Delocalization Patterns in University–Industry Interaction: Evidence from the Sixth R&D Framework Programme

Abstract

Increasing university–industry interaction (UII) and university contribution to the local economy are compatible—conventional wisdom would say. However, similar to other university activities, interaction with industry may be limited due to a lack of absorptive capacity in local firms. The data of those participating in the European Union's (EU's) Sixth R&D Framework Programme (FP6) were used to obtain values for the number and, notably, the budgets of UII projects at the regional level for the EU27. Two types of interactions were considered: inside and outside the region. Our analysis indicates that universities from regions whose firms have low absorptive capacity participate more often in FP6 projects with firms outside the region. Our results highlight the value of policies that facilitate firm R&D to enhance collaboration with regional universities.

Acknowledgements

This research has taken place under the framework of ERAWATCH, a joint initiative of the European Commission's Directorate General for Research and the Joint Research Centre-Institute for Prospective Technological Studies (IPTS). We extend our gratitude to René van Bavel for and Xabier Goenaga for their support and to Dilek Çetin for her patient advice regarding econometric issues. Laura de Dominicis, Gérard Carat and Peter Voigt, all colleagues at the IPTS provided invaluable advice on the interpretation of regional UII patterns. Sean Kask and Francesco Rentocchini were kind enough to help with technical advice during the revision process. We would also like to thank the participants at the EUNIP 2009 Workshop (especially our discussants Johan Willner and Javier García-Estévez), the 2010 BRICK-DIME-STRIKE Workshop (especially our discussant Aldo Geuna) and the 2010 Triple Helix International Conference the for their constructive comments and suggestions. As usual, any possible errors and omissions are the responsibility of the authors.

Notes

1 The management literature has nevertheless focused on how firms can increase the success of interaction with universities, with recommendations that are perfectly compatible with the concept of raising absorptive capacity, such as the creation of hybrid organizations (Andrisano et al., 2006; Rohrbeck & Arnold, 2006) or managing the different available instruments for interaction (Romero, 2007).

2 These authors sometimes refer to the “use of knowledge created in universities” as a proxy for UII with the same meaning that we give it here: partnerships, not spillovers. However, it is not clear from the nature of their dependent variable whether this excludes spillovers.

3 Actually, when talking about the localization of knowledge spillovers, Agrawal (2001) conducted a bibliographic review, according to which such localization occurs and indirectly implies that the degree of localization varies across regions. In the concept of “regional absorptive capacity”, the author found an interesting opportunity for future research to explain this variation. Still, the relationship is between R&D spillovers—not UII and absorptive capacity of firms in the region.

4 The relation between absorptive capacity and industry interaction with other partners in general (not only with universities) has also been explored. Boschma and ter Wai (2007), using a sample from one Italian Industrial District and 33 firms, stated that there was no influence of absorptive capacity on local networking, but a positive one on non-local networking. For Belussi et al. (2008), number of patents (which is related to absorptive capacity) was related to research collaboration with partners abroad and not with regional or national partners, based on one Italian region and 78 life science firms.

5 Sometimes, absorptive capacity of firms in the region may increase if multinationals locate their subsidiaries next to relevant university research (Abramovsky et al., 2007).

6 For projects with more than one university, we duplicated observations, attributing a distinct nationality in each duplicate project. We then added as many duplicate project observations as the discrete nationalities of participating universities.

7 We have also used the percentage of less refined variables, the number and value of projects with UII (instead of number and value of UIIs), and the results do not change (available upon request).

8 For some countries, there was a mismatch between the NUTS code reported in the database and contemporary NUTS classifications used for the same regions by Eurostat. This is probably due to comprehensive national coding revisions (as, for example, in the case of Bulgaria, Denmark, Romania, Sweden and Slovenia) and to smaller ad hoc changes (as, for example, in the German regions DEE2 and DEE3 which have been merged into DEE0).

9 The transformation introduces a complication, as the logarithm of zero is undefined. A common solution is to add a small positive number to all observations before taking logarithms. We added 0.0001, so that when INTERREG_C or INTERREG_M is equal to 0, ln INTERREG_C and ln INTERREG_M are equal to −9.21.

10 It is difficult to measure the absorptive capacity of firms in the region and, to the best of our knowledge, few studies have put forward tangible results. Roper and Love (2006) tested how the labour market characteristics of European regions shape regional absorptive capacity. To that end, they added to the usual innovation production function some explanatory variables of interaction effects between the labour market indicators and public and private technology investment. However, for the authors, these interaction effects capture “regional absorptive capacity” effects rather than “regional absorptive capacity” per se. All in all, there is no generally accepted method for quantifying regional absorptive capacity. For this reason, based on new economic geography, we prefer to talk about the absorptive capacity of firms in the region, and in our empirical analysis, we opt for a measure that is faithful to the origins of the concept within the firm while maintaining the regional focus: regional BERD.

11 As these variables take integer, mostly low, values, we retain them in their original form.

12 Our near-complete sample of UII across EU regions is somewhat marred by missing year-region observations for HERD, BERD and GDP. To counter this issue, we have followed the common convention of filling single missing year-region observations with the average value of the preceding and following years. This made the recuperation of a small number of observations for BERD (150) and HERD (49), but not for GDP, which had no missing values meeting the above criterion, possible.

13 Conditional fixed-effects models are not common practice as there does not exist a sufficient statistic allowing the fixed effects to be conditioned out of the likelihood. Unconditional fixed effects with dummy variables for the members of the cross-section produce biased estimates (available at http://www.stata.com/help.cgi?xttobit, accessed 19 January 2010).

14 We acknowledge here the convenience offered by the table-producing tool developed by Wada (2009).

15 Certainly, the marginal effects of the absolute measures in Table 5 are the same (column 1) or slightly smaller (column 2) than in Table 6. One may wonder how it is possible that variables in relative measures behave differently than those in absolute measures. One reason is that the ratio of the marginal effects in Table 6 over those in Table 5 is equal to 1, so the difference is not remarkable. More technical reasons are that (a) the Tobit estimation is not linear; (b) the absolute measures are nevertheless taken in logs, which introduce additional nonlinearity; and (c) one has to take into account the effect of the rest of the parameters in the model.

16 This may not happen necessarily, as the exclusion of a large number of projects means that point estimates are produced from an overall lower number of observations—potentially exacerbating the impact of “noise”.