The German biotechnology industry
Aiming at the expansion of knowledge creation and the innovation potential of domestic organizations, the German government employed various policy instruments and strategies. Facing increasing competition on a global scale “[…] financial aid for R&D and innovation activities in firms and research institutions, support of cooperation, networking and cluster formation, funding of technology-oriented start-ups, as well as institutional support for research institutions and knowledge transfer facilities” (BMBF 2006) became popular measures of policy makers to enhance organizational innovation capacity.
The biotechnology industry has been particularly strongly supported by the German government since the late 1990s, as it is assumed to be one of the key industries of the future (see for a more detailed description of the industry’s development in Germany in Fornahl et al., 2011). Fig. 1 elucidates that the number of joint R&D projects subsidized in the biotechnology industry substantially increased since the late 1990s with a peak in the period from 2007 to 2010.
Today biotechnology contributes to various fields such as public health, waste management, environmental protection and agriculture and is an important contributor to national and international economies. To cope with the comparatively less developed national biotechnology industry in the beginning of the 1990s, the governmental assistance program BioRegio was introduced in 1995 and it is assumed to have made an important contribution to the increasing growth rate of the biotechnology industry (Dohse 2000). The main aim was to develop biotechnological capacities by connecting relevant actors and supporting newcomers to foster the innovation processes and consequently achieve increasing competitiveness in the biotechnology industry. In the first phase of the competition, three regions (BioRegion Munich, BioRegio Rheinland, BioRegio Heidelberg) were supported with 90 million Euros for the time span between 1997 and 2005. Analyses have shown that these regions, compared to others, had higher average founding rates within the biotechnology industry (Staehler et al. 2007).
Moreover, the program enhanced cooperation activities between organizations not only in the subsidized regions but also on an inter-regional scale. The authors conclude that the BioRegio and the follow up BioProfil program where important measures that helped to develop a competitive biotechnology industry in Germany.
Consequently, the German biotechnology industry presents an ideal case for studying the evolution of knowledge networks in general and that of subsidized R&D networks in particular. Furthermore, its R&D activities are still dominated by basic science implying that universities represent central knowledge sources and key collaboration partners (Powell et al. 1996a; Niosi and Banik 2005; Meyer-Krahmer and Schmoch 1998). Moreover, universities with laboratories and teaching medical centres are frequently integral to therapeutic innovation by serving as bridges between research and new applications of pharmaceutical products. In particular, so-called bench-to-bedside or translational medicine stimulates the linking of universities’ and firms’ applied research.
Data source and network construction
To construct the network of subsidized R&D collaboration for this industry, we rely on information from the German subsidy catalogue (“Förderkatalog”). The catalogue contains organizations and projects subsidized by the Federal Ministry of Education and Research (BMBF), Federal Ministry for Economic Affairs (BMWi), Federal Ministry for Environment, Nature Conservation and Nuclear Safety (BMU), Federal Ministry of Transport, Building and Urban Development (BMVBS) as well as the Federal Ministry of Food, Agriculture and Consumer Protection. The database contains more than 160,000 running or already finished R&D projects subsidized by these ministries in the time span between 1960 and 2016. Information comprises the subsidized organizations, the amount of money the organization received, locations, industrial classification code (NACE), who jointly participated in which projects, as well as project duration. The information on participants in the joint R&D projects are used for the construction of the network (Broekel and Graf 2012). Due to uncertainty and potential risks, the level of organizations’ engagement in joint knowledge production processes are assumed be below a socially optimal level. For this reason, policy encourages collaborative research activities by means of subsidizing joint R&D projects, which make inter-organizational collaboration obligatory for receiving subsidies (see, e. g., Scherngell and Barber 2009, 2011). Due to regulations organizations agree to when being awarded a joint R&D grant, knowledge exchange becomes highly likely, implying that participation in subsidized R&D projects can be interpreted as links in inter-organizational knowledge networks (see for a discussion Broekel and Graf 2012).
Additional information were obtained through the web presentation biotechnologie.de, which is initiated by the BMBF. This includes information on organizations’ founding years, their numbers of employees, and specifications of research.
We collect information for all organizations belonging to the funding area and funding priorities (“Förderkatalog”/“Förderschwerpunkt”) biotechnology (B) in the Federal Government’s R&D planning system (“Leistungsplansystematik”) between 01.01.2007 and 31.12.2010. We consider only projects with at least two partners to reduce effects related to organizations’ general capability and willingness to collaborate. Hence, our network of subsidized R&D collaboration includes all organizations that participated in at least one joint R&D project classified into the funding priority area B in the respective time period. Links between organizations are established when two organizations jointly participate in the same project.
