Randomized Trial of Intensive Housing Placement and Community Transition Services for Episodic and Recidivist Homeless Families

Abstract

Objectives. We compared Home to Stay, a pilot of intensive housing placement and community transition services for episodic and recidivist homeless families, with a standard services approach.

Methods. Using intention-to-treat analyses, we conducted a modified randomized trial of 138 Home to Stay client families and a control group of 192 client families receiving standard shelter services.

Results. Home to Stay clients exited shelter more quickly than clients in the control group (Cox regression, P < .001), more commonly exited shelter with housing subsidies (75% vs 56%), stayed out of shelter longer (Cox regression, P = .011), and spent fewer total days in shelter (376 days vs 449 days). Home to Stay performed best with clients who entered shelter within 180 days of the pilot’s start date and had less impact on clients entering shelter before that time.

Conclusions. Relative to standard services, Home to Stay services can accelerate exit from shelter and reduce return to shelter and total sheltered days for episodic and recidivist homeless families. Standard shelter services may be able to narrow this performance gap by incentivizing work with all episodic and recidivist homeless families.

Homeless families living in temporary shelters or places not meant for human habitation¹ make up more than one third of all homeless people in the United States.^2,3 According to the 2012 Point In Time estimate, conducted under the auspices of the US Department of Housing and Urban Development Continuums of Care to provide a snapshot of homelessness on a single night, 239 403 members of families were homeless nationwide.¹ Family homelessness decreased by 1% nationally between 2009 and 2011, perhaps reflecting the effects of the Homeless Prevention and Rapid Rehousing Program funded through the 2009 American Recovery and Reinvestment Act.⁴ The 1.4% uptick in national family homelessness¹ between 2011 and 2012 may, however, indicate that the effects of the prolonged economic downturn are still worsening for the nation’s most vulnerable families, with homelessness functioning as a lagging indicator of the economy.^5,6

Homeless families largely consist of mothers aged younger than 30 years with multiple young children,^2,7 and most experience a range of circumstances associated with poverty, including low-paying jobs and unemployment, unstable housing conditions, and domestic violence.^2,7,8 Although a number of studies^9–12 have indicated no major differences between most mothers who become homeless and their socioeconomic counterparts who remain housed, strong evidence has indicated that those who return to homelessness (episodically homeless families) have significantly higher rates of behavioral health problems and other issues that require more intensive services.

New York City, the setting of this study, is mandated by law to provide shelter to every eligible homeless family or individual who seeks it. Families who apply for shelter are provided conditional shelter placements while the Department of Homeless Services (DHS) determines whether the family is homeless and therefore eligible for the city’s services. New York City DHS operates a family shelter system consisting of approximately 9900 units for families with children in 150 shelters, most operated by contracted social service providers. New York City DHS oversees these shelters, setting service guidelines including intake, assessment, and exit protocols; case-management expectations; and client responsibility guidelines. Oversight also includes health and safety monitoring, emphasizing infant safe sleeping and child welfare.

As of the start of the Home to Stay pilot study, on April 26, 2010, families exiting the shelter system were able to access 1-year Advantage housing subsidies, with conditional approval available for a second year. The Advantage program was designed to serve 3 subpopulations of homeless clients and incorporated 3 distinct certifications: Child Advantage served clients with child welfare involvement, Fixed Income Advantage served clients who were receiving federal disability payments, and Work Advantage served employed clients (and required a $50 monthly family contribution). The Advantage subsidies were modified in August 2010, when the family contribution was increased to 30% of monthly income and eligibility was restricted to families with adults who were employed or receiving federal disability benefits. State funds for these reduced-value subsidies were discontinued for new families in March 2011, at which point the city terminated the program, and families exiting the shelter system became heavily reliant on obtaining employment and market-rate housing or returning to live with friends or family. Monthly payments were terminated as of February 2012 for all Advantage subsidies still active at that time.

In April 2010, New York City DHS also operated an incentive payment system designed to reduce the time spent in shelter by families with children. Under this incentive system, family shelters received 10% more than the standard contract or service per diem rate for all sheltered days up to 180 days per family and 20% below the per diem rate for all sheltered days over 180 days per family. Incentives were deactivated in March 2011, when Advantage subsidies for families newly exiting shelter were discontinued.

The Home to Stay pilot was initiated to provide enriched housing placement and support services to episodic and recidivist homeless families who appeared likely to require such services, in addition to housing subsidies, to achieve long-term housing stability. Compared with standard shelter services, we hypothesized that Home to Stay services would increase the proportion of families exiting shelter to housing and reduce families’ time to exit shelter, time to return to shelter, and total sheltered days.

METHODS

We conducted a modified randomized trial. The experimental group was randomly selected from among all eligible participants; the control group was selected later from among those eligible participants remaining in the pool after the experimental group was selected. Sample size was determined by pilot services funding, which was available for 120 enrollees. We reached the enrollment limit with 139 participants assigned to the experimental condition. The control-to-experimental group ratio of 3:2, determined post hoc, was based on the available sample.

Participants were families with at least 1 custodial child living in the New York City family shelter system who were certified for Advantage subsidies and who had either (1) at least 2 prior stays in that system in the previous 5 years (episodic) or (2) at least 1 prior stay in that system in the previous 5 years that ended with the family moving into subsidized housing (recidivist).

Interventions

Control—standard services.

