Housing Assistance and Housing Insecurity: A Study of Renters in Southeastern Michigan in the Wake of the Great Recession
Abstract
This article examines the factors shaping longitudinal patterns of housing insecurity in the wake of the Great Recession, with a focus on whether housing assistance helped renters who received it. We use data from the first two waves (2009–10 and 2011) of the Michigan Recession and Recovery Study, a population-representative sample of working-age adults from Southeast Michigan. We use detailed reports from renters and other nonhomeowners to construct measures of instability and cost-related housing problems at both waves, and we compare the changes in these over follow up between housing assistance recipients and their income-eligible but nonrecipient counterparts. Our findings suggest that receiving housing assistance reduced the chance of experiencing housing insecurity problems over follow up regardless of baseline housing insecurity.
Introduction
Prior studies show that housing insecurity further disadvantages those who are already economically marginalized (Desmond 2012b, 2016; Pattillo 2013). Recent studies document the far-reaching influence of housing insecurity on the reinforcement and reproduction of contemporary economic inequality (Wildeman 2014). For example, over time housing insecurity severely undermines stable employment (Desmond 2016; Desmond and Gershenson 2016), erodes informal employment networks that could reduce spells of unemployment (Ziersch and Arthurson 2005), and curtails access to institutional and informal supports that mitigate material hardship (Greenbaum et al. 2008; Keene and Geronimus 2011). While scholarly understanding of the negative effects of housing insecurity on individuals is growing, there are still some important limitations to the evidence. In particular, we know relatively little about the change in housing insecurity that individuals experience or about how programs designed to reduce housing insecurity affect housing histories.
Understanding whether housing assistance helps to reduce housing insecurity for poor Americans is important in the contemporary context of the slow recovery from the largest macroeconomic recession in a generation. A key risk factor for housing insecurity is lack of housing affordability, and the recent recession exacerbated a long-term rise in the fraction of lower-income renters living in unaffordable housing (Collinson 2011; JCHS 2011). Moreover, very-low-income renters appear to have not benefited much from the economic recovery in the years immediately following the official end of the Great Recession in 2009. The number of severely cost-burdened households—those spending more than 50 percent of their income on housing—increased by 20 percent to a record high of 8.5 million US households between 2009 and 2011 (JCHS 2011).
Housing policies and programs directed at low-income people aim to alleviate the uneven distribution of access to secure housing between the poor and the nonpoor. Among other things, housing assistance from one of these programs can directly stabilize recipients’ ability to pay for their housing consistently by reducing their actual rental cost (Collinson, Ellen, and Ludwig 2016). For example, scholars have suggested that housing vouchers are effective in reducing housing insecurity, including shelter use and residential instability, among families who have been homeless or are at risk of homelessness (Culhane 1992; Wong, Culhane, and Kuhn 1997; Shinn et al. 1998; Gubits et al. 2015). However, recent studies of housing insecurity among other populations at heightened risk, including single mothers who received welfare benefits and unwed urban parents, do not find significant effects of housing vouchers on subsequent housing insecurity (Geller and Curtis 2011; Geller and Franklin 2014; Wildeman 2014; but see Phinney 2013). These recent studies are valuable because they offer a view beyond the experiences of those with the most severe housing problems; much of the research on the determinants of housing insecurity and recovery focuses on those who have already been homeless (e.g., Metraux and Culhane 1999; Zlotnick, Robertson, and Lahiff 1999; Fisher et al. 2014; Parsell, Tomaszewski, and Phillips 2014). However, even these newer studies capture only more economically disadvantaged parts of the population, and they exhibit several methodological limitations, including narrow measures of housing insecurity (but see Geller and Franklin 2014) and lack of an appropriate comparison group (i.e., most studies compare voucher recipients to all nonrecipients, even when those who did not have a voucher may not have been eligible). More studies are needed that examine the association between receiving housing assistance and subsequent change in housing insecurity and that consider broader populations of renters.
In this study, we examine the association between housing assistance receipt and change in housing insecurity among renters in one large metropolitan area in the wake of the Great Recession. We build on prior literature by developing a comprehensive measure of housing insecurity that incorporates multiple types of involuntary moves and an indicator of delayed rental payments, broadening our scope beyond those people who are already showing evidence of housing insecurity to include people who are at risk of involuntary moves. We use data from a population-based sample of individuals residing in southeastern Michigan that were collected over two survey waves in late 2009–early 2010 and in 2011, and we capture the change in housing insecurity spanning these years. We also examine the link between housing assistance receipt and change in housing insecurity while accounting for the potential influence of receiving cash-based forms of public assistance.
Literature Review
Measuring Housing Insecurity
In past research, scholars have used a variety of measures to indicate housing insecurity, from a simple measure separating those who are physically housed from those who are not to more detailed measures of housing status that capture unstable living arrangements (Eastwood and Birnbaum 2007; Rebholz, Drainoni, and Cabral 2009), length of stay at current residence (Coley et al. 2013), or recent involuntary moves (Pavao et al. 2007; Phinney et al. 2007; Reed et al. 2011). More recently, scholars have developed more comprehensive measures of housing instability that incorporate more than one dimension of housing insecurity (Kushel et al. 2001; Burgard, Seefeldt, and Zelner 2012; Rollins et al. 2012; Geller and Franklin 2014). For example, Sarah Burgard and colleagues (2012) measured multiple types of housing insecurity, including multiple moves, cost-related moves, doubling up, homelessness, being behind on rent, eviction, being behind on a mortgage, and foreclosure.
Scholars suggest that a comprehensive measure of housing insecurity that captures a wide spectrum of possible severity is more conceptually appropriate than measuring specific discrete events (Sosin, Pilliavin, and Westerfelt 1990; Kleit, Kang, and Scally 2016). They argue that housing insecurity manifests itself in the lives of low-income people as a complex, sequential pattern of multiple housing insecurity events over time rather than as discrete housing events that befall different subgroups. For example, less severe types of housing insecurity, such as a delay in rent payment or moving for cost reasons, can precede more dramatic events like eviction and homelessness. Those facing severe forms of housing insecurity also try to strategically avoid them by moving in with others to reduce housing cost burden (Pilkauskas, Garfinkel, and McLanahan 2014; Desmond 2016). Thus, a comprehensive measure of housing insecurity captures those who are experiencing varying degrees of housing-related hardship at some point in a dynamic process, distinguishing them from others who are not insecure in any way over the same period.
Although some prior measures of housing insecurity have successfully captured its multiple dimensions, most have only been single-point-in-time measurements, typically capturing whether respondents experienced any housing problems in the last year (Phinney et al. 2007; Geller and Curtis 2011; Geller and Franklin 2014; Wildeman 2014). If there were not much year-to-year variation in housing insecurity, then there would not be much to gain by interviewing respondents more than once in order to capture changes in housing insecurity. However, housing insecurity is a dynamic experience in the lives of low-income people (Edin and Shaefer 2015; Desmond 2016). For example, low-income workers often get behind on rent but then are able to get back on schedule as their employment status and work hours fluctuate. Measuring housing insecurity at only one point in time would miss this kind of volatility. Especially during and following economic recessions, the oscillation of the business cycle likely generates instability in employment status and working hours that could create rapid change in housing insecurity over relatively short intervals. Measuring housing insecurity over multiple time points may thus better capture important variation in such insecurity across low-income renters.
