European infrastructure networks and regional innovation in science-based technologies

Abstract

We analyse the innovative activity of European regions in the fields of biotechnology and semiconductor technology. We explain regional patenting levels from publication levels within each region and nearby regions to account for local knowledge spillovers. We extend this approach by including connectivity measures for each region in order to indicate their position in the pan-European networks of Internet backbone providers, airline routes and global banks. We hypothesise that a region's position in all these networks contributes to its innovation capability as these networks provide high-quality and relative cheap access to digital information (Internet), fellow researchers (airlines) and financial resources (banks). The results show that connectivity indeed supports a region's patenting level in science-based technologies. In particular, we found that connectivity through the Internet backbone and through global banks enhance innovative activity, while airline connectivity does not. A second conclusion holds that while local knowledge spillovers are found to be very strong for patenting in biotechnology, this effect is found to be absent in semiconductor patenting. This result indicates that the importance of geographical proximity in generating knowledge spillovers is highly technology-specific.

Keywords:

• global cities
• world cities
• infrastructure networks
• knowledge spillovers
• NUTS3

Acknowledgements

This research has been sponsored by the Netherlands Institute for Spatial Research (RPB) and The Netherlands Organisation for Scientific Research (NWO) under the NVN programme. We thank Hans van Amsterdam, Stephaan Declerck, Roderik Ponds, Jan Schuur and Joep van Vliet from the RPB for their assistance and advice. We also thank Jonathan Rutherford for providing the data on Internet backbone providers and Guillaume Burghouwt for providing the data extracted from OAG Airline Schedules, as well as Ron Boschma, Stefano Breschi, Raymond Florax, Alfonso Gambardella, Harry Garretsen, Paola Giuri, Mario Maggioni, Frank Neffke, Erika Uberti and two anonymous reviewers for their useful suggestions at different stages of our research. All errors remain ours.

Notes

Though recently a large-scale questionnaire has been carried out in the European Union, which is likely to yield important insights (Giuri et al. 2007).

Previously known as the Science Citation Index.

Publications from Applied Physics are even more often cited than publications from electrical and electronical engineering (Verbeek et al. 2002), yet Applied Physics is rather broad to take into account as a discipline.

The address information contained in publication data refers to the address of the organisation where the researcher is working. By contrast, in patent data the address information we make use of refers to the home addresses of the researchers involved.

Correlations among variables treated with different counting procedures always exceeds 0.99.

We were not able to locate the addresses within the greater urban areas of London and Manchester and as a result consolidated them into two new regions. Furthermore, we excluded some islands due to their remote locations and disproportional great geographical distances to other regions. These islands are: Guadeloupe Las Palmas (ES), Santa Cruz de Tenerife (ES), Guadeloupe (FR), Martinique (FR), Guyane (FR), Réunion (FR), Região Autónoma dos Acores (PT) and Região Autónoma da Madeira (PT). The outcome is a total number of 1316 NUTS3 regions instead of 1329.

Yet, for Germany it is known that labour market areas are a combination of the NUTS2 and NUTS3-level.

We ran the regressions models also for zero and 20% depreciation rate. As we found little differences between the results, we decided to maintain the 10% depreciation rate suggested by Peri (2005) as the preferred depreciation rate.

Using data on 40 Dutch NUTS3 regions made available by the Netherlands Institute for Spatial Research (Ponds, van Oort, and Frenken 2010) we found that the correlation between publications and public R&D is 0.925 for biotechnology and 0.846 for semiconductors, while correlation with private R&D is 0.595 for biotechnology and 0.658 for semiconductors.

Any other measure of centrality that takes into account the indirect connectivity between cities (like closeness or eigenvector centrality), would assume that the people involved in relationships between city A and B are also involved in relations between B and C. Otherwise, spillovers could not travel indirectly between A and C via B. We feel that this assumption is too strong, since we look at inter-city networks, not at inter-organisational or inter-personal networks.

The extraction from OAG data set refers to the route-schedule among European airports during the third week of September in 2002. The year considered refers to the year preceding both patent and publications outcomes, while the third week of September, among air transportation experts, is known to be less effected by seasonal flight schedules, including winter and summer holidays.

In order to test for spatial autocorrelation we ran Moran's I test. For both dependent variables, the statistics show positive and significant values. This motivates the inclusion of a variable capturing the local spillovers from nearby regions.

Tests of overdispersion and Vuong tests show positive and significant values.

High correlations are found for the dummy variables (5–7) and their respective connectivity measures (8–10). This does not pose a problem in our estimation because these variables are two alternative indicators that are not simultaneously included in any of the regressions.