Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction: Proximity and Firm Innovativeness

In this analysis we seek to disentangle the relationship between network positioning, geographical co-location and firm innovativeness.Footnote 1 The knowledge-based approach started to thrive in economics in the early 1980s. The neo-Schumpeterian school of thought explicitly emphasized the central role of knowledge and innovation as drivers of economic change and prosperity (Hanusch and Pyka 2007b). From this perspective, knowledge is no longer considered purely a public good but rather as a scarce and highly valuable firm-specific resource. Firms have to take action, spend time and resources, and develop routines and capabilities to successfully tap external stocks of knowledge. The creation of novelty is regarded as a collective process that involves knowledge exchange and learning processes among mutually interconnected economic actors, each of whom seeks to improve their own imperfect knowledge base and accomplish their very individual goals (Hanusch and Pyka 2007a). These actors are embedded in socio-economic systems of innovation (Lundvall 1992; Nelson 1992) that are complex and dynamic in nature (Fagerberg 2005). Quite recently it has been argued that the proximity concept originally proposed by Boschma (2005b) can be regarded as an integral part and extension of the evolutionary economic approach (Boschma and Frenken 2010).

The knowledge-based view in management science emerged just a few years later. Scholars started to emphasize the strategic nature of knowledge and pointed to the fact that both knowledge accessing and learning processes are crucial for firms to gain competitive advantages and outperform competitors (Nonaka 1991; Kogut and Zander 1992; Grant 1996; Coff 2003). According to this view, firms have to build up a certain degree of absorptive capacity (Cohen and Levinthal 1990), avoid learning races (Amburgey et al. 1996) and reduce causal ambiguities (Simonin 1999) to make successful use of external knowledge sources. These challenges are essentially attributable to the tacit nature of non-codified and context-specific knowledge (Polanyi 1958; Polanyi 1967) that underlies innovation processes at the micro-level. With the emergence of the proximity concept it has been argued that proximity in all its facets can enhance a firm’s ability to access new stocks of knowledge and generate novelty in the form of innovation (Amin and Wilkinson 1999).

We are surely not the first to address the direct relationship between proximity and firm innovativeness. An excellent overview of contemporary research in this area is provided by the study carried out by Knoben and Oerlemans (2006). Quite recently, a few excellent theoretical (e.g. Boschma 2005b; Torre and Rallet 2005; Visser 2009; Boschma and Frenken 2010) and empirical studies (e.g. Oerlemans et al. 2001; Oerlemans and Meeus 2005; Owen-Smith and Powell 2004; Whittington et al. 2009) have started to address both distinct and combined proximity affects. Nonetheless, we still face more questions than answers.

For instance, Boschma (2005a) calls for a clear analytical separation of distinct proximity dimensions, a more dynamically oriented proximity perspective and stronger recognition of both positive and negative effects of proximity on innovation. However, these issues have several far-reaching implications. Due to the conceptual ambiguity of the proximity concept, the underlying mechanisms are not clearly assigned to one specific proximity dimension. Thus, we are still lacking an in-depth understanding and a clear separation of the mechanisms that foster or hamper innovation processes at the firm level. In addition, we have a rather vague idea of how and why one proximity dimension affects another and the underlying logic of combined proximity effects is not yet sufficiently understood. Quite recently some pioneering studies have started to address these questions empirically (most notably: Owen-Smith and Powell 2004; Whittington et al. 2009). Nonetheless, it should be noted that most previous studies are cross-sectional in nature. This neglects the dynamic nature of the mechanisms that underlie the various types of proximity dimensions. Moreover, previous studies inherently imply that proximity is positive per se; the “dark side” of proximity is widely ignored. Finally, the majority of previous empirical studies are based on data from the biotech industry. However, knowledge exchange, learning and innovation processes can significantly differ across industries due to differences in the degree of the industry’s technological maturity, different industry life-cycle stages and differences in firm size distribution.

Consequently, this study contributes to the existing body of literature in several ways. Firstly, by focusing on two proximity dimensions – network proximity and geographical proximity – we seek to deepen our understanding as to how distinct, or potentially, combined proximity effects relate to firm-level innovation outcomes. Secondly, in response to Boschma’s critique we provide an evolutionary proximity framework and apply longitudinal data and panel estimation techniques to account for the dynamic nature of proximity in all its facets. In doing so, we seek to understand the underlying mechanisms that determine both distinct as well as combined proximity effects. Thirdly, we supplement existing research by providing new empirical evidence from a unique panel dataset for the entire population of 233 German laser source manufacturers between 1990 and 2010.

In a nutshell, inspired by the conceptual framework of Boschma (2005a) and the empirical study of Whittington et al. (2009) and supplemented by our own considerations, we raise the following research question: Are firm-level innovation outcomes of German laser source manufacturers related to network proximity effects, to geographical co-location effects or to combined proximity effects; and if so, are the effects positively or negatively related to firm innovativeness at later points in time?

2 Theoretical Background

2.1 The Multifaceted Character of Proximity

Over the past few years, scholars in the field of economics, sociology, geography and management science have significantly improved our understanding of how proximity can improve a firm’s ability to tap into new knowledge sources, learn to recombine existing knowledge stocks and finally generate new and commercializable goods and services. Firms are simultaneously exposed to a variety of proximity dimensions such as institutional proximity, organizational proximity, cultural proximity, technological proximity, network proximity and geographical proximity (cf. Knoben and Oerlemans 2006, p. 71). In the most general sense, proximity can be defined as “[…] being close to something measured on a certain dimension” (Knoben and Oerlemans 2006, pp. 71–72). One of the main issues that is common to all literature on proximity is the conceptual ambiguity of previous approaches. In this context Knoben and Oerlemans (2006, p. 71) criticize the fact that previous research has failed to provide a clear separation of proximity dimensions which is still reflected in conceptual overlaps across many proximity dimensions.