Recognizing that housing insecurity is dynamic, housing scholars have developed the concept of the housing pathway or career to examine how work and family domains (e.g., labor market status or relationship dynamics) influence housing insecurity. For example, William Clark, Marinus Deurloo, and Frans Dieleman (2003) examine the change in housing tenure and affordability over the entire life course and differential trajectories of tenure and affordability across income levels. One common housing career shows a pattern of upward mobility toward a stage of home ownership or higher-priced rental dwellings, while another distinctive housing career ends in low-priced rental units for those with very low household incomes and negative income growth (Clark et al. 2003). Although housing researchers have developed multiple concepts to capture longitudinal housing trajectories that unfold over the life course (May 2000; Clark et al. 2003; Skobba 2016), they have rarely examined relatively short-term housing trajectories or how they are influenced by the receipt of housing assistance. This limits our ability to understand the volatility of the lives of low-income renters and its implications for housing assistance. One exception is a study of a small sample of low-income mothers in a large Midwestern metropolitan area who reported on the type and duration of each housing accommodation they had experienced since they began living independently (Skobba 2016). Kim Skobba (2016) shows that receipt of a housing voucher partially alleviated housing insecurity induced by precarious employment and relationship disruption among these low-income renters. We build on this past evidence by using a larger survey sample of adults who represent the working-age population in a large metropolitan area and by generating measures of housing insecurity at two time points and considering the change between them, using detailed data that capture an array of housing insecurity experiences. We also assess the association between housing assistance receipt and change in housing insecurity, and we distinguish an appropriate comparison group of income-eligible nonrecipients whose demographic and socioeconomic characteristics are otherwise similar to those of housing assistance recipients.
Housing Assistance Programs and Housing Insecurity over Time
The United States operates many different types of housing assistance programs targeted to different populations. In general, recipients of housing assistance live in public housing or privately owned units that have received federal subsidies, or they receive a voucher to use on the housing market. From the 1930s through the early 1970s, public housing built and operated by local public housing authorities (PHAs) was the dominant form of low-income-targeted housing assistance. However, from the 1960s to the early 1980s, the federal government executed long-term contacts with for-profit and nonprofit developers and built privately owned subsidized housing that guarantees subsidies and imposes affordability restrictions on a certain number of units for a period of up to 30 years. The demolition of distressed public housing developments has gradually decreased the importance of public housing as a means of federal low-income housing assistance and has resulted in the reduction of total public housing stock by about 300,000 units over the past 20 years. Currently, the Housing Choice Voucher (HCV) program and privately owned subsidized housing (e.g., Section 8 Project-Based Rental Assistance [PBRA]) are the largest programs, assisting more than 3 million households (2 million through HCV; Collinson et al. 2016).
The HCV program in the United States (formerly known as the Section 8 Housing Voucher Program) was established after the Housing and Community Development Act of 1974. Housing vouchers aim to reduce housing insecurity among recipients by providing a significant amount of subsidy for rental payments. Voucher recipients contribute 30 percent of their income to housing costs, and the HCV program subsidizes the difference between that amount and the total allowed cost of rent, which is set annually by the US Department of Housing and Urban Development (HUD) as the Fair Market Rate (FMR) and further locally adjusted by public housing authorities (PHAs; HUD 2001). HUD requires local PHAs to set payment standards between 90 and 110 percent of the FMR for each unit size. PHAs can also set different payment standards for different parts of the FMR area. The FMR in 2010 for the metropolitan Detroit area, where respondents in our sample resided, was $796 for a two-bedroom unit. A hypothetical family of three whose income was at the upper bound of the extremely-low-income category ($18,850 for the Metropolitan Detroit area in 2010) was thus expected to contribute 30 percent of that monthly income ([$18,850/12] × 0.3 = $471.25 per month) toward rent. Their subsidy would not exceed the difference between the payment standard and 30 percent of their monthly income, so their maximum amount of subsidy in this scenario would be about $320 ($796 − $471.25). Since voucher recipients pay less out of pocket toward rent each month due to the subsidy, they are presumably less likely to fall behind on rent, be evicted or become homeless, or experience other forms of housing insecurity compared to their counterparts who are income eligible but not receiving a subsidy. Further, if a recipient’s income drops, the subsidy amount is adjusted upward, making it easier for recipients to weather income shocks than otherwise similar individuals not receiving assistance.
PBRA is the largest project-based rental assistance program in the United States. It serves more than 1.2 million low-income households. PBRA is tied to particular housing developments, and families cannot retain their rental assistance when they move to new locations. HUD makes an annual contract with private property owners to rent some or all of the units in the development to low-income households at an affordable rent. PBRA also provides a housing subsidy to the owner of the units that fills the gap between 30 percent of household income and the cost for operating and managing the contracted units. Although income eligibility for PBRA is set at having an income below 80 percent of Area Median Income (AMI), the median family income for an area estimated annually by HUD, federal regulation mandates that 40 percent of assisted units in the development are allocated to very-low-income households (i.e., income below 50 percent of AMI).
Although the substantial amount of subsidy that housing assistance programs provide should positively shape recipients’ housing security patterns, there is still limited evidence showing whether having housing assistance prevents recipients from developing new housing problems. On the one hand, because the subsidy is substantial, housing assistance recipients would appear to be protected from developing housing problems and would seem more likely to get out of trouble compared to those with similar resources but no housing assistance. However, there may be a further issue to consider that conditions the advantage of housing assistance recipients. The way that the HCV and PBRA system is structured means that program responses to recipients’ income changes could actually generate fluctuations in their housing security. Some scholars suggest that the housing assistance system only provides a limited safety net from unstable employment and associated income volatility during recessionary hard times (Ellen and O’Flaherty 2007; Collinson et al. 2016). This is because local public housing authorities annually recalculate the monthly subsidy for families by subtracting 30 percent of the estimated monthly income of voucher recipients from their total housing cost. Inaccurate annual income projections, which are more common for extremely-low-income persons (e.g., due to unstable working hours), could mean that in reality the financial burden imposed by their monthly contribution of rent might vary from month to month. In this scenario, a given month’s rent contribution could temporarily effectively increase to more than 30 percent of a housing assistance recipient’s actual income that month.1 Coupled with a low level of personal savings (Seefeldt 2015), as well as delays in processing changes to other benefits a recipient might receive (Seefeldt forthcoming), the structural lag in the housing assistance system’s response to a given recipient’s income decline might mean that they have to delay rent payments and face eviction. Moreover, housing assistance recipients cannot legally use the cost-sharing strategy of doubling up (Ellen and O’Flaherty 2007), which is relatively common among low-income families (Pilkauskas et al. 2014), although levels of enforcement of this rule may vary across local PHAs. For these reasons, it is possible that housing assistance recipients may not be more protected from volatility in their housing security than income-eligible families not receiving housing assistance.
Characteristics of Households Served and the Importance of an Appropriate Reference Group
Housing assistance recipients are demographically and socioeconomically distinct. To assess the association between a housing assistance receipt and housing insecurity over time, it is critical to obtain a comparison group with a set of characteristics similar to those of housing assistance recipients. Failure to account for factors that predict both housing assistance receipt and housing insecurity, such as income, participation in other means-tested social programs, or some key demographic characteristics, may lead to an overestimate of the influence of housing assistance. One of the eligibility criteria for participation in housing assistance programs is the applicant’s income level in relation to AMI. In order to be income eligible for HCV, for example, an applicant family’s income should be below 50 percent of the annually updated AMI after adjusting for family size, with some exceptional cases allowed to have incomes up to 80 percent of the AMI. However, housing assistance is not an entitlement in the United States, and in the case of housing voucher programs, only one-fourth of income-eligible families actually receive vouchers (Collinson et al. 2016). Thus, there are many income-eligible nonrecipients who could plausibly be used as a comparison group in examining the influence of housing assistance on housing insecurity changes. However, previous studies generally have compared housing assistance recipients to all nonrecipients in evaluating the effect of housing assistance on housing insecurity (Geller and Curtis 2011; Geller and Franklin 2014; Wildeman 2014), with the exception of experimental studies that have sampled respondents from among people who were participating in other means-tested social programs (Wood and Rangarajan 2004; Wood, Turnham, and Mills 2008). Past studies have partially addressed income differences between recipients and nonrecipients by including a measure of income in their models, but this may not be sufficient to create an appropriate comparison.
Additionally, it is reasonable to presume that many housing assistance recipients are also eligible for, and participate in, other means-tested social programs, since they meet the income-eligibility requirement for housing assistance programs. Means-tested benefit programs are a critical income stabilizer for the very-low-income population, and they help low-income people cope with typical day-to-day financial challenges, as well as with more serious recessionary hard times, since those with very low incomes have less access to credit and support from family (Harknett and Hartnett 2011; Desmond 2012a; Seefeldt 2015). Receipt of other means-tested benefits could also reduce housing insecurity by decreasing the chance of having an income shock and associated housing problems, including delayed rent payment, eviction, or homelessness (O’Flaherty 2009). In order to evaluate the influence of housing assistance alone, rather than the total influence of multiple means-tested social programs on housing insecurity trajectories, we adjust for cash assistance receipt in our analyses.
