Spatial Interaction Modelling of Cross-Region R&D Collaborations Empirical Evidence from the EU Framework Programmes

The focus of this study is on cross-region R&D collaboration networks in the EU Framework Programmes (FP’s). In contrast to most other empirical studies in this field, we shift attention to regions as units of analysis, i.e. we use aggregated data on research collaborations at the regional level. The objective is to identify determinants of cross-region collaboration patterns. In particular, we are interested whether geographical and technological distances are significant determinants of interregional cooperation. Further we investigate differences between intra-industry networks and public research networks (i.e. universities and research organisations). The European coverage is achieved by using data on 255 NUTS-2 regions of the 25 pre-2007 EU member-states, as well as Norway and Switzerland. We adopt a Poisson spatial interaction modelling perspective to analyse these questions. The dependent variable is the intensity of collaborative interactions between two regions, the independent variables are region-specific characteristics and variables that measure the separation between two regions such as geographical or technological distance. The results provide striking evidence that geographical factors are important determinants of cross-region collaboration intensities, but the effect of technological proximity is stronger. R&D collaborations occur most often between organisations that are located close to each other in technological space. Moreover geographical distance effects are significantly higher for intra-industry than for public research collaborations.

💡 Research Summary

The paper investigates the determinants of cross‑regional research‑and‑development (R&D) collaborations within the European Union’s Framework Programme (FP) system, focusing on regions rather than individual firms or projects. Using data from the pre‑2007 FP5 and FP6 programmes, the authors construct a collaboration matrix for 255 NUTS‑2 regions covering the 25 EU member states at the time, plus Norway and Switzerland. Each cell of the matrix records the number of joint projects between two regions, providing a count‑type dependent variable.

To model these counts the authors adopt a Poisson spatial interaction (gravity‑type) model. The expected collaboration intensity λij between regions i and j is expressed as a log‑linear function of (i) the R&D “mass” of each region (patent applications, research staff, etc.), (ii) separation variables capturing geographic distance, technological distance (derived from patent classification similarity), and (iii) control variables for language, cultural, and institutional differences. Separate specifications are estimated for the full sample and for two sub‑samples: intra‑industry collaborations and public‑research collaborations (universities and research institutes).

Key findings are: (1) Geographic distance negatively affects collaboration intensity, but the magnitude differs by sector – the distance elasticity is roughly 1.8 times larger for industry links than for public‑research links. (2) Technological distance exerts a far stronger deterrent effect: a one‑standard‑deviation reduction in technological distance raises expected collaborations by about 35 % across the whole sample, more than double the impact of physical distance. (3) Language and institutional commonality matter for industry collaborations but are less significant for public‑research networks, suggesting that academic partners can more easily overcome such barriers. (4) Model diagnostics (over‑dispersion tests, negative‑binomial and zero‑inflated alternatives) confirm that the Poisson specification provides the best fit for the data.

The authors conclude that policy aimed solely at improving physical infrastructure will have limited success in fostering cross‑regional R&D cooperation. Instead, initiatives that enhance technological proximity—such as supporting sector‑specific knowledge clusters, facilitating patent‑based matchmaking, and promoting joint research facilities—are likely to be more effective. For industry, reducing geographic frictions through regional collaboration platforms is important, whereas for public‑research entities the emphasis should be on aligning complementary technological capabilities. The paper also suggests future work incorporating temporal dynamics and network centrality measures to refine policy simulations.