Introduction

Population growth and development in coastal regions of the highly vulnerable U.S. Atlantic and Gulf Coasts have increased the likelihood of substantial damages from disasters like tropical storms and hurricanes. The number of deaths associated with hurricanes and tropical storms has declined dramatically over the last century due in part to improved forecasting and advanced warning systems that alert residents to evacuate to prevent catastrophic consequences. Therefore, factors that predict hurricane evacuation have been studied extensively. ^{Reference Huang, Lindell and Prater1-Reference Dow and Cutter4} Individuals typically make evacuation decisions in the context of actual and perceived risks as well as social and economic constraints. While some predictors of evacuation are consistent across studies, many predictors have had mixed associations with evacuation.

In the first major review of evacuation studies Quarantelli, ^{Reference Quarantelli5} developed a model of evacuation decision-making that included community context, threat conditions, social processes, patterns of behavior, and community preparedness based on a dozen large scale, random-sample population surveys of communities impacted by floods, tropical storms, hurricanes, tornados, and manmade accidents. In the next review, ^{Reference Baker2} Baker demonstrated through analysis of a large database constructed from surveys conducted following 12 hurricanes that made landfall in the U.S. between 1961 and 1989 that there were no consistent associations between demographic factors and evacuation. Peacock, Marrow, and Gladwin, ^{Reference Peacock, Morrow and Gladwin3} used data from Hurricane Andrew, a major Category 5 hurricane that made landfall near Miami, Florida in 1992, and found mixed results when assessing the association between race and evacuation. When demographic and household variables were included in their models and indicators of ‘risk’ excluded, African-Americans and Hispanics were less likely to evacuate than whites. However, when risk indicators were included, no statistically significant differences in evacuation were observed by race. In addition, household size and the presence of children or elderly in the household were negatively associated with evacuation. The inconsistencies in the association between demographic factors and evacuation were explored by Horney et al., ^{Reference Horney, MacDonald, Willigen and Kaufman6} who found that social factors could act as modifiers of the relationship between demographic variables and evacuation.

Mobile home residence has been consistently positively associated with evacuation across multiple studies. Dow and Cutter found that the most significant determinant of hurricane evacuation behavior is a personal perception of risk, which may be the mechanism by which the association between mobile home residents and evacuation operates. ^{Reference Dow and Cutter4} Individuals who live in mobile homes likely know their homes are not sturdy enough to withstand hurricane impacts. Families that reside in multi-unit buildings were more likely to evacuate than those who lived in single-family homes, who may remain to protect their homes from flooding or looting. ^{Reference Dow and Cutter4} In contrast, Horney, et al. found no association between individual actual or perceived flood risk and evacuation from Hurricane Isabel in 2003 or Hurricane Irene in 2011. ^{Reference Horney, Macdonald, Van Willigen, Berke and Kaufman7}

The assessment of the association between income and evacuation has also yielded mixed results in the observed literature, with some authors reporting a positive relationship, ^{Reference Peacock, Morrow and Gladwin3-Reference Noltenius9,Reference van Houwelingen, Arends and Stijnen21} and others an inverse association. ^{Reference Ng, Diaz and Behr10-Reference Smith and McCarty14} The justification for the latter finding has been that residents with higher incomes may perceive their homes to be of higher quality and able to withstand a strong storm or that they may be more likely to remain in their homes to protect valuables. A positive association between income and evacuation may be because wealthier residents may own higher risk coastal properties subject to waves and storm surges and may also evacuate relatively more easily due to their ability to finance expenses associated with evacuation.

