In this study, we consider R&D collaboration networks as a mechanism that modifies knowledge flows in space, and hence as another source of interaction among regional innovation processes. Our objective is to understand the relative role of spatial neighbors and network neighbors on patenting performance of regions. We make use of data on R&D collaborations supported by the European Union’s Framework Programs (FP) and empirically investigate the patent activity of 213 European regions in the field of ICT during 2003-2009. Concerning the short length of the time frame we adopt a static modeling strategy and specify a spatial Durbin Model. As spatial neighbors intersect with network neighbors we decompose neighbor regions into three sets: spatially proximate regions that are not collaboration partners, spatially proximate regions that are collaboration partners, and distant collaboration partners. We express the weight matrix as a convex combination of these three sets and by means of gridding we compare how model fit changes as we move from a purely space based view to a purely network based view to express the dependence structure. The weight matrix that performs the best accords 60% weight to distant collaboration partners, 30% weight to proximate collaboration partners and 10% weight to proximate regions with whom there is no FP collaboration. This result reveals that the interaction (proximate and distant) among European regions within FP networks in the field of ICT is key for understanding dependence among their patenting performances.
Using a network approach that circumvents well-known challenges in estimating peer effects, we show that interactions with a firm’s geographic neighbors play a significant causal role in corporate investment behavior, and a modest role in financial policies and firm performance. Moreover, these geography network effects are almost entirely driven by propagation effects through product market and supply chain networks. We corroborate our findings in a quasi-experimental framework that allows for spillovers in treatment effects. Our findings help rationalize industrial clusters (e.g., Silicon Valley), as they illustrate that agglomeration economies are substantial and operate predominantly within industry boundaries.