Evacuation behavior analyzed using discrete choice theory
In a combined research of the TU Delft and UC Berkeley the behavior of evacuees of Hurricane Irma was modelled and analyzed empirically using a variety of choice models, estimated on actual evacuation choice data.
The United States experiences several natural disasters per year. When major hurricanes are predicted, often large scale evacuations have to be carried out. Local governments do not always have the resources and knowledge to forecast evacuee behavior and to make a solid plan for safely and effectively evacuating vulnerable inhabitants of the area. Basic questions such as who evacuates, whereto, when, and using which route are very hard to answer reliably due to lack of effective evacuation models.
Researchers from the Faculty Technology Policy & Management and the Faculty of Civil Engineering & Geosciences hosted a researcher from UC Berkeley and together built and estimated a model that provides reliable answers to these questions.
They used a unique dataset of choices made by evacuees fleeing Hurricane Irma. The innovation of the choice model which they developed, is that it jointly captures correlations on all relevant choice dimensions. Examples of such choice dimensions are route choice, departure time choice and destination choice. One single evacuation choice is actually a bundle of many such dimensions, and failing to understand their mutual dependencies leads to biased predictions. For example, if a model fails to capture that evacuees with further away destinations are more likely to leave early, predictions will be off.
The developed portfolio choice model indeed highlighted and captured a myriad of dependencies in the data. By doing so, it can help governments to make more accurate predictions, and accommodate evacuees by for example placing shelters at the right locations at the right moment.
Authors paper: Stephen D.Wong, Adam J.Pel, Susan A. Shaheen and Caspar G.Chorus
