Transferability of Hurricane Evacuation Choice Model: Joint Model Estimation Combining Multiple Data Sources

Abstract

Inferences on hurricane evacuation behavior are usually drawn through developing empirical models. These models are estimated using data that are specific to a given hurricane context. One important issue, therefore, is whether such models are applicable to different hurricane contexts. This paper investigates this transferability issue of evacuation choice models across different hurricanes. Initially, we estimate three separate models of the binary decision to evacuate or not, using data sets from three hurricanes (Andrew, Ivan, and Katrina) that occurred at different periods. Then a joint model is estimated combining these three evacuation data sources. When estimating the model jointly, the differences among the scale parameters of the data sets are specifically accounted for. The results from joint and separate models are then statistically tested to evaluate whether evacuation decision model parameters are transferable across different hurricane contexts. The result from the statistical test suggests that the parameters of the evacuation choice models are transferable over different hurricane contexts in similar regions, an important implication for policymakers and emergency preparedness agencies.