Lag or error? - Detecting the nature of spatial correlation

Abstract

Theory often suggests spatial correlations without being explicit about the exact form. Hence, econometric tests are used for model choice. So far, mainly Lagrange Multiplier tests based on ordinary least squares residuals are employed to decide whether and in which form spatial correlation is present in Cliff-Ord type spatial models. In this paper, the model selection is based both on likelihood ratio andWald tests using estimates for the general model and information criteria. The results of the conducted large Monte Carlo study suggest that Wald tests on the spatial parameters after estimation of the general model are the most reliable approach to reveal the nature of spatial correlation.