Knowledge networks in joint research projects, innovation and economic growth across European regions

In particular, several empirical studies have examined the nature and the determinants of scientific cooperations among firms (Hagerdoon 2000; 2002; Miotti and Sachwald 2003; Caloghirou et al. 2006) or between firms and universities (Geuna 1998; Hayashi 2003; Laursen and Salter 2004; Arundel and Geuna 2004; Fontana et al. 2006; D’Este et al. 2011).

For a discussion of the game theoretic literature on the private incentives to cooperate in R&D, see Cassiman and Veugelers (2002).

Ertur and Koch (2006); Artelaris et al. (2010); Chapman and Meliciani (2018);

Rodríguez‐Pose (1999); Chapman and Meliciani (2012, 2017); Meliciani (2016).

Socio-economic clusters are based on Rodríguez-Pose (1998), who classifies EU-12 regions into four groups: (1) capital and urban areas, (2) regions affected by industrial decline, (3) intermediate regions and (4) peripheral regions, and on Chapman and Meliciani (2012), who extend this classification to the countries that joined the EU later (EU-27).

Di Cagno et al. (2014), using data from a panel of European countries participating in FP over the 1994–2005 period, find participation in EU funded projects helps laggard countries to reduce a part of their economic gap with more advanced countries (Macdissi and Negassi 2002 for France; Medda et al. 2006 for Italy).

See both the literature on endogenous economic growth, e.g. Aghion and Howitt (1992); Grossman and Helpman (1994) and evolutionary models, e.g. Nelson and Winter (1982); Fagerberg (1994).

EU ODP: https://​data.​europa.​eu/​euodp/​en/​home.

In terms of funding allocated, the most important issues are health, energy, transport, environment and, in the most recent FPs, climate change.

The NUTS classification subdivides the economic territory of the Member States. It ascribes to each territorial unit (NUTS) a specific code and name. The NUTS classification is hierarchical. It subdivides each Member State into NUTS level 1 territorial units, each of which is subdivided into NUTS level 2 territorial units, which in turn are subdivided into NUTS level 3 territorial units (source: REGULATION (EC) No 1059/2003).

The most used indexes of centrality in the literature to analyse networks are degree, strength, closeness, betweenness and eigenvector (Barrat 2004 for a review). Degree simply denotes the number of neighbours of the node (region in our analysis). Strength is the extension for the weighted networks. Closeness centrality represents the closeness of a given node with every other node of the network. Betweenness centrality measures the ability of the node occupying a critical gate-keeping position to act as an intermediary. Betweenness centrality of a given node is based on the number of shortest paths passing through the node. Eigenvector centrality is used to measure the influence of a node in the network. It assigns a relative index value to all nodes in the network based on the concept that connections with high indexed nodes contribute more to the score of the node than the connections with low indexed nodes (Saxena and Iyengar 2020). Given the strong positive correlation found for the centrality indices, we decided to use only one and the choice fell on strength both for its simplicity and because it takes into account both the connectivity (the number of connection of a node) as well as the intensity of the ties, measured by the weights of the edges.

We observe, as in other real-world networks, a negative correlation between the strength centrality and the local clustering coefficient (Table 8 in the appendix). As pointed out by Opsahl and Panzarasa (2009), a node with more neighbours is likely to be embedded in relatively fewer closed triplets and therefore to have a smaller local clustering than a node connected to fewer neighbours.

Time span of the analysis and the number of regions (nuts2) is influenced by the availability of EUROSTAT data.

Country dummies are included also in the specifications which include class dummies. In fact, some countries have levels of innovation that are higher than those experienced in other countries irrespective of the socio-economic group the region belongs too.

The GDP of the NUTS2 region variables has been deflated using the corresponding national GDP deflator (2010 = 100).

For details on how to implement this procedure, see Hole (2006).

We thank an anonymous referee for pointing this out.

In Appendix 1, Tables 9 and 10 report the estimates of Tables 4 and 5 without considering the intermediate group dummy and EU15 countries’ dummy, which are taken as a base level. This allows to statically test the respective coefficient differences of the groups.