A Multidimensional Study of Transit Ridership and Station Mode Shares in the United States: Nonlinear Effects, Data Aggregation, and Post-Pandemic Changes

Understanding the differences among public transit types allows for the development of more targeted policies at both local and regional levels. Examining how the built environment (BE) influences travel behavior (Delclòs-Alió et al.) and assessing data aggregation effects around different transit station types at the local level, along with identifying key predictors of ridership across transit modes at the regional level, offers valuable insights for policy efforts. Specifically, the dissertation comprises three studies that analyze BE-travel behavior associations and data aggregation effects locally, as well as variations in key predictors of rail and bus ridership at a regional scale within the United States. The findings emphasize the unique land-use and travel behavior associations for various public transit systems within transit catchment areas, the effects of data aggregation on BE-travel behavior models, and the critical predictors of rail and bus ridership at regional levels. The first study highlights nonlinear associations between BE attributes and commuting mode share within rail and Bus Rapid Transit (BRT) catchment areas, using data from approximately 2,790 transit stations across 34 U.S. metropolitan statistical areas. Applying a random forest model, this study reveals substantial differences between rail and BRT areas, with rail catchment areas showing greater sensitivity to BE factors in reducing car dependency. BRT, however, emerges as a viable alternative for sprawling areas lacking the compact development needed to support rail systems. The second study investigates how data aggregation influences the BE-mode share relationship around 2,794 rail and BRT stations, utilizing both inferential and machine learning approaches. Findings indicate that data aggregation affects BE-mode share models regardless of the analytical method. Optimal buffer sizes for capturing BE effects and minimizing sensitivity to data aggregation were identified as 800 meters for BRT stations and 1,000 meters for rail stations. Key BE features such as employment density, jobs per household, intersection density, residential density, distance from the central business district, job accessibility (active modes) demonstrated robustness against data aggregation for both rail and BRT stations. The third study examines changes in transit ridership predictors before and after the COVID-19 pandemic across 35 U.S. metropolitan areas. Using extreme gradient boosting on data spanning January 2019 to June 2023, the study identifies a shift from internal to external factors as key drivers of ridership post-pandemic. Socioeconomic factors, gasoline prices, telecommuting, population density, employment density and polycentric development emerged as influential for bus ridership post-pandemic, while traditional factors like vehicle revenue miles, fare, transit coverage, and service areas are more important for rail ridership. Additionally, the study uncovers unique threshold and interaction effects in the post-pandemic period, including positive interactions between African American population proportions and poverty rates for bus ridership, carless households and gasoline prices for bus ridership, and between VRM and polycentricity for rail ridership. This dissertation provides insights into the complex dynamics between BE, transit types, and travel behavior, offering valuable implications for urban transportation planning and policy development at multiple levels.