Fig. 2 shows the spatial distribution of subsidized joint R&D projects at the level of districts. Intensive R&D collaboration activities exist in the area surrounding Munich, Mannheim, and Heidelberg. Both regions were participants in the first round of the BioRegio competition mentioned above. The same is true for the area of Berlin and Potsdam, which belong to the receiving regions of the BioProfil program and show intense R&D project participation between 2007 and 2010. All three regions are characterized by a high number of biotechnology firms and universities as well as public research institutes. Moreover, the regions around Cologne, Aachen, Düsseldorf, Hamburg, and Hanover show increases in funding for joint R&D projects. Notably, all their universities are significantly involved in life science and applied biotechnology. The findings on the spatial distribution of funding are in line with other findings about the general distribution of important German biotechnology regions and organizations (see Zeller 2001; Ter Wal 2013).
The data distinguishes between the name of the recipient organization and the name of the executing organization. In most instances, both names are identical. However, in case of large firms or research institutes with various sub branches located in different regions, the recipient and executing entity may vary. To keep the level of observation equal for all organizations and to control for potential “head-quarter effects”, we apply the approach introduced by Broekel and Graf (2012). Hereby each organization is identified by the combination of the recipient’s name and the official municipality identification number of the district it is located in. Collaborations are defined by simultaneous participation in joint R&D projects at different points in time.
The final dataset consists of 316 unique German biotechnology organizations, which participated in at least one of 193 projects between 2007 and 2010. The average number of participants per project is 3.6 (ranging from 2 to 39). The network is constructed by projecting the two-mode network connecting all unique organizations with their joint R&D projects into a one-mode network consisting only of organizations and their direct relation. Organizations participating simultaneously in multiple projects may (indirectly) connect to other organizations that do not interact directly in a subsidized joint project. It is argued that through contracts and regulations all participants of a project are prompted to share relevant knowledge (see for a discussion Broekel and Graf 2012). Consequently, links between actors are undirected and represent mutual agreements on a collaboration project x
ij = x
ji. For each year (t) between 2007 and 2010, we construct a network matrix BT
ij(t) with values between x
ij = 1 and x
ij = 0 indicating an existing and non-existing link between actor i and j (dichotomized network), respectively.
Table 1 shows the general growth of the analysed network while. Table 2 shows the increasing density and average degree of the network for the respective years.
Fig. 3 depicts the overall share of links between the different organizational types. Links spanning institutional borders are less frequent compared to those within the same institutional framework. In particular, over one third of all links are among universities. The second largest share are links among firms (economy). The high share of links among universities underlines biotechnology’s strong orientation towards basic research, which tends to be very costly and complex with uncertain returns making public research facilities a necessity. However, knowledge diffusion between the private and public sphere takes place as about 10% of all links connect firms and universities. Given that just 6% of all links are established by joint projects between public research institutes (PRI) and firms, it implies that universities are more important for knowledge diffusion into the private economy than public research institutes.
Actor level variable
The most important variable at the actor level shows whether an organization is a university or not. The corresponding information is given in the database. In this paper, we combine public and private universities and technical colleges (“Fachhochschulen”). While the two types of organizations differ in some dimensions (Beise and Stahl 1999), their similarities preponderate in the context of the paper. The variable UNIVERSITY takes a value of 1 if an actor is a university or technical colleague and zero otherwise. Distinguished from that are public research institutes (PRI). The dummy variable RESEARCH summarizes a variety of state or publicly established research organizations like the Frauenhofer‑, Max Planck- or Helmholtz institutes and associations (e. g. German Centre for Aviation and Space Travel). Those organizations that are not universities or research organizations are primarily firms and miscellaneous organizations. They serve as the reference group to identify the effect of UNIVERSITY and RESEARCH.
Fig. 4 shows the share of subsidies attributed to the different types of organizations for four consecutive years. The figure indicates the share of organizations belonging to the group of economic (firms) and public research institutes are alternating between all four years. While during 2007 economic organizations dominate in terms of being recipients of funding, the share is decreasing in the following years. Private research organizations account for around 30% of all shares during all years except for a decline in 2009. Apart from the first year, universities are the main recipient of joint R&D funding, highlighting their central role in the subsidized biotechnological R&D network in Germany. The importance of universities as a recipient of public funding for joint R&D is further underlined by their rank in terms of project participation. Universities dominate the top 10 spots (see Table 3).
In addition to the organizational type, we also consider variance in organizations’ size. The size variable (EMPLOYEES) captures the number of organization employees. For computational reasons, we do not use the number of employees directly but categorize the numbers according to the EU’s definition of small and medium sized enterprises. The number of employees is rescaled to the categorical variable EMPLOYEES with the following categories: 1–9 employees (1), 10–49 employees (2), 50–249 employees (3), and more than 249 employees (4).
The third variable on the actor level is organizational experience. The variable captures if the respective organization was involved in any subsidized R&D projects before the analysed time frame. Hence organizations experience is coded as a binary variable with 1 if they participated in earlier subsidized R&D projects and 0 otherwise.
Dyad level variables
The first variable at the dyad level is institutional proximity (INST.PROX). Organizations share institutional proximity when they belong to the same type of institutional framework. As explained above, the database differentiates between universities, public research institutes, firms, and miscellaneous organizations. We construct a relational variable that takes a value of 1 if two organizations share the same institutional type and 0 otherwise.
