Structural dynamics of regional innovation patterns in Europe: the role of inventors’ mobility

ABSTRACT

The role of inventors’ mobility on the innovative capacity of the host region has largely been highlighted, measured and empirically proved. In this work, the perspective is a rather different one. The paper assesses the role that the flow of inventors and high-skilled technicians has on the region's capacity to modify its structural mode of innovation. By applying the regional patterns of innovation framework in a dynamic perspective, it is shown that inventors’ inflows across space produce structural dynamics in the mode regions innovate.

KEYWORDS:

• inventors’ mobility
• regional patterns of innovation
• structural dynamics

JEL:

• O3
• R11

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

SUPPLEMENTAL DATA

Supplemental data for this article can be accessed at https://doi.org/10.1080/00343404.2017.1379600.

ORCID

Roberta Capello http://orcid.org/0000-0003-0438-6900

Camilla Lenzi http://orcid.org/0000-0002-0880-3516

Notes

1. See Lissoni (2016) for a recent survey.

2. Inventors’ migration and diaspora have been proved beneficial not only for the recipient area but also for origin places (Kerr & Kerr, 2017, Kerr et al., 2016, Migueléz, 2016).

3. There is a longstanding tradition of using data on patents and their inventors to track mobility flows across firms and in space (e.g., Agrawal et al., 2006; Breschi et al., 2017; Miguélez, Moreno, & Suriñach, 2010). Even if this approach is not free of possible measurement bias (Ge, Huang, & Png, 2016), especially when tracking firm-to-firm mobility (Lenzi, 2013), it remains a valuable choice for detecting the spatial diffusion of knowledge and its potential impact on recipient economies. Ultimately, ‘knowledge always travels with people who master it’ (Breschi et al., 2010, p. 367).

4. In the science-based pattern, knowledge is mostly created by local actors, typically universities, R&D centres and large firms. Local relationships are generally enriched by interregional cooperation with selected partners, as highlighted in most of the literature dealing with knowledge and innovation creation and diffusion (Jensen, Johnson, Lorenz, & Lundvall, 2007; Mack, 2014). In the creative application pattern, entrepreneurial creativity and collective learning enable one to access external knowledge and use it for local innovation needs (Foray, 2009; Licht, 2009). Knowledge sources are mostly located outside the region, and knowledge exchanges are nourished more by cognitive and sectoral proximity (i.e., shared cognitive maps) than by belonging to the same local community (Asheim & Isaksen, 2002). In the imitative innovation pattern, instead, relationships among actors (generally between local firms and dominant firms, typically multinationals) are aimed at the adoption of innovations new for the area, as described in the literature dealing with innovation diffusion (Pavlínek, 2002; Varga & Schalk, 2004).

5. NUTS = Nomenclature des Unités Territoriales Statistiques. For further details on the variables used in the cluster analysis implemented to detect innovation patterns in European regions and the variables representing the key territorial features of the different groups of regions, see Capello and Lenzi (2013) and Appendix A in the supplemental data online.

6. As noted in the second section, regional innovation patterns are made of two main blocks: functional and relational elements. While the impact of incoming inventors on the former may be limited, it is reasonable to assume that inventors will establish professional connections primarily dedicated to the exchange of scientific, technical and formal knowledge within the region, across regions and back to the origin region.

7. Trippl et al. (2015) advance a similar claim in the context of the dynamics of regional innovation systems. In particular, they propose that the relative balance of the role of endogenous versus exogenous stimuli to regional innovation system renewal depends on the need, attractiveness and absorptive capacity of recipient regions.

8. See note 6 above.

9. The changes that occur between 2002–04 and 2004–06 are the outcome of a long process of adjustment that develops its final step in the short period considered. A recent paper (Capello & Lenzi, 2017) shows that such changes are linked to the accumulation, over the past, of structural characteristics fundamental to move to another pattern.

10. See Miguélez and Moreno (2013a, 2013b, 2013c, 2015) for a similar approach.

11. The regions in the European science-based area have been excluded from the analysis (i.e., 20 regions out of 262 NUTS-2 regions in the EU-27) because for them there is not the possibility to observe any move towards a more complex innovation pattern.

12. We computed Moran's I measure of spatial autocorrelation on the dependent variable: 0.122, p < 0.00. The significance of Moran's I provides a sufficient basis for adopting a spatial econometric framework, as discussed next.

13. The estimation of a spatial logit model would be possible but at the cost of a considerable loss in terms of sample size, as commented above.

