Improvement of spatial autocorrelation, kernel estimation, and modeling methods by spatial standardization on distance

Abstract

In a point set in dimension superior to 1, the statistical distribution of the number of pairs of points as a function of distance between the points of the pair is not uniform. This distribution is not considered in a large number of classic methods based on spatially weighted means used in spatial analysis, such as spatial autocorrelation indices, kernel interpolation methods, or spatial modeling methods (autoregressive, or geographically weighted). It has a direct impact on the calculations and the results of indices and estimations, and by not taking into account this distribution of the distances, spatial analysis calculations can be biased. In this article, we introduce a “spatial standardization”, which corrects and adjusts the calculations with respect to the distribution of point pairs distances. As an example, we apply this correction to the calculation of spatial autocorrelation indices (Moran and Geary indices) and to trend surface calculation (by spatial kernel interpolation) on the results of the 2017 French presidential election.