Spatial association measures for an ESDA-GIS framework: Developments, significance tests, and applications to spatio-temporal income dynamics of United States labor market areas, 1969--1999

This study is concerned with developing new spatial association measures (SAMs), elaborating generalized significance testing methods, proposing associated graphical and mapping techniques for an ESDA-GIS (Exploratory Spatial Data Analysis-Geographic Information Systems) framework, and applying those techniques to spatio-temporal income dynamics across U.S. labor market areas, 1969–1999. It is argued that SAMs play a central role in obtaining a seamless integration between ESDA and GIS where the cross-fertilization between them is highly achieved in such a way that ESDA takes advantage of GIS's data manipulation and visualization capabilities and a GIS utilizes ESDA's statistical integrity and computational efficiency.; Two sets of new SAMs are developed: global S and local S_i as univariate SAMs, and global L and local L_i as bivariate SAMs. Global S, spatial smoothing scalar, captures the degree of spatial smoothing when a geographical variable is transformed to its spatially smoothed vector in which each observation is re-computed in conjunction with its neighbors as defined in a spatial weights matrix. If a spatial pattern is more spatially clustered, it is given a higher value of S. Local S_i, defined as an observation's relative contribution to the corresponding global S, allows a researcher to detect spatial clusters with effectively avoiding the tyranny of reference observations that preexisting univariate SAMs have suffered from.; Global L and local L_i are devised to conform to two concepts of association involved in comparing two spatial patterns in a simultaneous fashion: pairwise point-to-point association and univariate spatial association. Whereas aspatial bivariate association measure, such as Pearson's correlation coefficient, is dedicated solely to the first type of association, global L captures numerical co-variances conditioned by topological relationships among observations to parameterize bivariate spatial dependence and to calibrate the degree of spatial co-patterning. Local L_i, a localized spatial correlation, captures the degree to which each location conforms to or deviates from the corresponding global L, and allows for exploring spatial heterogeneity in a bivariate relation. (Abstract shortened by UMI.)...