What predicts hurricane evacuation decisions? The importance of efficacy beliefs, risk perceptions, and other factors

With growing hurricane risks along U.S. coastlines, motivating evacuation among high-risk populations remains critical yet challenging. Some residents in danger zones do not or cannot evacuate, leading to preventable fatalities and injuries (e.g., storm surge drownings during Hurricane Ian). Prior research identifies many influences on evacuation (messages, demographics, experience, capacity/barriers, social influences, risk perceptions), but findings vary across studies due to differing contexts, measures, and operationalizations of constructs like experience and risk perception. Theories such as Protection Motivation Theory, the Extended Parallel Process Model, and the Protective Action Decision Model posit that protective behaviors are shaped by risk (threat) perceptions and efficacy beliefs (response and self-efficacy), alongside potential non-protective responses. This study examines, in a controlled hypothetical hurricane scenario, the relative importance of multiple situation-specific risk perceptions (cognitive and affective) and efficacy beliefs, and how these interact, compared with general hurricane-related factors. Research questions: RQ1—How do different types of situation-specific risk perceptions and efficacy beliefs predict evacuation intentions? RQ2—How do situation-specific risk perceptions and efficacy beliefs interact in influencing evacuation intentions? RQ3—How do situation-specific perceptions/beliefs compare with general hurricane-related factors as predictors of evacuation intentions?

The article reviews theoretical frameworks—Protection Motivation Theory, Extended Parallel Process Model, and the Protective Action Decision Model—highlighting roles for cognitive risk perceptions (likelihood, severity), affective responses (fear/worry), response efficacy (belief that an action reduces harm), and self-efficacy (ability to take action). Meta-analyses in health contexts show both risk and efficacy messages affect protective actions, with some findings that efficacy beliefs can exert stronger effects than risk perceptions. In climate adaptation and other hazard contexts, efficacy has sometimes been a dominant predictor of protective behaviors. Hurricane studies generally find risk perceptions associated with evacuation decisions, but measures vary widely (perceived hazard characteristics, likelihood/severity, impacts/safety, affect). Capacity and barriers (transportation, shelter, funds, work, pets, disabilities) are also important. Fewer studies have isolated the distinct roles of self- versus response-efficacy in near-term hazard threats like hurricanes, motivating the present focus on multiple dimensions of risk perceptions and efficacy beliefs.

Design and sample: Online survey (2015) of 1716 residents in coastal Connecticut, New York, and New Jersey, areas affected by Hurricane Sandy (2012). Sampling used GfK’s probability-based panel supplemented with non-probability opt-in recruitment within ZIP codes selected via NWS Maximum of MEOW (MOM) storm surge risk for a Category 2 hurricane. At the time of the survey, 61.7% lived in storm surge risk areas and 19.5% in the 100-year floodplain; 12.9% reported evacuating prior to Sandy. Survey structure: After initial questions on general hurricane-related perceptions and experiences (e.g., perceived residence in flood/evacuation zone, evacuation plan, prior home flooding, evacuated for Sandy, emotional distress), respondents completed four randomized experimental modules presenting different hypothetical coastal storm/hurricane scenarios. This article analyzes one module featuring a strong approaching hurricane (landfall in two days; up to 130 mph winds). Experimental messages: Within this focal module, respondents were randomly assigned to combinations of three message conditions (Hazard, Impact, Fear). All respondents also received uniform information about recommended behavior: evacuate if in an evacuation zone, with options (hotel, friends/family outside zone, shelter). Measures (situation-specific within the module): Intended behavioral response (evacuation intentions, 1–7). Cognitive risk perceptions—overall threat (likelihood home affected, severity at home, likelihood of getting hurt if staying home, all 1–7) and hazard-specific threats (strong winds, storm surge flooding, rain flooding; yes/no). Affective risk perceptions: worry and fear (1–7). Efficacy beliefs: self-efficacy to evacuate and response efficacy (evacuating reduces harm to self/family), both 1–7. Other responses: perceptions that information is overblown/misleading (1–7 agree–disagree). General factors: Perceived residence in flood/evacuation zone (yes/no/don’t know), evacuation plan (yes/no), past home flooding (yes/no/don’t know), evacuated for Sandy before landfall (yes/no), emotional distress due to Sandy (1–4). As a proxy for general propensity to evacuate, the analysis also used respondents’ average evacuation intentions across the other three modules (mean 4.66, SD 1.64). Analysis: Multiple linear regression models (SPSS v28.0.1.1) predicting evacuation intentions. Model 1 included individual/household controls (age, gender, race/ethnicity, education) and message condition indicators. Models 2–5 added sets of situation-specific variables (overall cognitive risk perceptions; hazard-specific threats; affective perceptions; efficacy beliefs). Model 6 included all situation-specific variables together. A parsimonious Model 6a retained only significant situation-specific predictors from Model 6. Interactions among key cognitive risk perceptions and efficacy beliefs were tested (Model 6b). Models 7–8 compared general hurricane-related factors to situation-specific predictors in explaining evacuation intentions; Model 9 added the average evacuation intentions across other modules. Collinearity checks via VIFs showed acceptable levels (generally <3.0). Pearson and Spearman correlations were computed; direct effects were the primary focus.

