Critical issues in the evaluation of spatial autocorrelation

Abstract

Spatial autocorrelation measures the degree to which a spatial phenomenon is correlated with itself in space. As such, it can be used as an indicator of the fundamental topological structure of the spatial relationship among geographic entities displayed on a map. Statistics of spatial autocorrelation are especially useful for characterizing the spatial pattern in the distribution of any phenomenon in question. This paper addresses three important issues pertaining to the evaluation of spatial autocorrelation: measurement of the study variable, definition of geographic units, and specification of spatial weighting functions. Theoretically, spatial autocorrelation is most adequately evaluated when the study variable is measured in either an interval or a ratio scale and geographic units are better delineated by polygons of homogeneous surfaces based on variables that are significant to the distribution of the study phenomenon. In evaluating spatial autocorrelation, weighting functions such as area, boundary length, distance, and their combinations must be examined carefully and specified whenever necessary.