transmission channels foster the diffusion of external knowledge into the system thereby reducing the risks of lock-in effects (Grabher 1993; Keeble et al. 1999).

Thus, the literature is in agreement about the necessity of external connections for a well-functioning RIS, but it is not very specific about which configurations of RIS in terms of internal density and external openness are the most conducive to sustainable regional development. Similar to multinational firms (with R&D facilities at different locations), that face a tradeoff between internal (within the firm) and external proximity (Blanc and Sierra 1999), we might expect RIS to perform best when comprised of a certain specific mix of internal interaction and external relations.1 In addition to these relational factors, the literature discusses a number of region-specific factors that may influence RIS performance, such as location with regard to other regions (Andersson and Karlsson 2004), size of the region and its degree of agglomeration, qualification of the regional workforce, endowment with universities and other public research organizations, the innovative milieu, regional industry specialization, etc. (clustering; Fritsch and Slavtchev 2010).

To summarize the argument, the innovation performance of a RIS may be particularly shaped by region-specific factors such as the level and structure of regional interaction, as well as by relationships with external actors, which may indicate the openness of a region (Graf 2011). In a nutshell, the systemic view may be boiled down to the hypothesis that the level and quality of the division of innovative labour have an important positive effect on the level and success of innovation activity and, therefore, on the performance of RIS. Hence, one may expect that tightly knit regional networks and the integration of local actors into global knowledge flows will create an excellent environment for an effective RIS.

There are several empirical studies on the region-specific determinants of RIS performance, but the effects of the general framework conditions, as well as the importance of the surrounding spatial environment, are more or less unexplored territory. We thus know nearly nothing about the relative importance of such environmental factors for the performance of RIS and the role they play in the effectiveness of region-specific determinants.2 For example, how do specific institutional settings affect knowledge transfer from universities and other public research organizations to the private sector? How does general macroeconomic prosperity affect the performance of RIS as compared to either a decline in prosperity or a high level of turbulence as occurred in East Germany during the 1990s?

In this study, we compare a number of key characteristics of systemic innovation processes between four case study regions. We focus on the relations between innovative actors (firms, public research institutions and individuals), the resulting regional innovation networks, and on links to actors external to the respective region. This information allows us to assess the systemic properties of the four RIS and to derive expectations about their relative performance. Comparing these results with the actual level of regional innovation activity leads to conclusions about the relative importance of a region's characteristics and its more macroeconomic environment.

3 Historical background: The two German innovation systems

There are two reasons why the environment for innovation may differ between East and West Germany. First, for more than 40 years, East Germany was a socialist regime characterized by a substantially different institutional framework, different macroeconomic conditions, and, most particularly, a different organization of innovation processes compared to West Germany. This ‘natural experiment’ left a substantial imprint on the East German RIS. Second, beginning in 1990, East Germany experienced a turbulent transformation process to a market economic system (Brezinski and Fritsch 1995), the effects of which are still clearly noticeable 20 years later and that created an environment for innovation activity quite distinct from the one prevailing in the West (Fritsch 2004; Kronthaler 2005). Hence, one may expect to find that these different general conditions will have a considerable impact on the performance of RIS in both parts of the country.

Until 1945, the end of the Second World War, the national framework conditions in what is today's Germany were identical. Right after the end of the war, the country was divided into four zones, each governed by one of the allied powers. In 1949, the Soviet zone became the German Democratic Republic (GDR), commonly referred to as East Germany; the other occupation zones became the Federal Republic of Germany (FRG), or West Germany. The FRG was set up as a capitalistic market economy and soon experienced vigorous economic recovery. In contrast, the GDR, East Germany, became a socialist-type centrally planned economy with an innovation system much like the Soviet model. Specifically, the innovation system in East Germany was characterized by a close orientation towards the linear model of the innovation process and pronounced bureaucratic steering (Fritsch and Werker 1999; Hanson and Pavitt 1987).

In 1961, the East German government instituted a border regime that more or less completely separated East Germany from the West and made any uncontrolled transfer of people, goods and resources nearly impossible. In the course of these developments, innovation activities in East Germany were largely cut off from those in the West. The East German government only rarely allowed Eastern scientists to travel to the West or communicate with Western colleagues. Innovation in the East was also hampered by the fact that the Western bloc imposed an embargo on certain goods (e.g., modern machinery, software), into East Germany (Kogut and Zander 2000). In 1990, the socialist regime in East Germany ended rather abruptly and both parts of the country were once again united, with a common currency and the rapid creation of an identical framework of formal institutions. Adjustment to these new conditions, however, was not nearly so rapid and is by no means complete even today, 20 years later.

4 Regional innovation systems compared: Dresden and Jena vs. Aachen and Karlsruhe

4.1 Selection of case study regions

Dresden and Jena are two East German regions that perform relatively well despite the challenges posed by the East German innovation system described above. At the turn of the millennium, 10 years after the transformation process began; they were the two East German lighthouses of innovation in terms of level of innovation activity as well as efficiency of their innovation systems (Fritsch and Slavtchev 2011). However, they still considerably lagged behind the West German level of innovation. For our analysis, we employ a matched-pairs approach and compare Dresden and Jena with two regions of about similar size and population density in the West that are characterized by relatively high efficiency of their RIS.3 Comparable size and population density are required because innovation theory, as well as empirical research, stresses the importance of agglomeration economies for innovation activity (Feldman and Audretsch 1999). The regions also had to have a renowned research university as well as a number of other public research organizations, such as institutes of the Fraunhofer and the Max Planck Society. According to these criteria, we chose Aachen and Karlsruhe for the comparison.