We follow the proximity framework proposed by Boschma (2005b) for several reasons. Firstly, the framework allows five proximity dimensions to be clearly defined and separated: cognitive, organizational, social, institutional and geographical proximity (Boschma 2005b, p. 62). He stresses the lack in understanding combined proximity effects and emphasizes that proximity in all its facets can both facilitate and impede knowledge access and learning processes over time. Thus the framework enables both the positive as well as the negative impact of proximity on firm-level innovation outcomes to be explained. Secondly, the proposed proximity framework can be regarded as an integral part of the evolutionary economic approach (Boschma and Frenken 2010, p. 121). The integration of the concept broadens the analytical scope of the evolutionary approach by explicitly considering several types of proximity dimensions (ibid). Moreover, it also paves the way for a more dynamic and process-oriented understanding of the proximity concept. Thirdly, proximity as an analytical concept allows multiple proximity dimensions to be incorporated into an explanatory framework (Boschma and Frenken 2010, p. 124). The proximity dimensions are clearly separated and thus independent of each other. This implies, however, that one can reduce as well as extend the list of relevant proximity dimensions without changing the meaning of each dimension (ibid).

In summary, the analytical proximity framework originally proposed by Boschma (2005b) and then extended upon by Boschma and Frenken (2010), clearly promotes a more process-oriented understanding of how changes in proximity affect innovation outcomes over time. It allows the interplay between selected proximity dimensions to be analyzed and provides potential explanations for both complementary and substitutional effects. Finally, the framework offers the possibility of analyzing the relatedness between firm innovativeness and individual proximity dimensions and provides a solid basis for examining whether combined proximity effects are positively or negatively related to firm innovation outcomes at later points in time.

2.2 Network Proximity and Firm Innovativeness

Now we take a closer look at the network proximity dimension. This type of proximity is frequently referred to as relational or social proximity (Coenen et al. 2004). The social proximity concept is strongly influenced by social capital and embeddedness literature (Laumann et al. 1978; Granovetter 1985; Uzzi 1996; Uzzi 1997; Granovetter 2005). According to this perspective, economic actions and outcomes are influenced by the context in which they occur (Uzzi 1996; Gulati 2007). Boschma (2005b) defines social proximity “[…] in terms of socially embedded relations between agents at the micro-level.” The use of this proximity dimension requires an in-depth specification of at least three constituent features: the agents, the type of relations that connect these agents and the system boundaries that define the scope and size of the overall network.

For the purpose of this paper we focus on interorganizational innovation networks (Pyka 2002) consisting of all German laser source manufacturers (LSMs) and laser-related public research organizations (PROs) that were actively operating in the field of laser source research and production between 1990 and 2010. In the most general sense, these actors can be interconnected in multiple ways and can exchange knowledge either through informal or formal relationships. According to Pyka (1997, p. 210) the former encompasses “[…] any action that can contribute to disclosure, dissemination, transmission and communication of knowledge.” The latter addresses a broad variety of structural forms ranging from short term contractual alliances and minority alliances, characterized by an intermediate degree of hieratical control, to long-term equity alliances such as joint ventures (Gulati and Singh 1998). All formalized partnerships, however, stipulate that all parties involved agree upon more or less formalized obligations, rights and common goals. We focus on a specific type of innovation network that is constructed on the basis of nationally or supra-nationally funded R&D cooperation projects.Footnote 2 The reasoning for this is straightforward. All project partners have to agree on contract clauses that aim to improve knowledge exchange among project partners and initiate innovation activities (Fornahl et al. 2011; Scherngell and Barber 2009). At the same time the concretization of node and tie dimensions outlined above specifies the boundaries of the network under investigation.

Previous research provides strong evidence that not only structural network characteristics such as network density, structural holes, or structural equivalence (Gulati et al. 2000, p. 205) but also a firm’s structural position within the overall industry network can significantly affect various dimensions of firm-level performance (Baum et al. 2000; Stuart et al. 1999; Zaheer and Bell 2005). It has been demonstrated that a firm’s occupation of strategically important network positions can improve its ability to access external knowledge sources (Grant and Baden-Fuller 2004; Buckley et al. 2009) and facilitate interorganizational learning processes (Hamel 1991; Schoenmakers and Duysters 2006; Nooteboom 2008). In a similar vein, previous studies have explored the importance of structural network characteristics and various types of network positions in a firm’s innovative performance (Powell et al. 1996; Stuart 1999; Stuart 2000; Fornahl et al. 2011). The potential benefits of a firm’s network position are closely related to the overall network topology. Some scholars have argued that brokerage positions in sparsely connected networks are the most beneficial – the “structural hole theory” (Burt 1992) – whereas others have stressed the importance of high nodal degrees at the actor level and closely interconnected overall network structures – the “closure theory” (Coleman 1988).Footnote 3 For the purpose of this study we focus particularly on the latter stream of research.

However, not only the positive but also the negative impact of a firm’s network embeddedness on performance outcomes has been the subject of debate over the past few years (Boschma 2005b). By now it is well-recognized that a firm’s position within the innovation network can have a positive impact on its innovative performance at subsequent points in time (Uzzi 1997). However, after a certain point the positive effects of social proximity may move in the opposite direction and have an adverse effect on learning and innovation (Boschma 2005b, p. 66). This phenomenon is referred to as “overembeddedness” (Uzzi 1997). The main argument behind this concept is straightforward. Too much social proximity can cause a lock-in effect in a sense that actors remain in an established web of habitual partnerships. Such a closed network system generates opportunity costs because the actors involved isolate themselves from other firms and organizations with fresh and novel ideas (Boschma 2005b, p. 66). Others have pointed to the fact that some organizations face considerable difficulties in dissolving old relationships and forming new network ties (Kim et al. 2006). These authors coined the term “network inertia” to address the persistent organizational resistance to changing its interorganizational network. Kim and colleagues (2006, p. 706) argue that change is mainly influenced by four types of constraints: internal constraints, network tie-specific constraints, network position-specific constraints, and external or environmental constraints. This concept has its intellectual roots in the structural inertia theory in organizational ecology (Hannan and Freeman 1984) and is clearly evolutionary in nature. The process-oriented network inertia concept explains negative innovation outcomes of firms not primarily because of overembeddedness issues but rather because some firms cannot react and adapt fast enough to new conditions and needs. In other words, a firm may face a situation in which the formation of connections to new cooperation partners that would bring in innovation stimuli is likely to be seriously delayed or even entirely impaired whereas obsolete linkages simply cannot be resolved. Following this reasoning one could argue that a firm can face a situation in which it is not necessarily overembedded but rather missembedded.