Finally, housing assistance recipients are demographically different from those who do not receive housing assistance. Using a population-based sample of urban low-income individuals, researchers find that while having a larger number of children is positively associated with having voucher-based housing assistance, being married reduces that likelihood (Park, Fertig, and Metraux 2014). Barbara Sard and Thyria Alvarez-Sánchez (2011) document demographic characteristics of HCV recipients that could make their risk of housing insecurity different from that experienced by nonrecipients. Although the share of vouchers going to families with children declined from 2000 to 2010, 52 percent of voucher households have children, while more than 20 percent of voucher recipients are elderly. To address these factors, we adjust for a set of demographic variables, including marital status, number of children, and age, and we also adjust for income and match on income eligibility.
Data and Method
Data
We use the first two waves of the Michigan Recession and Recovery Study (MRRS), a stratified, random sample of working-age adults drawn from the general population of the three counties (Macomb, Oakland, and Wayne) surrounding Detroit. We conducted wave 1 interviews between October 2009 and April 2010 with 914 respondents, with a response rate of 82.8 percent. We re-interviewed 847 respondents between April and August of 2011, with a response rate of 94 percent of those who completed wave 1 interviews. We limit our analytic sample to those who identified as renters or “others” (those who were not paying rent and did not have a mortgage or own their home; N = 421) and excluded four cases with missing data on independent and dependent variables used in multivariable analyses. This yields an analytic sample of 417.
The Detroit-Warren-Livonia Metropolitan Statistical Area (MSA), from which our sample is drawn, had a high rental vacancy rate of 17.7 percent, compared to the median of 9.1 percent among the 75 largest US MSAs (HUD 2015b). The Detroit metropolitan area is somewhat unique in that it has both high unemployment and a higher-than-average proportion of high-interest mortgages, which led to a higher foreclosure rate compared to other US MSAs during the recent recession (Dwyer and Lassus 2015). The high rental vacancy rate in our study area, however, could have absorbed some part of the foreclosure shock on the rental housing market. It is also noteworthy that there is considerable variation in the rental vacancy rate across the three counties in which our respondents resided. Wayne County, where 91.1 percent of housing assistance recipients in our study resided, had a vacancy rate of 14.5 percent, while Macomb and Oakland Counties had much lower rates at 7.0 percent and 8.3 percent, respectively (HUD 2015b). The HCV program and PBRA comprise more than half of all the HUD programs in Wayne County (43.3 percent and 33.6 percent, respectively), followed by public housing (15.8 percent), which is similar to the national level. HUD (2015a) data indicate that the percentage of blacks receiving any type of HUD program assistance was 78 percent, while the percentage was 85 percent when we only look at HCV recipients. For project-based rental assistance, the percentage of blacks was lower, at 70 percent (national averages were 43 percent, 47 percent, and 28 percent, respectively).
Measures
Dependent Variables
At wave 1, we asked about residential moves in the year prior, and at the wave 2 interview, we asked about moves between wave 1 and wave 2 (representing a period of about 17 months on average), as well as about reasons respondents gave for moving and about other types of housing insecurity experiences. We created a comprehensive measure of housing insecurity, using a detailed set of questions about housing and residential mobility (see Table A1 in the online appendix). Because rates of any one type of housing insecurity problem were low, as this is a population-based sample and not a sample of very disadvantaged people, we decided to aggregate in order to capture all individuals facing housing problems of a relatively serious nature, so that we could conduct multivariable analysis. In constructing this housing insecurity measure, we focused on separating what seemed to be voluntary from what apparently were involuntary residential moves (involuntary moves were defined as those that appear to have occurred for cost reasons), and we included information about whether respondents were behind on rent as an additional indicator of risk for housing insecurity, since it was likely caused by limited financial resources. At each wave, we classified respondents as housing insecure if they reported any moves for cost reasons; had completed foreclosure (only at wave 2, since we only considered renters at wave 1, but a small number became homeowners between waves); had experienced eviction, homelessness, or moving in with others to share expenses; or were behind on rent. Otherwise, the respondent was classified as being housing secure.
Key Predictors
In order to examine the difference in housing insecurity over both waves between housing assistance recipients and their income-eligible nonrecipient counterparts, while separating out income-ineligible respondents, we created a measure that combines information on housing assistance status and the federal income eligibility criterion of 50 percent of AMI. MRRS respondents were asked the following question about housing assistance status at wave 1: “Do you get any help on the monthly rent for this apartment or house from any federal, state, or city government housing programs, including any federal Section 8 certificate or voucher?” To narrow down possible types of housing assistance they were receiving, we retrieved physical addresses of subsidized housing stock from the Michigan State Housing Development Authority (MSHDA) website and compared those with respondents’ residential addresses at wave 1. This comparison suggested that our interview question on housing assistance status captured only non–public housing residents.
HUD annually estimates the median family income for an area and adjusts that amount for different family sizes. We used AMI in 2008 for the Detroit-Warren-Livonia MSA, which includes all three counties in our sample. The reference category in our analyses is housing assistance nonrecipients who meet the eligibility criterion of household income below AMI 50 percent. Thus, when we compare housing assistance recipients with this reference category, we can estimate the extent to which having housing assistance is related to subsequent housing insecurity in a population of similar individuals. When respondents with household incomes above AMI 50 percent are compared with the reference group, we can estimate the extent to which income below or above AMI 50 percent influences subsequent housing insecurity among those not receiving housing assistance.
Control Variables
Based on the prior literature, we adjusted for several variables in multivariable analysis, including receipt of any other social program (Temporary Assistance for Needy Families [TANF], Supplemental Security Income [SSI], Unemployment Insurance [UI]),2 race (black vs. nonblack), marital status (married or not), number of children, age, education level (more than high school vs. high school or less), and county of residence (Wayne vs. other). Adjusting for these characteristics helps to address potential selection bias in estimating the association between having housing assistance and experiencing housing insecurity between waves 1 and 2 of the MRRS. In order to control for varying durations between survey waves for different respondents, we included duration in months between wave 1 and wave 2 as a continuous variable. Additionally, some of those who reported receiving housing assistance at wave 1 of the MRRS reported not receiving it at wave 2 (15 of 69), and we coded these respondents as having lost housing assistance, while others who did not report housing assistance at wave 1 reported having it at wave 2 (14 respondents), and we coded them as having gained housing assistance at wave 2.
Analytic Strategy
We first conducted descriptive analyses to examine whether characteristics of housing assistance recipients differ from those of income-eligible nonrecipients or those who are not eligible for and do not receive housing assistance, and to explore the composition and pattern of housing insecurity at wave 1 and wave 2 of the MRRS. Afterward, we examined the prevalence of each type of housing insecurity by income eligibility and housing assistance receipt. We then estimated two multivariable models predicting housing insecurity at wave 2, while controlling for wave 1 housing insecurity. We used wave 2 weights in all analyses to adjust for the wave 1 sampling design and attrition by wave 2.
We first estimated a logistic regression model predicting housing insecurity problems at wave 2. In the second model, we estimated the association between having housing assistance and housing insecurity at wave 2 using propensity score methods. When trying to draw causal inference using observational data, the simple comparison of treatment group and control group can be problematic when the distribution of covariates associated with the outcome varies for treatment and control groups (Morgan and Winship 2014), here, housing assistance recipients versus income-eligible nonrecipients. To address imbalance in covariates between the two groups in observational studies, one can match respondents who are similar on their observed characteristics but who differ on the treatment variable and then assess whether having housing assistance was associated with a differential outcome over follow up. Since our major focus is on the comparison between voucher recipients and income-eligible nonrecipients, we have estimated the average causal effect of housing assistance receipt on housing insecurity at wave 2, considering only respondents whose household income meets the income eligibility criterion.