All four case study regions are defined as German planning regions (Raumordnungsregionen). To represent functional entities, planning regions normally comprise several NUTS 3-level districts,4 namely, a core city and its surrounding area. While districts are administrative regions, planning regions are more often used for spatial analysis and policy development, particularly regarding public infrastructure. Planning regions tend to be somewhat larger than labour market regions or travel-to-work areas. We consider planning regions as more suitable than districts for an analysis of RIS for two reasons. First, a single district, particularly a core city, is probably too small to include the most important sites of innovation-related local interaction. The second reason is of a methodological nature: since patents are assigned to the residence of the inventor, taking just a core-city as a region would lead to an underestimation of patenting activity since many inventors have their private residence in surrounding districts.5Figure 1 shows the location of the four case study regions. Aachen and Karlsruhe are situated close to other regions with a high level of innovation activity (e.g., Bonn and Cologne in the case of Aachen; Stuttgart and Mannheim in the case of Karlsruhe), but the two East German regions are more isolated in this respect.6 This is particularly true of Jena, which represents a ‘cathedral in the desert’ even within its planning region (Graf 2006).

Figure 1.

The case study regions

4.2 Characteristics and general performance of case study regions

The size of the four case study regions ranges from nearly 800,000 inhabitants in the Jena region to about 1,250,000 inhabitants in the Aachen region (Table 1). All four regions have a long tradition in manufacturing industries: electronics and mechanical engineering in Dresden; optics and precision mechanics in Jena; electronics and electrical engineering in Aachen;7 and electrical engineering, mechanical engineering, and vehicles construction in Karlsruhe. The fact that the two East German regions have a much smaller establishment size in the manufacturing sector is probably a result of the transformation process, during which large entities were split-up, often followed by further employment decline due to unfavourable economic performance. Moreover, many East German establishments are small because they are relatively young, having been set up only after German Reunification. The higher start-up rates in the two East German regions reflect an adjustment to the level of entrepreneurship in the West German part. The share of R&D employees8 is considerably lower in the Eastern regions but both regions show a relatively high share of employees with a tertiary degree.

Table 1.  Innovation and performance indicators for case study regions (ROR level)
 East GermanyWest Germany
  • Notes: a Private universities and university hospitals excluded.

  • b

    As of 2008.

Number of population1,032,659788,2361,247,2701,087,776
Number of employees (private sector)289,647198,501271,232324,759
Average establishment size (number of employees) overall7.436.867.879.68
Average establishment size (number of employees) in manufacturing14.2514.8019.3224.45
Average establishment size (number of employees) in services5.524.374.775.66
Share of employees in manufacturing in total private-sector employment25.9130.1337.5742.46
Start-up rate private sector7.177.787.055.50
Share of R&D employees3.162.443.693.98
Share of employees with tertiary degree12.6510.837.838.07
Third-party funds per professor (in 1,000 €)a72.9239.73169.25109.11
Third-party funds from private firms per professor (in 1,000 €)a14.495.56169.2535.18
Third-party funds from German Science Foundation (DFG) per professor (in 1,000 €)a17.7516.8943.8536.89
Third-party funds per professor (in 1,000 €) in departments of engineering and natural sciences onlya119.8166.33234.99131.20
Number of Fraunhofer Institutesb10132
Number of Max Planck Institutesb3300
Patents of private firms per 1,000 employees 1995−20010.770.581.371.44
Patents of private firms per 1,000 R&D employees 1995–200121.6322.8646.1138.92
Efficiency of the RIS 1995–2000 (Fritsch and Slavtchev 2011)0.3540.3940.7690.613

The amount of third-party funds per professor may indicate several things. First, since external funds are predominantly allocated by means of highly competitive procedures, the amount of third-party funds per professor can be regarded as an indicator of research quality. This is particularly true of funds from the German Science Foundation (DFG), which are designated for basic research. Funds from private firms signify university-industry linkages that may result in significant knowledge spillovers.9 An important difference between the two Eastern and the two Western regions is the lower level of third-party funds per professor in the East. Since departments of engineering and natural sciences tend to have the highest levels of external funding, we restrict this indicator to these departments only. Aachen is the clear leader with respect to this indicator, Karlsruhe and Dresden take the middle position, and Jena lags behind, with less than 30 percent of Aachen's level of funding.

