Twin Cities Bus Ridership: A Spatial Bayesian Analysis

Transit agencies often use simple models to predict transit demand. In this study, we explore a series of models for Metro Transit’s bus ridership in the Twin Cities region of Minnesota which potentially improve upon the current industry standard. Specifically, we implement Bayesian models which incorporate (1) demographic predictors; and (2) spatial components which are not often included in transit demand models. We consider regularized horseshoe priors to determine which demographic predictors are most relevant to ridership but ultimately do not include the shrinkage priors. We incorporate spatial structure using the BYM2, a reparameterization of the classical Besag-York-Molli ́e spatial Bayesian model. We present an application of the BYM2 to 2017 Metro Transit bus ridership. The BYM2 is an areal data model which uses an Intrinsic Conditional Autoregressive (ICAR) prior and allows us to build a more nuanced understanding of bus ridership. We discuss specific interpretations of the BYM2 for Metro Transit and the benefits of statistically rigorous models at transit agencies.