Modeling hurricane evacuation behavior using a dynamic discrete choice framework

Predicting evacuation-related choices of households during a hurricane is of paramount importance to any emergency management system. Central to this problem is the identification of socio-demographic factors and hurricane characteristics that influence an individual’s decision to stay or evacuate. However, decision makers in such conditions do not make a single choice but constantly evaluate current and anticipated conditions before opting to stay or evacuate. We model this behavior using a finite-horizon dynamic discrete choice framework in which households may choose to evacuate or wait in time periods prior to a hurricane’s landfall. In each period, an individual’s utility depends not only on his/her current choices and the present values of the influential variables, but also involves discounted expected utilities from future choices should one decide to postpone their decision to evacuate. Assuming generalized extreme value (GEV) errors, a nested algorithm involving a dynamic program and a maximum likelihood method is used to estimate model parameters. Panel data on households affected by Hurricane Gustav, which made landfall in Louisiana on 1 September 2008, was fused with the National Hurricane Center’s forecasts on the trajectory and intensity for the case study in the paper.