Jointly modeling bus and taxi ridership: A Bayesian multivariate spatial analysis accounting for cross‐ridership correlation

Abstract

AbstractUnderstanding the spatial patterns of transit ridership and their ascribable factors is critical for urban and transportation planning. Existing studies predominantly examine a single type of transit ridership but neglect important cross‐ridership correlation. This study aims to develop a Bayesian multivariate spatial modeling approach for jointly analyzing multiple ridership. In particular, cross‐ridership correlation is estimated using correlated spatially unstructured residuals, correlated spatial random effects, or the combination of both. The proposed approach is applied to a case study of Wuhu, China, where the correlation between bus and taxi ridership is accounted for in the modeling of transit ridership and their built environment determinants. Results show that accounting for cross‐ridership correlation significantly improves model‐fitting, but the fashion used to induce such a correlation matters. Furthermore, the role of built environment in determining transit ridership could be over‐estimated without considering cross‐ridership correlation. This study contributes to statistical modeling of transit ridership by applying a spatial multivariate approach, which can improve our understanding of the underlying processes of multiple ridership, their interactions, as well as their associations with physical environments.