ABSTRACT
The analysis of multivariate local spatial autocorrelation is challenging due to the difficulty of separating spatial effects from pure attribute correlation. This chapter introduces a Bivariate Local Moran, a multivariate Local Geary and a Local Neighbor Match Test as methods to disentangle spatial and attribute correlation in a local statistic.
As its global counterpart, a Bivariate Local Moran captures the association between a given variable at one location and a different variable at the neighboring locations. The interpretation is similar to the standard Local Moran in terms of clusters and spatial outliers but needs to be carried out with caution since the intrinsic bivariate correlation between the variables (in situ) is not controlled for. The concept is also not symmetric.
The Multivariate Local Geary is based on squared difference dissimilarity, measured as a distance in multi-attribute space to the neighbors in geographic space.
The Local Neighbor Match Test is based on the overlap between the connectivity graphs of k-nearest neighbors in geographic space and in multi-attribute space. Significant overlap between the two signifies that points (observations) that are close in multivariate attribute space are also close in geographic space.
