ANALYZING REDISTRIBUTION OF FEDERAL DISASTER AID THROUGH MACHINE LEARNING
Publication or External Link
Natural disasters are on the rise and will be costly for both the United States government and its citizens. The record-breaking year of 2020 left $1 billion worth of damages in the wake of twenty-two different events (FEMA, 2023). As costs due to disasters increase in the coming decades, the livelihoods of all citizens, especially those most vulnerable are at risk. It is known that natural disasters exacerbate current standing social vulnerabilities and inequities. Federal disaster aid programs in place are intended to assist those who cannot solely finance their own recovery efforts. This study looks to analyze FEMA’s Public Assistance (PA) program, Individual Assistance (IA) program, and Hazard Mitigation Assistance (HMA) program. It is important that these systems put in place are distributing federal resources as intended because they are funded via people’s taxpayer dollars. This study looks to explore the relationship between disaster aid that is awarded at the county level with respect to the federal income taxes residents of that county pay to the federal government. This is expressed through the creation of the donor-donee ratio. This study also contributes to the literature a new metric of burden, the ratio of expected annual disaster losses of a county and its gross domestic product, which can beutilized as a proxy for coping capacity. The burden metric provides additional useful insight as it is tabulated by FEMA directly and published in their National Risk Index (NRI) at the county level. Over the last decade, this research examines the donor-donee ratio and burden metric over the years 2010 to 2019. Results of mapping the donor-donee ratio and burden metric indicate there is spatial heterogeneity between counties in the United States. The redistribution of federal aid is not only heterogeneous but there are distinct regional patterns where further research could investigate their causality. To investigate the relationship between the redistribution of aid and coping capacity by proxy, this study utilized supervised machine learning to characterize counties. Significant outcomes of the machine learning indicate that most counties across the country received moderate funding and were evaluated as having a moderate burden as well. This does suggest that to some level the redistribution of aid is working as intended. Although upon further digging, it was found that counties that experience high-cost, less frequent events, contain over 50% of the country’s population and lie in metropolitan areas. Upon the application of a logistic regression model, it was found that these counties while associated with higher income, are also associated with higher mobile homes residence. As the risk of these higher costs events increases over the years, it is imperative that vulnerable communities are receiving adequate funding to increase their resilience to future hazards. This study highlights the flows of federal disaster dollars and where these programs allocate funding.