Chapter 12 Special Weights Operations
In this third chapter devoted to spatial weights, I consider some more specialized topics, in the sense that they are less frequently encountered in empirical practice. There are two main subjects.
First, I discuss two situations where the actual values for the spatial weights take on a special meaning. So far, only the presence or absence of a neighbor relation has been taken into account. In this chapter, this is generalized to weights that are transformations of the pairwise distances between neighbors, i.e., inverse distance functions and kernel weights.
The resulting weights files primarily provide the basis for creating new spatially explicit variables
for use in further analyses, such as in spatial regression specifications.86 The actual value of the weights themselves
are not used in measures of spatial autocorrelation or other exploratory analyses in GeoDa. As mentioned before, only the existence of a neighbor relation is taken into account.
There are two important applications for these spatial transformations. One pertains to the construction of so-called spatially lagged variables. Such variables are used in the exploration of spatial autocorrelation and in spatial regression analysis. In a second set of applications, the spatial weights are included as part of rate smoothing operations, as an extension of the methods covered in Section 6.4.
To illustrate these techniques, I continue to use the point layer from the Italy Community Banks sample data set and the Oaxaca Development polygon layer.
The distance functions in
GeoDaprovide an alternative and more user-friendly way to calculate the weights included in PySAL and GeoDaSpace (see Anselin and Rey 2014 for details).↩︎