The importance of effect measure modification when using demographic variables to predict evacuation
INTRODUCTION: Previous studies have identified a number of demographic characteristics (e.g., age, income, level of education, household composition, and race or ethnicity) that affect hurricane evacuation. However, the magnitude and direction of these associations vary widely, even when the area of landfall or the intensity of the storm is similar. We propose that the associations of demographic characteristics and hurricane evacuation are modified by psychosocial factors such as social cohesion, social capital, and social control. Additional variability may be the result of the changing prevalence of these demographic variables over time or between study locations. METHODS: Ninety census blocks in three eastern North Carolina counties affected by Hurricane Isabel were selected probability proportionate to population and seven interviews were conducted at random locations within each of the selected blocks. Risk differences (RD) and 95% confidence intervals (CI) were produced for stratified data to test for heterogeneity. RESULTS: There was statistical evidence of effect measure modification on the additive scale of the effect of home type, homeownership, age, race, gender, marital status, and having children under age 18 living at home on hurricane evacuation based on Wald p-values of the interaction terms of ≤ 0.20 and strata-specific RDs which crossed the null value. Social cohesion, volunteerism, property preparation, church attendance, neighbor's evacuation, and the number of local friends and family modified the RDs for the demographic characteristics. CONCLUSIONS: The associations between demographic characteristics and hurricane evacuation failure are modified by social factors. Effect measure modification on the additive scale may help explain the inconsistency of previously published results and is the appropriate measure for targeted interventions that can increase evacuation among certain groups that are missed when average risks are calculated across the population.