14. In terms of the spatial specification, statistical tests reported in the next section justify the choice of SLX specification with respect to more complex and general ones (e.g., spatial lag models), signalling the joint significance of the spatially lagged independent variables but not of the spatially lagged dependent variable. Therefore, the estimates commented on in the next section are based on the LPM model with SLX specification. Moreover, results of the SLX specification have been compared with the spatial Durbin error model (SDEM) one, when relevant, to control for the spatial diffusion of shocks to the model disturbances. In fact, the inclusion of the spatially lagged independent variables could not rule out the existence of spatial diffusion of shocks to the model disturbances, which can be easily detected and controlled in a linear setting by estimating an SDEM.

15. In this case, in fact, binary non-linear models offer limited tools to assess the quality of the instrumental variables approach implemented (i.e., the standard battery of tests to assess the endogeneity, orthogonality, strength and relevance of instruments reported in Appendix C in the supplemental data online), whereas linear approaches guarantee more flexibility in this respect (Cameron & Trivedi, 2010).

16. In particular, inventors’ mobility has been instrumented by using two sets of variables. The former accounts for the degree of urbanization in the recipient region, measured through the regional NUTS-2 share of the urban fabric over total land, the regional NUTS-2 share of water areas over total land, derived from CORINE Land Cover 1990 (sourced from the European Observation Network for Territorial Development and Cohesion – ESPON) and the presence of the capital city in the NUTS-2 region. Previous analyses have indeed shown the relevance of these predictors in directing skilled individuals’ mobility patterns (Breschi & Lenzi, 2016; Miguélez & Moreno, 2015). The latter account for the geographical position of the region in the EU space, measured through the region's centroid latitude and longitude.

17. Following a reviewer’s suggestion, we tested the robustness of the results by including the number of resident inventors in the region per 1000 inhabitants to check whether larger regions (in terms of inventors) attract more inflows of inventors (i.e., the absorptive capacity argument presented in the second section). Unfortunately, this variable raises problems of multicollinearity and has been finally discarded. The results are available from the authors upon request.

18. We acknowledge that this result may sound surprising. We investigated whether this might be related to multicollinearity, but the computation of the variance inflation factor (VIF) for each variable excluded this risk, as it is never greater than 2.5, the value generally used (as a rule of thumb) to detect serious multicollinearity problems. Following a reviewer's suggestion, we tried to find an alternative variable to R&D to check the robustness of our results. The best alternative option we found for the period under consideration was the share of employment in high-tech sectors (both services and manufacturing), as defined according to the Organisation for Economic Co-operation and Development’s (OECD) classification. Other variables, such the human resources in science and technology (HRST) series, available from EUROSTAT, unfortunately presents considerable gaps in the years under consideration which prevent their use. In particular, we used the average employment share in high-tech sectors in the years 2000–02, as retrieved from EUROSTAT. This variable unfortunately is never significant and, therefore, does not improve our estimates with respect to the R&D one. The results are available from the authors upon request.

19. When including interactions, all the interaction terms, i.e., interaction effects and simple effects, should be included unless there is a good (theory-based) reason not to do so (Brambor, Clarck, & Golder, 2006). Interpretation should be done in relative terms with respect to the reference case (in this case the applied science area) and not in absolute terms.

20. Estimates of Tables 2 and 3 are robust to the use of a binary contiguity matrix (the results are available from the authors upon request).

21. We also tested the robustness of our results against the possibility that ongoing processes of structural change occurring in regional economies can affect the probability of changing the current innovation pattern. For this purpose, we built three dummy variables dividing regions into three groups following three alternative processes of structural dynamics: Mediterranean regions (regions in Spain, Portugal, Italy and Greece); transition regions (regions in Eastern European countries that joined the EU after 2004 and Eastern Germany regions); and dynamic innovation regions (the remaining ones). The results are qualitatively unchanged and available from the authors upon request. We thank an anonymous reviewer for suggesting this robustness check.

22. By replicating the endogeneity test on the dummy variables for the regional patterns of innovation, it turns out that the null of exogeneity cannot be rejected also in this case (see Appendix C in the supplemental data online). Therefore, there are no conceptual reasons to expect that the interaction terms (between the mobility variable (exogenous) and the dummies for regional innovation patterns) could be endogenous. Accordingly, the expanded model with interaction variables (equation 2) is simply estimated through LPM and SLX specification, with further checks for spatial dependency, as shown in Table 3.