The two East German regions are well equipped with non-university public research institutes of the Max Planck Society, which focus on basic research, and of the Fraunhofer Society, which have the mandate of transferring results of basic research to private-sector innovators. Dresden, in particular, is home to the remarkably high number of 10 institutes of the Fraunhofer Society. Using patents as a measure of innovation output, the two West German regions perform much better than their Eastern counterparts. This becomes particularly clear if one takes the number of patents per R&D employee as an indicator, which can be viewed as a measure of the productivity of R&D activity. Estimates of the efficiency of German RIS in the 1995−2000 period by Fritsch and Slavtchev (2011) reveal a much better performance of Aachen (0.769) and Karlsruhe (0.613) compared to Dresden (0.354) and Jena (0.394).10

This first inspection of innovative resources and performance in the four case study regions reveals the impact of a socialist heritage and the subsequent transformation process in the two East German regions. All four regions are similar with regard to the prerequisites for innovation activity on the resource side, but the two West German regions clearly perform better. Following the arguments of the RIS approach (see Section 2), one should expect to observe a higher interaction intensity in Aachen and Karlsruhe as compared to their East German counterparts. In the following section, we analyse the networks of inventors in the four regions in order to test this proposition.

5 Regional networks of inventors

The innovation systems approach suggests that the division of innovative labour is of crucial importance for innovation performance (Lundvall and Johnson 1994; Capello and Faggian 2005; Malmberg and Maskell 2006). For our empirical study, we operationalize the concept of division of innovative labour by looking at the structure of interaction within regional networks of innovators.

5.1 Method: Social network analysis and patent data

An empirical analysis of social networks requires relational data on the vast majority of the actors constituting the system. Patent data meet this requirement as they reveal information about the persons involved in the underlying innovative activity (the inventors), and the patent applicants (firms, individuals, research organizations), who own the rights to exploit the invention (the innovators11). The data are publicly accessible, consistent, and complete in the sense that any innovative effort that was judged worth a patent application is included. Since a patent application tends to represent a certain minimum standard of newness of an invention and of the respective R&D, the quality of the links between actors that we identify on the basis of patent statistics should be comparable across regions. Of course, patent data are not perfect and have some widely acknowledged flaws. Most importantly, there are alternative mechanisms of appropriating the returns to innovative activity (Cohen et al. 2000), and the propensity to patent varies substantially between sectors (Arundel and Kabla 1998; Mairesse and Mohnen 2003). We overcome this last problem by selecting regions that specialize in manufacturing industries in which patenting is presumed to be of similar importance.

Methodically, patent networks can be constructed by relating patent applicants (innovators) to the respective inventors. Generally, it seems more plausible to assume knowledge flows between individuals who know each other from joint research projects rather than between patent applicants. Therefore, it is common practice in the analysis of innovation networks based on patent data to link the inventors directly (Balconi et al. 2004; Singh 2005; Fleming et al. 2007); however, these connections can also be used to identify channels of knowledge transmission between the innovators by linking them via common inventors (Cantner and Graf 2006; Graf and Henning 2009; Breschi and Lissoni 2009). We follow the latter approach because it seems more appropriate to assume organizations (innovators) as constituents of the regional system.12

Our analysis of the networks of innovators is based on patent applications at the German Patent Office that were disclosed between 1995 and 2001. To assess inventive activity in a region, the first best solution would be to use the address of the research lab where the R&D was performed. Unfortunately, patent statistics do not provide this information. Using the applicant's address does not help in this respect because a number of (particularly large) firms have several and regionally dispersed subsidiaries in which the research may have been conducted, but apply for patents centrally on behalf of headquarters. The common solution to this difficulty is to use inventor residence instead of that of the innovator, under the assumption that people normally live fairly near where they work (Jaffe et al. 1993; Breschi and Lissoni 2009). Consequently, we base the regional networks on all patent applications with at least one inventor residing in the respective region. Innovators are defined as external to the region if they have applied for at least one patent naming an inventor living in the focal region, but are themselves not located in that region.13

We assume two innovators to be related if at least one inventor has developed a patent for both innovators. In other words, a relation is established between innovators A and B if we find the same inventor named on a patent applied for by A and on a patent applied for by B. There are two ways this could occur:

  • 1First, the innovators jointly apply for a certain patent. In this case, we assume a previous research co-operation and there are as many linkages between all co-applying innovators as there are inventors.
  • 2Second, the same inventor is named on two distinct patent applications submitted by different innovators. In this case, we assume mobility of the inventor between the innovators.

Both types of linkages are based on the concept of knowledge transfer via personal relationships (e.g., Almeida and Kogut 1999). The main idea is that organizations, namely, firms or research institutes, interact via scientists who know each other either from working on joint projects (co-operation) or as they move from one organization to another (mobility). Of course, mobility does not only encompass individuals changing jobs between existing organizations but also captures spin-off processes in which new entities are formed by employees of incumbents.

During the period under analysis, German patent law allowed university professors to patent under their own names instead of under the name of the university, and thus the number of university patent applications is underestimated in our data. The number of patent applications from public research is further underestimated because universities may trade intellectual property rights for financial support in university-industry co-operation projects, that is, a private firm sponsors the research carried out in the university's lab in exchange for the exclusive right to patent the invention. This means that not only will public research patent activity be underestimated but, even more importantly, a number of university-industry co-operations leading to patent output will not be identified as co-operative activity.