2.3 Geographical Proximity and Firm Innovativeness

Next, we focus on the spatial or geographical dimension of proximity. Oerlemans and Meeus (2005, p. 94) point out that this body of research can be grouped into two categories: one which focuses on spatial (or face-to-face) interaction and interactive learning (Saxanian 1990; Maskell and Malmberg 1999) and one focusing on spatially mediated knowledge spillovers (Feldman 1993; Audretsch and Feldman 1996, 2003). For the purpose of this paper we stick to the latter perspective for two reasons. On the one hand, the knowledge spillovers perspective allows us to define geographical proximity in a quite restrictive manner and isolate the spatial proximity dimension from other proximity dimensions (Boschma 2005b, p. 69). On the other hand, the knowledge spillover perspective acknowledges the partially non-rival, dynamic and cumulative character of knowledge (Oerlemans and Meeus 2005, p. 94) and puts forward the argument that knowledge tends to spill over locally between firms of the same industry – so-called intra-industry or MAR externalities (Marschall 1890; Arrow 1962; Romer 1986) – or between firms of different industries – so-called inter-industry or Jacobs externalities (Jacobs 1969). Due to the aim and the scope of this study we focus specifically on intra-industry knowledge spillover.

This perspective stresses that proximity influences a company’s ability to benefit from knowledge spillover stemming from research and development activities taking place outside the boundaries of the firm (Audretsch 1998). In the early 1990s a vibrant field of research started to address the spatial dimension of knowledge and innovation by introducing novel methods for measuring the extent of local knowledge spillovers and innovative activities (Audretsch 1998, p. 22). Empirical studies from this strand of research suggest that physical proximity of firms to external knowledge sources enhances innovative and economic performance (Jaffe 1989; Audretsch and Feldman 1996, 2003; Audretsch and Dohse 2007). However, they have also demonstrated that spatial knowledge accumulation effects and innovation activities can be strongly determined by the knowledge intensity of the industry and stage of the industry life cycle (Audretsch 1998).

Keeping in mind the previous considerations, we define geographical proximity as a concept that “[…] refers to the spatial or physical distance between economic actors, both in its absolute and relative meaning” (Boschma 2005b, p. 69). This perspective is consistent with a related concretization that defines geographical proximity as “[…] kilometric distance that separates units (e.g. individuals, organizations, towns) in geographical space” (Torre and Rallet 2005, p. 49) and, at the same time, provides a solid basis for taking a closer look at the link between geographical co-location, knowledge and innovation. The main argument is that any firm located in an agglomeration area can benefit from local knowledge spillovers as long as geographical openness of the agglomeration is ensured (Boschma 2005b, p. 69). The mechanisms that generate the knowledge spillover are not relational in nature. In other words, knowledge spillovers can occur regardless of whether firms in a region are interconnected by a formal relationship or not (Boschma 2005b, p. 69). Geographical proximity has to be defined in such a restrictive manner as to allow a clear separation of other proximity dimensions (ibid), especially relational knowledge transfer mechanisms such as informal social relationships at an interpersonal level.

However, geographical proximity does not have a positive effect on knowledge transfer and learning processes per se. Boschma (2005b, p. 70) stresses the risk that spatial lock-in effects and a lack of openness to the outside world can result in situations in which local knowledge quickly becomes outdated and knowledge-based agglomeration effects become increasingly eroded over time. Thus, firms in closed agglomeration areas become increasingly inward-looking and isolate themselves from the other actors in the industry. In summary, the geographical proximity dimension defined in this way addresses positive but also negative consequences of local knowledge spillovers on firm innovativeness due to a firm’s co-location to other organizations.

2.4 Addressing Combined Proximity Effects and Firm Innovativeness

In the real world, it is very unlikely that only one proximity dimension affects a firm’s innovative performance in isolation. Instead, firms are simultaneously exposed to multiple mutually interdependent proximity dimensions. Below we focus exclusively on combined proximity effects between two proximity dimensions: network proximity and geographical proximity.

According to Whittington et al. (2009, pp. 97–98), there are theoretically three ways in which combined proximity effects can affect a firm’s innovation outcome. Firstly, independent effects of network and geographical proximity on innovation would imply that both proximity dimensions influence innovation through autonomous mechanisms. Secondly, substitutional effects of network and geographical proximity on innovation are based on the notion that one proximity dimension can compensate for a lack of another proximity dimension. Finally, complementary effects of network and geographical proximity on innovation imply that these two dimensions are supplementary in nature due to extra-additive effects. Unfortunately, Whittington and colleagues (2009) do not clearly address and separate the mechanisms that underlie these combined proximity effects. This is precisely the point at which our conceptual framework comes into play.

3 Conceptual Framework and Hypotheses Development

3.1 Specifying Distinct and Combined Proximity

Our conceptual framework (cf. Fig. 12.1) builds upon the theoretical considerations outlined above and aims to contribute to an in-depth understanding of the mechanisms that underlie both distinct and, especially, combined proximity effects. The first requires a concretization of the elementary building blocks in our framework. The framework consists of four elements – (I) network proximity (II) geographical proximity, (III) combined proximity, and (IV) innovation outcomes.

Fig. 12.1
figure 1

Conceptual framework – distinct and combined proximity and firm-level innovation output (Source: Author’s own illustration)

Full size image

To start with, we outline our notion of network proximity. We argue that a firm with a high number of direct partners, irrespective of whether these direct partners are themselves well-connected or not, has a high level of network proximity. We focus on a firm’s degree of connectedness for the following reasons. Firstly, a firm’s nodal degree is quite a simple and straightforward network concept that reflects the full range of its external knowledge channels. Secondly, densely connected actors are highly visible and well-recognized by other actors in the network (Wasserman and Faust 1994, p. 179).

