Alternative Approaches to Spatial Autocorrelation: An Improvement Over Current Practice

Abstract

THE ASSUMPTION THAT THE CORRELATION BETWEEN THE TEMPORAL ERROR TERMS ∈ _t AND ∈_t-n DECLINES AS n INCREASES CAN BE JUSTIFIED BY APPEALING TO FIRST OR SECOND ORDER MAR-KOV PROCESSES OR TO SPECTRAL ANALYSIS, BUT A SIMILAR ASSUMPTION CANNOT BE JUSTIFIED FOR SPATIAL ERROR TERMS. THIS INTRODUCES AMBIGUITY IN THE SPECIFICATION OF THE WEIGHTING MATRIX W. IN THIS PAPER WE PROPOSE THE JOINT GENERALIZED LEAST SQUARES, EQUICORRELATED ERROR TERMS, RANDOM ERROR COMPONENT MODELS, AND RANDOM COEFFICIENT REGRESSION MODELS AS ALTERNATIVE SOLUTIONS. THESE APPROACHES ARE SUBJECT TO FEW OR NO PERSONAL BIASES, YET THEY ARE ABLE TO RESOLVE THE PROBLEM OF SPATIAL AUTOCORRELATION.