1 Introduction

The perception that innovative activities exhibit a strong regional component as well as insights into the supportive role of co-location and regional networking on innovation led to a shift in modern innovation policy towards the funding of clusters or regional networks (Eickelpasch and Fritsch 2005; Koschatzky 2000). The concepts of Marshallian agglomeration externalities, the cluster approach, and the regional systems of innovation approach (Porter 1990; Cooke et al. 1997; Baptista and Swann 1998; Galliano et al. 2015) provide the theoretical basis for modern regionally oriented innovation policy and stress the beneficial role of geographical as well as other types of proximity among private and public actors for knowledge production and exchange, innovation, and productivity. The main arguments in favor of co-location concern the ease of local actors to form collaborative linkages and their efficiency in terms of knowledge exchange. Moreover, this supportive effect of geographical proximity might be reinforced by the interplay with other types of proximities, including non-spatial proximities (Boschma 2005; ter Wal and Boschma 2009; Crescenzi 2014; Torre and Wallet 2014). While the ex-ante constituent effects of geographical proximity—along other proximity dimensions—on the formation of research alliances has been well examined (Hazir and Autant-Bernard 2011), little attention has been paid to the actual consequences of geographical co-location of alliance partners for subsequent performance (Crescenzi 2014). In addition, concrete conclusions and implications can hardly be drawn from the few studies on this topic, as they reveal a quite ambiguous picture and give rise to the question about unobserved factors that mediate the relationship between geographical proximity and alliance performance.

Despite this rare evidence and our vague understanding of the role of geographical proximity for the performance of research alliances, cluster policies focus almost exclusively on fostering regional linkages without considering other contextual factors. Therefore, the aims of this study are to analyze the role of geographical proximity within publicly funded clusters and thereby add insights as to whether R&D partners within cluster programs should be located more or less distant from each other. We approach this question by providing evidence on the relevance of geographical proximity for project performance. In addition, we elaborate on further contextual factors that mediate this relationship. So, we go beyond explaining why linkages have become apparent and analyze how these linkages perform and what explains their variation.

To do so, we use an original and unique dataset from a survey with project managers of collaborative R&D projects that were funded within the German “Leading-Edge Cluster Competition” (LECC). The program aims at funding joint R&D-projects and support regional networking in selected cluster regions in Germany. The clusters differ substantially in terms of geographic reach, so this sample should serve well for our purposes.

With the analysis of this rich data, we attempt to shed some light on the complex and multifaceted relationship between geographical proximity and the outcome of publicly funded R&D projects. Overall, we find that the relationship between geographical proximity and project success is by no means univocal but rather mediated by various technological, organizational and institutional aspects. Our findings suggest that the nature of knowledge determines the degree to which collaborators prefer or perceive it necessary to be co-located. The relevance of geographical proximity increases in contexts where knowledge is novel to the organization and the innovation endeavor is more radical, while this effect is less pronounced for projects in basic research. In addition, we find significant actor specific differences concerning the role of geographical distance for project satisfaction. Firms’ project satisfaction decreases significantly compared to that of research institutes with increasing distance to their collaboration partners. In line with existing studies (Gulati 1995; Gulati and Gargiulo 1999; Mowery et al. 1998; Ahuja 2000; Singh 2005) that underpin the importance of social proximity for successful cooperation, we observe that common project experience is a strong predictor of project satisfaction. Contrariwise, we cannot observe a substitutive relationship between geographical proximity and social proximity. With regard to final project results, we find that both geographical proximity and project satisfaction support the cross-fertilization effects between LECC projects and other projects.

The paper is organized as follows: in Sect. 2, we provide a general overview of the related literature and present major findings from prior studies on the relation between proximity and project performance. Building on that, we derive our research hypotheses in Sect. 3. Section 4 will introduce our basic methodology. Subsequently, the hypotheses are tested in Sect. 5. The final section concludes, discusses our results and highlights policy implications and potential avenues for further research.

2 Proximity and performance

The early 1990s witnessed an upsurge of studies that fathomed the factors behind the phenomenon of regionally clustered innovative activities and their uneven distribution across space (Jaffe et al. 1993; Audretsch and Feldman 1996; Porter 1990, 1998). The discovery of the beneficial effects of co-location of economic actors equally affected academia and policy makers in the development of new regional concepts and policy programs.

The economic benefits of co-location had already been described by Alfred Marshall (1890) in his study on the externalities that arise from agglomeration of specialized firms in industrial districts. According to him, the basic advantages that arise from the dense location of similar actors stem from the exploitation of regional synergy effects and opportunities for resource sharing. Co-located economic agents share access to specialized labor and supplier markets and benefit from the proximity to important customers and local markets. These ideas experienced a renaissance after Porter made the idea of agglomeration of companies and organizations from related industries popular and subsumed them under the concept of clusters (Porter 1998). In contrast to Marshall, Porter emphasized the vital role of increased cooperation and competitive pressure as explanatory factors for superior innovative and economic performance of spatially concentrated actors. The role of geographical proximity on networking, learning and innovation gained attention in later concepts, such as the regional innovation systems approach, which explicitly focused on explaining the regional production of knowledge and innovations (Cooke et al. 1997; Braczyk et al. 1998). The idea behind regional innovation systems is that a region’s innovation potential is strongly contingent on the interplay of several actors of knowledge production and usage, the linkages among them and the involved region-specific institutions. Another ongoing debate in a related stream of literature concerns the optimal regional industry structure and the exploitation of agglomeration externalities, that is, specialization versus diversification, in order to benefit from co-location (Frenken et al. 2007; van Oort et al. 2015; Galliano et al. 2015).

The main ingredient common to all these concepts that constitutes the importance of geographical proximity for innovative capabilities is the observation that local knowledge spillovers are spatially bounded (Jaffe et al. 1993; Mansfield and Lee 1996; Crescenzi 2014). Technological know-how is sticky since it has tacit components (Polanyi 1966; Cowan et al. 2000). Therefore, its diffusion requires continuous face-to-face interactions, especially in the early stages of an industry when newly generated knowledge is highly complex and specific and therefore hard to codify (Breschi and Lissoni 2001; Audretsch and Feldman 1996). In this regard, geographical proximity has been pointed out to be supportive for knowledge transfer by decreasing the costs of traveling, of obtaining face-to-face contacts and for partner search (Breschi and Lissoni 2001).

