Multicollinearity in geographically weighted regression coefficients: Results from a new simulation experiment

Multicollinearities among the coefficients obtained from the method of geographically weighted regression have been identified in recent research. This is a serious issue that poses a critical challenge for the utility of the method as a tool to investigate multivariate relationships. The evidence regarding the ability of GWR to retrieve spatially varying processes remains mixed due to partial and inconclusive experiments. The objective of this paper is to provide stronger support to the thesis that multicollinearities are inherent to the method. This objective is accomplished by: 1) Investigating multicollinearity in situations where the underlying process is stationary and non-stationary; and 2) Using advanced visualization to report results for a range of outcomes as opposed to the average of r replications as in previous research. Extensive simulation experiments that test two different implementations of GWR provide evidence of spurious multicollinearity between local regression coefficients. This suggests that extreme caution should be exercised when drawing conclusions regarding spatial relationships retrieved using this modeling approach.