Similarly, geographical proximity has to be specified. One way to accomplish this task is to focus on the physical distances between firms in a well-specified population. Thus, a firm with a short average distance to all other firms at the same stage of the industry value chain has a high level of geographical proximity (Whittington et al. 2009). However, not only firms but also universities or other public research organizations (PROs) are an important source of new technological knowledge (Agrawal 2001, p. 285). It has been argued that PROs follow quite different rules for the dissemination and use of scientific findings than profit-oriented firms (Owen-Smith and Powell 2004, p. 7). In other words, there is a qualitative difference as to whether an LSM is co-located to other firms at the same stage of the industry value chain, or to other laser-related public research organizations carrying out either basic or applied research. In line with Whittington et al. (2009) we consider a second type of geographical proximity that reflects a firm’s average distance to all laser related PROs in the sample. The combined proximity dimension captures a firm’s simultaneous positioning in both the network space and the geographical space.

Finally, a firm’s innovation output has to be clarified. The Oslo Manual (OECD 2005) differentiates between four types of innovation: “product innovation”, “process innovation”, “organizational innovation”, and “marketing innovation”. We focus here on all kinds of novel ideas generated by laser-related firms that are truly new to the market and thus at least theoretically patentable.

3.2 The Link Between Proximity Effects and Firm Innovativeness

Previous research has significantly contributed to our understanding of how a firm’s structural embeddedness and network positioning affects the innovation generating process (Shan et al. 1994; Powell et al. 1996; Ahuja 2000; Owen-Smith and Powell 2004; Gilsing et al. 2008). These findings leave us to suppose that a firm’s network proximity is positively related to firm-level innovation outcomes (cf. Fig. 12.1, Arrow 1). At least three theoretical arguments substantiate this assumption.

Firstly, it is of vital importance, especially in science-based industries (Grupp 2000), to have access to external knowledge stocks and to be able to acquire new knowledge stocks (Al-Laham and Kudic 2008). A high degree of connectedness provides access to complementary knowledge sources (Grant and Baden-Fuller 2004) and opens up opportunities of interorganizational learning processes (Hamel 1991). Secondly, firms with a high nodal degree are most visible in the network. Agency theory (Spence 1976, 2002) implies that an above-average nodal degree signalizes an advantageous cooperation opportunity to other network actors. As a consequence, a well-embedded firm is likely to get more cooperation offers than other actors (Hanneman and Riddle 2005). A broad opportunity-set of potential cooperation partners increases the probability of finding the right partner when required. Finally, well-embedded firms are characterized by a comparably high level of alliance experience that allows alliance capabilities to be built up over time (Kale et al. 2000, p. 750; Schilke and Goerzen 2010). The implementation of cooperation routines saves costs (Zollo et al. 2002) and increases managerial efficiency over time (Goerzen 2005). As a consequence, well-experienced firms have a higher chance of completing innovation projects successfully than firms that are cooperating for the first time. In summary, these considerations result in the formulation of our first hypothesis:

H1

A firm’s network proximity is positively related to its innovative performance at later points in time.

Next, we turn our attention to the relationship between a firm’s geographical proximity to other firms at the same stage of the industry’s value chain and its subsequent innovation outcomes (cf. Fig. 12.1, Arrow 2a). The following theoretical arguments substantiate the assumption that geographical co-location can generate localized knowledge spillovers. Knowledge spillovers provide valuable information and increase a firm’s awareness of new industrial and technological trends. Consequently, firms who benefit from knowledge spillovers increase their general technological understanding. According to Breschi and Lissoni (2001) at least three underlying mechanisms are responsible for knowledge spillovers at the local level: local labor markets (Almeida and Kogut 1999; Zucker et al. 1998), local technology markets (Lamoreaux and Sokoloff 1999) and the low propensity of skilled workers to relocate in space (Breschi and Lissoni 2009). In particular, it has been argued that these mechanisms enable knowledge transmission and facilitate knowledge reuse among firms of the same industry at the local level. These considerations substantiate the assumption that a firm can benefit from regional knowledge spillovers due to its geographical closeness to other firms at the same stage of the industry value chain. Consequently we elaborate on our hypothesis:

H2a

A firm’s geographical proximity to other laser source manufacturers is positively related to its innovative performance at later points in time.

Now we address a firm’s proximity to other laser-related public research organizations (cf. Fig. 12.1, Arrow 2b). Jaffe (1989) and Acs et al. (1992) provide interesting empirical results concerning the contribution of knowledge spillovers from public research organizations (PROs). Using patents (Jaffe 1989) and direct counts of innovation outputs (Acs et al. 1992), these studies provide evidence for the positive impact of knowledge spillovers from universities on corporate innovation activity. According to Jaffe (1989), this effect is particularly significant in the areas of drugs and medical technology, electronics, optics and nuclear technology. At least two theoretical arguments come into play in this context. Firstly, knowledge flows out of PROs, especially universities, much more readily than it does from privately owned firms (Jaffe et al. 1993; Owen-Smith and Powell 2004, p. 7). Firms tend to retain information or implement secrecy strategies to protect their knowledge stock whereas universities and PROs tend to disseminate research findings instantly. Secondly, shortages in qualified personnel can become an existential problem for a firm. These issues become all the more significant for firms operating in science-driven and highly interdisciplinary fields of research. A firm’s geographical co-location to technical universities and universities of applied science increases the firm’s chance of hiring well-qualified graduates such as engineers, scientists and other experts. Drawing on these considerations we formulate the following hypothesis:

H2b

A firm’s geographic proximity to laser-related public research organizations is positively related to its innovative performance at later points in time.

