Skip to Content

College of Arts & Sciences
Hazards & Vulnerability Research Institute

Analysis of Evacuation Behaviors and Departure Timing for October 2016's Hurricane Matthew


Advisor:  Dr. Susan L. Cutter



During a hurricane, an individual’s risk perception does not remain static. Spatial and temporal variations of a hurricane will shift perceptions of risk, and complicating this dynamic are information-seeking processes increasingly reliant on the individual’s self-motivated interpretation of information sources. Initial evacuation resistance or willingness could change even after evacuations are ordered, affecting evacuation preferences and departure times. Because Hurricane Matthew’s continually shifting track had virtually nonstop coverage, evacuation decisions were also being modified as residents either grew more or less certain of their safety.

This research investigates the evacuation behaviors associated with Hurricane Matthew in October of 2016. It assesses the relationships between selected variables acquired from an online survey and evacuation departure times – generally speaking, the differences in evacuation behaviors between Florida, Georgia, and South Carolina survey respondents. Descriptive statistics are provided for several variables of interest, followed by spatiotemporal analysis of evacuation departures using Esri ArcGIS® software’s space-time pattern mining tools. To assess the relationship of a subset of variables and evacuation during the entire study period (5 days), a binary logit model is estimated, and subsequently, to investigate the relationship of several variables and evacuation by day, four additional binary logit models are estimated and discussed.

Results indicate that approximately 62% of the Florida sample, 77% of the Georgia sample, and 67% of the South Carolina sample evacuated. Under the logistic model analyzing the overall time period, an observed evacuation depended on not having prior hurricane experience, having received an evacuation order, an increased level of communication about the evacuation order, believing one lived in an evacuation zone, not knowing if one lived in an evacuation zone, having fewer pets in the household, and higher household income in 2015. However, the state (FL, GA, SC) was the only variable which was significant for each of the four days’ logistic regressions. Essentially, for each day, there were consistent differences in evacuation between the states.

Some of the results concerning the influence of some variables on evacuation behavior have been confirmed in past research, while others continue to emphasize the case-specific nature of every hurricane event through inconsistently influential variables. The descriptive results provided makes clear that a number of variables are taken into consideration in a household’s decision to evacuate, but the results from the subsequent analyses highlights that an authoritative evacuation order is the primary triggering variable. The results also show the significance of the state variable on coordinating an evacuation at the household level, which stresses the criticality of recommending protective action in a timely manner.


PDF icon Thesis