In terms of Principal Components Analysis or Multiple Discriminant Analysis, I have used dummy variables to represent a group of like-features that are located in the same geographic region. For example, in the analysis of two metropolitan cities, the census tracts that make up city one are classified as 0 and city two as 1. This binary classification groups the census tracts into two regions.
I am wondering how I could store geographic coordinates (centroid?) of these census tracts so that I could introduce a spatial component to the multivariate analysis. This is dissimilar to the approach mentioned above since I am not grouping any features; but rather, I am interested in determining whether or not spatiality is a statistically significant independent variable (in terms of regression), or if spatiality is a significant part of a component/factor (in terms of PCA/FA).
This would be simple enough if the variable was a single measured value; but alas, we have x and y coordinates to think about.