On the one hand, the independent effects of network proximity and geographical proximity on innovation imply that both proximity dimensions influence innovation through completely unrelated mechanisms and transmission channels. In other words, the effects of positioning in the geographical and network space do not influence each other. On the other hand, it is plausible to assume that these two proximity dimensions can affect each other in various ways. Substitutional effects of network proximity and geographical proximity on innovation are based on the notion that a firm may compensate for its disadvantages in one proximity dimension through an advantageous position in another proximity dimension. Or to put it another way, firms located in remote geographical regions may compensate for their location disadvantages by fostering cooperation. In contrast, complementary effects of network proximity and geographical proximity on innovation enhancement are regarded as mutually reinforcing.

We argue that network proximity and geographical proximity are not independent but rather complementary in nature (cf. Fig. 12.1, Arrow 3a and 3b). Accordingly, being well-embedded in both proximity dimensions implies that firms can gain extra-additive innovation effects due to their advantageous positioning in both the geographical space and the network space. To exemplify this point, firms benefiting from local knowledge spillovers, in terms of improved accessibility to professionals and graduates on the local labor market, may have better qualified employees. This would allow these firms to generate greater value from their interorganizational partnerships. Against the backdrop of these considerations we formulate the following two hypotheses:

H3a

Combined proximity effects of a firm’s network proximity and its geographical proximity to other laser source manufactures are complementary in nature.

H3b

Combined proximity effects of a firm’s network proximity and its geographical proximity to other laser-related public research organizations are complementary in nature.

4 Data Sources, Methodological Issues and Variable Specification

We employed a unique panel dataset for the full population of 233 German laser source manufactures between 1990 and 2010 to answer the questions raised above. Four main data sources were used to conduct this study: patent data, industry data, geographical data and network data.

Patent dataFootnote 4 was used to measure innovative performance at the firm level. A lot has been written about the empirical challenges of measuring innovation processes. Despite the methodological constraints related to the use of patents to measure innovation performance (Patel and Pavitt 1995), patent indicators are commonly used for analyzing innovation processes (Jaffe 1989; Jaffe et al. 1993). The use of patent data as a proxy for firm innovativeness must be viewed critically for several reasons (cf. Fritsch and Slavtschev 2007, p. 204). Nonetheless, there are good reasons that advocate the use of patent data, especially in longitudinal settings (cf. Brenner and Broekel 2011, p. 13). In accordance with previous network studies (Ahuja 2000; Whittington et al. 2009; Stuart 1999), we used annual patent counts as a proxy for innovation output. Our database (cf. Sect. 6.1.2) includes patent applications as well as patents granted by the German Patent Office and by the European Patent Office. DEPATISnet (the German Patent and Trade Mark Office’s online database) and ESPACEnet (the European Patent Office database) were used to check results for integrity and consistency. We ended up with patent data-based innovation indicators: annual count of patent applications [pacnt] and annual count of patent grants [pgcnt].

Industry data Footnote 5 came from a proprietary dataset containing the entire population of German laser source manufacturers between 1969 and 2005 (Buenstorf 2007). Based on this initial dataset we used additional data sources to gather information about firm entries and exits after 2005. We ended up with an industry dataset encompassing 233 laser source manufacturers throughout the period under observation.

In addition, we used two methods to identify 145 universities and public research organizations that carried out laser-related activities. We started with the “expanding selection method” according to Doreian and Woodard (1992). Using this approach, we identified 138 laser-related public research organizations. This method, however, is limited insofar as it completely ignores non-cooperating laser-related PROs. Thus, we applied a second methodological approach to solve this problem and supplement our sample. Based on a bibliometric analysis we identified all German public research organizations which published laser papers, conference proceedings or articles in academic journals over the past two decades. Several raw data sources were tapped to conduct this analysis. We ended up with a final list of 145 laser-related PROs for the time spanning 1990 and 2010. Then, entry and exit dates were retrieved for all of the PROs in the dataset.

Industry data was used for two reasons. On the one hand, we had to specify the boundaries of the network. On the other hand, two basic firm-level control variables were recorded and included in our panel dataset: a linear firm age variable [firmage] as well as a squared firm age variable [firmage_sq].

Geographical data for all LSMs and PROs in the sample was reconstructed over the entire observation period from 1990 to 2010 (cf. Sect. 4.2). Data from Germany’s official company register (“Bundesanzeiger”) was used to reconstruct the firms’ current addresses and address changes for the entire observation period. We employed the ESRI ArcMap 10.0 Software package and a freely accessible geo-coding application to gather GPS coordinates (latitudes and longitudes) on an annual basis for each firm in the sample. Based on this data we set up two types of localized density measures (LD).Footnote 6 The shortest distance on a curved surface can be calculated by using a simple geographical distance formula (cf. Sect. 5.3.1, Eq. 5.12). Unlike Sorenson and Audia (2000, p. 435) we calculated the distances in kilometers by using the natural earth radius constant (c = 6,378 km) and we split the overall population into two sub-populations, LSMs and PROs. Inspired by Whittington et al. (2009) we calculated the shortest distance on a curved surface not only for each LSM to all other LSMs but also for each LSM to all PROs in our sample. After these preparatory steps, both geographical co-location measures were calculated using the localized density formula (cf. Sect. 5.3.1, Eq. 5.13). Thus, we ended up with two types of localized density measures for each firm in the sample, each of which was calculated on an annual basis i.e. [coloclsm] and [colocpro].

We also used a simple Herfindahl-Hirschman Index (Acar and Sankaran 1999) in order to set up two geographical concentration indices at the industry level. To do so, we proceeded as follows: First, we used the planning region scheme (“Raumordnungsregionen”), commonly used in Germany to classify of territorial units for statistical purposes. This divides the territory into 97 geographical areas. Next, we generated a count variable for both types of organizations – LSMs and PROs – that represented the number of organizations per planning region and year. Then we calculated for each planning region i (with i = 1…97) the relative proportion of organizations on an annual basis. Finally, two concentration indices were established by applying an HHI formula (cf. Sect. 5.3.2, Eq. 5.14). The indices moved in the direction of zero if the organizations under observation are equally dispersed throughout the geographical space; the HHI had comparably large values if some organizations were widely dispersed whereas others showed a pronounced tendency of crowding together. We ended up with two normalized indicators that allowed us to quantify the intensity of LSM crowding [hhi_lsm] and PRO crowding [hhi_pro] in the geographical space.

