Abstract
This article empirically analyses the impact of agglomeration economies on the clustering of German and European firms using partial proportional-odds models. Firms are grouped according to industry and divided into departments. At the industry level, I find evidence for inter-industry economies derived from the New Economic Geography (NEG) framework for European firms in general and German knowledge-intensive firms in particular. At the department level, Marshallian Externalities such as the hiring of skilled labour and technological spillover, and therefore intra-industry economies, are empirically confirmed for European and German departments like Human Resources and R&D but rarely for others.
Acknowledgements
This article was partly written at the Bavarian State Institute of Higher Education Research and Planning (IHF) in Munich and Centro de Investigación y Docencia Económicas (CIDE) in Mexico City. I would like to thank one anonymous referee for helpful comments.
Notes
1 Updated information is given at http://www.hightech-strategie.de (see Bundesministerium für Bildung und Forschung) and http://www.cluster-bayern.de/ (see Bayerisches Staatsministerium für Wirtschaft, Infrastruktur, Verkehr und Technology) respectively.
2 See Ottaviano and Thisse (Citation2004) for a discussion of the early works of economic geographers and location theorists.
3 NEG-models are discussed in detail by Fujita et al. (Citation1999). Moreover, Fujita (Citation2010) reviews the literature and gives an overview of the NEG framework.
4 For a micro foundation of knowledge spillovers and therefore technology diffusion within NEG modeling, see Hafner (Citation2011).
5 A similar approach is given by Eberts and McMillan (1999) in distinguishing categories of agglomeration economies: internal economies of scale, economies within industries (‘localization economies’) and economies between industries (‘urbanization economies’).
6 The 2006 Innobarometer survey was carried out in the 27 member states of the EU, in two candidate countries (Croatia and Turkey) as well as in Norway, Switzerland and Iceland.
7 If all five cluster characteristics apply to the firm according to top managers’ opinion, then the maximum score of five points is given. If none apply, the score is zero.
8 Note that the estimated coefficients for the independent variables would not vary significantly if the equations were estimated separately using, for example, binary logistic regressions.
9 Detailed information for German firms is given in .
10 gives specific information about the different sector categories and their classifications.
11 In general, logistic regression models do not have an equivalent to the R-squared that is found in Ordinary Least Squares (OLS) regressions as a goodness-of-fit measure. The coefficients estimated are maximum likelihood estimates and have nothing to do with minimizing the variance of the dependent variable. There are a wide variety of pseudo R-squared that can be used to evaluate the goodness-of-fit of logistic regression models, but one should interpret them with great caution. I will therefore refrain from reporting them here.
12 A summary of the responses for German firms is shown in .