The non-stationary influence of geography on the spatial agglomeration of production in the EU

In this paper, we investigate the relative importance of geographic features on the location of production in the EU. Specifically, we want to quantify how much of the spatial pattern of GDP can be attributed to only exogenous first nature elements (physical and political geography) and how much can be derived from endogenous second nature factors (man-made agglomeration economies). In order to disentangle both effects empirically, and to learn how they are interrelated, we control for second nature. We use a methodology based on an analysis of variance (ANOVA), which is applied to a panel of 1,171 European NUT-3 in 2006. We demonstrate that -due to a high degree of spatial non-stationarity present in the data- results can be biased if spatial autocorrelation and spatial heterogeneity, as well as multicollinearity and endogeneity, are not properly taken into account.