Network dataFootnote 7 was gathered from two official databases on publicly funded R&D collaboration projects. The first source was the Foerderkatalog database provided by the German federal government which encompasses information on a total of more than 110,000 completed or ongoing subsidized research projects and provides detailed information on the starting point, duration, funding and characteristic features of the project partners involved. In total, we were able to identify, for the entire population of 233 German laser source manufacturers, 416 R&D projects with up to 33 project partners from various industry sectors, non-profit research organizations and universities. The second raw data source was an extract from the CORDIS project database which includes a complete collection of R&D projects for all German companies which were funded by the European Commission between 1990 and 2010. In total, this database extract consisted of a project dataset with over 31,000 project files and an organization dataset with over 57,100 German organizations and roughly 194,000 international project partners. Based on this raw data, we identified 154 R&D projects with up to 53 project partners for the entire sample of German laser source manufacturers.

Finally, both cooperation data sources were used to construct interorganizational innovation networks and to calculate network indicators on a yearly basis. We decided in favor of the degree centrality concept in order to quantify a firm’s network position (cf. Sect. 5.2.1, Eq. 5.1). The degree centrality measure ranges from 0 to 1 and can be compared across networks of different sizes (Wasserman and Faust 1994, p. 179). We applied the data described above to calculate several network measures. Firstly, we calculated degree centrality measures on an annual basis for each actor in the sample [ctr_degree]. Then, two network level variables were calculated and included in the dataset to control for the structural network characteristics at the overall network level: overall network size [nw_size] and overall network density [nw_density]. Standard algorithms implemented in UCI-Net 6.2 were used to calculate the network measures (Borgatti et al. 2002).

Finally, we take a brief look at the variable description and basic summary statistics (cf. Table 12.1). We have a total of 2,645 firm-year observations in the time span between 1990 and 2010. The average number of observations per firm amounts to 11.35. Table 12.2 shows the correlation coefficients for all variables in our empirical models.

Table 12.1 Descriptive statistics – distinct and combined proximity effects
Full size table
Table 12.2 Correlation matrix – distinct and combined proximity effects
Full size table

5 Model Specification, Estimation Strategy and Findings

Patents take non-negative integer values. We estimated a count model in line with Ahuja (2000), Stuart (2000) and Whittington et al. (2009).Footnote 8 In doing so, we employed panel count data techniques (cf. Sect. 6.1.2) and adopted the following estimation strategy to test our hypotheses. First we estimated a Poisson model (Hausman et al. 1984) in order to obtain an initial idea of the relationship between distinct proximity effects, combined proximity effects and a firm’s patenting activity. We tested the significance of overdispersion using the procedure proposed by Cameron and Trivedi (1990) and rejected the null hypothesis of no overdispersion with a p-value of 0.000. As our endogenous variables exhibited strong overdispersion, we then turned to a negative binomial model specification. Like Whittington et al. (2009) we estimated all models using a negative binomial specification. In the next step we estimated both fixed effects and random effects models. We used the Standard Hausman Test (Hausman 1978) to decide which results to interpret. The basic idea was to test the null hypothesis that the unobserved effect is uncorrelated with the explanatory variables (Greene 2003, p. 301). If the null hypothesis cannot be rejected, both fixed effects estimates as well as random effects estimates are consistent and the model of choice is the random effects model due to its higher explanatory power. Under the alternative, random effects and fixed effects estimators diverge and it is argued that the latter model is the appropriate choice (Cameron and Trivedi 2009, p. 260). Finally, we ran several consistency checks to ensure robustness of the reported results. We set up several empirical settings with different time lags and we used data on patent grants as an additional innovation measure to ensure robustness of our results.

The presentation of our empirical findings was organized as follows. We specified a total of three empirical settings to test our hypotheses. The presentation and discussion of our estimation results was centered on a panel data count model for annual patent application counts with a time lag of 2 years, estimated by using both types of estimation techniques: fixed effects (cf. Table 12.3) and random effects (cf. Table 12.4). Next, we set up an additional empirical setting with a 3-year time lag structure to ensure the robustness of our results (cf. Table 12.5, fixed effects; Table 12.6, random effects). Finally, we employed an alternative innovation proxy to cross-check results and substantiate our initial findings. More precisely, we used data on annually granted patents, again with a time lag of 2 years (cf. Table 12.7; fixed effects: Table 12.8; random effects).

Table 12.3 Estimation results – distinct and combined proximity effects; negative-binomial panel data count model, patent applications, time lag (t − 2), fixed effects
Full size table
Table 12.4 Estimation results – distinct and combined proximity effects; negative-binomial panel data count model, patent applications, time lag (t − 2), random effects
Full size table
Table 12.5 First robustness check – distinct and combined proximity effects; innovation proxy: patent applications, time lag (t − 3), fixed effects
Full size table
Table 12.6 First robustness check – distinct and combined proximity effects; innovation proxy: patent applications, time lag (t − 3), random effects
Full size table
Table 12.7 Second robustness check – distinct and combined proximity effects; innovation proxy: patent grants, time lag (t − 2), fixed effects
Full size table
Table 12.8 Second robustness check – distinct and combined proximity effects; innovation proxy: patent grants, time lag (t − 2), random effects
Full size table

Each of the three empirical settings outlined above comprise eight models. In addition to a baseline model (BL Model), there were two models addressing distinct geographical proximity effects (Model I and Model II), one model addressing network proximity effects (Model III), and a fully specified model that incorporated both combined proximity effects (Model VII). In addition, we specified three additional models to check whether the results remained stable when estimating distinct proximity effects together (Model IV, Model V) and when estimating combined proximity effects separately (Model VI). The results were reported in accordance with Standard Hausman Test results and interpreted on the basis of the fully specified models (Model VII).

