Predictors of hurricane evacuation decisions: A meta-analysis

Authors: Shakhawat Tanim*, University of South Florida, Brenton Wiernik, University of South Florida , Steven Reader, University of South Florida, Yujie Hu, University of Florida
Topics: Hazards and Vulnerability, Transportation Geography, Human-Environment Geography
Keywords: evacuation model, meta-analysis, hurricane, evacuation decisions
Session Type: Virtual Paper
Day: 4/9/2021
Start / End Time: 3:05 PM / 4:20 PM
Room: Virtual 7
Presentation Link: Open in New Window
Presentation File: No File Uploaded

Extensive research has investigated determinants of hurricane evacuation decisions using logistic regression models, logit models, and correlation analyses. Meta-analysis reviews of these models have generally been limited to narratives that only discuss the identified statistically significant predictors. However, if the results from previous models are going to be used for purposes of predicting how many and which people evacuate in a hurricane, then a statistical meta-analysis is required to determine useable "effect sizes" for the factors considered. To our knowledge, only one such study currently exists (Huang et al., 2016), but that study had some methodological limitations. This study attempts to address methodological limitations of current statistical meta-analysis, performing a statistical meta-analysis of 33 hurricane evacuation models for 22 predictors while including moderator analysis. The mean effect sizes of each predictor were reported in the result with the prediction interval. Predictors mobile home, evacuation order, hurricane category, evacuation plan, and disabled person's presence showed a strong positive effect on evacuation decisions. Other predictors, age, window protection, and race, showed a strong negative effect. Moderator analysis found that models with larger mean effect sizes for real hurricane models showed substantially larger effect than hypothetical hurricanes. Models considered in Florida were found to have a more consistent effect compared to other states' models, for most cases. Finally, models with a higher number of covariates tend to make the effect sizes smaller than models with a lower number of covariates.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login