ABSTRACT
Given the sheer number of cross-border regions (CBRs) within the EU, their socio-economic importance has been recognized both by policy-makers and academics. Recently, the novel concept of cross-border regional innovation system has been introduced to guide the assessment of integration processes in CBRs. A central focus of this concept is set on analyzing the impact of varying types of proximity (cognitive, technological, etc.) on cross-border cooperation. Previous empirical applications of the concept have, however, relied on individual case studies and varying methodologies, thus complicating and constraining comparisons between different CBRs. Here a broader view is provided by comparing 28 Northern European CBRs. The empirical analysis utilizes economic, science and technology (S&T) statistics to construct proximity indicators and measures S&T integration in the context of cross-border cooperation. The findings from descriptive statistics and exploratory count data regressions show that technological and cognitive proximity measures are significantly related to S&T cooperation activities (cross-border co-publications and co-patents). Taken together, our empirical approach underlines the feasibility of utilizing the proximity approach for comparative analyses in CBR settings.
Acknowledgements
We thank Denise O’Brien from the Census Enquiries Section of the Central Statistics Office Ireland for her help with the data and the anonymous reviewers for their helpful comments. An earlier draft of this paper was presented at the 18th Uddevalla Symposium in Sønderborg, Denmark, 11–13.6.2015.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Teemu Makkonen http://orcid.org/0000-0002-1065-1806
Timo Mitze http://orcid.org/0000-0003-3799-5200
Notes
1 Additionally, for methodological and data related reasons: (1) Maas–Rhein is treated as three pair-wise (Belgian–German, Belgian–Dutch and Dutch–German) CBRs, (2) in the case of Tornio River Valley, Livonia, Country of Lakes and Neman, the parts of the CBRs belonging to non-EU counties (Belorussia, Norway and Russia) are excluded and (3) in the case of Neiße–Nisa–Nysa and SaarLorLux, the Czech and French parts, respectively, of the CBRs are delineated to belong outside of ‘Northern Europe’.
2 The EU classification for NACE codes (NACE Rev. 2) was updated in 2008 (Eurostat, Citation2008), which imposes a break in the time-series data on regional employment figures in the EU between the contemporary data and the older data that has been compiled on the basis of earlier classification procedures.
3 Unfortunately, for patent applications (due to the RegPAT database utilized) we are restricted to using NUTS-3 categories, which in some cases do not fit the delineated CBRs. Thus, in some cases the results related to co-patents might overestimate the ‘real’ depth of cooperation. However, NUTS-3 categories are still a preferred solution in the case of CBRs instead of using the commonly applied, but significantly larger, NUTS-2 regions.
4 Due to the limitations inherent with using the inventor field in identifying cross-border patents, the information in the applicant field to account for ‘true’ collaborations is utilized (Bergek & Bruzelius, Citation2010). These figures are significantly lower than in the case of using inventor level data. However, the inventor level data arguably gives more information concerning labour market mobility than actual cross-border S&T cooperation. For example, in the case of Øresund a significant number of Danes have moved to the Swedish side of the border due to housing market price differentials, but continue to work in and, thus, commute to Denmark (Makkonen, Citation2016).
5 For the regression analysis we do not transform the outcome variables into population-based intensities but include the latter variable as an additional regressor in the econometric model.
6 The estimated coefficients are transformed to incidence-rate ratios (IRRs) and both heteroscedasticity-robust and bootstrapped standard errors are calculated to address the problem that asymptotic inference in small samples can be unreliable (Efron, Citation1981).