The Influence of the Structure of Spatial Weight Matrix on Regression Analysis in the Presence of Spatial Autocorrelation

Abstract

Many remedial methods for the spatial autocorrelation of error terms in regression model have been developed. However, the spatial weight matrix, which may play an important role in estimating the parameters, is often designed by the analysts a priori. The empirical results suggest that the parameter estimates depend heavily on the spatial weight matrix. They also suggest that the spatial weigh matrix which presents the strong correlation among neighborhoods and is very different from what is lead from the analogy of gravity law and generally used, may lead to the maximum of log-likelihood.