Building on that, more recent studies have challenged the view that solely being co-located to innovative actors is a sufficient precondition for the exploitation of the fruitful effects of local knowledge spillovers. These studies emphasize the crucial role of the embeddedness in regional networks to gain access to the prolific regional knowledge pool and to be connected to appropriate partners (Giuliani 2007). It is not only geographical proximity but also its interplay with other types of non-spatial proximities that drive the formation of these linkages and their efficiency in terms of knowledge exchange (Boschma 2005; ter Wal and Boschma 2009; Crescenzi 2014; Torre and Wallet 2014). More concretely, the probability to form research collaborations is positively affected by the regional proximity of actors certainly due to cost advantages, as well as through fostering the establishment of social proximity and cognitive proximity between potentially connected actors. Closely co-located actors are more prone to connect with each other as they have a higher awareness of each other and can more easily observe their respective capabilities and opportunities compared to those of more remote actors (Hazir and Autant-Bernard 2011). Over time, repeated interpersonal contacts and efficient knowledge exchange are responsible for the emergence of two non-spatial proximities, cognitive proximity between partners, on the one hand, and social proximity (trust) among them, on the other (Boschma 2005). The cognitive dimension manifests in a common knowledge base and appropriate absorptive capacities that are decisive to warrant common understanding and learning entailing efficient knowledge transfer and higher potentials to innovate (Cohen and Levinthal 1990; Nooteboom et al. 2007; Boschma 2005; Crescenzi 2014). Social proximity between collaboration partners serves as a control mechanism to reduce the risk of undesired knowledge flows and the danger of opportunistic behavior (Breschi and Lissoni 2003; Boschma 2005; Cantner and Graf 2011). However, in contrast to geographical proximity, the positive effects of these two main non-spatial proximities are not infinite: scholars have emphasized that the positive effects might revert once actors are too close. Especially too much cognitive proximity might also impede learning and innovation due to redundancy of knowledge (Nooteboom et al. 2007). The existence of an optimal level of proximity has been labeled as the proximity paradox (Broekel and Boschma 2012; Cassi and Plunket 2014) or goldilocks effect (Fitjar et al. 2016).

Empirical studies on this issue have emphasized various types of proximity as constituent factors for the formation of research collaboration (Katz 1994; Cantner and Meder 2007; Cassi et al. 2014; Balland et al. 2013; Singh 2005; Cassi and Plunket 2012; Broekel and Boschma 2012). While focusing on geographical proximity, Hazir and Autant-Bernard (2011) refer to this as the ex-ante effect of proximity on the collaboration decision, as actors expect higher returns from collaboration with proximate partners and therefore connect to them. Most work in this field studies either the collaboration propensity conditional on geographical proximity along with other proximity dimensions or explain how geographically distant partnerships are characterized. For instance, Cantner and Meder (2007) analyze German co-applications for patents from all topical areas to investigate whether geographical and cognitive proximity increase the likelihood to collaborate. They find that both proximity dimensions increase the probability to appear on a co-patent.

In addition, D’Este and Iammarino (2010) investigate the frequency of university-firm relationships in the UK and the spanned geographic distance therein. They explain the frequency of collaborations by the distance between partners and regress geographic distance on several partner characteristics. They observe that geographical proximity fosters the frequency of interaction between industry and academia in applied research (engineering disciplines) but not in basic research. Another interesting finding is that partners’ expertise might substitute for geographic distance. The benefits of expertise seem to outweigh the costs of collaboration over larger distances. However, they find that this effect decays when the distance becomes too large. Following this study, Garcia et al. (2013) ascertain whether similar patterns can be observed for industry-university linkages in Brazil. They also control for the quality of research output when explaining the geographic distance between research partners. In line with D’Este and Iammarino (2010), they find that partners are more prone to look outward for higher expertise, but again this relationship is rather curvilinear and only holds up to an intermediate level of distance.

While there is vast empirical evidence on the interplay between (geographical) proximities and the formation of cooperation, there is sparse evidence on the role of geographical proximity for project outcomes, i.e. the ex-post effects of proximity on collaboration. Geographical proximity is found to be positively correlated with firm performance in terms of economic and innovative outcomes (Oerlemans and Meeus 2005), with survival rates of SMEs (Staber 2001) or with continuation, respectively successful finish of research projects (Lhuillery and Pfister 2009). No proximity effects are observed on cooperation satisfaction or the longevity of industry-research partnerships (Mora-Valentin et al. 2004). However, these studies do not account for other types of proximity, such as social or cognitive ones. The study by Broekel and Boschma (2012) is the only one that considers multiple types of proximities. They find a somewhat paradoxical effect of geographical proximity on performance of collaborations in the Dutch aviation industry: while co-location seems to be a crucial driver of link formation, it does not affect subsequent innovative performance. This is what they call the proximity paradox. Cassi and Plunket (2014) confirm this finding, in that geographical proximity is crucial for link formation, but seems to be irrelevant for the outcomes of the collaborations as measured by forward citations to patents in the field of genomics. Fitjar et al. (2016) even find a reverse effect: geographical distant partners are more likely to be innovative and to introduce new products.

In sum, the ambiguous and sparse evidence on the role of geographical proximity for project success questions the necessity primarily to foster regional linkages in modern innovation policy. In light of recent findings on the danger of regional technological lock-in and the vital role of extra-regional linkages in their prevention, one may ask whether this policy perspective is too restricted and even outdated (Bathelt et al. 2004). In order to give an answer, it is necessary to analyze whether there are main confounding factors that condition the supportive role of geographical proximity on project performance. In this respect, the relevance of geographical proximity for the successful implementation of R&D projects seems to be still a relevant research issue (Hazir and Autant-Bernard 2011). Building on this, we investigate research relationships that have already been formed and analyze how project managers evaluate project performance contingent on their project partners’ geographical proximity as well as further confounding, mediating or moderating factors. So, our focus is not on explaining why certain linkages have been formed, but rather on how these linkages perform.