• Message conditions and demographics alone explained little variation (Model 1 adjusted R² ≈ 0.02).
• Situation-specific cognitive risk perceptions of overall threat (likelihood home affected, severity at home, likelihood of getting hurt) added substantial explanatory power (Model 2; +0.58 adjusted R²; total ≈ 0.60).
• Hazard-specific threat perceptions (winds, storm surge, rain) and affective perceptions (worry, fear) each predicted evacuation intentions when entered alone with controls (Models 3–4; adjusted R² ≈ 0.19–0.20), but were not significant when overall cognitive risk and efficacy variables were included together (Model 6, p = 0.07–0.94), suggesting shared or mediated variance.
• Efficacy beliefs were the strongest predictors: adding self- and response-efficacy (Model 5) yielded adjusted R² ≈ 0.67. In the full situation-specific model (Model 6; adjusted R² ≈ 0.73), the strongest predictors (unstandardized coefficients) were: response efficacy (0.51***), likelihood of getting hurt (0.20***), severity at home (0.20***), self-efficacy (0.16***), and likelihood home affected (0.06*).
• Interactions (Model 6b; adjusted R² ≈ 0.73): • Self-efficacy × Response efficacy positive (≈ 0.048***): self-efficacy mattered more when response efficacy was moderate–high; when response efficacy was low–moderate, self-efficacy had limited influence. • Likelihood of getting hurt × Response efficacy negative (≈ −0.050***): either high perceived harm if staying or high response efficacy was sufficient to elevate evacuation intentions; when both were low, intentions were low. • Visual analyses (heatmaps/boxplots) showed some respondents with low risk perceptions still had high response efficacy and high evacuation intentions—counter to some theoretical expectations that low risk appraisals preclude protective action consideration.
• General vs situation-specific predictors: General hurricane-related factors (Model 7) explained modest variance (adjusted R² ≈ 0.16). When combined with situation-specific variables (Model 8; adjusted R² ≈ 0.73), only perceived residence in a flood zone and having evacuated for Hurricane Sandy remained significant; others (evacuation plan, perceived evacuation zone, prior home flooding, Sandy distress) were no longer direct predictors.
• Propensity to evacuate: Adding average evacuation intentions from other modules (Model 9; adjusted R² ≈ 0.75) showed this proxy significantly predicted intentions (0.21***), and response efficacy remained a strong predictor (albeit somewhat reduced), suggesting response efficacy reflects both general and situation-specific beliefs about evacuation effectiveness.
• Non-protective responses: Perceiving information as overblown/misleading was weakly negatively correlated with evacuation intentions and with efficacy, but overall correlations were small. Regression models for these perceptions had very low explained variance (adjusted R² ≈ 0.04). Data did not support a strong pattern of high-risk/low-efficacy leading to reactance or boomerang effects.

Findings indicate that in a near-term hurricane threat scenario, situation-specific response efficacy—belief that evacuating will reduce personal/family harm—and perceived likelihood of personal harm if staying home are the dominant drivers of evacuation intentions. These variables outperformed affective responses (worry/fear), hazard-specific threat perceptions (wind/surge/rain), and general hurricane-related perceptions/experiences. Interactions showed self-efficacy matters primarily when response efficacy is at least moderate; and either high perceived harm or high response efficacy can elevate intentions. Notably, some respondents with low risk perceptions still reported high response efficacy and high intentions to evacuate, suggesting risk appraisals may not always precede efficacy appraisals in this context and that experience or anticipation of evolving risk might sustain high response efficacy. These insights refine applications of PMT, EPPM, and PADM for approaching natural hazards. Practically, risk communication that clearly conveys potential harms and the effectiveness of evacuation may be more impactful than fear arousal per se and should be complemented by strategies that bolster self-efficacy and reduce evacuation barriers. Although results derive from a hypothetical scenario, high explained variance suggests partial generalizability to real events, with the caveat that actual behavior will be constrained by real-world capacities and impediments.

This study advances understanding of hurricane evacuation decision making by comparing multiple dimensions of risk perceptions and efficacy beliefs and their interactions, alongside general hurricane-related factors, within a controlled hypothetical scenario. The main contributions are: (1) identifying evacuation-related response efficacy and perceived likelihood of personal harm if staying as the strongest predictors of evacuation intentions; (2) showing that situation-specific perceptions/beliefs explain substantially more variance than general experiences/attributes; (3) clarifying interactions indicating the conditional role of self-efficacy on response efficacy and that high response efficacy can motivate evacuation even with lower risk perceptions. Implications for risk communication include emphasizing specific potential harms and how evacuation reduces them, and implementing interventions that increase capacity and self-efficacy. Future research should: measure both general and situation-specific efficacy more comprehensively; examine barriers/costs and their links to self-efficacy; test communication strategies that raise response efficacy over time; use longitudinal designs to assess causal pathways; and evaluate generalizability across diverse populations and hazard contexts.

• Hypothetical scenario: Intentions may differ from real-world behaviors; affective responses and practical impediments likely play larger roles during actual events.
• Cross-sectional survey: Causality cannot be inferred; temporal sequencing of risk and efficacy appraisals is unresolved.
• Sample/context: Coastal CT/NY/NJ residents several years post–Hurricane Sandy; generalizability to other regions/populations and hazards may be limited.
• Measurement gaps: Did not directly measure evacuation costs/impediments; few respondents reported low self-efficacy, limiting power to assess its effects; response efficacy was measured only for the focal module (not across all scenarios); social/collective decision processes were not examined.
• Operationalization variability: As with prior literature, constructs like risk perception encompass multiple correlated elements; potential mediated pathways were not tested explicitly.