Geographical proximity (GEOG.PROX) measures the spatial distance between two organizations based on their locations’ latitude and longitude coordinates given in the data. Similar to Boschma et al. (2011), we take the natural log of the difference between the maximum distance found between any pair of organizations in the data and the individual dyadic distance of the focal pair of organizations. The result ranges between 0 and 6.44, whereby 0 indicates the maximal and 6.44 the minimal distance.
Alternatively, to capture the importance of shared regional context, embeddedness into the same regional institutional innovation system and the advantages of geographical proximity, we construct the variable (SAME.NUTS). It obtains a value of one if two organizations are located within the same region and zero otherwise. A comparison of the two variables allows the investigation of the relative relevance of the regional context (SAME.NUTS) in contrast to “pure” spatial distance (GEOG.PROX).
Organizational proximity (ORG.PROX) is approximated by differentiating between executing and receiving organizations. As mentioned above, the subsidies database includes information on the organization receiving the grant and the organization actually conducting the research. In line with Broekel and Graf (2012), two organizations are argued to be part of the same mother organization when they share the name of the receiving organization and, hence, are characterized by organizational proximity. The variable ORG.PROX obtains a value of 1 if two organizations have the same mother organization and 0 if otherwise.
In order to account for cognitive proximity (COG.PROX), a more complex approach is needed. Following Broekel and Boschma (2012), we estimate cognitive proximity between organizations based on their sector membership given in the subsidies data. The data includes information on two-digit NACE codes for all subsidized organizations. In the case of universities, universities of applied science, and some research organizations three-digit NACE codes are available.
Based on this information, we construct a measure of technological similarity (cognitive proximity) for each pair of sectors included in the data. As each organization is assigned to a single sector, the measure can then be applied to describe inter-organizational technological similarity. For the construction, we follow the approach of Broekel and Brachert (2015). In a first step, we count the co-occurrence of six-digit research areas (“Leistungsplansystematik”) as reported on each individual projectFootnote 4 acquired by two sectors’ organizations in the preceding 10 years. The idea behind it is that the more frequently two sectors are subsidized through the same research areas (six-digit “Leistungsplansystematik”) scheme, the more likely they are active in the same technologies. Hence, the greater their technological overlap, the smaller is their technological distance. The precise estimation is done in accordance with Teece et al. (1994) and Bryce and Winter (2009). The number of six-digit area codes’ co-occurrences on projects acquired by two sectors’ organizations (i and j) is denoted as J
ij will naturally increase with the number of acquired subsidized projects. It is therefore compared to the number of co-occurrences that can be expected if projects are randomly assigned across all sectors’ organizations. K is the number of six-digit research areas and n
i represents the total number of individual projects organizations of sector i are active in. n
j is the corresponding number for sector j. The expected number of projects in the same technological class acquired by sector i and j is (x
ij), which can be seen as a hypergeometric random variable whose mean and variance can be estimated as shown in Bryce and Winter (2009, p. 1575 f.). The sectors’ size is approximated by the total number of acquired projects in this sector. The final index τ
ij is then estimated as the standardized difference between the observed and expected numbers of co-occurrences. It refers to the “raw” number of co-occurrences of two sectors’ organizations. In order to avoid size biases and allow for easier interpretation, it is standardized and divided by the maximal similarity score in the data. Negative values imply strong dissimilarity and hence their interpretation is the same as in the case of zero values. They are set to zero, implying that the final similarity index COX.PROX ranges between 0 and 1 with values close to one indicating maximal technological similarity.
We approximate social proximity by two organizations’ collaboration history. The underlying assumption is that organizations have a distinct staff running R&D projects. Hence, two organizations share social proximity (SOC.PROX) if both jointly participated in at least one subsidized R&D project in the 10 years prior to the investigated time period. The variable SOC.PROX is 1 if that is the case and zero otherwise.
To gain insights about potential knowledge diffusion from universities across institutional boarders, we define the two variables UNI.FIRM and UNI.PRI. UNI.FIRM takes the value of 1 in case of a university collaborating with a firm and 0 otherwise. Similarly, UNI.PRI. takes the value of 1 when a link is established between a university and a public research institute and 0 otherwise.
Finally, we add four interaction terms. UNIVERSITY*EXPERIENCE and UNIVERISTY*EMPLOYEES are included for testing the influence of the two actor level characteristics (experience and number of employees) on the probability of universities establishing links. UNI.FIRM*GEO.PROX and UNI.PRI*GEO.PROX are interaction terms between two dyad level variables and approximate the effect of geographical distance on specific types of university links.
The Siena model is estimated using the 7 dyadic variables approximating the different types of proximity, 3 node level variables capturing organization type (UNIVERSITY) and their size (EMPLOYEES), as well as organization experience (EXPERIENCE). In addition, we consider two network structural effects, density and preferential attachment as well as four interaction terms. The model estimates the year-to-year evolution of the network for the time period 2007–2010.