![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
Wanzenböck I., Scherngell T. and Lata R. Embeddedness of European regions in European Union-funded research and development (R&D) networks: a spatial econometric perspective, Regional Studies. This study focuses on the embeddedness of regions in research and development (R&D) networks within European Union Framework Programmes by estimating how distinct regional factors affect a region's network positioning. Graph theoretic centrality measures in terms of betweenness and eigenvector centrality are calculated at the organizational level to reflect the relevant network structure before aggregation to the region level. Panel spatial Durbin error models (SDEM) reveal that region-internal knowledge production capacities, a region's level of economic development as well as spatial spillovers are important determinants for a region's positioning in the European Union-funded R&D network, but their significance differs depending on the centrality concept.
Wanzenböck I., Scherngell T. and Lata R. 欧洲区域在欧盟资助的研究发展（R&D）网络中的镶嵌性：空间计量经济的视角，区域研究。本研究透过评估特出的区域因素如何影响一个区域在网络中的位置，聚焦区域在欧盟架构计画的研究发展网络（R&D）中的镶嵌性。图论的中心性将以居中性测量之，而特徵向度中心性则在组织层级进行计算，以反映聚集至区域层级之前的相关网络结构。面板空间杜宾误差模型（SDEM），揭露出区域内部的知识生产能力、一个区域的经济发展层级，以及空间上的外溢，对该区在欧盟资助的研究发展网络中的位置而言是重要的决定因素，但它们的重要性，则根据中心性的概念而有所不同。
Wanzenböck I., Scherngell T. et Lata R. L'ancrage des régions européennes dans les réseaux de recherche et de développement (R et D) financés par l'Union européenne: le point de vue économétrique spatial, Regional Studies. Cette étude porte sur l'ancrage des régions dans les réseaux de recherche et de développement (R et D) dans le contexte des programmes-cadres de l'Union européenne en estimant comment des facteurs régionaux distincts influencent le positionnement des réseaux selon la région. Au niveau organisationnel on calcule des mesures théoriques graphiques de la centralité en termes de l'intermédiarité et de la centralité des vecteurs propres afin de refléter la structure de réseau concernée avant l'agrégation au niveau régional. Des modèles d'erreurs de Durbin spatiaux en panel laissent voir que la capacité de production de connaissances sur le plan régional interne, que le niveau de développement économique régional ainsi que des retombées spatiales constituent des déterminants clés du positionnement d'une région dans le réseau de R et D financé par l'Union européenne, mais que leur importance varie suivant la notion de centralité.
Wanzenböck I., Scherngell T. und Lata R. Positionierung europäischer Regionen in von der Europäischen Union finanzierten Netzwerken für Forschung und Entwicklung (F&E): eine räumliche ökonometrische Perspektive, Regional Studies. Regionen und ihre Positionierung im europäischen, durch die EU-Rahmenprogramme geförderten F&E-Netzwerk bilden den Schwerpunkt dieser Studie. Es wird der Versuch unternommen, die Netzwerkposition von Regionen anhand von regionsspezifischen Einflussfaktoren zu erklären. Hierzu werden graphentheoretische Zentralitätsmaße wie Betweenness-Zentralität und Eigenvektor-Zentralität auf Ebene der beteiligten Organisationen berechnet und auf Regionen aggregiert. Unter Einsatz von räumlich ökonometrischen Modellen in Form von räumlichen Durbin-Fehlermodellen kann gezeigt werden, dass regionale Faktoren der Wissensproduktion, der wirtschaftliche Entwicklungsstand einer Region sowie auch räumliche Spillover-Effekte signifikanten Einfluss auf die regionale Positionierung im europäischen F&E-Netzwerk haben. Deren Wirkungskraft unterscheidet sich jedoch hinsichtlich der unterschiedlichen Zentralitätsmaße.
Wanzenböck I., Scherngell T. y Lata R. Integración de las regiones europeas en las redes de investigación y desarrollo (I+D) subvencionadas por la Unión Europea: una perspectiva econométrica espacial, Regional Studies. Este estudio trata sobre la integración de las regiones en las redes de investigación y desarrollo (I+D) dentro de los programas marco de la Unión Europea al calcular en qué medida influyen los factores regionales en el posicionamiento en la red de una región determinada. Se calcula la centralidad de las medidas de la teoría de gráficas en lo que respecta a la centralidad de intercalación y vectores propios de ámbito organizativo con el objetivo de reflejar la estructura de redes relevante antes de agregar estos factores a una región. Los modelos espaciales Durbin para la corrección de errores de panel indican que las capacidades de producción de conocimiento en las regiones, el nivel de desarrollo económico de una región así como los desbordamientos espaciales son determinantes importantes para el posicionamiento de una región en la red de I+D subvencionada por la Unión Europea, pero su importancia difiere en función del concepto de centralidad.