3 Hypotheses

The performance of R&D projects can be measured in many ways. A successful project is mostly understood as one that meets predefined goals. Innovation as the output of research projects and R&D productivity are the most obvious indicators for success (Brown and Svenson 1988). However, there is a lag between research conduct and the time until the innovative output becomes apparent in observable data (such as patents or products). The repetition or longevity of a research collaboration, as well as mutual knowledge transfer, can also be viewed as a project success (Hamel 1991; Lhuillery and Pfister 2009). Furthermore, the satisfaction of project managers with the project processes can be an early indicator for project success that is correlated with later innovative outputs (Mora-Valentin et al. 2004). We try to combine several output measures, namely, self-reported project satisfaction and subsequent innovative output, to analyze the role of geographical proximity for the success of research cooperation.

Three interrelated research questions constitute the framework for our analysis: Do cooperating actors perceive geographical proximity necessary in order to be successful? Does geographical proximity yield higher satisfaction in cooperative projects? Does geographical proximity indirectly (via project satisfaction) and directly increase success chances in terms of final project results?

We suggest that technological and organizational specificities of collaborative research projects govern the necessity for geographical proximity and that geographical proximity along with other factors increases project satisfaction and, in turn, the final project results. Our main assumption is that geographical proximity eases coordination and knowledge transfer within collaborations and increases the probability of success via decreasing the costs of personal contacts, leading to better communication and knowledge exchange conditions, and the creation of trust (Boschma 2005). However, the context of the research projects in terms of research orientation, exploration of new knowledge and the familiarity with research partners determines the need for continuous personal interaction and might render the argument for the advantages of geographical proximity obsolete.

3.1 Novelty and the relevance of geographical proximity

To be more specific, we assume that geographical proximity is especially relevant for project success, if the project focus is on exploring a radical novelty rather than a mere advancement of previous results. Therefore, when we consider novelty, we relate it to the exploration of new opportunities rather than the continuation or exploitation of prior generated knowledge (March 1991). Because knowledge in explorative research is highly complex and specific, it is hard to codify and to share without permanent personal communication and interaction. Since, as pointed out earlier, geographical proximity eases personal interaction and knowledge exchanges, we assume that explorative and novel research projects are more reliant on close geographical linkages and that geographical proximity becomes more relevant for project success with a higher degree of novelty of the project. Since novelty can be measured along several dimensions and we operationalize the concept of novelty accordingly.

As a first dimension, research endeavors can be characterized as novel when they are targeting radical novelties that significantly differ from prior research results. So for radicalness of the knowledge produced as the first dimension of novelty we suggest:

H1a

The relevance of geographical proximity for project success increases with the radicalness of the novelty.

A second dimension of novelty relates to the familiarity with the technology applied in the research project. Actors who are unfamiliar with the technology utilized in the project might require face-to-face interaction with their partners more frequently to increase learning. Therefore, we assume that respondents who work with a technology that is new to them value geographical proximity to their partners higher. Hence, we propose:

H1b

The relevance of geographical proximity for project success increases with the novelty of the applied technology within the project.

A third dimension of novelty concerns whether projects establish new research lines or represent a continuation of activities from prior projects. Contrary to radicalness and familiarity with the applied technology, geographical proximity might be less relevant for projects that perpetuate activities from prior related projects since certain routines and processes or institutions are already established. Therefore, we assume that:

H1c

The relevance of geographical proximity for project success decreases with the number of prior related projects.

3.2 The link between proximity and project satisfaction

Building on the above, we explore how geographical proximity is associated with project performance. As a first step, we consider project satisfaction as the intermediate outcome. Based on the above argument, we presume that geographically close partners tend to be more satisfied with their projects since communication and knowledge exchange is eased by geographical proximity.

H2a

Project satisfaction increases with geographical proximity between partners.

In the same vain, we expect that social proximity also directly effects cooperation satisfaction. We assume a positive relationship between social proximity and project satisfaction.

H2b

Project satisfaction is positively associated with social proximity (acquaintance of partners).

For the direct relation formulated in hypothesis 2a and b, we additionally consider other confounding factors and moderation effects. First, this relationship might be moderated by the perceived relevance of geographical proximity for project success. For respondents who deem co-location to their partners as irrelevant, the actual distance to their partners should not affect project satisfaction. Vice versa, we expect that actors, who evaluate geographical proximity to partners as essential while their project partners are remotely located, will be less satisfied with the project.

H2c

The link between project satisfaction and geographical distance is moderated by the relevance of geographical proximity for project success.

Another important factor driving project satisfaction is the acquaintance of partners through prior project experience, i.e. social proximity. Multiple studies have pushed forward arguments for a substitutive relationship between geographical proximity and social proximity. In our study, we assume that collaboration with distant partners is easier when they have previously worked together and could establish communication routines and trust. When partners are socially proximate they already exhibit a certain level of trust and are not reliant on frequent interaction and observation of the partner’s behavior. Therefore, we assume that already known partners are unaffected by geographic distance in their satisfaction with the overall collaboration.

H2d

The relation between geographical proximity and project satisfaction is moderated by social proximity between the partners.

3.3 Project performance

Finally, and based on the arguments that already led to hypothesis 2a, we expect that projects between geographically proximate partners are more successful than between distant partners. However, we assume that, in addition to a direct effect of proximity on success, there is also an indirect effect via increased project satisfaction. It seems plausible to expect that more satisfied researchers display higher productivity and better outcomes. Also, project satisfaction captures latent problems/hurdles within the projects, which might hinder the success of the project. Therefore, we assume:

H3a

Project outcome is positively correlated with geographical proximity.

and

H3b

Project outcome is positively correlated with project satisfaction.

4 Methodology

4.1 Data

The “Leading-Edge Cluster Competition” was a national, technology open cluster funding program launched by the German Federal Ministry for Education and Research (BMBF) in 2007, which aimed at funding collaborative R&D projects in selected cluster regions.

The cluster definition within the official program documentation was quite flexible and focused on research excellence and a common strategy of its members. Applicants were expected to possess strong expertise in their focal technology field and exceed a critical mass of internationally operating firms and reputable research institutes. Furthermore, they should hold a strong international market position, have a dynamic research focus, and exhibit potentials for increasing their profile and competitiveness (Rothgang et al. 2014).

