Multi-Hazard Exposure and Recorded Disaster Impacts in Indonesian Villages: Associations with a Village-Level Mitigation Indicator

Abstract

Multi-hazard exposure may place cumulative pressure on local communities, but national village-level evidence remains limited. This cross-sectional study examined associations between the number of hazard types recorded in Indonesian village-level administrative units and two 2020 disaster-impact outcomes: affected residents and affected agricultural land. It also assessed whether a composite mitigation indicator modified the exposure-impact association. Data came from the 2021 Village Potential Statistics (Podes), a national census covering 84,096 village-level administrative units. Multi-hazard exposure was constructed from 11 recorded hazard types, while the mitigation indicator summarized five preparedness and response measures. Ordinary least squares models with heteroskedasticity-robust standard errors were estimated for natural-log-transformed outcomes, and Poisson regression was used as a count-data sensitivity analysis. Multi-hazard exposure was positively associated with affected residents (b = 1.301, p < .001) and affected agricultural land (b = 0.721, p < .001). The mitigation indicator was also positively associated with the outcomes. In the interaction model, the exposure-by-mitigation coefficient was positive (b = 0.166, p = .002). Under the reported coding, this coefficient means that the estimated exposure-impact slope increased as the mitigation indicator increased; it is therefore inconsistent with a protective buffering interpretation. Positive mitigation coefficients may reflect risk-based placement of preparedness measures, omitted village characteristics, reporting capacity, or the indicator's emphasis on the presence rather than the effectiveness of measures. The findings support village-level multi-hazard screening but do not establish causal effects or mitigation effectiveness. Population- and area-normalized outcomes, regional and spatial controls, count-model diagnostics, and longitudinal or quasi-experimental designs are required for stronger policy evaluation.