Since we have a relatively small number of cases, we have used propensity score weighting, which re-weights all the observations in our analytic sample with a propensity weight (Busso, DiNardo, and McCrary 2014). Propensity score weighting produces unbiased estimates when treatment selection depends on covariates included in the propensity score model (Morgan and Winship 2014). Among other contributing factors to receiving a housing voucher, federally mandated income targeting can be critical (AMI below 30 percent). We have included a dichotomous variable indicating whether a respondent’s income is below AMI 30 percent in the propensity score model predicting housing assistance receipt. We also include the same set of demographic variables as in the earlier regression models, including respondent’s race, marital status, number of children, age, education, and other social program participation. To further reduce bias, previous literature also recommends including variables related to the outcome even when they are not associated with treatment selection, since the purpose of the propensity score approach is to control imbalance in covariates associated with outcome (Brookhart et al. 2006; Austin 2011). Thus, we have also included months elapsed between waves of the MRRS in the propensity score model. We also included survey weights as a predictor in the propensity score model, since they can additionally capture place of residence, demographic characteristics, and variables related to the probability of survey response (DuGoff, Schuler, and Stuart 2014).
We first tried commonly used propensity weights, called inverse-probability treatment weights (IPTW), which give the inverse of the propensity score to treated respondents and the inverse of one minus the propensity score to observations in the comparison group. However, the IPTW not only created extremely high values of propensity score weights for some respondents, which is one of the caveats in IPTW (Morgan and Winship 2014), but, more importantly, significant imbalances in covariates remained after weighting. Thus, instead of IPTW, we have used overlap weights recommended by Fan Li, Kari Morgan, and Alan Zaslavsky (2016). The overlap weight of one minus the propensity score is assigned to a treated respondent, and the propensity score itself is assigned to respondents in the comparison group. Weighted means of covariates in the outcome model suggest a significant reduction in imbalance across covariates, as shown in Table A2 in the online appendix. We multiplied these overlap weights with the survey weight before using them in the propensity score weighted model so that our results could be generalized to the population of working-age adults in our study area (DuGoff et al. 2014). These procedures address potential confounding issues while still retaining the representativeness of the population-representative data.
In order to illustrate the practical significance of coefficients from these two different multivariable models, we also present the average marginal effect (AME) for each independent variable. Using the example of the black race coefficient, these AMEs were generated by first calculating a predicted probability for each respondent while treating them as though they were black, and then nonblack, while leaving all other independent variables at their actual values. Afterward, we estimated the AME by averaging the difference between the two predicted probabilities for each respondent (individual-level marginal effect) across all respondents (Williams 2012).
Results
Table 1 presents percentages or means for characteristics of the sample overall in the first column and then compares the characteristics of housing assistance recipients at wave 1 of the MRRS with those of income-eligible nonrecipients in the middle column and income-ineligible nonrecipients in the final column. We present p-values for t-tests of differences between groups, with significance denoted with asterisks and income-eligible nonrecipients as the reference group. Table 1 shows that housing assistance recipients were disproportionately likely to be African American (96 percent compared to 54 percent of income-eligible nonrecipients). They were more likely to participate in other social programs than income-eligible nonrecipients (73 percent vs. 45 percent) and were more likely to reside in Wayne County (91 percent vs. 68 percent).
| Overall | Income-Eligible Nonrecipients | Income-Eligible Recipients | Income-Ineligible Nonrecipients | |
|---|---|---|---|---|
| Number of observations | 417 | 198 | 69 | 150 |
| Wave 1 SES and demographics: | ||||
| Respondent is black (%) | 39.8 | 54.0 | 96.0*** | 20.3** |
| Respondent is married (%) | 25.2 | 22.6 | 7.8 | 29.8 |
| Number of children in household | 1.0 | 1.1 | 1.5 | .8 |
| Respondent’s age (in years) | 35.3 | 35.8 | 39.8 | 34.2 |
| Respondent has some college experience (%) | 53.5 | 36.9 | 46.9 | 68.5*** |
| Wave 1 other social program participation | 33.4 | 45.3 | 73.1** | 18.0** |
| Months between waves | 17.3 | 17.5 | 17.0 | 17.2 |
| Wave 1 housing insecurity (%) | 39.1 | 43.6 | 36.6 | 35.6 |
| Housing assistance status change (%): | ||||
| Lost housing assistance | 21.7 | |||
| Gained housing assistance | 6.1 | 1.3 | ||
| Wayne County resident (%) | 55.2 | 68.1 | 91.1* | 39.5* |
Figure 1 shows the percentage of respondents who had experienced housing insecurity by type of housing insecurity problem. At wave 1, about 39 percent of respondents were experiencing or had recently experienced any of the problems comprising our housing insecurity indicator, with being behind rent the most common (18 percent), followed by moving in with others (15 percent), moving due to cost (14 percent), homelessness (4 percent), and eviction (3 percent). At the wave 2 interview, the percentage of respondents who had any housing problems fell to 34 percent, but the dip was not statistically significant, nor were there statistically significant changes in the prevalence of any specific housing problems. Population-weighted prevalence of each type of housing insecurity (n = 417). The percentage reported in the center of 95% confidence interval (CI) is a point estimate of the prevalence of each type of housing insecurity. Nine percent of respondents (n = 36) had become homeowners by the follow-up interview; they were asked the same set of housing insecurity–related questions except for those about eviction and being behind on rent because they were no longer eligible for those problems. Instead, they were asked about being behind on mortgage in the last year. Only renters or other nonhomeowners were asked about eviction at the follow-up interview. Wave 1 housing insecurity captures respondents’ statuses 1 year prior to the interview conducted between October 2009 and April 2010. Thus, wave 1 housing insecurity information was gathered in the period several months before and after the official end of the Great Recession. Wave 2 housing insecurity captures the period between the wave 1 and wave 2 interviews, with the average length of 17.3 months. “Any Housing Insecurity” indicates respondents who reported any of the specific problems illustrated in the figure.
The cross-wave comparison of the prevalence of housing problems in figure 1 summarizes housing insecurity at two points in time. An individual-level prospective view reveals more dynamic patterns of housing insecurity. Figure 2 shows that in the overall analytic sample, 66 percent of respondents reported either persistent housing insecurity (insecure-insecure category, 19 percent) or consistent security (secure-secure category, 47 percent) from wave 1 to wave 2. However, 15 percent of respondents without housing insecurity at wave 1 developed insecurity by wave 2 (secure-insecure category), and 20 percent of respondents who were insecure at wave 1 had resolved those problems by wave 2 (insecure-secure category). When we break down the overall sample by income eligibility and housing assistance status, we can see the following patterns in the housing insecurity change among the three groups. In the income-eligible nonrecipient group (shown in the second set of columns in fig. 2), 36 percent of respondents had avoided housing insecurity over the entire period, and 17 percent of respondents had been insecure at wave 1 but had not experienced insecurity since then. In the housing assistance recipient group (shown in the third set of columns), 53 percent of respondents were in the secure-secure category, followed by the insecure-secure category (19 percent), the insecure-insecure category (18 percent), and the secure-insecure category (11 percent). Respondents whose income made them ineligible for federal housing assistance showed a pattern generally similar to that of the housing assistance recipients. Typologies of stability or change in housing insecurity between baseline and wave 2, overall and by income-eligibility and housing assistance status (n = 417). The percentage reported in the center of a 95% confidence interval (CI) is a point estimate of the prevalence of housing insecurity. The Secure-Secure category indicates those who were not housing insecure near the official end of the Great Recession in June 2009 and did not experience housing insecurity at wave 2. Respondents in the Secure-Insecure category did not have housing insecurity at wave 1, but they developed it in the post-recession period. The Insecure-Secure category includes respondents who had experienced housing insecurity problems near the end of the Great Recession but who had resolved those housing problems in the follow-up period. The Insecure-Insecure category includes those who reported housing insecurity at wave 1 and at wave 2.