5.2 Overall structure of inventor networks

Looking at the number of patent applications in each region reveals that actors in Aachen and Karlsruhe filed many more patents than actors in Jena or Dresden, with the number of patent applications in Karlsruhe being almost triple that of Jena (Table 2). In terms of the number of applicants (the network actors), the differences are not quite as pronounced, with Aachen and Karlsruhe having roughly twice as many applicants as Jena. The number of patent applications per actor (Dresden 3.21 patents per applicant, Jena 3.02, Aachen 3.46, Karlsruhe 4.16), is probably a result of the smaller average size and corresponding lower levels of R&D of actors in the two East German regions. Another significant difference between the East German and West German regions is the relative importance of patent applications by public research institutions. In Dresden and Jena, roughly every fourth patent was filed by a university or other public research institute; this share is only 9 percent in Karlsruhe and 15 percent in Aachen.14

Table 2.  Characteristics of the inventor networks in case study regions
 East GermanyWest Germany
  • Notes: a Only relations with at least one internal actor involved.

  • b

    Based on degree centrality.

Number of patents3,7202,0945,5086,072
by type of applicant (%)
− individual19.218.825.519.5
− public23.725.714.88.7
− firm57.255.559.771.8
by location of applicant (%)
− same region70.375.065.866.8
− same Federal State7.23.221.918.8
− rest of Germany21.421.210.211.9
− abroad1.
Number of actors1,1586941,5911,460
by type (%)
− individual40.238.350.046.7
− public5.
− firm54.
by location (%)
− same region51.856.261.260.2
− same Federal State11.16.317.916.8
− rest of Germany35.135.916.419.0
− abroad2.
Total Number of linkagesa4,1063,6144,0363,754
− share internal (%)30.842.449.325.6
− share external (%)69.257.650.774.4
Number of co-operation linkagesa2,5702,1002,3741,906
− share internal (%)31.041.750.526.7
− share external (%)69.058.349.573.3
Number of mobility linkagesa1,5361,5141,6621,848
− share internal (%)30.543.347.724.5
− share external (%)69.556.752.375.5
Share of mobility linkages (%)37.441.941.249.2
Network measures    
Number of components549309910875
Size of main component359259254344
Share in main component (%)31.037.316.023.6
Share of isolates (%)35.132.642.548.3
Density (valued) (%)0.440.920.210.25
Density (binary) (%)0.190.390.100.11
Mean degree (valued)5.0696.3833.3823.627
Mean degree (binary)2.2252.6891.6121.604
Average distance within main component3.3743.1034.4234.032
Mean degree (valued)3.0213.8391.8981.879
Mean degree (binary)0.8050.8960.5690.518
Mean degree (valued)2.0482.5451.4831.748
Mean degree (binary)1.4351.8301.0461.101

In Table 2 we also present variables that measure the structure and intensity of interaction within the regional networks.15 Under the assumption that knowledge can only ‘flow’ between actors who are linked either directly or indirectly, the largest connected part, the networks main component,16 should indicate the share of actors that have (potential) access to the bulk of local knowledge. It is remarkable, and in sharp contrast to our initial proposition, that the inventor networks in the two East German regions are much more integrated than those in the two West German regions. In Dresden, as well as in Jena, the share of actors in the main component is much higher and the share of isolated actors, that is, those who have no connections with other actors, is much lower than in Aachen and Karlsruhe (Table 2). Since the actors in the two East German networks have a larger average number of links to other actors (mean degree17), the network density, namely, the share of realized links over all possible links, is also higher in the Eastern regions.18 The considerably greater number of relationships in the two East German RIS holds for both types of links, those based on co-operation and those related to inventor mobility. While a higher level of mobility links in the East German RIS could have been expected as a result of the turbulent transformation process during which relatively many persons had to change employers,19 the higher number of co-operative links is surprising, given the disruptive effects of the transformation process on personal ties and networks (Albach 1994).

The higher average number of co-operative links observed in the two East German RIS may be an artifact of the more open attitude toward R&D co-operation under its socialist regime. Obviously, many of the relationships established under the past system proved stable enough to survive the radical reorganization of the East German RIS caused by the transition.20 If we distinguish between the two types of linkages, co-operation and mobility, we do not observe much of a difference with respect to the shares of internal and external relations. In addition to the higher level of interaction in the two Eastern regions, there is another notable structural difference between the two parts of the country: the networks in Dresden and Jena are far more centralized21 than those in the West, that is, linkages are more concentrated between few key actors.

The higher degree of integration in the two East German regions holds not only for the networks as a whole but also for their main components (Figure 2). The networks in East Germany seem to be more tightly knit than their West German counterparts. Accordingly, the average distance between actors within the main component22 is also smaller in the two East German regions. Especially when comparing Jena and Aachen, we observe a dense pattern of relationships with a large number of central actors in Jena, whereas in Aachen, the network appears less dense, with no easily identifiable centre. The main component of the inventor network in Dresden is clearly dominated by two public research actors (the Technical University of Dresden and the Fraunhofer Society), that have by far the highest number of linkages. In the other regions, public research actors do not play such a dominant role.

Figure 2.

