The Effects of Spatial Autocorrelation in Spatial Data Analyses

Abstract

The use of Ordinary Least Square (OLS) models for spatial data analyses frequently comes with theproblem of spatial autocorrelation in disturbance causing over/under estimated standard error of coefficient. Thesebiased standard errors make inferential tests invalid and the model inefficient in OLS framework. Since spatialautocorrelation mostly comes from intrinsic features of spatial data dependence, the problems of spatial autocorrelationare widely and frequently noted in literature. However, this study points out that previous notions on spatialautocorrelation in OLS framework may be insufficient if not wrong. Using eigenvectors and hexagon shapetransformation in controlled experiments, this study presents exact mechanism and effects of spatial autocorrelation inOLS models. Results indicates that standard errors of coefficients are decided by 1) spatial pattern of correspondingvariable, 2) correlation among the exogenous variables, and 3) parameter correlation in the model rather than simplyspatial autocorrelation in disturbance.