Figure 3 provides information on the association of our key independent variable of income eligibility and housing assistance receipt and each type of housing insecurity at wave 2 that is included in our aggregated measure used in the main analysis. Severe forms of housing insecurity, including eviction and homelessness, were rare in our population-based sample of renters, resulting in nonsignificant differences in the point estimates, with large confidence intervals across the three subgroups. Figure 3 also shows that the point estimate of the aggregated measure of any housing insecurity was not significantly different for housing assistance recipients and income-eligible nonrecipients before controlling for other covariates that are potentially associated with housing assistance receipt. Respondents who were not income eligible for housing assistance had a significantly lower rate of overall housing insecurity than both income-eligible nonrecipients and housing assistance recipients. Population-weighted prevalence of each type of housing insecurity by income eligibility and housing assistance receipt. The percentage reported in the center of 95% confidence interval (CI) is a point estimate of the prevalence of each type of housing insecurity.
In Table 2, we present coefficients, standard errors, and average marginal effects from logistic regressions and a propensity score model. Results for the first model show that housing assistance receipt reduced the likelihood of housing insecurity at wave 2. The average marginal effect suggests that housing assistance recipients were about 22.3 percent less likely to experience housing insecurity at wave 2, compared to income-eligible nonrecipients, net of other characteristics adjusted for in the model. This first model also reveals that losing housing assistance is associated with an increased chance of experiencing housing insecurity problems at wave 2 of 21.3 percent (based on the average marginal effects). This first model also shows that number of children and wave 1 housing insecurity were positively associated with housing insecurity at wave 2. In the second set of columns, results for the propensity score model further confirm our finding for a significant protective influence of housing assistance receipt against housing insecurity at wave 2. Housing assistance recipients were about 31.0 percent less likely to experience housing insecurity problems at wave 2 when compared to income-eligible nonrecipients when using the propensity score approach.
| Any Housing Insecurity Problems over Follow Up | |||||||
|---|---|---|---|---|---|---|---|
| Logistic Model | Propensity Score Model | ||||||
| Coefficient | SE | AME | Coefficient | SE | AME | ||
| Income eligibility/housing assistance (reference = income-eligible nonrecipients): | |||||||
| Housing assistance recipients | −1.635** | .490 | −.223 | −1.605** | .443 | −.310 | |
| Not income-eligible | −.692 | .442 | −.140 | ||||
| Wave 1 SES and demographic characteristics: | |||||||
| Black | .537 | .498 | .133 | .198 | .372 | .037 | |
| Married | .415 | .512 | .080 | −.256 | .676 | −.048 | |
| Number of children | .301* | .133 | .057 | .283+ | .142 | .053 | |
| Age (in years, centered) | .001 | .012 | .000 | −.011 | .018 | −.002 | |
| More than high school education | −.097 | .262 | −.034 | .421 | .244 | .079 | |
| Wave 1 other social program participation | .209 | .398 | .034 | .413 | .349 | .077 | |
| Month elapsed between waves (centered) | .112 | .139 | −.022 | −.067 | .177 | −.013 | |
| Wave 1 housing insecurity | .980** | .266 | .190 | 1.421** | .388 | .290 | |
| Housing assistance change over follow up: | |||||||
| Lost housing assistance | 1.102+ | .540 | .213 | .962+ | .555 | .180 | |
| Gained housing assistance | −.169 | .755 | −.016 | −1.190 | .875 | −.204 | |
| Constant | −1.689** | .506 | −1.338* | .581 | |||
| N | 417 | 267 | |||||
We conducted a set of additional sensitivity analyses to assess the robustness of these results. Respondents with housing assistance could have more serious housing insecurity problems and be at greater risk of future insecurity than their income-eligible counterparts, since some housing assistance programs specifically target those who are experiencing a high level of housing insecurity. In models not shown here, we adjusted for some forms of pre–wave 1 housing insecurity with indicators of ever having been evicted or foreclosed upon (if they had ever owned a home), and our results were substantively unchanged. We also estimated a logistic regression model with an interaction between housing assistance receipt category and housing insecurity at wave 1 and distinguished, for example, respondents who were stably insecure from those who developed new housing insecurity at wave 2. Results show that regardless of wave 1 housing insecurity, housing assistance receipt is associated with a reduced likelihood of housing insecurity at wave 2. It is also plausible that changes in characteristics and experiences occurring between wave 1 and wave 2 were the cause of divergence in housing insecurity at wave 2 for recipients and income-eligible nonrecipients rather than housing assistance receipt. To capture the influence of a financial shock, we added an indicator of more than a 25 percent decrease in the respondent’s household’s income-to-needs ratio, but substantive results remained the same.
The household income of housing assistance recipients was significantly lower than that of their income-eligible nonrecipient counterparts. When we used a lower income cutoff to construct the comparison group of income-eligible nonrecipients, the coefficients associated with receipt of housing assistance increased slightly, suggesting that our estimate of the influence of housing assistance on housing insecurity may be conservative. Results were also robust to an alternative calculation of household income eligibility that accounted for childcare costs, as well as to a broader measure of social program participation that included participation in the Supplemental Nutrition Assistance Program (SNAP), a program that could save money on food-related consumption and leave more funds for rental payments. Finally, we conducted an additional analysis that examined whether the receipt of housing assistance and income eligibility for assistance predicted attrition, using a model predicting whether respondents participated at wave 2 or not with all covariates included in the main analytic model. We find no significant difference in the likelihood of attrition among income-eligible nonrecipients, housing assistance recipients, and income-ineligible respondents. Additionally, wave 1 housing insecurity did not significantly predict attrition.
Discussion
We examine the association between housing assistance and subsequent housing insecurity among renters and other nonowners in the Detroit Metropolitan region in the years immediately following the Great Recession of 2007–9. Our findings suggest that having housing assistance is associated with a significantly lower risk of housing insecurity over about 17 months of follow up, when comparing recipients to their income-eligible nonrecipient counterparts, and estimates are similar across different model specification and modeling approaches.
While it is important to consider these results in light of the previous literature, it is challenging to compare our results directly to findings from previous population-based studies on housing insecurity. The influence of housing assistance on subsequent housing insecurity is not the central question of many prior studies, which primarily focus on examining housing insecurity either in the context of mass incarceration (Geller and Curtis 2011; Geller and Franklin 2014; Wildeman 2014) or welfare reform (Phinney 2013). Federal housing assistance, whether public housing residence or receipt of a housing voucher, is included as a control variable in some of these prior studies (Geller and Curtis 2011; Geller and Franklin 2014; Wildeman 2014) or is mentioned conceptually as one of the protective factors contributing to housing security among the low-income population (Phinney 2013). However, none of those studies developed and applied a sophisticated methodology to estimate the relationship between housing assistance receipt and housing insecurity.
Previous population-based studies that have focused on housing insecurity have used narrower measures, such as homelessness (Wildeman 2014) or residential mobility (Phinney 2013). A few prior studies have used more comprehensive measures of housing insecurity, similar to the approach used in our study (Geller and Curtis 2011; Geller and Franklin 2014). In particular, Amanda Geller and Marah Curtis (2011) and Amanda Geller and Allyson Franklin (2014) studied housing insecurity 4 years after respondents were asked whether they were receiving any federal housing assistance. They find that housing assistance is not associated with housing insecurity at follow up. It is important to consider that the median length of time people receive housing assistance is around 5 years (Kucheva 2012), and there might have been a very low percentage of respondents who retained housing assistance from their baseline interview over the entire period of follow up in these studies. Our finding of a significant protective influence of housing assistance receipt may differ from the weaker findings of past studies because of a more appropriate reference group and our relatively short follow-up period, which resulted in potentially higher retention of housing assistance by our respondents over follow up than in prior studies with longer follow-up periods.
In estimating the association between housing assistance and subsequent housing insecurity, we addressed several methodological limitations of prior population-based studies. First, and most important, we use income-eligible nonrecipients as a comparison group rather than simply adjusting for income differences. Second, we establish that the differential likelihood of housing insecurity over follow up for housing assistance recipients and nonrecipients is not a function of differential levels of cash assistance receipt across groups or of changes in housing assistance receipt over follow up. Propensity score modeling yields very consistent findings to the more conventional regression model results, and it provides a more explicit attempt to account for differences in the characteristics of recipients and nonrecipients that could drive both housing assistance receipt and subsequent housing insecurity.
