Aggregate and Disaggregate-level Models of Public Transit Ridership in Bengaluru, India

Abstract

Ridership forecasting models are valuable tools for public transit agencies to assess and quantify the possible impacts of their operational strategies and service improvements on transit demand and revenue. Although a rich body of literature exists on this topic, there has not been much study on transit ridership forecasting for cities in emerging economies such as India. Further, the literature in this area is saddled with one or more of the following methodological and substantive issues: (a) the difficulty of relating spatially aggregate demand data (typically available with transit agencies) to disaggregate, transit stop-level catchment characteristics, (b) inadequate consideration of inter-route interactions such as competition and complementarity for transit riders, (c) neglect of the endogeneity and nonlinearity of the influence of service frequency on ridership, (d) ad-hoc ways to address the overlapping of catchment areas among closely spaced stops, (e) inadequate efforts toward quantifying the impact of service reliability on ridership and revenue, and (f) limited efforts on understanding the role of subjective perceptions of service quality on transit usage (and the extent of usage). The overarching goal of this dissertation is to advance the formulation of bus transit ridership models to address the methodological and substantive issues discussed above. To this end, first, the study develops a direct demand modeling framework for forecasting bus transit ridership at a spatially disaggregate stop-route level. The proposed model is applied to a rich empirical dataset put together from various sources, including the bus ticket sales and service data from Bengaluru Metropolitan Transport Corporation (BMTC). Empirical findings from the model indicate that while service frequency is one of the significant drivers of transit ridership even after controlling for endogeneity, routes with low current frequency (1 to 3 buses per hour) in Bengaluru benefit the most from increasing frequency than those with high current frequency. In addition, the results suggest that ignoring endogeneity between service frequency and ridership would result in overestimating the impact of service frequency on ridership. Next, the study expands the above direct demand modeling framework to examine both theoretically and empirically the relationships between transit ridership, service frequency (or headways), and service variability (variability in headways) using a tri-variate equations framework that considers endogeneity relationships between demand and supply. In addition, auxiliary supply models are developed to understand the determinants of service frequency and service variability in Bengaluru. The proposed models are applied for policy analyses to quantify the ridership impacts of service modification strategies, such as an increase in service frequency and an improvement in service reliability, on bus transit ridership in Bengaluru. Empirical results from these models demonstrate that headway variability negatively impacts transit ridership and passenger-kilometres, with the adverse effect exacerbating as variability increases. This trend is in contrast to the diminishing benefits of increasing service frequency at high current frequency levels. Further, the empirical results and policy simulations indicate that transit agencies can potentially gain greater ridership and revenue if they can lower headway variability rather than by merely adding more buses to high-frequency routes. Finally, the study develops a disaggregate, integrated choice and latent variable model of individual-level transit usage and extent of usage to complement the direct demand modeling frameworks discussed above. Specifically, the model considers the influence of individuals’ demographics and their subjective perceptions of transit service quality on the usage and extent of usage of bus transit. The role of subjective perceptions is incorporated using a generalized heterogeneous data modeling framework that treats individuals’ perceptions as latent explanatory variables identified using Likert scale measurements of the perceptions. In doing so, the study addresses the methodological issue of disentangling individuals’ unfamiliarity from their informed opinions underlying the Likert scale measurements of individuals’ subjective perceptions. The proposed model is applied to examine individuals’ transit usage and the extent of usage using empirical data from Bengaluru, India. The empirical results shed light on the role of individuals’ demographic characteristics and perceptions of transit service quality on their usage and extent of usage of bus transit in Bengaluru. Additionally, comparing the empirical model with another model that did not consider individuals’ unfamiliarity with transit service highlighted the importance of separating the influence of unfamiliarity from that of informed perceptions of service quality on transit usage. To the best of our knowledge, this dissertation is the first attempt to (a) develop a comprehensive demand modeling framework for analyzing bus transit ridership in an Indian city using empirical data from a large-scale transit system, (b) address several methodological and substantive issues identified in existing bus transit ridership studies, (c) examine both theoretically and empirically the endogeneity and non-linear effects of two important transit service attributes – service frequency and variability in headways – on demand for public transit, (d) shed light on individuals’ use and extent of use of bus transit in an Indian city while also focusing on the role of their subjective perceptions, and (e) address the issue of separating unfamiliarity from informed opinions when analyzing the role of subjective perceptions on individuals’ transit usage. Although the application of the models in this dissertation is for Bengaluru, the proposed methodological frameworks and empirical strategies are potentially applicable for studying bus transit demand in other cities in India and other countries. Specifically, the models can be used as descriptive tools to understand the factors that influence bus transit ridership, and as predictive tools to forecast bus ridership in response to a variety of what-if scenarios of interest to transit agencies – as long as empirical parameters are estimated using relevant data from the city of interest. In this context, it will be useful to apply the models to other cities to examine their applicability and the generalizability of substantive findings.