抄録
Bangladesh’s coastal belt experiences recurrent tropical cyclones, yet most evacuation research has emphasized whether households evacuate rather than when they depart. This study examines evacuation timing during Cyclone Remal (26 May 2024) among 215 households that moved to designated cyclone shelters in Khulna and Bagerhat districts. Evacuation start time is modeled as an ordered outcome with three categories (during/within 1 hour before landfall, 1–3 hours before landfall, and 3–6 hours before landfall) using hierarchical ordinal logistic regression. Explanatory variables are organized into six domains: social structure, social context, warning, perceived threat, constraints, and shelter facilities and management. The results show that earlier evacuation is primarily associated with lower income day labor households, warnings issued within 6 hours & 6-12 hours in advance, larger cattle ownership, stronger threat perception, prior government assistance, and credible shelter management. By contrast, perceived false alarms and reliance on walking to shelters are linked to later departure. These findings suggest that national level emergency agencies responsible for cyclone risk management should treat warning quality, assistance and shelter management as core components of evacuation policy.