Following recommendations of an expert jury, the Federal Ministry appointed 15 Clusters in three waves (2008, 2010, 2012) to be labeled as “Leading-Edge Clusters” and to receive funds amounting up to 40 million euros per cluster over a five year period. The funds were distributed to organizations in the winning clusters to conduct R&D projects in collaboration with cluster partnersFootnote 1 under a common leading cluster theme. Within the scope of the BMBF funded research project “Evaluation of the German LECC”, surveys were conducted between 2010 and 2013, with the beneficiaries of the ten selected clusters of the first two waves.Footnote 2 As part of these surveys, project managers were asked to evaluate processes and activities within the LECC-funded projects. To analyze cooperative processes, we consider only those respondents who participated in a joint research project (i.e. we excluded information from individual projects). These joint research projects can be understood as collaborations that are divided into subprojects concerned with specific aspects relevant to the common themes. The respondents, either employees of a research institute, a university or a firm, were the managers of these subprojects. Therefore, our dataset includes multiple responses within the same joint projects. This allows us to calculate relative distance measures within one joint project as well as to observe deviations in satisfaction levels of respondents within the same project. We exploit this unique dataset and complement information on project activities and outcomes with information on respondent’s geographical location. Even though the data were collected in consecutive interrogation rounds at different points in time, several items were not repeatedly reported and, therefore, our data are of cross-sectional nature.

4.2 Sample characteristics

In total, our sample comprises 475 consistent responses across all interrogation rounds by project managers of 101 joint projects. Table 1 provides an overview of the sample characteristics and the distribution of responses across clusters and actor types. The responses are almost equally distributed across actor types (last column). When annulling size differences and aggregate answers of large firms and SMEs, a dominance of firms prevails in the data set (two-thirds of the respondents are enterprises). The number of responses per cluster is very uneven (last row), ranging from a minimum of 23 to a maximum of 98 cases. This can partly be explained by the fact that the second wave clusters comprise a larger number of beneficiaries.

Table 1 Distribution of answers across clusters and actor type
Full size table

4.3 Variables

In order to analyze the interplay between geographical proximity, project satisfaction and project performance, we estimate three models with different dependent variables capturing three interrelated topics: the relevance of geographical proximity for project success, project satisfaction and project results. The description of the variables including selected summary statistics can be found in Table 6 in the “Appendix”.

4.3.1 Dependent variables

Perceived relevance of geographical proximity for project success (self-reported)

In the first model, we aim to explain under what circumstances project managers perceive geographical proximity between project partners to be relevant for the success of the research project (relev.geo.prox). The managers were asked to evaluate on a scale from 1 (“I strongly disagree”) to 5 (“I strongly agree”) whether geographical proximity is a central precondition for their project success.

Project satisfaction

One way to define the success of research projects is by measuring the satisfaction with the project processes by the actors. We assume that projects with satisfied project managers will result in higher projects outputs (in terms of innovation and cross-fertilization). Thus, project satisfaction represents an intermediate result of the research endeavor. For this reason, we explain self-reported satisfaction with various aspects of project implementation in the second model. More precisely, the project managers were asked to indicate their degree of satisfaction on a scale from 1 (low) to 5 (high), differentiating between partner types with respect to the following aspects: cooperation in general for cooperations with public research institutes or universities (sat.coop.univ) and with companies (sat.coop.firm), know-how transfer into their own organization (sat.knowhow.trans.uni, sat.knowhow.trans.firm), information transfer between the partners (sat.info.trans.uni, sat.info.trans.firm), and coordination among partners (sat.coord.univ, sat.coord.firm). Since some of the projects were still running while the survey was conducted, we assume that project satisfaction indicates prospective project success, which manifests in concrete project outputs at later stages. One advantage of measuring project success in this way is that successful projects can be identified earlier as compared to projects that are evaluated by means of patents or other concrete output measures.

Project results

Besides project satisfaction, we are also interested whether projects with more satisfied project managers also result in higher innovative performance. Project success is proxied by indicators for cross-fertilization effects (cross-fertilization) and for innovative performance (innovation). Concerning cross-fertilization, respondents were asked if project results could already be used as inputs for other projects within the organization (from 1—strong disagreement to 5—strong agreement). Innovation output is captured by a binary variable (0 = no, 1 = yes), which indicates whether novel and significantly improved products, services and processes have been launched by the respondent organization as a result of the project work.

We assume project satisfaction and project results to be strongly correlated. This could be simply due to the fact that both proxies might capture the same underlying factor and face the danger of being highly endogenous. The separate collection of information on project satisfaction and project output in different interrogation rounds—project satisfaction was asked in 2010/2011, the project results in 2013—reduces this risk. The data collection process, along with the project progress, is shown in Fig. 1.

Fig. 1
figure 1

Timeline of data collection and project progress

Full size image

4.3.2 Independent variables

Novelty

We assume that the degree of novelty of the project determines the relevance of partners’ geographical proximity for successful project accomplishment. To test this, we divide novelty into three sub aspects. First, we measure the degree of radicalness of the targeted innovation production (radical.inn). Respondents were asked to indicate on a scale from 1 (strongly disagree) to 5 (strongly agree) whether the project aimed at generating a radical novelty. Second, the familiarity with the knowledge applied in the project might shape the necessity for geographically close interaction. This aspect (new.tech) is measured by the respondents’ agreement to the item “The technology used in this project is completely new to us” (same 5 point Likert scale as before). Third, we also want to consider internal aspects of novelty by asking whether there have been prior projects to the current project (prev.proj). This variable is of a binary nature, indicating whether the current project continues work from previous projects (one) or not (zero). Of these three novelty aspects, only radical.inn and new.tech are correlated (See Results section and Tables 8, 9 in “Appendix”).

