Can geographically weighted regressions improve regional analysis and policy making?

Policy design in a regional context requires explicit recognition of spatial heterogeneity in community characteristics as well as in the heterogeneity of how these characteristics impact the target variables. By providing only a "global" measure for the entire space, standard approaches such as ordinary least squares or (most) spatial econometric models tend to compromise spatial heterogeneity in favor of average estimates and efficiency. More assessment is needed of whether the gains of simplicity and statistical efficiency offset the losses from ignoring spatial heterogeneity. Using data for about 1,900 rural Canadian communities as a backdrop, the authors address this issue using a geographically weighted regression approach. The authors find that for about two-thuds of the variables, standard approaches would have significantly understated the spatial differences in the impact of selected variables. Standard analysis would not have uncovered this information, suggesting that subsequent policy inferences would be poorly suited to many local settings.