Local versus regime convergence regression models: a comparison of two approaches

Two types of econometric models have typically been used in geographical research: global and local. The former produces parameter estimates that represent an average type of economic behavior; that is, for each variable there is one regression coefficient for the entire sample. The latter suggests that the relationship of interest can vary widely over space allowing the regression parameters to change across spatial entities. Commonly, the regression results obtained from these two different approaches are compared to each other to determine which model performed better. The aim of this paper is twofold. First and foremost, it is to show that a comparison between global and local models is problematic and misleading since it overlooks a critical group of regression models known as regime models. This means that this type of comparison should be extended by including all different regression approaches (i.e. global, local and regime). Sometimes, regime models can perform as well as, or even better, than (more complex) local models and this is something that should be taken into account by empirical geographical studies. Secondly, the paper investigates the convergence process in the regions of the (enlarged) European Union (EU) over the period 1995–2005. The results of this paper show that a regime (convergence) model seems to perform better than any other model employed, implying a very low convergence rate among European regions and highlighting the heterogeneous spatial impact of the EU economic integration process. © 2014, Springer Science+Business Media Dordrecht.