Geographically Weighted Regression (GWR) is a local regression model thatcalibrates model coefficients based on local observations. Compared to ordinary least square regression models, in which model coefficients depict a global relationship between a dependent variable and a set of independent variables, the coefficients in a GWR model are local and vary from location to location. These locally varying coefficients provide an opportunityfor investigating the strength of the relationship between the dependent and independent variables and identifying variables that may not be significant for certain areas. This paper examines the significance of variablesin a GWR model to identify new model structures for subregions. A GWR model is first estimated using the 2000 CTPP data for Broward County, Floridato investigate potential variables that affected public transit use for home-based work trip purpose, which included demographic, socioeconomic, land use, transit supply quality, and pedestrian environment variables. Based on the local significance of the independent variables, two subregional GWR models are then calibrated to include variables that are locally significant. A comparison between the subregional GWR models and the original GWR model showed that the subregional GWR models performed better than the regional GWR models in terms of model accuracy.
Abstract