Integrating Open Government Data and Causal Machine Learning to Evaluate the Effectiveness of Four FEMA Flood-Mitigation Investments Across County Contexts

Flooding is among the costliest U.S. natural hazards. Across the harmonized 1990–2020 records used here, FEMA distributed approximately $26 billion in flood mitigation grants to reduce future losses. FEMA evaluates proposed mitigation projects through benefit-cost analysis (BCA), but reported BCAs are not standardized across counties, since they depend heavily on protected property values and documentation capacity rather than a mitigation category’s historical effectiveness. This study measures how the effectiveness of FEMA-supported flood mitigation varies across county contexts to show which viable mitigation categories perform better under which local conditions. Using 1990–2020 FEMA funding records linked to subsequent flood losses (via SHELDUS) across 506 high-risk counties, we apply Double Machine Learning with Mundlak adjustment and year fixed effects to control for baseline county differences, causal forests, which repeatedly split counties based on their characteristics to estimate how effects differ across contexts, and K-means clustering to group counties by response pattern. Negative estimates indicate lower subsequent flood losses and range from −0.052 to −0.070 on the study’s log(1 + x) scales, but the size of that association varies systematically across county contexts: Structural Flood Control estimates are more negative in exposed, lower-income rural counties; Property Acquisition estimates in lower-density rural counties with high floodplain development and expected loss; and Critical Infrastructure Protection and Preparedness/Capacity Building estimates in counties with greater exposure, insurance participation, and newer housing. Study validation included placebo tests, future-treatment lead tests, and model stability diagnostics. This study offers a reproducible framework for accurately estimating effectiveness measures across mitigation categories. It complements FEMA’s existing process by providing context-specific historical evidence for counties, helping FEMA agencies make grant allocation decisions more effective and equitable.