Main components of regional networks
Note: Actors located within the region (headquarter or subsidiary) are marked in grey/red (electronic version), external actors are in black. Squares indicate a private actor; public research organizations are circles. The size of a node is proportional to the number of patents filed.

5.3 Internal vs. external relations

According to the RIS approach, the functioning of a regional innovation system depends on interaction within the region and on the connections regional actors have with the ‘outer world’ of external knowledge sources (Bathelt et al. 2004; Graf 2011). As such, we could expect to find differences in the openness and geographic reach of regional networks that may explain the differences in performance. We use a number of measures to analyse the importance of extra-local linkages (Table 2). Since the regional networks are defined based on patents naming at least one inventor located in the respective region, applicants might be from different locations. This could be due to commuting inventors, inventors working for firms that have their headquarters in a different region, or co-applications by actors from different localities. Analysing the number of patents by location of the patent applicant, we find that between two-thirds and three-quarters of all patents are from local applicants, with the highest share in Jena and the lowest share in Aachen. Counting only actors not their patents, the share of locals is lower in all regions, which is not surprising since we account for all patents by locals, but only for that fraction of patents by external actors that are invented in collaboration with local inventors.

A rather interesting observation from both measures (patents and applicants), is the different degree of integration of the regions into their surrounding space. The two West German regions have a much larger share of links to actors located within the same Federal State,23 indicating that they have a considerably higher level of interaction with neighbouring regions than do their East German counterparts, which appear to be almost isolated spots in their surrounding spatial environment. The two West German regions also seem better integrated into international knowledge flows, having much higher shares of actors that are located abroad.

Having classified actors as either internal or external, we now analyse the relations between actors according to their location. Internal linkages are linkages between actors located within the same region. External linkages are those between an internal and an external actor.24 The outward orientation, as measured by the share of external relationships, is highest in Karlsruhe and lowest in Aachen, which is surprising given the large share of external actors in the Aachen network.

Analysing the number of external linkages of local actors gives a first impression of a region's integration into interregional knowledge flows. A large number of interregional linkages, however, does not necessarily mean that the system can effectively integrate external knowledge because it says nothing about how this knowledge is disseminated in the region.25 Therefore, we take a more micro perspective and identify those actors who play the role of gatekeeper for the RIS by absorbing external knowledge and passing it on to local actors.

5.4 Gatekeeper

‘Gatekeepers’, that is, actors who are well integrated into global knowledge flows as well as related to regional actors, play a key role in connecting the RIS to the ‘outer world’ (Giuliani 2005; Giuliani and Bell 2005; Graf 2011). Gatekeepers serve two functions in a regional innovation system: external knowledge sourcing and diffusion of knowledge inside the local system, thereby acting as knowledge brokers (Giuliani 2005; Wink 2008).

To identify regional gatekeepers, we plot the actors in each network according to the number of their internal and external relations (Figure 3). The four parts of Figure 3 are scaled identically to make them comparable. Private actors (firms and individuals) are represented by squares; public research institutes and universities are shown as circles. The size of the symbol reflects the number of patent applications submitted by that actor. For most actors, internal and external contacts seem to go hand in hand, but with different intensities. Actors in Jena appear more inward oriented than those in Dresden, Aachen, and especially, Karlsruhe. According to our gatekeeper index, defined as a brokering position between internal and external actors (Gould and Fernandez 1989), gate-keeping activity is strongly concentrated in a few innovators (see Table A1 in the Appendix). For example, in Dresden, the two actors with the highest numbers of relationships, the Technical University of Dresden and the Institutes of the Fraunhofer Society, score 30 and 10 times higher, respectively, than the third actor on the list. In Karlsruhe, the top gatekeeper (Forschungszentrum Karlsruhe) has a score three times higher than that of the runner-up (Bosch) and 10 times higher than the local university (Technical University of Karlsruhe). In Aachen, the same distributional characteristics are seen, but at a lower level, while in Jena, the gatekeeper function seems to be performed by a variety of actors.

Figure 3.

Gatekeepers in the regional networks of innovators

Common to all regions is that the local university and large research institutes, such as the Fraunhofer Institutes, are at the top of the gatekeeper index (Table A1 in the Appendix). Certain private firms, such as Carl-Zeiss in Jena and Bosch in Karlsruhe, also act as gatekeepers.

5.5 Dynamic perspectives

The analysis in the previous sections covered a seven-year period. The advantage of such a long time period is that the network structures are very visible; a disadvantage is that changes in this structure over time cannot be analysed. To investigate the dynamics of regional innovation networks, we now divide the observation period into three overlapping sub-periods, 1995–1997, 1997–1999 and 1999–2001. Networks constructed for shorter time periods tend to be much smaller than networks for longer time periods because of the smaller number of patent applications filed and the consequently fewer links.