Geographic distance

To analyze the correlation between geographical distance and project satisfaction, we employ several objective distance measures. Based on the respondents’ locations, we compute the average distance in km to all partners (managers of subprojects) within the joint project (avg.dist). To also differentiate between projects that are clustered close in space as compared to projects with core-periphery structures, we calculate a relative distance measure that takes into account the distance of each respondent to a pre-defined geographical core or center of the joint project (dist.center). We identify those cities as project centers where the majority of partners are located. We assume that this center hosts the core activity of the project work due to the clustering of project partners. This measure is the de facto geographical proximity among project partners and will be coupled with the perceived importance of geographical proximity by the project managers in the analysis (relev.geo.prox).

Social proximity

Social proximity comprises the “social context” of the economic relations (Boschma 2005). Basically, it is understood as the level of trust between the partners and has been identified as a crucial factor in mediating the positive effects of geographical proximity on collaboration (Gulati and Gargiulo 1999; Breschi and Lissoni 2009). In research on networks or bilateral collaborations, social proximity is often operationalized in terms of the number of previous mutual collaborations or shared inventors (inventor mobility) (Gulati 1995; Breschi and Lissoni 2009) or some other proxy (prior indirect ties, reputation) that indicates trust or familiarity among project partners (due to prior experience) (Gulati and Gargiulo 1999). In our case, the respondents were asked about the share of partners in the current project with whom they have collaborated previously. Specifically, we measure social proximity (social.prox) on a scale from 0 to 3, where 0 indicates that none of the current partners is known, 1 that a minority are already known, 2 that the majority are already known, and 3 that all the partners in the current project are known from previous projects.

Controls

Apart from these main variables of interest, we include additional variables to control for factors that might influence our dependent variables. When talking about the importance of geographical proximity, one has to control for the general goal of the project as the perceived relevance of geographical proximity for project success might differ for projects that aim at establishing regional infrastructure (qualification programs, start-up climate) as compared to ones that explicitly aim at producing novel knowledge. Therefore, we differentiate between projects that aim primarily on the development of new product and process innovations (goal.prod.inno, goal.proc.inno), the support of start-ups (goal.startup) or the development of qualification and educational programs (goal.education). Since projects might pursue different goals simultaneously, each of these variables indicates the relative importance of each goal on a five point Likert scale. Closely related to that, it has been found that effects of geographical proximity on success are less pronounced for research endeavors that are basic rather than applied (Mansfield and Lee 1996). For this reason, the basic nature of each research project is proxied by the respondent’s binary indication regarding the potential of the project results to be implemented directly in new products/processes (applied results).

In the second model, further confounding factors that might drive the variance in perceived project satisfaction are project size as measured by the number of organizations collaborating in one project (project size), whether the respondent was the initiator of the project (project initiator), the general importance (project importance) of the project for the respondent in terms of network activities (i.e. to identify low engagement in joint projects due to deviating targets) and whether the project would have been dismissed without funding (project dismissal). Larger projects might receive lower satisfaction scores, since they require higher coordination, communication and transaction costs. Likewise, projects that are more important for the respondent organization might be evaluated better.

In explaining project results in terms of the generation of innovation and cross-fertilization effects, we also control for R&D-input measured by the number of highly skilled employees in the project (Human Capital Input—high skilled).

Moreover, in all three models we control for actor type (firm or public research institute) and cluster specific effects (to account for unobserved differences between clusters, such as technology, potential governance, overall network structure, etc.).

4.4 Estimation strategy

The relations that we aim to analyze are highly intertwined. Geographical proximity between the research collaborators as our main variable of interest is assumed to be a crucial determinant of project satisfaction, which in turn should affect later project outcomes. The relation between geographical proximity and project success is, in turn, mediated by other factors. For this reason, we follow a three stage estimation strategy in which the predicted values of the previous step are integrated as independent variables in the subsequent step. Since all our dependent variables represent a set of choices (response categories), we apply discrete choice models. In these models, one estimates the probability for a certain choice dependent on the characteristics of the individual respondent. For the n response categories, we estimate the following models:

Step I:

In the first model, we estimate the conditions (novelty, project goals) under which geographical proximity is seen as a necessity for the successful accomplishment of the project. Since the response categories are ordered along ascending agreement, we estimate an ordered logistic regression model. For each response category j from 1 to n − 1,Footnote 3 the ratio between the probability that the observed response is below category j and the probability that the response score is above the category j is calculated (left hand side) (Wooldridge 2002). In this step, the categories range from 1 to 5. To be more specific, we regress the response for the perceived relevance of geographical proximity relev.geo.proxi on the radicalness of the project (radical.inni), the familiarity with the technology applied (new.techi) and whether the current project is based on previous project activities (prev.proji). The last summation term represents further control variables

$$ \begin{aligned} ln\left[ {\frac{{P\left( {relev.geo.prox_{i} \le j} \right)}}{{1 - P\left( {relev.geo.prox_{i} \le j} \right)}}} \right] = & \beta_{0} - \left( {\beta_{1} radical.inn_{i} + \beta_{2} new.tech_{i} + \beta_{3} prev.proj_{i} + \mathop \sum \limits_{k = 1}^{n} \gamma_{k } c_{ik} } \right) \\ j = & 1,2, \ldots , n - 1 \\ \end{aligned} $$
Step II:

In the second model, we regress the satisfaction of project managers with certain aspects of the project work (sat.coopi) on their geographic distance (geo.disti) to partners within the joint project, the predicted values of the perceived relevance of geographical proximity from the first model (relev.geo.proxi), their social proximity (social.proxi) and other confounding factors. The variable sat.coopi represents the various aspects of project work that the respondents were asked to evaluate: general cooperation satisfaction, know-how transfer, information transfer, and coordination. The response options again range from 1 (low satisfaction) to 5 (high satisfaction). The variable geo.disti is measured in two ways: avg.dist and dist.center. Just as in step I, the last term represents the further control variables

$$ ln\left[ {\frac{{P\left( {sat.coop_{i} \le j} \right)}}{{1 - P\left( {sat.coop_{i} \le j} \right)}}} \right] = \beta_{0} - \left( {\beta_{1} geo.dist_{i} + \beta_{2} relev.geo.prox_{i} + \beta_{3} social.prox_{i} + \beta_{4} geo.dist_{i} *relev.geo.prox_{i} + \beta_{5} geo.dist_{i} *social.prox_{i} + \beta_{6} geo.dist_{i} *firm_{i} + \mathop \sum \limits_{k = 1}^{n} \gamma_{k } c_{ik} } \right) $$
Step III:

