Transit Ridership Model Based on Geographically Weighted Regression

Abstract

This paper describes the development of a geographically weighted regression (GWR) model to explore the spatial variability in the strength of the relationship between public transit use for home-based work (HBW) trip purposes and an array of potential transit use predictors. The transit use predictors considered include demographics and socioeconomics, land use, transit supply and quality, and pedestrian environment. The best predictors identified through model estimation include two global variables (regional accessibility of employment and percentage of households with no car) and three local variables (employment density, average number of cars in households with children, and percentage of the population who are black). The models were estimated on the basis of the 2000 Census Transportation Planning Package data for Broward County, Florida. Model testing indicates the GWR model has improved accuracy in predicting transit use for HBW purposes over linear regression models. The GWR model also indicates that the effects of the independent variables on transit use vary across space. The research points to future research to explore different model structures within a geographic area.