Despite these advances, our results should be considered in the context of several limitations. Previous research has identified limitations of self-reported data on program participation. Underreporting of housing assistance receipt could lead us to underestimate its positive influence, because respondents who had such assistance but did not report it would be included in the control group, which could decrease overall housing insecurity in the control group, thereby reducing the gap in outcomes between those with and without assistance. Also, recipients may have failed to report exactly what types of housing assistance they received (Shroder 2002). We used the residential addresses of respondents to verify that none were living in public housing, but we were not able to further differentiate project-based housing vouchers from tenant-based ones. Also, although HCVs are the most common form of housing vouchers administered by local PHAs, it is also plausible that respondents may have received housing assistance from other sources (e.g., PBRA or the HUD Veteran Affairs Supportive Housing Program) that have different regulations, such as different income eligibility criteria.
Our study is also limited by our use of typical survey items that ask about housing problems in the last year rather than about shorter intervals and more detail, given the empirically documented within-year fluctuation in income among low-income people (Bania and Leete 2009; Hannagan and Morduch 2015). However, even when housing assistance does not fully mitigate the negative influence of monthly income fluctuation on housing security for an extended period of time, housing assistance could significantly delay the actualization of housing insecurity arising from abrupt decline in income over several months. Thus, our results might be an underestimation of the positive influence of housing assistance that could be better understood in future studies that use more frequent assessment (e.g., monthly) of housing problems. This measurement issue is likely to arise for other aspects of the social safety net as well, which could be incorporated in future studies with larger samples of respondents or a focus on those who are eligible for social programs or close to eligibility cutoffs.
Our sample is drawn from three counties in the Detroit Metropolitan area, where housing assistance programs are separately administered across 30 highly fragmented local PHAs (HUD 2015a). We had only a small number of respondents per PHA, so we could not explore the effects of living in any specific PHA jurisdiction. Future studies should examine whether the positive relationship between housing assistance receipt and subsequent housing security differs according to PHA-level administrative practices, to identify program parameters under PHA discretion that can improve the effectiveness of housing assistance programs. Also, nearly all of the housing assistance recipients in our sample lived in Wayne County, where residents are more likely to be income eligible for housing assistance; this means that our results are more likely to reflect the case of Wayne County and its specific history of racial residential segregation from the suburban areas in surrounding counties. In a sensitivity analysis not shown here that constrained the analytic sample to Wayne County residents only, our results were consistent with those shown here, but future studies with larger samples should consider the value of housing assistance for those living in more and less advantaged communities.
In spite of these limitations, our study documents a link between housing assistance receipt and reduced housing insecurity in the years immediately following the Great Recession of 2007–9. Two recent evaluation studies of the housing voucher program and its influence on housing insecurity, conducted in 2000–2005 (Wood and Rangarajan 2004) and 1998–2003 (Wood et al. 2008), captured conditions prior to the Great Recession. These two experimental studies have been widely cited as empirical evidence that housing voucher programs should be expanded to reduce housing insecurity among low-income renters (Fischer 2015), and we offer more recent empirical evidence that suggests the same protective effect. Regrettably, there was a dramatic reduction of about 100,000 housing vouchers with the 2011 Budget Control Act, and the program has only recently started to slowly recover the vouchers lost to this sequestration cut (Rice 2015). Our results provide further empirical support for the continued growth of this program. Our findings are also relevant to recent scholarship that frames housing insecurity, including eviction and government inaction to address it, as a cause of poverty (Desmond 2016). Previous studies document the many negative consequences of housing insecurity. During and following economic recessions, labor market instability leads to housing instability as income shocks and increased volatility in income cause missed rent payments. Unfortunately, housing assistance programs suffered during a period of growing need, when households were exposed to very high levels of labor market insecurity. It is an empirical question whether and to what extent housing insecurity among low-income renters that arose in the recent recession exacerbated economic inequality in the United States. The findings of this and other studies suggest, however, that housing support programs could potentially mitigate some of the pernicious consequences of housing insecurity.
Huiyun Kim is a doctoral candidate in the joint program in social work and sociology at the University of Michigan. His research focuses on the intersection of housing instability among low-income populations and US rental assistance programs. His current project examines how the locally fragmented administration of the Housing Choice Voucher program in metropolitan areas poses challenges to the federal initiative to end homelessness.
Sarah A. Burgard is an associate professor of sociology and epidemiology and a research associate professor at the Population Studies Center, all at the University of Michigan. Her research focuses on the ways in which employment conditions and socioeconomic disadvantage across the life course shape health and health disparities.
Kristin S. Seefeldt is an assistant professor at the School of Social Work and the Gerald R. Ford School of Public Policy at the University of Michigan. Her research focuses on poverty and social welfare policies in the United States.
Data collection for this study was supported by funds provided to the National Poverty Center (NPC) by the Office of the Assistant Secretary for Planning and Evaluation at the US Department of Health and Human Services, the Office of the Vice-President for Research at the University of Michigan, the John D. and Catherine T. MacArthur Foundation, and the Ford Foundation. This research was supported by a Research Partnership Summer Award from the School of Social Work at the University of Michigan. We thank Sandra Danziger, Lucie Kalousová, Michael Evangelist, Sarah Seelye, and Anne Blumenthal for their thoughtful comments. We also thank the editor and three anonymous reviewers for their thought comments, including one who went well beyond the call of duty to offer insightful comments. An earlier version of this paper was presented at the Population Association of America Meeting in Washington, DC, in 2016.
Notes
1. In the case of income decline, prompt reporting of income change can significantly benefit recipients if it means they will be able to reduce their contribution to housing costs. However, a legal case study about conflict over the adjusted income calculation between voucher recipient and PHAs documents possibly significant administrative delay in income adjustment and its potential effect on housing security of recipients (Daniels v. Housing Authority of Prince George's City, 940 F. Supp. 2d 248 [Dist. Ct. D. Md. 2013]). Also, the lagged response of PHAs is even more critical in relation to housing insecurity when there is no emergency safety net (e.g., an emergency fund for housing assistance recipients to prevent eviction from delayed rental payment) for housing assistance recipients (Ross and Pelletiere 2014).
2. Thirty-nine percent of housing assistance recipients participated in Supplemental Security Income (SSI), followed by TANF (Temporary Assistance for Needy Families; 31 percent), and UI (Unemployment Insurance; 11 percent). Income-eligible nonrecipients showed a similar pattern, but with a much lower percentage of TANF participation: SSI (26 percent) was followed by UI (15 percent) and TANF (9 percent). Not surprisingly, none of the income-ineligible respondents received TANF, and only a small percentage received UI (12 percent) or SSI (7 percent).