In step three, we finally want to elaborate whether projects with more satisfied participants exhibit a higher success probability. Therefore, we relate the predicted values of overall project satisfaction (sat.coopi) from the second step and the geographic distance to the partners (geo.disti) to the project results (resultsi) in terms of cross-fertilization effects (cross-fertilization) and innovative performance (innovation). The response categories j for cross-fertilization range from 1 to 5 and we apply an ordered logit model as well. Since the responses for innovation are binary (0-no innovation, 1-innovation), we employ a binary logistic regression model. As an analog to the first two steps, the last term represents further control variables

$$ ln\left[ {\frac{{P\left( {results_{i} \le j} \right)}}{{1 - P\left( {results_{i} \le j} \right)}}} \right] = \beta_{0} - \left( {\beta_{1} sat.coop_{i} + \beta_{2} geo.dist_{i} + \mathop \sum \limits_{k = 1}^{n} \gamma_{k } c_{ik} } \right) $$

5 Results

The projects analyzed in this study were all funded within the LECC, which had a strong focus on fostering regional linkages but also allowed collaboration among distant partners. As such, one might expect the project partners in our sample to be more closely located than in a representative sample of publicly funded research projects. In the following, we analyze the distribution of distances among project partners to get an overview of the “regionality” of linkages. However, one should keep in mind that co-location of research partners is also present without policy prescription. Garcia et al. (2013) find that the likelihood of interaction decreases sharply beyond a distance of about 100 km between the partners, which equals approximately one hour of travel time between collaborators. D’Este and Iammarino (2010) study the spatial profile of research partnerships between universities and business firms in the UK and observe an average distance of 268 km and a median distance of 148 km. In line with Mansfield and Lee (1996), they find partnerships for applied research to be in closer proximity than those for basic research, where research excellence seems relatively more important than geographic proximity.

As displayed in Fig. 2, the geographical distance between participants in the funded R&D projects in our sample conforms to prior findings, with the majority of project partners being located within (median of avg.dist) 107 km of each other. Beyond this threshold, the number of distant project members drops sharply. Additionally, the highly skewed distribution of the average distance (red line) and its concentration at rather small values (75% of observations are below 166 km) reflects the strong regional focus of the LECC. Far distant partners can inflate the average distance measure of the respondents to their partners. Therefore, we also calculated the distance of each respondent to the identified geographical core of the joint project (dist.center). The distribution of this measure is represented by the blue line in the same figure. The median distance of partners to the center equals 20.1 km, which also mirrors the selective support of regional linkages by the program. We can see, however, that the mean of avg.dist and the mean dist.center vary substantially across clusters. (See Table 7 in the “Appendix”.) BioRN, centered in Heidelberg, is the most compact cluster, with an average distance of 31.6 km while, on the other extreme, Solarvalley is the most dispersed cluster, with several linkages to distant partners and an average distance of 222.1 km.

Fig. 2
figure 2

Distribution of two distance measures (avg.dist and dist.center) with respective median (dashed line)

Full size image

Furthermore, Garcia et al. (2013) have also stressed, that geographical proximity particularly plays a role in industry-university collaborations. In their study, the majority of collaborations of this type were formed with partners that were less than 100 km away. When subdividing our sample by the type of collaboration (research-industry, inter-academia, interfirm) and comparing them in terms of their average distance between the partners in one project, reveals a somewhat deviating picture (Table 2). Collaborations that exhibit some degree of institutional proximity, i.e. between actors of the same type as shown in column 2 and 3, are more proximate to their partners. In contrast, collaborations between research institutes and firms are more likely to include more distant partners. However, the number of industry-research collaborations in our sample is far higher than for the other cases.

Table 2 Distance between project partners by collaboration type (absolute numbers of cases per collaboration category)
Full size table

These results are also mirrored in the self-reported evaluations of the project managers when asked whether geographical proximity is an important precondition for project success. Figure 3 shows the distribution of answers across agreement levels. In general, slightly more than half of the respondents (52%) confirm the need of being closely located to each other in order to be successful. However a non-negligible share of respondents is rather neutral or disagrees with this statement.

Fig. 3
figure 3

Necessity of geographical proximity for project success (own analysis based on LECC surveys)

Full size image

To elaborate further on what drives this heterogeneity concerning the perceived relevance of co-location, we regress the categorical responses on certain peculiarities of the research projects, such as the novelty of the project activities, the applicability of the results as well as the targeted goals and control for actor and cluster specific effects. Table 3 presents the estimation results of our first model. We start by including our main variables of interest and then stepwise introduce the dummies for actor type and cluster to check the robustness of our findings.

Table 3 Estimation results step 1: dependent variable is the perceived relevance of geographic proximity for project success (coefficients of ordinal logistic regression)
Full size table

Basically, we find mixed results for the hypothesized positive relation between novelty of the collaborative research endeavor and the perceived relevance of geographical proximity to warrant success. Concerning the extent of novelty production and the familiarity with the technology applied, we find partial support for our hypotheses 1a and 1b. The perceived relevance of geographical proximity for successful project implementation increases with the exploratory nature of the project activities in terms of producing more radical innovations (radical.inn) as well as applying new technologies (new.tech). But this effect disappears after controlling for specific project goals, type of respondent and cluster. Instead, we observe that, for members of projects focusing on the development of process innovations, geographical proximity is of minor importance. This relation holds in all model specifications.

With regard to the organizational aspect of novelty, we find that projects that were established as continuation of prior project activities are more likely to rate geographical proximity more important for project success. The coefficient of prev.proj does not show the expected sign and the result is not robust to the inclusion of actor and cluster dummy variables. Consequently, we find no support for hypothesis 1c.

Another interesting and strong finding is related to the applicability of project results (applied results). In line with prior studies on collaborations (D’Este and Iammarino 2010; Mansfield and Lee 1996), we can assert that members of projects with a focus on basic research are less reliant on geographical proximity to their partners as compared to actors in applied research projects. Probably the solving of more applied problems in the development of a ready to implement product and/or process requires more frequent interaction due to experimentations and observations processes which, in turn, are facilitated by geographical proximity of the collaborators.

