ABSTRACT
This chapter contains the fundamental statistical components of the geographically weighted regression (GWR) framework. It begins with GWR and then moves to multiscale geographically weighted regression (MGWR), which is the focus of the book. The chapter covers model specifications and calibration issues and discusses how data are “borrowed” from other locations in order to perform a series of local regressions. Specific topics covered include kernel functions, kernel types, bandwidth selection, local hypothesis testing, measuring model fit, testing for spatial variability in the local parameter estimates, and detecting possible multicollinearity. The backfitting method of calibrating MGWR models is described as are the ramifications of covariate-specific bandwidths, which distinguish MGWR from GWR.