Table A2 in the Appendix presents network statistics for the four regions in each sub-period. Figure 4 shows development of the network components. Selected network indicators are illustrated in Figure 5. First, all networks increase in size and with regard to the number of actors in the main component. This increase is not only in absolute terms but also with regard to the share of actors within the main component. At the same time, the share of isolates is decreasing, which is in line with the general tendency in science toward increasing collaboration and larger teams (Wuchty et al. 2007). There are differences between the regions with regard to the average number of links per actor (mean degree) and the level of network centralization. For example, there is a sharp increase of the mean degree in Jena and, to a lesser extent, also in Dresden, but the values for the two West German regions, Aachen and Karlsruhe, remain more or less constant. Both East German regions also show a strong tendency toward increasing centralization of the network, which is far less pronounced in Aachen and Karlsruhe. Jena is the only case study region where the average distance within the main component is decreasing, indicating an increasing degree of integration of this part of the network. While no clear trend in this respect can be found for Dresden and Aachen, we see an increase of the average distance within the main component of the Karlsruhe region, indicating disintegration.

Figure 4.

Component distribution in the four networks in three subperiods

Figure 5.

Dynamics of the regional networks

In all four regions there is a considerable increase in the number of relations to external actors (Figure 6). The highest level of dynamics in this respect occurs in the two East German regions: in Jena, the number of external relations almost doubled from 520 to 956 and in Dresden it increased by about 50 percent, whereas the increase in both Aachen and Karlsruhe was about 25 percent. Development in the share of external relations is most pronounced in Dresden, where the share of external relations rose from 60 percent in the mid 1990s to 75 percent by the turn of the millennium. There is a slight decrease in the share of external relations in Jena, and a small increase in Aachen and Karlsruhe.

Figure 6.

Development of external orientation

6 The subsequent performance of the RIS

It is plausible to assume that main effects of the quality of a RIS on its performance do not become immediately apparent, but occur with a considerable time-lag. It is therefore of interest to compare the performance of the four RIS in the subsequent period. We have already shown that during the years 1995–2001, the two East German RIS had a much lower efficiency than their West German counterparts (see Table 1, based on Fritsch and Slavtchev 2011). In terms of patents of private firms per 10,000 employees or per 1,000 R&D employees, neither East German region reached much more than about half the values of Aachen and Karlsruhe (Table 3). Between this first period and the subsequent period of 2002–2005, we find much higher growth rates in the number of patents per employee or per R&D employee in the two East German regions, indicating considerable convergence in the levels of regional patent productivity. Notwithstanding these relatively high growth rates, in many respects the two East German regions had not yet attained the levels achieved by the West German regions. An exemption is the number of patents of private firms per 1,000 employees, where Dresden reached the level of the Aachen region in the 2002–2005 period.

Table 3.  Indicators for the development of patent productivity in the four RIS under study
 East GermanyWest Germany
Patents of private firms per 1,000 employees 1995–20010.770.581.371.44
Patents of private firms per 1,000 employees 2002–20051.470.871.481.87
Change (%)90.9150.08.0329.86
Patents of private firms per 1,000 R&D employees 1995–200121.6322.8646.1138.92
Patents of private firms per 1,000 R&D employees 2002–200533.8532.4945.7644.61
Change (%)56.5042.13−0.0714.62

It would be well in line with the systemic view of innovation processes to assume that at least a part of the relatively high growth rates in patent productivity in the two East German RIS during the 2002–2005 period resulted from the rather advantageous characteristics of their networks, that is, the higher intensity of interaction and corresponding knowledge flows. Even if we cannot rule out that other factors may also have played an important role in this respect, the characteristics of the two East German RIS that we found suggest further strong improvements in the performance of the two East German RIS.

7 Discussion

Our comparative analysis has clearly shown that RIS are in no way islands but that regional innovation processes are to a considerable degree dependent on their wider spatial environment and the governing macroeconomic conditions. We found that even though the two leading RIS in the eastern part of Germany showed a much larger degree of interaction than two roughly comparable RIS in West Germany, the latter were clearly more efficient in terms of patenting and innovation.

At least at first sight, this result is in sharp contrast to the typical argumentation of RIS studies that proclaim unequivocal benefits from networking. We have three possible explanations for our observations that should add to the understanding of RIS. A first explanation for the relatively poor performance of the East German RIS is based in the ongoing transformation of the East German economy during the period of analysis. Obviously, innovation processes in Dresden and Jena have − to a considerable degree − been hampered by economic problems and the resulting reorganization of the East German economy. This clearly indicates that macroeconomic conditions at the sub-national level can play an important role in the performance of RIS. A second explanation may be provided by the different degrees of embeddedness of the RIS in their proximate geographic environment. The two East German RIS, Dresden and, particularly, Jena, are ‘cathedrals in the desert’, whereas the two RIS in the West, Aachen and Karlsruhe, are much more embedded in their surrounding spatial environment and have a higher share of relationships to actors located abroad. Third, it may be argued that it is not the level, but the quality of co-operation that is decisive and that our findings may be biased by the fact that the quality of the co-operative links of actors in the two East German regions is systematically lower than the quality of the links enjoyed by their West German counterparts. One might particularly suspect that in the East German case study regions, social aspects could have superseded economic imperatives (Uzzi 1997, p. 59), so that social proximity outweighs economic reasoning, leading to economically suboptimal collaboration decisions. Although we cannot completely exclude such influences, we have no indication that this is the case, especially since it has been found that these two regions outperform comparable regions in East Germany because of their intensity of interaction (Graf and Henning 2009). Moreover, because our network analysis is based on patent statistics, these links are all productive in that they have led to at least one patent application. Nevertheless, it would be quite important to learn more about the quality of collaborative linkages and to assess its influence on the performance of the respective RIS.

