GIS Workshops - Geographically Weighted Regression

  Intro to GIS  |   Coordinates Systems  |   Exploratory Spatial Data Analysis  |   Points and Surfaces  |   GWR  |   Spatial Regression  |   Spatial R

Geographically Weighted Regression (GWR) is a specialized tool for exploring how the relationships among variables in a multivariate regression change across the study area. In traditional regression analysis we generate coefficients associated with our independent variables which we assume are constant (more formally we assume they are stationary) across all of our observations.GWR tests this assumption by generating a series of 'local' regression models that give greater weight to nearer observations and less weight to those that are more distant.GWR permits us to analyze patterns of change (or fixity) in the relationships among our variables across our study area. This can be useful as a means of identifying missing variable problems or can be an end in itself. This class offers a start to finish walkthrough of the specification of a GWR model in ArcGIS and the interpretation of its outcomes. We will also consider some of the major pitfalls in data preparation and in model specification that can lead to errors or system failure.

Prerequisite: The workshop assumes some prior knowledge of GIS. In particular, the workshop on Exploratory Spatial Data Analysis is highly recommended.


Chris Fowler
Raitt Hall 218M
(206) 920-1686


  • Why GWR?
  • Mapping OLS Residuals
    • Basics of OLS Regression
    • Spatial autocorrelation and clustered residuals
  • The mechanics of GWR
    • Bandwidth and Kernel selection
    • Types of GWR
  • Further steps
    • Significance Testing
    • Interpreting Results
    • Common Errors


Further Reading

  • Fotheringham et al. 1998
    • A good general citation for the technique as explained by its developers
  • Brunsdon et al. 1999
    • Suggestions on how to test for significance in coefficient nonstationarity as well as an introduction to mixed GWR models

To register for the workshops, please complete the Registration Form.

Clustered Residuals from a Standard OLS Regression (Poverty in the Southern U.S.)

Nonstationarity in the coefficient for Percent Hispanic (Poverty in the Southern U.S.)