Causal Inference of Longitudinal Behavioral Responses to Mobility Interventions

Abstract

Public Transport Networks (PTN) are complex dynamical systems in which supply and demand continuously interact. Mobility interventions are introduced to address market failures and critical operating conditions, such as congestion, yet their behavioral effects are often difficult to evaluate causally. The Box-Tiao model provides a useful econometric framework for intervention analysis, but offers limited insight into causal mechanisms and heterogeneous responses over time. This study proposes a flexible multivariate causal learning framework based on graphical models to dynamically estimate intervention effects at both the population and individual levels. The framework generalizes the classical Box-Tiao model, which can be recovered as a marginal case. We apply the approach to the public transport fare incentive program introduced in Hong Kong in 2014. The results show that the proposed method can identify intervention effects, estimate the associated parameters, and uncover the direct influence of user characteristics on incentive adoption.