Modeling post-disaster return mode choice of evacuees following large-scale evacuations: A national stated-preference survey-based analysis using machine learning and causal inference

Disasters caused by natural hazards pose significant threats to life and property, often requiring evacuation. While evacuation behavior has been widely studied, return mode choice during the re-entry phase following large-scale natural disaster evacuations remains understudied. This study addresses that gap using a national stated-preference survey of over 1000 U S. residents, modeling return mode choice with logistic regression, tree-based ensemble models, and propensity score-based causal inference. Three returnee groups were analyzed: (1) all returnees, (2) those with personal vehicle access, and (3) those not using personal vehicles. For the first two groups, awareness of emergency public transit services was associated with a lower likelihood of returning by personal vehicle. For the third group, such awareness increased the likelihood of using public transportation. Personal vehicle access was the strongest predictor of return mode, though not all vehicle owners chose to drive—especially among low-income individuals. Other influential factors included evacuation destination, delayed return due to safety concerns, prior disaster experience, and disaster type. Among those without vehicle access, individuals from households with elderly, disabled, or medically vulnerable members were less likely to use public transportation, suggesting concerns about accessibility. These findings underscore the need for inclusive return-phase transportation planning and improved awareness of transit services. These insights highlight the critical need to incorporate return-phase planning into emergency transportation strategies, with particular attention to vulnerable populations. Enhancing public awareness of available transit options may be a cost-effective and high-impact approach to encouraging modal shifts and alleviating traffic congestion during post-disaster recovery.