Concerning actor and cluster heterogeneity, we find no significant differences in the respondent behavior between research institutes and firms. It is not very surprising that controlling for cluster membership reduces the variation explained by the technological and novelty aspects of the projects, since the cluster technologies differ in terms of novelty and radicalness. This can also be seen in the significant bilateral correlations of some of the cluster dummies with the new.tech and radical.inn variables (Tables 8, 9 in the “Appendix”).

After identifying the circumstances that guide the perceived relevance of co-location for project success, we are interested whether projects with local partners indeed outperform the ones with distant partners. Therefore, we use the predictions for perceived relevance of geographical proximity of step 1 (Model 3Footnote 4) along with the de-facto geographical proximity to explain project satisfaction as an intermediate outcome of the project work. Table 4 provides the estimated parameters for our second model.

Table 4 Estimation results step 2: dependent variables are project satisfaction in cooperation with research institutes and firms in general and along various dimensions (knowledge transfer information transfer; coordination) (coefficients of ordinal logistic regression)
Full size table

Overall, our estimates do not support the presumed direct relationship between the distance of collaboration partners and project satisfaction (hypothesis 2a). Neither the single average distance (avg.dist) nor the single distance to the project center (dist.center) turn out to play a significant role for most of the project aspects such as general cooperation satisfaction (satisfaction with cooperation), knowledge transfer (satisfaction with know-how transfer), information transfer (satisfaction with information transfer), as well as the coordination of project members (satisfaction with coordination). Distance only becomes relevant with regards to overall satisfaction in cooperation with firms. However, the coefficients do not show the expected signs. Checking for a threshold distance [both the mean and the sophisticated one hour travel distance (100 km)] by compiling the distance values to the binary information distant (one) or close (zero) did not yield different results. Although we ran both regressions for binary avg.dist and dist.center, the table only contains the model modification for dist.center.bin (column 4).

In contrast to geographical distance, the individual effect of social proximity (social.prox) on project success is significant for the overall cooperation satisfaction, with a more pronounced effect for collaborations with firms (sat.coop.univ and sat.coop.firm, column 1, 8, 9). This conforms to the ample evidence provided by a multitude of prior studies (Mora-Valentin et al. 2004; Breschi and Lissoni 2009). Projects that involve more familiar partners have higher chances to contain highly satisfied partners than projects where completely new partners interact. Consequently, our findings underpin our hypothesis 2b.

Finding only partial support for a direct link between distance and satisfaction is hardly surprising, since the relation between co-location of partners and project satisfaction is very complex and mediated by project peculiarities, as seen in our step one estimations. Thus, geographic proximity might affect satisfaction levels through multiple channels. First, the preference for being closely located might determine whether distant project members appoint high satisfaction scores or not. If respondents deem proximity to their partners as irrelevant, we would expect that the satisfaction scores do not decrease with geographic distance, and vice versa (hypothesis 2c). The inclusion of a joint effect of the perceived relevance of proximity and the actual distance of the partners on project satisfaction (geo.dist *predict.relev.geo.prox) do not support this hypothesis.

Second, the substitutive relationship between geographical proximity and social proximity has been stressed by multiple studies (Agrawal et al. 2008; Singh 2005; Breschi and Lissoni 2003; ter Wal and Boschma 2009; Boschma 2005). In our study, we assume that collaboration with distant partners is easier when they already have worked together in the past and have already established communication routines and trust and therefore do not evaluate the collaboration with distant partners worse than with close ones (hypothesis 2d). However, we only find weak evidence for an interaction effect between social proximity and geographic distance (geo.dist * social.prox) on cooperation satisfaction. Solely with respect to know how transfer and information transfer in collaboration with research institutes (columns 5 and 6) does a significant relation become apparent, showing that socially proximate partners are more likely to award higher scores to distant partners as compared to formerly unknown partners.

Third, if we scrutinize the influence of geographical distance on project satisfaction by actor groups, we find that the interaction effect of distance with the actor dummy (firm) is significant and negative. This means that, if the distance to the partners increases, companies are less satisfied with the collaboration. This effect is most pronounced for overall satisfaction levels (sat.coop.univ and sat.coop.firm, column 2–4, 8, 9) and independent of the type of cooperation partner (regardless of whether they should evaluate cooperation with research institutes or other firms). Moreover, the observed significant relation is robust to the modification of the distance measures (column 3 and 4). From this we can conclude that the respondent companies in our sample are more reliant on being close to their cooperation partners as compared to the research institutes.

Apart from these major findings, satisfaction levels over all project aspects are primarily driven by the main motif of the respondents to participate in the project (project importance). Project managers who rated the project to be of minor importance in their organization’s project portfolio are less satisfied with all cooperation aspects (except sat.knowhow.trans.firm and sat.coord.firm).

Also, respondents within larger projects in terms of number of collaboration partners (project size) are comparably less satisfied with the overall cooperation—at least with research institutes—than those in smaller projects. Other controls, such as initiating the project (project initiator) or the necessity of public funding (project dismissal), show no robust significant influence.

In the last step, we want to clarify whether geographical proximity has a direct effect on project results and whether project satisfaction is indeed an appropriate indication for later projects success in terms of producing valuable results.Footnote 5 Therefore, we regress two success variables on both geographic distance and the predicted cooperation satisfaction from step II (from Model 2 for RI and Model 8 for Firm Table 4) while controlling for the application of project results (applied results), human capital input, actor type and cluster differences. The first success variable relates to the cross-fertilization effects of the funded projects on other projects in the same organization (cross-fertilization). The second output variable captures whether project activities already resulted in novel products, services or processes (innovation). Since the two success variables are of different scale, we first estimate an ordered logit model for cross-fertilization and then a binary logistic regression model for innovation. The resulting parameter estimates can be found in Table 5.

