https://orcid.org/0000-0002-4013-8421
![Sample our Built Environment journals, sign in here to start your access, Latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5926/TOC-Subject-Sample-2021-Built-Environment.jpg"})
ABSTRACT
Focusing on the French cluster policy, this study aims at evaluating how cluster policies influence the structure of local inventor networks. Based on a panel data of four periods and 94 NUTS-3 French regions, we implement various model specifications to make explicit processes that may drive the impact of the policy. The results suggest that spatial interdependences and region-specific trends should be considered to make correct causal inference about such place-based policies. Overall, the results do not provide clear evidence supporting that the French cluster policy has strengthened the cohesion, efficiency and resilience of local inventor networks.
ACKNOWLEDGEMENTS
The article is based on an earlier working paper (see https://halshs.archives-ouvertes.fr/halshs-02482565). This research work benefitted from the comments made by the anonymous referees, as well as from the GeoInno2020 conference participants and discussions with the AQR research group members.
DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.
Notes
1. The European Observatory for Clusters and Industrial Change (2019).
2. Falck et al. (Citation2019) implement a similar approach to evaluate a German place-based innovation policy programme. They rely on Ahlfeldt et al. (Citation2018) using individual (firm) and regional data and study the effects on direct and indirect beneficiaries (firms located in the same region as direct beneficiaries). Their focus, however, is on R&D expenditure rather than the network dimension.
3. The main French funding instrument of the competitiveness cluster policy.
4. Mar and Massard (Citation2021) found that joint treatment is the most effective way of increasing SMEs’ economic performances.
5. The choice of this three-year lag is justified by the average duration of R&D projects supported by clusters through the FUI, which is one of the main instruments for financing collaborative R&D projects in French clusters.
6. The inventor-disambiguated patent data we relied on – which was provided by the French patent office – only includes applicants and inventors with a French address. The dataset is freely accessible at: https://data.enseignementsup-recherche.gouv.fr/explore/dataset/fr-esr-brevets-france-inpi-oeb/.
7. Information on the R&D projects that have been granted public support is more scattered (and therefore not used here). The DGE centralizes information on FUI projects. The National Research Agency (ANR) records all ANR projects, some of which are directly linked to the cluster policy. Between 2007 and 2013, 1082 FUI projects and 909 ANR projects were supported by the French clusters. To a lesser extent, national and regional offices also hold some information on public subsidies to R&D cluster projects.
8. A 10% threshold is also used in Appendix A in the supplemental data online, resulting in 51 treated and 43 untreated firms. Our preferred specification relies, however, on the 25% threshold because there is a clear cut-off (with almost no NUTS-3 regions between 20% and 25%), and because more than 25% of the cluster participants belong to the first NUTS-3 region in more than 90% of the clusters.
9. Only one region recorded no cluster participant for a single year after policy implementation.
10. Acknowledging that the inclusion of such a ‘scale effects’ control variable may increase multicollinearity biases, we carried out a robustness test by dropping this variable from the regressions. This test confirmed that our results and conclusions of the study hold.
11. See https://competitivite.gouv.fr/.
12. Technological fields are based on the aggregation of International Patent Classification (IPC) codes proposed by the World Intellectual Property Organization (WIPO) (Schmoch, Citation2008). This classification aggregates IPC codes into 35 technological fields.
13. After performing a variance inflation factor (VIF) analysis, we found out that the regression analysis may be biased due to multicollinearity mainly generated by the variable ‘Number of nodes’ (which has a VIF > 10). In order to check the extent of this potential bias, we carried out a robustness test consisting of dropping this variable into the baseline models. These estimations highlight the absence of effect of cluster policies on the cohesion and efficiency of local inventor networks (as we also concluded in the paper). Given this conclusion, we decided to keep the ‘Number of nodes’ because this variable is a proxy for the size of resources (i.e., human capital) available in a given region, which is a key determinant of both innovation output and regional system innovation. We also implemented this test on the spatial models, and the same conclusion holds: there is a limited effect of cluster policies on the cohesion, efficiency and resilience of local inventor networks; the positive and significant effects are mainly found on network assortativity.