| Types of Housing Insecurity | Items and Coding Strategy |
|---|---|
| Moved for cost | [wave 1] Respondents were first asked: “How long have you lived here in this house/apartment?” Respondents who had stayed less than 1 year in their current home were asked: “Did you move because you could no longer afford that home?” If respondents answered affirmatively, they were coded as having moved for cost. If respondents either had not moved in the last 12 months or had moved but not for cost, they were coded as not having moved for cost. [wave 2] Respondents were asked: “Since we last talked to you, how many times have you moved?” If respondents had moved more than once over follow up, they were asked: “Why did you decide to move? Please tell me all that apply.” If respondents chose either “could no longer afford the previous home” or “home was foreclosed upon,” they were coded as having moved for cost. |
| Evicted | [wave 1] Respondents were asked: “In the last 12 months, have you been evicted at any time?” If respondents answered affirmatively, they were coded as having been evicted. [wave 2] Respondents were asked: “Since we last talked to you, have you been evicted at any time?” If respondents answered affirmatively, they were coded as having been evicted. |
| Homeless | [wave 1] Respondents were asked: “In the last 12 months, have you ever been homeless?” If respondents answered affirmatively, they were coded as having been homeless. [wave 2] Respondents were asked: “Since we last talked to you, have you ever been homeless?” If respondents answered affirmatively, they were coded as having been homeless. |
| Moved in with others | [wave 1] Respondents were asked: “Have you moved in with anyone in the last 12 months to share household expenses?” If respondents answered affirmatively, they were coded as having moved in with others. [wave 2] Respondents were asked: “Since the last time we interviewed you, have you moved in with anyone to share or because you couldn’t afford your own place?” If respondents answered affirmatively, they were coded as having moved in with others. |
| Behind on rent/mortgage | [wave 1] Respondents were asked: “In the last 12 months, have you ever gotten behind on rent?” If respondents answered affirmatively, they were coded as having been behind on rent. [wave 2] Renters at follow up were asked: “Since we last talked to you, have you ever gotten behind on rent?” If respondents answered affirmatively, they were coded as having been behind on rent/mortgage. Respondents who became homeowners at follow up were first asked: “Do you own this house outright or do you have a mortgage or land contract on the property?” For mortgaged homeowners, the following question was asked: “Are you paying off this loan ahead of schedule, behind schedule, or are your payments about on schedule?” If the mortgaged homeowners answered they were “behind on mortgage,” they were coded as behind on rent/mortgage. |
| Survey-Weighted Original Sample | Propensity Score–Weighted Sample | |||||
|---|---|---|---|---|---|---|
| Housing Assistance Recipient | Income-Eligible/Nonrecipient | Standardized Mean Difference | Housing Assistance Recipient | Income-Eligible/Nonrecipient | Standardized Mean Difference | |
| Household income below AMI 30% | .88 | .30 | 1.38 | .84 | .83 | .04 |
| Black | .96 | .35 | 1.68 | .94 | .95 | −.03 |
| Married | .08 | .26 | −.52 | .08 | .11 | −.11 |
| Number of children | 1.50 | .91 | .36 | 1.50 | 1.48 | .01 |
| Age | 39.78 | 34.83 | .36 | 37.73 | 38.21 | −.04 |
| More than high school education | .47 | .54 | −.14 | .42 | .46 | −.07 |
| Other social program participation | .73 | .30 | .92 | .71 | .65 | .12 |
| Month elapsed between waves | 17.01 | 17.36 | −.24 | 17.30 | 17.25 | .03 |
| N | 69 | 198 | −.06 | 69 | 198 | −.08 |
References
Austin, Peter C. 2011. “An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies.” Multivariate Behavioral Research 46 (3): 399–424. Bania, Neil, and Laura Leete. 2009. “Monthly Household Income Volatility in the US, 1991/92 vs. 2002/03.” Economics Bulletin 29 (3): 2100–2112. Brookhart, M. Alan, Sebastian Schneeweiss, Kenneth J. Rothman, Robert J. Glynn, Jerry Avorn, and Til Stürmer. 2006. “Variable Selection for Propensity Score Models.” American Journal of Epidemiology 163 (12): 1149–56. Burgard, Sarah A., Kristin S. Seefeldt, and Sarah Zelner. 2012. “Housing Instability and Health: Findings from the Michigan Recession and Recovery Study.” Social Science and Medicine 75 (12): 2215–24. Busso, Matias, John DiNardo, and Justin McCrary. 2014. “New Evidence on the Finite Sample Properties of Propensity Score Reweighting and Matching Estimators.” Review of Economics and Statistics 96 (5): 885–97. Clark, William A. V., Marinus C. Deurloo, and Frans M. Dieleman. 2003. “Housing Careers in the United States, 1968–93: Modelling the Sequencing of Housing States.” Urban Studies 40 (1): 143–60. Coley, Rebekah L., Tama Leventhal, Alicia D. Lynch, and Melissa A. Kull. 2013. “Relations between Housing Characteristics and the Well-Being of Low-Income Children and Adolescents.” Developmental Psychology 49 (9): 1775–89. Collinson, Robert. 2011. “Rental Housing Affordability Dynamics, 1990–2009.” Cityscape: A Journal of Policy Development and Research 13 (2): 71–103. Collinson, Robert, Ingrid Gould Ellen, and Jens Ludwig. 2016. “Low Income Housing Policy.” 59–73 in Economics of Means-Tested Transfer Programs in the United States, vol. 2, edited by Robert A. Moffitt. Chicago: University of Chicago Press. Culhane, Dennis P. 1992. “The Quandaries of Shelter Reform: An Appraisal of Efforts to ‘Manage’ Homelessness.” Social Service Review 66 (3): 428–40. Desmond, Matthew. 2012a. “Disposable Ties and the Urban Poor.” American Journal of Sociology 117 (5): 1295–1335. ———. 2012b. “Eviction and the Reproduction of Urban Poverty.” American Journal of Sociology 118 (1): 88–133. ———. 2016. Evicted: Poverty and Profit in the American City. New York: Crown. Desmond, Matthew, and Carl Gershenson. 2016. “Housing and Employment Insecurity among the Working Poor.” Social Problems 63 (1): 46–67. DuGoff, Eva H., Megan Schuler, and Elizabeth A. Stuart. 2014. “Generalizing Observational Study Results: Applying Propensity Score Methods to Complex Surveys.” Health Services Research 49 (1): 284–303. Dwyer, Rachel E., and Lora A. Phillips Lassus. 2015. “The Great Risk Shift and Precarity in the US Housing Market.” ANNALS of the American Academy of Political and Social Science 660 (1): 199–216. Eastwood, Elizabeth A., and Jeffrey M. Birnbaum. 2007. “Physical and Sexual Abuse and Unstable Housing among Adolescents with HIV.” AIDS and Behavior 11 (2): 116–27. Edin, Kathryn J., and H. Luke Shaefer. 2015. $2.00 a Day: Living on Almost Nothing in America. Boston: Houghton Mifflin Harcourt. Ellen, Ingrid Gould, and Brendan O’Flaherty. 2007. “Social Programs and Household Size: Evidence from New York City.” Population Research and Policy Review 26 (4): 387–409. Fischer, Will. 2015. “Research Shows Housing Vouchers Reduce Hardship and Provide Platform for Long-Term Gains among Children.” Report, Center on Budget and Policy Priority, Washington, DC. http://www.cbpp.org/research/housing/research-shows-housing-vouchers-reduce-hardship-and-provide-platform-for-long-term .Fisher, Benjamin W., Lindsay S. Mayberry, Marybeth Shinn, and Jill Khadduri. 2014. “Leaving Homelessness Behind: Housing Decisions among Families Exiting Shelter.” Housing Policy Debate 24 (2): 364–86. Geller, Amanda, and Marah A. Curtis. 2011. “A Sort of Homecoming: Incarceration and the Housing Security of Urban Men.” Social Science Research 40 (4): 1196–1213. Geller, Amanda, and Allyson Walker Franklin. 2014. “Paternal Incarceration and the Housing Security of Urban Mothers.” Journal of Marriage and Family 76 (2): 411–27. Greenbaum, Susan, Wendy Hathaway, Cheryl Rodriguez, Ashley Spalding, and Beverly Ward. 2008. “Deconcentration and Social Capital: Contradictions of a Poverty Alleviation Policy.” Journal of Poverty 12 (2): 201–28. Gubits, Daniel, Marybeth Shinn, Stephen Bell, Michelle Wood, Samuel Dastrup, Claudia D. Solari, Scott R. Brown, Steven Brown, Lauren Dunton, Winston Lin, Debi McInnis, Jason Rodriguez, Galen Savidge, and Brooke E. Spellman. 