We start the discussion with Tables 12.3 and 12.4. The baseline model consists of firm level, network level and industry level variables. At the firm level we included two very basic variables i.e. “firm age” and “firm age squared” in the model. Network size and network density variables were incorporated to control for the structural network topology at the overall network level. Finally, two geographical concentration measures were considered to account for geographical concentration patterns of LSMs and PROs at the overall industry level. Three findings stand out:

Firstly, the firm-level variables have no significant effect on a firm’s patenting activity in terms of patent application counts in t − 2. Or to put it another way, a young firm’s patenting activity does not significantly differ from the patenting behavior of a mature firm. This result is robust over all empirical settings (cf. Tables 12.3, 12.4, 12.5, and 12.6) and fully confirmed by the patent grant model (cf. Tables 12.7 and 12.8).

Secondly, spatial concentration patterns of LSMs and PROs at the industry level turned out to have a significant impact on firm-level innovation outcomes. Estimation results from both fixed effects and random effects models (cf. Tables 12.3 and 12.4) indicate that a high level of geographical clustering of LSMs at the industry level is negatively related to a firm’s patenting activity at later points in time. As above, the patent grant model fully confirms this finding (cf. Tables 12.7 and 12.8). These results suggest that being part of an industry with a high level of geographical clustering among firms at the same stage of the value chain negatively impacts firm innovativeness at later points in time.

In contrast, geographical concentration of PROs at the industry level reveals exactly the opposite. Again, estimation results for both patent applications (cf. Tables 12.3 and 12.4) and patent grants (cf. Tables 12.7 and 12.8) are significant and robust over almost all model specifications. In other words, being part of an industry that is characterized by a high level of PRO clustering is positively related to a firm’s innovative performance measured by its patenting activities at later points in time.

Thirdly, it turns out that network size, measured by the number of participating laser source manufacturing firms, is negatively related to firm innovativeness at the 0.05 significance level (cf. Tables 12.3 and 12.4). This overall network size effect is fully consistent with the results reported by fixed effects and random effects models for grants with a t-3 time lag. Hence, firm-level innovativeness is negatively related to the increasing size of the industry’s innovation network. A look at the network density measure, however, reveals a somewhat ambiguous picture. On the one hand, none of the patent application models with a 2-year time lag structure show statistically significant estimates (cf. Tables 12.3 and 12.4). The same is true for the patent grant models with a comparable time lag structure (cf. Tables 12.7 and 12.8). On the other hand, patent application models with a 3-year time lag structure indicate a negative relationship between overall network density and firm level patenting performance at the 0.01 significance level.

In summary, our findings suggest that geographical concentration patterns at the industry level, as well as the structural network topology itself, turn out to affect the innovativeness of the firms involved. In addition, the spatial concentration patterns at the overall industry level, especially PRO clustering, seem to have an earlier impact on firm innovativeness than structural network characteristics, especially network density effects.Footnote 9

Now we look at geographical proximity effects. To start with, we examine a firm’s geographical co-location to other firms at the same stage of the industry value chain. Estimation results from patent application models with a 2-year time lag indicate a negative relatedness between geographical proximity and firm innovativeness at later points in time (cf. Tables 12.3 and 12.4). Coefficient estimates from the fixed effects model are highly significant at a 0.01 significance level and results from the random effects model confirm this relationship at a 0.05 significance level. Robustness checks do not contradict these findings but they also fail to provide additional empirical support. As a consequence, we have, in the very least, modest empirical evidence for a negative co-location effect of an LSM’s geographical proximity to other LSMs. Or to put it another way, being near to other LSMs is not beneficial per se; instead, geographical proximity can also hamper a firm’s innovativeness in terms of its patenting activity. This result supports the theoretical argument stated by Boschma (2005b, p. 70) according to which spatial lock-in effects and a lack of openness to the outside world can result in a situation in which negative agglomeration effects prevail.

Next, we place our attention on a firm’s geographical co-location to other laser-related research organizations. The fixed effects patent application model with a time lag of 2 years (cf. Table 12.3, Model VII) reports a positive and significant coefficient estimate for the co-location variable at the 0.1 significance level. The positive relationship between a firm’s patenting performance and its geographical closeness to other laser-related public research organizations implies the presence of purely regional knowledge spillovers. However, empirical evidence for this relationship is fairly weak. Similar to what was mentioned above, robustness checks do not contradict this finding but they also do not reveal additional empirical evidence. In a nutshell, the potential emergence or existence of pure knowledge spillover effects due to a firm’s geographical proximity to laser-related PROs is, in the very least, doubtful and the positive effect on firm innovativeness should not be overestimated.

Estimation results for network proximity effects reveal a much clearer picture. Similar to what was mentioned above, we start the discussion by focusing on the patent application models with a 2-year time lag (cf. Tables 12.3 and 12.4). Results provide strong empirical evidence for a positive and highly significant relationship between a firm’s nodal degree and its subsequent innovation output. Coefficient estimates are comparably high and turn out to be significant at the 0.05 level when using fixed effect estimation techniques. Random effects estimation techniques reveal even stronger empirical evidence at the 0.01 significance level. This result is fully confirmed by almost all model specifications, except for the fixed effects setting (cf. Table 12.5). The implications are straightforward: the more direct partners a firm has, the higher its innovative performance at later points in time. To recap, we found strong empirical support for a pronounced and highly significant relationship between distinct network proximity effects and firm innovativeness in the German laser industry.

Last but not least, we address the relationship between combined proximity effects and firm innovativeness. Coefficient estimates for combined proximity variables have to be interpreted as follows according to Whittington et al. (2009, p. 98): (a) insignificant estimates: the effects of geographical proximity and network proximity are independent; (b) positive significant estimates: geographical proximity effects and network proximity effects are complementary in nature; (c) negative significant estimates: geographical proximity effects can be substituted by network proximity effects and vice versa.

