Predicting Household Participation in Housing Choice Voucher and Federal Housing Administration Mortgage Insurance Programs

Predicting Household Participation in Housing Choice Voucher and Federal Housing Administration Mortgage Insurance Programs

Stephanie Hawke, Economist, Housing Finance Analysis Division, Office of Policy Development and Research, and Mariya Shcheglovitova, Extension Assistant Professor, University of Vermont

This article is the third in a series examining HUD's homeownership and rental assistance programs. The first two articles compared household-level demographics among program participants and compared neighborhood-level data to understand the geographic distribution of households participating in these programs. This article identifies the key household- and neighborhood-level drivers of program participation.

Whereas most existing empirical work considers these topics separately, this series highlights the value of examining federal rental and homeownership assistance programs together. The tenant-based housing choice voucher (TBV) program is the largest and longest-running rental assistance program in the nation, and the Federal Housing Administration (FHA) mortgage insurance program extends credit to borrowers who may not qualify for traditional mortgage products, helping remove barriers to homeownership facing underserved communities. Analyzing combined TBV and FHA data reveals that HUD programs have a broad national reach, serving all 50 states, the District of Columbia, and Puerto Rico and more than 85 percent of counties and census tracts. Taken together, the two programs are present in most U.S. census tracts, helping people afford rental housing and become homeowners. Analyzing FHA and TBV data together, however, also reveals differences in the populations and localities these programs serve.

Our first article discussed our findings that the TBV program primarily serves extremely low-income renters, with Black renters making up the largest race/ethnicity group served, and that, although FHA-backed mortgages serve significantly more minority homebuyers than traditional U.S. mortgage products do, most of the homebuyers in this program are White.

Although income targeting means that federal rental and homeownership assistance programs serve different groups, these programs also reflect the persisting societal racial, geographic, and economic disparities in access to quality, stable, and affordable housing and opportunities to build intergenerational wealth through homeownership. The 2024 State of the Nation's Housing Report from the Joint Center for Housing Studies of Harvard University indicated that wealth gaps between homeowners and renters have widened over the past year, and racial wealth gaps and racial differences in homebuying opportunities have persisted. Locational trends in federal rental and homeownership program participation also reflect these disparities; our second article describes how these program trends mirror broader national trends for owner- and renter-occupied households. FHA-backed mortgage recipients live in wealthier suburbs with higher homeownership rates and shares of White residents, whereas TBV recipients live in poorer, more racially concentrated urban areas where rental households are more prevalent.

This article presents results from a statistical model that uses the individual demographic and census-tract-level variables from our previous two articles to understand the factors that drive participation in the TBV rental assistance program and FHA-backed mortgage program. We begin by describing the data and our approach to the analyses, building on our work in the previous articles, and then present the model results. We conclude by considering future research directions and policy interventions to encourage housing stability and homeownership.

Data

This article analyzes household participation in two major housing programs: HUD's TBV program and the FHA mortgage insurance program. We constructed a dataset that integrates household-level data from HUD's internal TBV and FHA mortgage insurance datasets with socioeconomic characteristics at the census tract level from the American Community Survey (ACS) 2022 5-Year Estimates. This approach allowed us to identify both household- and tract-level factors that correlate with participation in either the TBV or FHA program. For more information about the dataset construction and our approach to operationalizing these variables, please see the first and second articles in this series.

Table 1 summarizes statistics for variables at the household and census tract levels included in the model. Model variables include all household and neighborhood indicators described in the previous articles in this series except for adjusted annual household income.

Table 1. Model Variables

		Mean	Standard Deviation
Dependent Variable
Program	FHA participation*	0.42	0.49
	TBV participation*	0.58
Independent Variables
Household	Black head of household*	0.33	0.47
	Hispanic head of household*	0.15	0.35
	White head of household*	0.49	0.50
	Age of head of household	48.13	15.95
	Number of dependents	0.88	1.34
	Gender of head of household*	0.67	0.47
Census Tract	Percent of residents, Black	0.19	0.24
	Percent of residents, Hispanic	0.21	0.24
	Percent of residents, White	0.59	0.00
	Poverty rate	0.17	0.12
	Percent of occupied units, Renter	0.21	0.14
	Tracts that are suburban*	0.31	0.46
	Tracts that are urban*	0.57	0.49
	Tracts that are rural*	0.12	0.32
	Population (in thousands)	4.15	0.00

*This variable is included in the model as a binary indicator. For example, the two program variables are binary response indicators of program participation; participants in FHA mortgage insurance are coded as 1, and participants in TBV are coded as 0, with the means giving the respective total shares of households in each program in our dataset. Suburban, urban, and rural are binary indicators of whether the census tract falls within a county matching each designation. Gender of head of household is coded 0 for male and 1 for female.

