The goal of this workshop is to introduce a new modeling technique for local spatial analysis— geographically weighted regression (GWR). This technique allows local, as opposed to global, spatial models to be calibrated and interesting variations in relationships to be measured and mapped. A. Stewart Fotheringham, Martin Charlton, and Chris Brunsdon are the pioneers in this field (Fotheringham, Charlton, and Brunsdon, 2002) and developers of the GWR package (current release is Version 3). They will be the lead presenters in the Geographically Weighted Regression Workshop. The GWR website (http://ncg.nuim.ie/ncg/GWR/) provides a brief primer on GWR, recent references, and details about acquiring the GWR software.
The standard procedure in the vast majority of empirical analyses of spatial data is either to calculate a global statistic or to calibrate a global model. The term “global” implies that all the spatial data are used to compute a single statistic that is essentially an average of the conditions that exist throughout the study area in which the data have been measured. Such a procedure is flawed when the relationships being measured vary over space. Geographically Weighted Regression is a statistical technique that allows variations in relationships over space to be measured within a single modeling framework. The output from GWR is a set of surfaces that can be mapped and measured, where each surface depicts the spatial variation of a relationship. The technique is based on regular regression modeling but can be extended in many different ways. It provides a great richness in the results obtained for any spatial data set and should be useful across all disciplines in which spatial data are used. This modeling approach challenges many of the global statements of spatial relationships that have been made in the academic literature. The authors have written Windows-based, user-friendly software for GWR, which will also be supplied to participants.
The workshop will be a combination of lectures and practical, computer-based sessions. Topics to be covered include local statistics and local models, the basics of GWR with examples, statistical inference and GWR, GWR and spatial autocorrelation, extensions to the basic GWR framework and concept, applications of specialized GWR software, and visualization of the output in ArcGIS. Exercises will be provided to participants but they will also be expected to bring their own spatial data sets for experimentation with GWR. A suitable data set will have between 250 and 2500 observations; it will have a dependent variable, a set of appropriate independent variables, and each observation in the dataset will have spatial coordinates. Participants will present the results of their GWR analyses on their own data sets at the conclusion of the course.