Acknowledgements
The authors are grateful to Manfred M. Fischer (Vienna University of Economics), Michael Barber (AIT) and Yuanjia Hu (University of Macau) for valuable comments made on an earlier version of the manuscript. They also thank two anonymous referees for valuable comments.
Funding
This work was funded by the FWF Austrian Science Fund [Project Number I 886 G11] and the Multi-Year Research Grant (MYRG) – Level iii [RC Reference Number MYRG119(Y1-L3)-ICMS12-HYJ] by the University of Macau.
Notes
1. In theoretical and empirical network research, network mechanisms related with the notions of ‘preferential attachment’ or ‘assortativity’ and ‘triadic closure’ (e.g., Newman, Citation2003; Ter Wal and Boschma, Citation2009; Rivera et al., Citation2010), are regarded to play a key role when analysing networks and network formation processes from a dynamic perspective.
2. Connectivity in R&D networks depends also on the capacity of actors to build up and maintain relations with a diverse set of collaboration partners. From a network formation perspective, decisions in this regard are no longer subject to the utility of single R&D collaborations, but more likely based on the costs and opportunities involved in cultivating a variety of network relations. Moreover, the willingness of actors to disclose and further distribute knowledge through network channels affects the decision of whether to form a new tie or to draw on already familiar R&D partners. Hereby, organization-specific strategies and motives for external knowledge sourcing might influence the ego-network structure of organizations, depending on institutional conditions in terms of appropriability of research results (Hazir, Citation2014), or the general development of the knowledge base in the research field (Ter Wal, Citation2013). However, these reflections on network formation cannot be considered in a comprehensive way in the current study given its regional focus.
3. Since their launch in 1984, the overall objectives of the FPs have been to strengthen the scientific and technological bases of the European scientific community and the European economy to foster international competitiveness, and the promotion of research activities in support of other EU policies (e.g., Scherngell and Barber, Citation2009). However, since FP5 a stronger focus on integrating national and regional research communities in different thematic fields across Europe is clearly noticeable. In the FPs, project proposals are to be submitted by self-organized consortia. Funding is open to all legal entities established in the member states of the EU – e.g. individuals, industrial and commercial firms, universities, research organizations, etc. – and can be submitted by at least two independent legal entities established in different EU member states or in a member state and an associated state. Proposals to be funded are selected on the basis of criteria including scientific excellence, added value for the European Community, the potential contribution to furthering the economic and social objectives of the community, the innovative nature, the prospects for disseminating and exploiting the results, and effective transnational cooperation.
4. In principle, a cross-region R&D collaboration matrix that contains FP collaborations intensities between regional entities (e.g., Scherngell and Barber, Citation2009) could also be used to measure a region's centrality. However, such an approach would suffer from the drawback that organizations – rather than regions – are the relevant actors that are participating in the FPs, and therefore organizations constitute the appropriate unit of observation for such kinds of analyses.
5. EUPRO is constructed and maintained by the AIT Austrian Institute of Technology. It contains systematic information on project objectives and achievements, project costs, project funding and contract type as well as on the participating organizations including the full name, full address and type of organization for FP1 to FP7 (e.g., Scherngell and Barber, Citation2011).
6. In the definition of the distinct thematic areas of the FPs this paper basically follows the study of Hoekman et al. (Citation2012) including the following programme lines in the distinct FP: FP4 programmes ENV2C, MAST3, JOULE and THERMIE, FP5-EESD, FP6-SUSTDEV for Sustainable Development; FP4-BIOTECH2, FP4-BIOMED2 and FP4-FAIR, FP5-Quality of Life, FP6-Food, FP6-LIFESCIHEALTH for Life Sciences; FP4-ACTS, FP4-ESPRIT4 and FP4-TELEMATICS 2C, FP5-IST, FP6-IST for the thematic priority ICT (see also Rietschel et al., Citation2009). The thematic areas included make up 72.5% of total funding in FP5 and 63.3% of total funding in FP6 (Hoekman et al., Citation2012). Some basic network statistics of the R&D network observed for each thematic priority are provided in Appendix B.