Note: The tract population variable was normalized by dividing the total population by 1,000.

Methods

To explore the relationship between program participation and the theoretical drivers identified in previous studies, we employed a two-step regression approach. This method integrates household- and census-tract-level analyses, allowing us to assess how broader neighborhood characteristics influence individual household decisions and outcomes.

Step 1: Analysis at the Census Tract Level

In the first step, we conducted a linear regression to predict the percentage of FHA and TBV program participants within each census tract. The dependent variable in this model is the proportion of FHA participants relative to the total program participants in the tract.

Independent variables include the following:

• Percent Black, Hispanic, and Asian residents: Measure of racial/ethnic composition.
• Poverty rate: Indicator of socioeconomic disadvantage.
• Percent renter-occupied units: Proxy for housing market characteristics.
• Urban/rural indicator: Captures geographic and contextual differences.
• Normalized tract population: Controls for tract size and density differences.

This regression identifies the tract-level predictors of program participation while leaving some unexplained variation in the form of residuals. These residuals represent unobserved factors influencing the distribution of program participants and are carried forward into the next step.

Step 2: Analysis at the Household Level

The second step builds on the tract-level results by incorporating the residuals from Step 1 into a logistic regression model predicting the likelihood of individual households participating in either the FHA or TBV program. The dependent variable is binary, indicating whether a household participates in the FHA program (1) or the TBV program (0).

Independent variables in this model include the following:

• Household characteristics:
  • Number of dependents.
  • Race/ethnicity of the head of household (Black, Hispanic, or Other).
  • Age and gender of the head of household.
• Tract-level residuals from Step 1: Captures unobserved neighborhood-level influences not directly measured in Step 1.

This two-step approach allows us to separate the effects of neighborhood characteristics from household-level determinants, offering a comprehensive view of the drivers of program participation. Table 1 summarizes the predictor variables, and figure 1 illustrates the conceptual framework of the two-step regression approach.

Figure 1. Two-Step Regression Model

Results and Discussion

The results from our two-step model indicate a complex and durable relationship between federal rental and homeownership program participation and our theorized variables. In this section, we explore our findings for each step.

Table 2. Step 1 Census-Tract-Level Regression, Predicting the Percentage of FHA Households in a Census Tract

	Coefficient	Std Error	P>|z|	95% conf interval
Percent Black	-0.022	.0012	0.000	[–.0244, –.0193]
Percent Hispanic	-0.071	.0012	0.000	[–.0739, –.0689]
Percent Asian	-0.683	.0031	0.000	[–.6892, –.6772]
Poverty rate	-0.553	.0029	0.000	[–.5587, –.5472]
Percent renter	-1.624	.0024	0.000	[–1.629, –1.620]
Urban*	-0.037	.0006	0.000	[.0354, .0379]
Rural*	-0.097	.0009	0.000	[–.0623, –.0586]
Population	0.026	.0002	0.000	[.0251, .0258]
Constant	0.822	.0011	0.000	[.8203, .8248]

Note: N = 2,154,373; R² = 0.3659.

The results from Step 1 yield an R² of 0.3659, indicating that these tract-level characteristics can explain approximately 37 percent of the variance in the distribution of FHA versus TBV participation across tracts. This finding suggests that local context shapes a substantial portion of the variation in program participation.

The Step 1 findings reveal an inverse relationship between the proportions of minority populations within a tract and the percentage of FHA households, and this relationship is consistent across Black, Hispanic, and Asian groups. Similarly, tracts with higher poverty rates tend to have lower rates of FHA participation. We observe that urban tracts have higher FHA participation rates than do suburban areas, whereas rural tracts have lower FHA rates than do suburban tracts.

One of the more striking findings is the strong negative association between renter density and FHA participation, with a 1 percent increase in renter rate corresponding to a 1.6 percent decrease in FHA participation, supporting the findings in our previous article. Figures 2 and 3 illustrate the relationship between renter density and TBV and FHA participation; the maps show the clustered concentration of both federally assisted renters (figure 2) and renters overall (Figure 3), whereas homeowners show more broadly dispersed spatial patterns overlapping with and surrounding areas of high rental density. This finding may reflect how housing choice in the United States is constrained by complex interactions among individual economics, historic neighborhood development patterns, local municipal development and zoning policies, and established structures of racial and economic segregation and disinvestment. As others have observed, these interactions can sustain a widening racial wealth gap and perpetuate access to credit and homeownership for middle- and upper-income White households at the expense of other households.