Table 5 Estimation results model 3: dependent variables are cross-fertilization effects and innovation production (coefficients of ordinal and binary logistic regression)
Full size table

Overall, we find that the relation between project satisfaction and project outcome only holds for potential cross-fertilization effects and not for the probability of introducing an innovation. The estimations support hypothesis 3b in that projects that receive a higher rating on the satisfaction scale are more likely to report project results that can be applied in and fertilize other projects (cross-fertilization). This effect is robust against the inclusion of all control variables including actor type and cluster dummies. Likewise, and in accordance with hypothesis 3a, geographical proximity is also only relevant for projects in terms of the production of the cross-usage of results but not for innovative outcomes. Here, the average distance to partners hampers the appearance of cross-fertilization effects. This effect does not appear for the distance to the project center (dist.center). However, the responses also vary significantly between applied and basic research projects, between research institutes and companies, as well as between the individual clusters. Project managers who do research in rather applied areas are more likely to report cross-usage of project results in other projects. Furthermore, research institutes are more likely to report that project results add value to other projects as compared to companies. Since we can assume that the main activities of research institutes are within earlier phases of the innovation process, this result is not surprising. The projects within the LECC are required to be at a pre-market stage and effects for firms might show somewhat later. In contrast, projects with higher satisfaction ratings do not necessarily manifest in superior innovative performance (innovation). The reporting of innovative outcome is also quite heterogeneous across clusters and actor types. Managers of applied research projects are again more likely to report innovations, and research institutes are also more likely to introduce a novel product, service or process as a result of the project as compared to respondent firms. Consequently, we find only partial support for hypothesis 3b.

6 Conclusion

The purpose of this study was to add empirical evidence on the relationship between geographical proximity among collaboration partners and their performance. While the constituent role of geographical proximity for the formation of research alliances came to the fore on the innovation research agenda, the consequences for subsequent performance of joint research were underexplored.

Our findings also serve to reassess the justification for the strong focus of innovation policies on fostering regional networking. Such policies for local collaboration are based on the assumption that geographical proximity has beneficial effects on research collaborations. However, this assumption becomes debatable when confronted with the empirical evidence presented in this study.

To address this matter, we utilized a unique dataset based on a survey conducted with beneficiaries of the German “Leading-Edge Cluster Competition“, one of the main national cluster funding programs in recent years. Specifically, we analyzed the simultaneous effects of geographical along with technological aspects, social proximity, and actor heterogeneity on intermediate outcomes in terms of project satisfaction and final project outputs, such as cross-fertilization effects and the introduction of a product or process innovation.

We find that geographical proximity among collaboration partners is not a universal precondition for project success. In fact, the picture on how the individual respondents perceive the necessity of being closely located in order to be successful is quite heterogeneous. Our findings suggest that the nature of knowledge involved determines the degree to which collaborators are reliant on being closely located to each other. Geographical proximity between partners is deemed especially important in exploration contexts when projects aim at the production of radical novelty or experiment with new technologies. By contrast, but in line with prior findings, this effect is less pronounced for projects focusing on basic research (Mansfield and Lee 1996; D’Este and Iammarino 2010; Garcia et al. 2013). Furthermore, we find significant actor type specific differences concerning the role of geographical distance to the project partners for project satisfaction levels. Firms are less satisfied in collaborations with distant partners than research institutes. In line with prior studies, we observe that prior common work experience is a good predictor of project satisfaction levels. Again, by contrast, we only find little evidence for the often suggested substitutive relationship between geographical proximity and social proximity. However, we only looked at possible substitution effects of both types of proximity at any level of each of the proximity dimensions. A deeper analysis on the interrelations of proximity depending on the prevalent level of each dimension in the fashion of Fitjar et al. (2016) would be an interesting follow up to our study. They analyze in detail the levels at which geographical proximity can be substituted or complemented by other types of proximities.

With regard to final project results, we find that both geographical proximity and project satisfaction foster the cross-fertilization of other projects. Conforming to the findings of D’Este and Iammarino (2010), our results lead us to the conclusion that the link between geographical proximity and project success is rather complex and characterized by strong interdependencies with other contextual factors. Consequently, not only should the connection to co-located partners be supported, but it needs to be ensured that the “right” actors are chosen. Our results speak against a one-fits-all type of policy that merely strengthens regional linkages, since thereby other important contextual factors might be overlooked and the policy program will not yield the ex-ante expected effects (Crescenzi 2014; Koschatzky 2000). In consideration of the relative importance of other proximity dimensions and contextual factors, policy makers should include these factors into their decision. Regional proximity per se might not always be a warrant for successful research, as the benefits of the expertise might outweigh the cost for the collaboration with a distant partner (Garcia et al. 2013). Moreover, geographical proximity can be even detrimental when regional knowledge has been exploited and there is no access to external knowledge via ‘global pipelines’ (Bathelt et al. 2004). Extra-regional connections might serve as a source for new knowledge to overcome these critical situations.

Furthermore, policy has to find a balance between funding research among new, unknown partners for the reason of access to novel knowledge and exploiting the benefits of joint R&D among old acquaintances based on established trust and institutions. Therefore, the stage of the technology of projects and the prevailing network structures should be taken into consideration, as the growth of regions specialized on old technologies might be hindered by the mere focus on regional networking.

Besides these findings, the analysis in this paper faces some limitations and, accordingly, leaves room for further research endeavors. The main limitation of this study is the focus on publicly funded R&D projects without an appropriate control group. The extent of the generalizability of our results needs to be tested on the basis of comparable data from non-funded projects. Moreover, the static nature of the analysis does not allow for any conclusions on causal mechanisms or statements about the development of the necessity for proximity over time. More dynamic approaches are needed to understand further whether the mechanisms of proximity exhibit stability over time and how their interrelations change when collaborations end or persist.

Finally, we have to acknowledge that the performance measures are based on self-reported information by the project managers. When interpreting the results, one has to bear in mind that there is a risk of a positive bias in their replies. However, the responses show a reasonable amount of variation and substantial shares of low assessments of project performance, which indicates a rather realistic view of their own projects. In fact, it would be an interesting extension of our analysis and a further check for reliability to mirror our results to the actual performance (patents, publications etc.) of the funded R&D projects. However, due to the time lag between research project and observable outcome, it is still too early to get reliable secondary data on the performance of these projects.