We conclude that focusing on a single region without accounting for geographic embeddedness and general macroeconomic conditions is not a sufficient approach to explaining RIS performance. This is particularly true when comparing RIS in different countries. There is no doubt that regions do differ with regard to their innovation performance and that regional conditions play an important role in explaining such differences, but this insight should not result in ignoring the effect that the wider spatial environment, particularly the national innovation system in which the regions are embedded, has on their performance. We therefore agree with Lundvall's (2007, p. 100) claim that investigations of regional, sectoral, and global innovation systems “have important contributions to make to the general understanding of innovation” but “are not alternatives to the analysis of national systems”.

To the degree that innovation activity benefits from co-operative links, division of innovative labour, and networking, the two East German regions in our sample are on the right track and have good prospects to perform equally well or even better than their West German counterparts in the future. This can be particularly expected after the phasing-out of the East German transformation process and a general economic recovery in this part of the country. Our analyses also show, however, that such a process will take a considerable amount of time, possibly decades, before the effect of conducive regional conditions has fully developed.

An important limitation of our comparative analysis is the small number of cases investigated. Future research involving more cases would therefore be very helpful in painting a more comprehensive picture of RIS. Another limitation is our reliance on patent data, considering their well-known shortcomings. Work based on other kinds of information about regional innovation networks, such as non-patenting-related interaction, must ensure that the information employed is comparable across regions. In this respect, patent data have the advantage that they reflect the same minimum standard of newness of an invention across all regions. Although our cases were selected to match in various dimensions, we cannot completely rule out the possibility that our interregional comparison is to some degree affected by a technology bias in the sense that interregional differences in the level of co-operation may be due to the fact that interaction is more important in some technological fields than in others.

An important policy conclusion that can be drawn from our analysis is that R&D co-operation and networks per se may not be sufficient to make RIS productive. The identification of other important factors that contribute to a well-functioning RIS is left for further research. Some of these factors, such as larger firm sizes in the West German regions that could result in scale effects in the production of knowledge, may also be found at the regional level. However, our analysis strongly indicates that these other factors will not be solely region specific, but should be looked for in the wider geographic environment and general economic conditions on a national as well as on a sub-national level.


  • 1

    This proposition may be regarded as in line with the lively Burt-Coleman debate in organization research. Coleman (1988) argues that the benefits of dense networks arise through increasing trust, whereas Burt (1992, 2004) emphasizes the importance of structural holes for the acquisition of diverse external knowledge. The validity of these arguments seems to depend on the direction of firms' search processes, with structural holes being more important in R&D aimed at new technologies for radical innovation (exploration) and density more important in the context of R&D based on a given technology aiming at incremental innovation (exploitation; Rowley, Behrens, and Krackhardt 2000). Provided that RIS are comprised of actors that follow quite heterogeneous search paths, it is plausible to assume that both types of activity will be present within a region. Since we have no information on the degree to which one of the two types of innovative activity prevails in the RIS under inspection, we are unable to account for these aspects in our empirical analysis.

  • 2

    A number of studies investigate the effect of different characteristics of national innovation systems on their performance (Hall and Soskice 2001; Dosi et al. 2006), but largely ignore regional conditions below the level of the nation-state.

  • 3

    Wanting to achieve comparability with regard to size and population density means that the two West German RIS with the highest levels of innovation efficiency, Munich and Stuttgart, were not selected because they are much larger than the two East German regions.

  • 4

    NUTS is an abbreviation of Nomenclature des Unités Territoriales Statistiques. This regional definition was established by Eurostat more than 30 years ago to provide a single uniform breakdown of territorial units for the production of regional statistics for the European Union; see URL: regions en.html.

  • 5

    A number of studies chose larger spatial units such as whole German Federal States for their analysis of RIS (see, e.g., the contributions in Cooke et al. 2004). Given the considerable differences in the efficiency of RIS (Fritsch and Slavtchev 2011), we believe that planning regions are more appropriate for such an analysis. This does, however, in no way mean that the wider spatial environment of a planning region is deemed to be irrelevant for innovation performance of planning regions. In fact, our analysis leads us to the conclusion that this wider spatial environment may have a considerable effect.

  • 6

    The cities in Figure 1 are represented by the respective city regions, which reflect their geographic size.

  • 7

    The Aachen region has experienced a considerable shift from coal mining to more manufacturing industries since the 1970s. In the period of our analysis, the mining sector no longer played an important role in the region's economy. The economies of the other three case study regions have not undergone such a dramatic change in their sector structure.

  • 8

    Employees are classified as working in R&D if they have a tertiary degree and work as engineers or natural scientists.

  • 9

    Although we have no information about the location of the respective private firms, we know from other studies (e.g., Fritsch and Schwirten 1999) that industry-university co-operation tends to be concentrated in the university's vicinity.