Estimation results for combined proximity effects of a firm’s geographical co-location to other LSMs and its network centrality provide sound empirical evidence for a complementary proximity effect. Coefficient estimates are positive and highly significant at the 0.01 level (cf. Tables 12.3 and 12.4). Patent application models with a 3-year time lag (cf. Tables 12.5 and 12.6) as well as the patent grant specification (cf. Tables 12.7 and 12.8) fully confirm this result at the 0.05 significance level. This finding reveals some interesting implications. The negative relatedness outlined above between a firm’s geographical proximity to other LSMs and its innovativeness only persists as long as these firms do not cooperate. Combined proximity effects – composed of a firm’s co-location to other LSMs and a firm’s network centrality measured by its nodal degree – are suggestive of complementary effects. Combined proximity effects of a firm’s co-location to other PROs and its network proximity turn out to be substitutional in nature. Again we have sound empirical support for this finding (cf. Tables 12.3, 12.4, 12.6, and 12.8).

6 Discussion and Implications

What do our empirical findings tell us in relation to our previously formulated hypotheses? Our first hypothesis (H1) suggests a positive relationship between a firm’s number of direct linkages and its patenting output at later points in time. We found strong empirical evidence for the relevance of distinct network proximity effects on the innovative performance of German laser source manufacturers. In other words, degree centrality, which measures the number of direct ties, turns out to be highly relevant for a firm’s innovative performance at later points in time. This finding is in line with the results reported by Whittington and colleagues (2009) for the US biotech industry. They found a highly positive relatedness between a firm’s eigenvector centrality and patenting performance.

With regard to distinct geographical proximity, Whittington et al. (2009) reported significant positive effects of co-location between US biotech firms and non-significant effects of geographical proximity to PROs. These results imply that the co-location to other biotech firms, rather than to research organizations, is what drives firm innovativeness. Our results for the German laser source industry paint quite a different picture. We have argued that geographical proximity – more precisely, co-location between a firm and other LSMs (H2a), or co-location between a firm and other PROs (H2b) – is positively related to firm innovativeness. Against our initial expectations, estimation results for co-location between laser source manufacturers turned out to have a significant negative effect on firm-level innovation outcomes. In other words, co-location between laser source manufactures reduces the innovative performance which leads to the rejection of Hypothesis H2a. This unexpected result is highly relevant for several reasons.

Firstly, it contradicts the empirical findings of Whittington et al. (2009). This indicates that distinct geographical proximity effects seem to be industry specific and follow a completely different logic in the German laser industry. Secondly, we found strong empirical evidence for the “dark side” of geographical proximity supporting the spatial lock-in arguments proposed by Boschma (2005b). Obviously, the positive knowledge spillover mechanisms do not unleash their effects. In contrast, the explanation for the negative co-location effect is straightforward. A firm’s geographical proximity to competitors may create an atmosphere characterized by reticence, aversion and preconceived notions. As a consequence, a firm can become inward looking (Boschma 2005b) and may choose secrecy strategies (Liebeskind 1996) instead of opening itself up to other LSMs. Especially non-cooperative firms may face a situation in which their knowledge base erodes and becomes outdated (Boschma 2005b). In summary, non-cooperating firms trapped in an inward looking geographical surrounding are likely to be adversely affected in their efforts to innovate.

Our empirical findings for a firm’s co-location to laser-related PROs are in line with our initial expectations. In contrast to Whittington and colleagues (2009) we found at least modest empirical support for Hypothesis H1b. Our result suggests that German laser source manufacturers may benefit from being located near laser-related public research organizations. Positive externalities in terms of scientific knowledge spillovers can be explained as follows. Firstly, it is commonly accepted that universities are an important source of new technological knowledge (Agrawal 2001, p. 285). It has been argued that especially the transfer of highly codified technological knowledge is facilitated by geographical proximity (Audretsch et al. 2004, p. 195). Others have put forward the argument that knowledge flows out of PROs, especially universities, much more readily than it does from privately owned firms (Jaffe et al. 1993; Owen-Smith and Powell 2004, p. 7). Secondly, firms located close to PROs may have a higher chance of attracting and hiring highly qualified graduates. It is plausible to assume that these employees bring new and creative ideas with them which affects a firm’s ability to innovate. Our results suggest that at least one of these two transmission channels seems to create what we would call “pure” scientific knowledge spillovers.

Finally, our findings on combined geographical proximity and network proximity confirm our expectations that combined proximity effects are not independent. However, the story turns out to be more complex than initially expected. To start with, we take a look at Hypothesis H3a. We have assumed that combined proximity effects of a firm’s network proximity and its geographical proximity to other laser source manufactures are complementary. Indeed, we found sound empirical support for Hypotheses H3a. The interpretation is straightforward. Being close to firms at the same stage in the value chain in an inward looking and non-cooperative environment seems to hinder non-cooperative actors in their efforts to innovate. In contrast, the complementary nature of combined proximity effects suggests that highly cooperative actors benefit from their geographical closeness to other LSMs. The results clearly show that being well-positioned in both types of proximity dimensions can lead to mutually reinforcing effects which, in turn, are positively related to a firm’s innovative performance at later points in time.

Surprisingly, combined proximity effects of a firm’s network proximity and its geographical proximity to other laser-related public research organizations turned out to be substitutional in nature. As a consequence we have to reject Hypothesis H3b. Nonetheless, this finding has some important implications. Firstly, scientific knowledge seems to be, at least to some extent, accessible via alternative transmission channels. In other words, firms can tap scientific knowledge by means of geographical spillovers or through cooperative linkages. Secondly, the two proximity dimensions seem to be exchangeable within certain limits. To illustrate this point, a peripheral firm located far away from laser-related research facilities can compensate for this geographical disadvantage by intensifying its cooperation activities. Following the same logic, firms located near laser-related PRO agglomerations are less dependent on having a high number of formal R&D partnerships. However, against the backdrop of the modest empirical support for the existence of “pure” scientific knowledge spillovers and the comparably strong empirical evidence for the existence of network effects, the latter implication should not be overstated.

To conclude with, the study provides us with some interesting insights and opens up at the same time several interesting research questions. Both, the limitations of this analysis (cf. Sect. 13.2) and fruitful avenues for further research on proximity and innovation (cf. Sect. 14.2) are the subject of discussion in the following chapters.