Baker’s review demonstrated that dissemination of information could predict evacuation. ^{Reference Baker2} Residents of high-risk coastal areas may be more likely to evacuate when they receive evacuation information from non-media sources like governmental or other public officials or family and friends. They may also be more likely to evacuate when public officials are proactive about issuing evacuation orders. Peacock, Marrow, and Gladwin, ^{Reference Peacock, Morrow and Gladwin3} made similar observations following Hurricane Andrew. On the other hand, other studies have reported that media sources exerted a significant influence on hurricane evacuation decision. ^{Reference Stein, Dueñas-Osorio and Subramanian15} Stein, et al. studied the determinants of evacuation following Hurricane Rita, which made landfall in Galveston, Texas, in 2008 and observed that heterogeneity in evacuation was explained by the evacuation designation of their area of residence. Another study in Florida found that residents who live in officially designated evacuation areas responded differently to a set of information cues, incentives, and risk factors than evacuees who live outside of these areas. ^{Reference Peacock, Brody and Highfield16}

It has been hypothesized that racial and ethnic minorities are less likely to evacuate because of differences in social and family networks, risk perceptions, language and communication difficulties, and inadequate access to the resources required for evacuation. ^{Reference Fothergill, Maestas and Darlington17} However, some prior studies have reported lower evacuation rates in minorities ^{Reference Peacock, Morrow and Gladwin3} ; Some have found lower rates among some minority groups but not others ^{Reference Riad, Norris and Ruback18} ; While others have found no significant differences. ^{Reference Bateman and Edwards19}

Findings related to the association between length of residence in an area at-risk of hurricanes and evacuations are also mixed. While some authors found a positive association, ^{Reference Baker2,Reference Horney, Macdonald, Van Willigen, Berke and Kaufman7} (e.g., longer-term residents were less likely to evacuate), others found a negative or no association with evacuation behavior. ^{Reference Peacock, Morrow and Gladwin3} Length of residence could impact evacuation decision in various ways. Newer residents could decide to evacuate because of their heightened perceived risk, but they could also choose not to evacuate because of their inexperience with risk in the area. On the other hand, long-term residents could decide to evacuate because of prior experience with hurricane landfalls in the area, or choose not to, because of their perception that the area is safe (e.g., false expectations paradox). ^{Reference Baker2}

Overall, numerous studies across different fields have been published assessing associations between demographic and other factors and evacuation in response to approaching hurricanes. ^{Reference Peacock, Morrow and Gladwin3,Reference Stein, Dueñas-Osorio and Subramanian15,Reference DeYoung, Wachtendorf, Farmer and Penta20} Except for a few of these factors, associations with evacuation reported in the literature are highly variable, with studies reporting positive, negative, or null associations. ^{Reference Horney, Macdonald, Van Willigen, Berke and Kaufman7,Reference Huang, Lindell, Prater, Wu and Siebeneck8,Reference Ng, Diaz and Behr10,Reference van Houwelingen, Arends and Stijnen21} Previous summary studies conducted on this subject were predominantly narrative reviews, which often yielded invalid estimates by not weighting average effect sizes. ^{Reference Huang, Lindell and Prater1} Hence, this study aims at enumerating factors that predict evacuation to increase the efficiency and effectiveness of evacuation in hurricane emergencies. It may also be reasonable to assert that determinants of hurricane evacuation differ across different disaster events of dissimilar intensity or location. Primary study results may vary by region impacted (e.g., Atlantic or Gulf Coast), the severity of the hurricane (e.g., Category on the Saffir-Simpson scale), evacuation zone, ^{Reference Stein, Dueñas-Osorio and Subramanian15} or year of publication. ^{Reference Greenland22} For instance, the year of publication may impact study results if research methods, hurricane severity, or study participants changed over time. In order to identify factors that explain heterogeneity of evacuation decision in hurricane conditions, this study aimed at investigating the moderators of evacuation across studies. The isolation of moderators may enhance disaster planning by emphasizing specific at-risk groups that could be targeted for evacuation communications. ^{Reference Blair, Burg and Foran23} Information on sources of heterogeneity may also inform the meta-researcher on the combinability of primary studies, whether summary statistics should be reported in a meta-analysis, and if so, how they should be produced (e.g. by reporting stratified estimates across levels of the moderator). ^{Reference Field and Gillett24}