7. Other point centrality measures commonly used in SNA are degree and closeness centrality. Degree centrality focuses only on connections directly attached to a vertex and is therefore rather a measure for local centrality (e.g., Wasserman and Faust, Citation1994). In contrast, closeness centrality is based on the shortest distance to all other vertices in the network and indicates how close a distinct vertex is to all other vertices in the network. Closeness centrality would be an informative measure with respect to the transfer of knowledge from a global network perspective. However, especially for the case of FP network at the organizational level, calculation would involve considerable difficulties since a path between pairs of unconnected nodes would be infinite, and thus is not defined for disconnected networks (e.g., Faust, Citation1997).
8. The bipartite graph is used to calculate betweenness centrality for each organization u, enabling the weighted character of the network to be kept when calculating standard SNA measures. Using the one-mode or unweighted representation would cause loss of information. For eigenvector centrality similar results will appear for the bipartite graph and its unipartite projection (Faust, Citation1997).
9. For practical purposes, betweenness centrality is based on the adjacency matrix At. A common notation used in this context is the eigenvector equation as given by: λ x = A x, where x is a vector of centralities with x = (x1, x2, …) denoting the eigenvector of the adjacency matrix A with eigenvalue λ (Bonacich, Citation1987).
10. For a detailed list of regions, see in Appendix B. Despite substantial size differences and interregional disparities of some regions, NUTS-2 units are widely recognized as an appropriate level for modelling and analysis purposes (e.g., Fischer et al., Citation2006; LeSage et al., Citation2007).
11. Random effects specifications in measuring network embeddedness at the regional level seems reasonable, in particular due to unobservable effects at the micro-level related, for example, to organization-specific decisions to engage in R&D collaborations (e.g., Autant-Bernard et al., Citation2007; Paier and Scherngell, Citation2011). Moreover, in the present case the units of observation for n = 241 regions, in contrast to t = 9, is relatively large, leading to a substantial loss of degrees of freedom. Further, observations of certain independent variables are quite invariant in time, and for this reason they could not have been included in the estimation (e.g., Baltagi, Citation2008). Note that the random effects specification of the empirical model is underlined by the significant Baltagi–Song–Koh test (Baltagi et al., Citation2007) pointing to random unobservable individual specific effects. The two-sided test points to the joint existence of time-series and spatial error correlation, providing statistical justification for the random effects spatial error model.
12. The row standardized version of W is used, allowing interpretation of the spatially lagged independent variables to be the weighted average impact on region i by their neighbouring regions. The simplified and straightforward interpretation of both direct and indirect effects is a specific advantage of the SDEM (Le Sage and Pace, Citation2009). In contrast to spatial lag (SAR) or spatial Durbin model (SDM) specifications (e.g., LeSage and Fischer, Citation2008; Fischer et al., Citation2009a, Citation2009b), which take global multipliers (induced by feedback loops between regions i and j) into account, the SEM and SDEM do not contain spatial lags of the dependent variable. Thus, interpretation of parameter estimates is less complicated since they directly reflect direct and indirect effects. Further, common inference statistics such as the standard deviation and t-statistic can be used to examine significance of parameter estimates (LeSage and Pace, Citation2009).
13. The classification of high-technology sectors is based on Eurostat.
14. The index is defined by:
where sip is the region's i share of patents in a specific IPC class p; and
is the mean of IPC class p. Patents were taken into account at a three-digit level corresponding to the International Patent Classification (IPC).
15. For the construction of the industrial diversity variable, five different main economic sectors are included, namely agriculture, manufacturing, construction, private services and the non-market service sector. Similar to technological specialization, the index of specialization to account for industrial diversity is defined as:
where oip is the region's i share of gross value added in a specific sector p (indexed p = 1, …, 5); and
is the mean of sector p for n = 241 regions.
16. Several robustness checks for the specification of the general model were carry out. First, the number of publications was included in order to proxy the presence of a strong university/scientific sector within a region, and thus to control for possible organization-specific idiosyncrasies in terms of R&D collaboration behaviour. Second, to control for geographical effects, specific dummy variables for Eastern and Southern European countries as well as for EU-15 member states were included. As coefficients were insignificant and did not change the estimators of the remaining variables, the study refrains from including publication intensity and geographical dummies for reasons of simplicity.
17. The estimates of the sector dummy variables provide statistical evidence that the probability for gaining higher network embeddedness is higher in the Sustainable Development and Life Sciences sector than in ICT (which is the reference sector) – a result that may traced back to the fact that ICT is more often characterized by industrial and small and medium-sized enterprise (SME) involvement as well as by university–industry relations than Sustainable Development and especially Life Sciences involving specialized and often university-led knowledge production that is more often led by a few scientific core players. Note that estimation results for additional, individual sector-specific models are given in Appendix A.