Figure 2. Dot Density Map of Participation in Rental and Homeownership Assistance Programs

Figure 3. Dot Density Map of Renter- and Owner-Occupied Households

Note: Dots are placed randomly within each census tract and do not represent specific locations of households. TBV and FHA data are from HUD's internal databases. Housing tenure data is from 2022 ACS 5-year estimates. Dot densities are scaled differently to account for the different sample sizes between the ACS and the sample of households participating in federal rental and homeownership assistance programs. In Figure 2, one dot represents 30 households; in Figure 3, one dot represents 300 households.

The Step 2 model, which includes the census-tract-level residuals from Step 1, yields a pseudo R² of 0.7521, indicating a strong model fit and suggesting that including both household and tract-level predictors effectively captures the variance in program participation. This finding suggests that program participation is a complex interaction between household characteristics and location. In essence, we need to understand the people and the place to understand program participation.

Table 3. Step 2 Household Level Logistic Regression Model To Predict Program Participation

	Coefficient	Std Error	P>|z|	95% conf interval
Black head of household	–2.140	.008	0.000	[–2.156, –2.123]
Hispanic head of household	–0.752	.009	0.000	[–0.772, –0.735]
Other race head of household	0.518	.013	0.000	[0.770, 0.821]
Age of head of household	–0.039	.0002	0.000	[–0.039, –0.039]
Number of dependents	–0.187	.003	0.000	[–0.185, –0.174]
Gender of head of household	–1.335	.007	0.000	[–1.493, –1.467]
Census tract residuals	11.738	.021	0.000	[11.767, 11.849]
Constant	1.826	.015	0.000	[3.384, 3.453]

Note: N = 2,153,299; pseudo R² = 0.7521.

The Step 2 results indicate that Black and Hispanic households are significantly less likely to participate in FHA (or more likely to participate in TBV) than White households are. A significant negative relationship also exists between the age of the household head and FHA participation, indicating that older-headed households tend to participate in the TBV program. In addition, households with more dependents are more likely to be TBV participants, as are female-headed households.

We find that the tract residual variable in the Step 2 model is significant, indicating that unexplained variation from the census-tract-level model (Step 1) is a predictor of program participation in the household-level model. This finding suggests that characteristics at the census tract level, such as local policy and housing market conditions, are affecting the distribution or use of TBVs and FHA-backed mortgages and are not captured by our model.

Household-level variables show distinct trends in who participates in federal rental and homeownership programs. TBV households are more likely to be older; have a female, Black, or Hispanic head of household; and have a higher number of dependents than do FHA households. Tract-level variables also show distinct trends, the most notable of which is the relationship between race and program participation as well as the share of renter- and owner-occupied households in a census tract.

Conclusions and Policy Considerations

This series of articles compares HUD's TBV program and FHA-backed mortgage program to describe their historical context and demographic and geographic trends. Our final article in this series highlights the challenges of pinpointing an easily identifiable set of drivers for program patterns. Instead, our model suggests potential unmeasured tract-level characteristics — such as specific local policies or housing market conditions — that may influence the distribution of FHA versus TBV participation. Federal rental and homeownership assistance programs are evolving to meet the needs of low-income tenants and homeowners, and HUD research highlights the impacts of these program interventions.

• HUD has explored a series of neighborhood mobility programs intended to offer resources for voucher holders to move to higher-opportunity neighborhoods. HUD's newest mobility demonstration program, the Community Choice Demonstration, is expected to enroll more than 15,000 participants through 2028 in PHA-supported programs that help voucher holders access and remain in higher-opportunity neighborhoods and increase landlord participation in the HCV program.
• At the state and city level, HUD research has shown that municipalities with source of income discrimination laws exhibit higher voucher utilization rates and some improvements in neighborhood conditions for voucher holders.
• HUD's Housing Choice Voucher (HCV) Homeownership Program allows HCV program participants to transition their rental vouchers into monthly assistance toward mortgage payments. The administration of homeownership vouchers, however, is a direct trade-off with TBV vouchers; a homeownership voucher takes the place of a renter subsidy, and less than 0.5 percent of vouchers currently are used for homeownership.
• FHA has shown that including positive rental history (PRH), which typically is not reflected in credit scores, increases access to credit for first-time homebuyers, borrowers with less wealth, and Black and female borrowers. FHA has incorporated PRH into their Technology Open to Approved Lenders (TOTAL) Mortgage Scorecard. As of August 31, 2024, more than 6,000 endorsements that ordinarily would have required manual review were accepted through TOTAL thanks to the inclusion of PRH.

Future studies can build on this research to better understand strategies that help federally assisted tenants access quality, safe, and affordable housing in neighborhoods with amenities and opportunities and increase the number of pathways to homeownership.