2015. “Family Options Study: Short-Term Impacts of Housing and Services Interventions for Homeless Families.” Report, US Department of Housing and Urban Development, Office of Policy Development and Research, Washington, DC. Hannagan, Anthony, and Jonathan Morduch. 2015. “Income Gains and Month-to-Month Income Volatility: Household Evidence from the US Financial Diaries.” New York University Wagner Research Paper no. 2659883, New York University. Harknett, Kristen S., and Caroline Sten Hartnett. 2011. “Who Lacks Support and Why? An Examination of Mothers’ Personal Safety Nets.” Journal of Marriage and Family 73 (4): 861–75. HUD (US Department of Housing and Urban Development). 2001. “Housing Choice Voucher Program Guidebook.” http://portal.hud.gov/hudportal/HUD?src=/program_offices/public in dian_housing/programs/hcv/forms/guidebook .———. 2015a. “Picture of Subsidized Households, 2010.” Online database. https://www.huduser.gov/portal/datasets/picture/yearlydata.html .———. 2015b. “Rental Vacancy Rates for the 75 Largest Metropolitan Statistical Areas: 2005 to 2014.” http://www.census.gov/housing/hvs/data/rates.html .JCHSH (Joint Center for Housing Studies of Harvard University). 2011. “Rental Market Stresses: Impacts of the Great Recession on Affordable and Multifamily Lending.” http://www.jchs.harvard.edu/research/publications/rental-market-stresses-impacts-great-recession-affordability-and-multifamily .Keene, Danya E., and Arline T. Geronimus. 2011. “‘Weathering’ HOPE VI: The Importance of Evaluating the Population Health Impact of Public Housing Demolition and Displacement.” Journal of Urban Health 88 (3): 417–35. Kleit, Rachel Garshick, Seungbeom Kang, and Corianne Payton Scally. 2016. “Why Do Housing Mobility Programs Fail in Moving Households to Better Neighborhoods?” Housing Policy Debate 26 (1): 188–209. Kucheva, Yana Andreeva. 2012. “The Dynamics of Participation in Subsidized Housing Programs in the US.” PhD diss., University of California, Los Angeles, Department of Sociology. Kushel, Margot B., Eric Vittinghoff, and Jennifer S. Haas. 2001. “Factors Associated with the Health Care Utilization of Homeless Persons.” Journal of the American Medical Association (JAMA) 285 (2): 200–206. Li, Fan, Kari Lock Morgan, and Alan M. Zaslavsky. 2016. “Balancing Covariates via Propensity Score Weighting.” Available at arXiv:1404.1785. May, Jon. 2000. “Housing Histories and Homeless Careers: A Biographical Approach.” Housing Studies 15 (4): 613–38. Metraux, Stephen, and Dennis P. Culhane. 1999. “Family Dynamics, Housing, and Recurring Homelessness among Women in New York City Homeless Shelters.” Journal of Family Issues 20 (3): 371–96. Morgan, Stephen L., and Christopher Winship. 2014. Counterfactuals and Causal Inference. New York: Cambridge University Press. O’Flaherty, Brendan. 2009. “What Shocks Precipitate Homelessness?” Discussion Paper no. 0809–14, Department of Economics, Columbia University, New York. Park, Jung Min, Angela Fertig, and Stephen Metraux. 2014. “Factors Contributing to the Receipt of Housing Assistance by Low-Income Families with Children in Twenty American Cities.” Social Service Review 88 (1): 166–93. Parsell, Cameron, Wojtek Tomaszewski, and Rhonda Phillips. 2014. “Exiting Unsheltered Homelessness and Sustaining Housing: A Human Agency Perspective.” Social Service Review 88 (2): 295–321. Pattillo, Mary. 2013. “Housing: Commodity versus Right.” Annual Review of Sociology 39:509–31. Pavao, Joanne, Jennifer Alvarez, Nikki Baumrind, Marta Induni, and Rachel Kimerling. 2007. “Intimate Partner Violence and Housing Instability.” American Journal of Preventive Medicine 32 (2): 143–46. Phinney, Robin. 2013. “Exploring Residential Mobility among Low-Income Families.” Social Service Review 87 (4): 780–815. Phinney, Robin, Sheldon Danziger, Harold A. Pollack, and Kristin Seefeldt. 2007. “Housing Instability among Current and Former Welfare Recipients.” American Journal of Public Health 97 (5): 832–37. Pilkauskas, Natasha V., Irwin Garfinkel, and Sara S. McLanahan. 2014. “The Prevalence and Economic Value of Doubling Up.” Demography 51 (5): 1667–76. Rebholz, Casey, Mari-Lynn Drainoni, and Howard Cabral. 2009. “Substance Use and Social Stability among At-Risk HIV-Infected Persons.” Journal of Drug Issues 39 (4): 851–70. Reed, Elizabeth, Jhumka Gupta, Monica Biradavolu, Vasavi Devireddy, and Kim M. Blankenship. 2011. “The Role of Housing in Determining HIV Risk among Female Sex Workers in Andhra Pradesh, India: Considering Women’s Life Contexts.” Social Science and Medicine 72 (5): 710–16. Rice, Douglas. 2015. “Housing Agencies Restoring Vouchers: Let’s Finish the Job in 2016.” Policy brief, Center on Budget and Policy Priority, Washington, DC. http://www.cbpp.org/blog/housing-agencies-restoring-vouchers-lets-finish-the-job-in-2016 .Rollins, Chiquita, Nancy E. Glass, Nancy A. Perrin, Kris A. Billhardt, Amber Clough, Jamie Barnes, Ginger C. Hanson, and Tina L. Bloom. 2012. “Housing Instability Is as Strong a Predictor of Poor Health Outcomes as Level of Danger in an Abusive Relationship Findings from the SHARE Study.” Journal of Interpersonal Violence 27 (4): 623–43. Ross, Lauren M., and Danilo Pelletiere. 2014. “Chile’s New Rental Housing Subsidy and Its Relevance to US Housing Choice Voucher Program Reform.” Cityscape 16 (2): 179–92. Sard, Barbara, and Thyria Alvarez-Sánchez. 2011. “Large Majority of Housing Voucher Recipients Work, Are Elderly, or Have Disabilities.” Policy brief, Center on Budget and Policy Priority, Washington, DC. http://www.cbpp.org/research/large-majority-of-housing-voucher-recipients-work-are-elderly-or-have-disabilities .Seefeldt, Kristin S. 2015. “Constant Consumption Smoothing, Limited Investments, and Few Repayments: The Role of Debt in the Financial Lives of Economically Vulnerable Families.” Social Service Review 89 (2): 263–300. ———. Forthcoming. “Waiting It Out: Time, Action, and the Process of Securing Benefits.” Qualitative Social Work. Available online in advance of publication. Shinn, Marybeth, Beth C. Weitzman, Daniela Stojanovic, James R. Knickman, Lucila Jimenez, Lisa Duchon, Susan James, and David H. Krantz. 1998. “Predictors of Homelessness among Families in New York City: From Shelter Request to Housing Stability.” American Journal of Public Health 88 (11): 1651–57. Shroder, Mark. 2002. “Does Housing Assistance Perversely Affect Self-Sufficiency? A Review Essay.” Journal of Housing Economics 11 (4): 381–417. Skobba, Kim. 2016. “Exploring the Housing Pathways of Low-Income Women: A Biographical Approach.” Housing, Theory, and Society 33 (1): 41–58. Sosin, Michael, Irving Piliavin, and Herb Westerfelt. 1990. “Toward a Longitudinal Analysis of Homelessness.” Journal of Social Issues 46 (4): 157–74. Wildeman, Christopher. 2014. “Parental Incarceration, Child Homelessness, and the Invisible Consequences of Mass Imprisonment.” ANNALS of the American Academy of Political and Social Science 651 (1): 74–96. Williams, Richard. 2012. “Using the Margins Command to Estimate and Interpret Adjusted Predictions and Marginal Effects.” Stata Journal 12 (2): 308–31. Wong, Yin-Ling Irene, Dennis P. Culhane, and Randall Kuhn. 1997. “Predictors of Exit and Reentry among Family Shelter Users in New York City.” Social Service Review 71 (3): 441–62. Wood, Michelle, Jennifer Turnham, and Gregory Mills. 2008. “Housing Affordability and Family Well-being: Results from the Housing Voucher Evaluation.” Housing Policy Debate 19 (2): 367–412. Wood, Robert G., and Anu Rangarajan. 2004. “The Benefits of Housing Subsidies for TANF Recipients: Evidence from New Jersey.” Report, Mathematica Policy Research, Princeton, NJ. Ziersch, Anna, and Kathy Arthurson. 2005. “Social Networks in Public and Community Housing: The Impact on Employment Outcomes.” Urban Policy and Research 23 (4): 429–45. Zlotnick, Cheryl, Marjorie J. Robertson, and Maureen Lahiff. 1999. “Getting off the Streets: Economic Resources and Residential Exits from Homelessness.” Journal of Community Psychology 27 (2): 209–24.
