Exploring Spatial Variation in Relationship between Station Level Metro Ridership and Influencing Variables

This paper describes the development of ordinary least square (OLS) model, full geographically weighted regression (GWR) model and mixed GWR model to explore the spatial variation in relationship between metro ridership and an array of influencing variables at station level. The influencing variables considered are categorized into four types: land use diversity, accessibility to other traffic modes, station context and others. The best predictors identified through mixed GWR model estimation include one global variable (proportion of entertainment jobs in total employment) and five local variables (number of bus stop, employment density, number of high income workers, whether it is a transfer or terminal station). The models were estimated based on ridership data of Chicago metro stations and smart location database in 2010. Comparing the results of mixed GWR, full GWR and OLS model, we find that GWR models fit the data better and the residuals are less spatially correlated than OLS model and mixed GWR model performs best among them. This paper points to future research to consider the mixed spatially varying relationship when forecasting transit ridership.