Quantifying service-reliability-based day-to-day evolution of travel choices in public transit systems with smart transit card data

This study develops day-to-day models to explore how unreliability of public transit service affects the day-to-day evolution of travel choices. We consider two dynamical processes that incorporate transit service unreliability, i.e., travelers' learning and perception updating process (LPUP) and proportional-switch adjustment process (PSAP). The conditions for existence, uniqueness, and stability of the fixed point of each model are analytically derived. These conditions are then examined using real-world public transit data from the Greater Sydney area. We find that with some aggregations, the system stability conditions are satisfied in both models. The observed weighted average flow change between two successive days is around 6.5% over the observation period, which may reflect the system stochasticity rather than instability. Among a series of empirical findings, it is noteworthy that in the Sydney case, the value of service schedule delay is around 3.27 times that of in-vehicle time.