Research Brief: How Equitable Are Disaster Recovery Programs?

Study shows that some county level social indicators are predictive of disaster recovery fund allocations, even when controlling for damage

Research briefs are my summaries of recently published peer-reviewed research on disasters and climate change that I think are especially important for disaster and climate policy. These summaries are based on my own analysis of the science - all mistakes are mine. Please be sure to download the original article, linked below.

Reference: Emrich, Christopher T., Sanam K. Aksha and Yao Zhou (2022). Assessing distributive inequities in FEMA’s disaster recovery assistance fund allocation. International Journal of Disaster Risk Reduction 74. Full-text is available here.

Topic and Research Question(s): The U.S. disaster recovery system is partly based on the distribution of aid to qualifying households. Decades of research shows that disasters tend to have unequal impacts, with ‘socially vulnerable’ households suffering greater losses. A key question is whether our system of federal aid helps to correct these imbalances, or is recovery aid also unequally distributed?

This paper examines the distribution of aid from FEMA’s Individuals and Households (IHP) program, which in one part of the Individual Assistance (IA) program that “provides financial and direct services to eligible individuals and households affected by a disaster, who have uninsured or under-insured necessary expenses and serious needs.” IHP assistance is meant to meet basic needs of households and to supplement other forms of recovery assistance. The IHP program includes such assistance as funds for temporary housing, providing a temporary housing unit, funds to repair or replace owner occupied homes, funds to repair or replace personal property and vehicles or assist with medical, child care, and funeral costs, and others - depending on the nature of the disaster and state approval.

This paper looks at the distribution of IHP housing assistance and ‘other needs’ assistance (e.g. childcare expenses, medical and dental expenses, funeral and burial expenses, damage to vehicles, damage to household items) and the relationship to different indicators of social vulnerability. It asks three research questions:

1. Are social indicators of vulnerability associated with receipt of lower amounts of disaster recovery assistance when controlling for damage?

2. Which social vulnerability indicators are most influential in predicting IHP assistance?

3. Do links between social vulnerability indicators and recovery assistance funding indicate distributive inequities?

The paper uses the computed data inputs to Cutter et al.’s (2003) social vulnerability index (SoVI), which is a widely used and robustly tested indicator set that currently includes 28 variables representing 6 vulnerability concepts: employment structure, housing, population structure, race/ethnicity, socioeconomic status, and special needs. These variables are measured using data from the U.S. Census, ultimately providing a geographical understanding of social vulnerability at the county-scale.

Data and Methods: This paper draws on two major sources of data:

1. Data on disaster declarations and the IHP program¹ from the Open FEMA Dataset for the years 2010-2018; and

2. Data on social vulnerability indicators from the U.S. census 5-year ACS datasets.

The authors created a composite county-level dataset of federally declared disasters during the study period (2010-2018)² and created several impact and recovery support variables from the IHP dataset. They then joined these to social vulnerability variables and used Ordinary Least Squares regression to identify relationships between social vulnerability variables and IHP assistance variables.

Key Findings:

The paper presents results for year-by-year models and all-year models. There are too many results to summarize adequately here, so I will focus on a few key findings from the all-year (combined) models that relate social vulnerability factors to 1) total IHP assistance, 2) housing repair and replacement assistance, and 3) other needs assistance. Be sure to read Table 4 in the paper, which includes all of the model results by variable. The R-squared value (the percent of model variance explained by social vulnerability indicators) varied: the adjusted R-squared value for total assistance was .482, for housing repair and replacement assistance was .471, and for other needs assistance was just .269.

• Percent Hispanic, Percent Native American, and Percent without access to automobiles are significant across all three models

• Percent Black, Percent Hispanic, and Percent Native American are negatively associated with total IHP approved per applicant and repair/replace per applicant

• Percent without access to an automobile was positively associated with total IHP assistance and housing repair and replacement, but negatively associated with other needs assistance

• The amount of elderly (65+) and young (under 5) variable was negatively associated with total assistance and housing repair/replace

• Percent renters and percent mobile homes was negatively associated with housing repair/replace but positively influenced other needs assistance

• The share of households who are cost-burdened is negatively associated with total IHP approved and other needs assessment, as well as variables associated with per-capita income and wealth households

So, there is a lot going on here.³ On some measures, the programs are leading to systematically inequitable distributions - for example, counties with large shares of Native American residents tend to receive systematically less recovery aid than those with smaller shares. On others, though, there are important achievements in equity, as defined by the authors. For example, counties with higher shares of Hispanic residents tend to receive higher shares of other needs assistance funds, and variables like percent female population and lower education predict higher levels of assistance.

What Questions Does It Raise For Me?

1. The most important follow-on question is, why? For the measures that showed systematic differences in receiving fewer (or greater) recovery resources, why is that the case?Are the programs themselves biased? Is it an issue of bias in the application and screening process? Differences in household ability to apply for aid, or appeal rejections? Bias in the allocation of recovery funds based on disaster event?

2. Clearly the models are telling different stories based on individual variables. What happens when they are combined? For example, does a county with relatively better-off non-white populations fare better than those with less-well-off non-white populations?

3. One of the key limitations of most uses of the SoVI, in my view, is the county-level aggregation of populations that obscures hugely important intra-county differences in population geographies, housing, etc. I would love to see a follow-up of this study that chooses representative counties and disaster declarations and pursues qualitative or mixed-methods research that explains the outcomes observed in the model. Would additional place-level variables help to explain the variance not attributed to social vulnerability indicators?

4. In the past several years FEMA has made some important commitments to program equity, which it defines as “The consistent and systematic fair, just and impartial treatment of all individuals.” FEMA has also made changes to how it defines eligibility, which has resulted in over 100,000 more households receiving assistance in FY 2021 than would have otherwise. If this analysis were re-run to include 2021 and 2022 disaster events, would the results change?

5. Finally, is this the right way to measure the research question, “are federal disaster recovery programs equitable?” It is certainly a unique research design that benefits from a rich literature and method associated with SoVI. But can it get to the heart of the question? I’m eager to hear from readers what they think.

1. It is important to note that this paper only focuses on home owners, because “renters are automatically ineligible for repair and replacement assistance - the largest component of total IHP assistance.”

2. The model assessed the interactions of variables across all years 2010-2018 except 2013-2014, because of the low numbers of total applicants which negatively impacted statistical power.

3. From the paper: “The three “All Year” model outputs show that social vulnerability variables such as percent Native American, percent African American, percent housing burdened popuations, percent social security benefit recipients, and percent age-dependent population (under 5 and over 65 years) have systematically received lower amounts of Individual Assistance recovery support (total IHP, housing repair/replace, and ONA). Meanwhile, in these models, variables such as percent rich, per capita income, percent female population, percent less than 12th grade education, percent service sector employment, percent unemployment, and total damage per applicant have equitably received federal individual assistance. Several variables, including percent Hispanic population, percent mobile homes, percent renters, and percent without automobiles have depicted mixed equity across the models (Fig. 4). In these instances, at least one model depicts inequitable distribution of recovery resources while other show equitable distribution when controlling for total damage per applicant.”