  • 10

    These estimates are based on a knowledge production function with the number of patents as R&D output and the number of R&D employees as R&D input; for details, see Fritsch and Slavtchev (2011). A value of 0.769 in the case of Aachen means that this RIS reaches 76.9 percent of the value for the RIS with the highest R&D productivity. Dresden and Jena reach only 35.4 percent and 39.4 percent, respectively, of that level.

  • 11

    We use the term ‘innovator’ instead of the more formal term ‘applicant’, since the main reason for patenting is to prevent copying of a successful innovation. Of course, not all patented inventions lead to marketable products or process innovations.

  • 12

    Hence, relationships between inventors within the same institution are not explicitly accounted for, but a preliminary study showed a close correspondence of network structures between the two approaches.

  • 13

    To correct for the misleading effects of headquarter patenting, the list of these (presumed) external actors was carefully checked to allocate them correctly whenever they could be identified as a local (and therefore internal) subsidiary.

  • 14

    Public research entities include universities and technical colleges (Fachhochschulen), as well as non-university publicly funded scientific institutes. The latter are in most cases members of one of large German scientific institutions: the Max Planck Society, the Leibniz Association, or the Fraunhofer Society.

  • 15

    For details on the calculation of network statistics, see Wassermann and Faust (1994).

  • 16

    A network component is defined as a subset of all network nodes that are directly or indirectly connected.

  • 17

    Mean degree is reported for valued and binary versions of the networks. The valued network accounts for the intensity of relations in terms of the number of common inventors, while the latter is solely based on the number of connections the actors have.

  • 18

    If g is the size of the network as measured by the number of actors and di is the degree, i.e., the number of connections of actor i (i= 1, . . . , g), then the density d of the network is defined as the number of all active linkages divided by the number of possible linkages within the network inline image. The density measure is somewhat problematic when comparing networks of different size as the number of possible links increases geometrically while the actual number of links usually does not since inventors are constrained in their capacity to have contact with other actors.

  • 19

    The share of mobility links over all links ranges from 49.2 percent in Karlsruhe to 37.4 percent in Dresden (Table 2). The figures do not indicate a more pronounced role for mobility in the East as might have been expected as a result of the East German transformation process.

  • 20

    Another explanation could be that the higher share of East German firms with R&D co-operation is motivated by greater scarcity of resources, which may result from their average relatively poor economic performance. There have been numerous policy programmes aimed at promoting R&D co-operation among East German firms but most of these were not implemented until the late 1990s, at the end of our period of analysis, and, therefore, cannot have resulted in patent applications.

  • 21

    Network centralization is given by inline image, where Cd(i) is the normalized degree centrality.

  • 22

    The average distance is the mean distance in terms of number of actors between any two actors in a component.

  • 23

    Federal States are an important level of administration and policy in Germany. We do not investigate relationships with actors in the surrounding planning region because this would mean including regions in other countries (for the cases of Aachen, Dresden and Karlsruhe), for which comparable information is missing, or regions in West Germany (for the case of Jena). The results for actors in the same Federal State may be distorted by the fact that the two East German states of Saxony (Dresden) and Thuringia (Jena) are much smaller than Baden-Württemberg (Karlsruhe) and North-Rhine-Westphalia (Aachen) in terms of population and economic activity so that opportunities to relate to other actors in the same state are smaller in the East.

  • 24

    Links between external actors are not considered here as they have little to do with the regional network.

  • 25

    It might well be the case that some actors hold the bulk of external relations but are not sufficiently integrated into the RIS to transfer the external knowledge to other actors in the system.


Table A1.  Gatekeepers according to brokerage score (top 5)
Technische Universität Dresden2,884Friedrich-Schiller-Universität Jena1,580Forschungszentrum Jülich290Forschungszentrum Karlsruhe1,042
Fraunhofer Dresden1,001Institut für Photonische Technologien598Fraunhofer Aachen221Robert Bosch GmbH, Karlsruhe329
Siemens AG Dresden104Carl Zeiss Jena GmbH515RWTH Aachen Universität57Universität Karlsruhe (TH)100
Leibniz-Institut für Festkörper- und Werkstoffforschung Dresden81Hans-Knoell-Institut für Naturstoff-Forschung481Gartzen, Johannes35Fraunhofer, Karlsruhe94
zentrum Dresden-Rossendorf80Jenoptik AG477FEV Motorentechnik23LuK Lamellen und Kupplungsbau52
Table A2.  Development of the regional networks over three subperiods
Number of nodes284376413536552636684814904666678752
Number of components164217212318332362473540600468473513
Size of main component61831098098121396886468994
Share of main component (%)21.522.126.414.917.819.
Number of isolates125166162236248280374416468373390414
Share of isolates (%)
Density (dichotomized)0.0050.0040.0050.0030.0020.0020.0010.0010.0010.0020.0010.001
Mean degree3.8594.3465.0703.4593.0943.7862.4442.6632.3702.5531.9262.261
Mean degree (dichotomized)1.4651.5111.9611.3841.2101.4400.9681.1621.0841.0550.8290.891
Average distance3.6392.7452.8152.9723.4962.9984.3563.5003.9603.4544.1713.941