Social service staffing and delivery vary widely across the 150 New York City DHS family shelters, but shelter caseworkers are typically assigned mixed caseloads of approximately 25 families. Caseworker qualifications are determined by each shelter provider, and caseworkers receive on-the-job training through the provider agency, augmented by specialized DHS-provided training. Caseworkers are generally supervised by more experienced staff, each of whom oversees 4 or more caseworkers; many large shelters also employ a director of social services.

Homeless families with children entering shelter are placed into apartment-style units for the duration of their stay. The families are assigned caseworkers who work in collaboration with other city agencies to encourage and assist families in accessing public benefits, pursuing employment, and obtaining supportive services and with shelter housing specialists to locate appropriate, permanent housing. Caseworkers and families work together to develop detailed plans for exiting shelter and returning to self-sufficiency. Caseworkers meet with clients biweekly to review their progress, reassess and address any potential barriers to employment or housing, and make referrals for any required services. Once permanent housing is obtained and a client family moves out of shelter, services from the homeless shelter cease, although community-based prevention and aftercare are available should the client family require further services.

Experimental—Home to Stay.

The Home to Stay model was designed to rapidly obtain and maintain housing for episodic and recidivist homeless families through intensive, temporary support services coupled with a time-limited housing subsidy. The Home to Stay pilot was a partnership between a New York City–based charitable foundation, 4 New York City–based nonprofit service providers, and the New York City government. Services focused on 3 strategies: moving families out of shelter rapidly using a locally funded, temporary housing subsidy; securing sufficient household income to enable families to pay market rent on expiration of the subsidy; and connecting families to community-based services that would help them to maintain housing stability after the termination of Home to Stay services.

Caseworkers met with client families at homeless shelters to encourage them to voluntarily enroll in Home to Stay services. Each enrolled family was assigned a single caseworker who followed them from shelter into permanent housing to ensure continuity of services across that transitional period. Initial services focused on helping families to secure permanent housing and exit shelter as quickly as possible. Once client families were placed in housing, services focused on their obtaining a monthly household income equal to at least 200% of the family’s rent obligation, obtaining a permanent housing subsidy, or both within 1 year of shelter exit. Because the availability of permanent subsidies was extremely limited, services primarily focused on maximizing income from public benefits for all eligible household members and obtaining or increasing employment income for all adult household members.

Common elements of the case management model for all service teams included caseloads of 10 to 15 client families per worker, early and aggressive engagement to enroll clients while they were in shelter, flexible scheduling that accommodated clients’ other time demands, individualized service plans informed by an assessment of clients’ needs and strengths and were developed collaboratively with clients, financial literacy services integrated into case management, and frequent contact appropriate to clients’ needs (beginning with at least 4 contacts per month, including at least 1 in situ face-to-face contact). Home to Stay clients also remained on the caseloads of their standard services caseworkers and housing specialists. The program elements differentiating Home to Stay from standard services were more frequent client contact, smaller caseloads, flexible scheduling, integrated financial literacy services, and continuity of services across the transitional period from shelter into housing.

Key service milestones were tracked at the case and team levels to help program managers monitor progress and make efficient use of staff time. These outcomes included engagement speed and enrollment success rate, time to housing placement, and progress toward household income and graduation targets. Program-level outcomes were also tracked and shared across teams to encourage friendly competition and help managers monitor their comparative performance.

Randomization

New York City DHS required that no eligible family be denied the possibility of receiving Home to Stay services at any time. We designed the following randomization procedure to meet this requirement while preserving the study’s validity. All baseline and outcome data were extracted by New York City DHS evaluation staff from citywide administrative databases maintained in the course of general shelter operations. We used no mechanism for blinding to condition.

New York City DHS evaluation staff used randomly generated numbers to select experimental group clients from all eligible clients. Lists of potential experimental group clients were passed to Home to Stay service staff for outreach and enrollment, and the process was continued until 120 families were successfully enrolled in services. The total number of families originally assigned to the experimental group by this process was 139, but we later discovered that 1 experimental group family exited shelter before the pilot start date and removed them from the sample. The final experimental group consisted of 138 client families. All eligible families not randomly assigned to the experimental group were added to a cumulative control group pool. Families in the control pool remained eligible for random selection to the experimental group and were removed from the control pool if so assigned. When the experimental group was complete, New York City DHS evaluation staff randomly selected control families from the final control pool, stratified by shelter entry date to match the proportions found in the experimental group. The control pool sufficed to supply a 3:2 control group to experimental group sample ratio within each stratification block, and we randomly assigned 210 families within stratification from the pool to the control group. We later discovered that 6 families assigned to the control group exited shelter before the pilot start date and removed them from the sample before analysis.

In accordance with New York City DHS requirements, eligible client families remained available for lottery assignment to Home to Stay services as caseload openings occurred, even after study randomization was closed. Twelve families initially assigned to the control group were enrolled in Home to Stay while the study was underway and removed from the control sample before analysis. Control group families remaining in shelter longer than the norm or returning sooner were more likely to be offered Home to Stay services, making this removal a conservative measure favoring the control group. The final control group included 192 of the original 210 families.

Analysis

We conducted all analyses on an intention-to-treat basis. Planned analyses addressed experimental and control group comparability and the 4 outcome variables. We assessed intervention group comparability by using the t test for 9 a priori characteristics: days from shelter entry to pilot start date, episodic homelessness, recidivist homelessness, number of adults in household, number of children in household, age of head of household at pilot start, work-based housing subsidy, fixed income–based housing subsidy, and child welfare–based housing subsidy (the last 3 variables were mutually exclusive).

We used Cox regression to assess the effect of intervention group on time to exit shelter after April 25, 2010 (limited to exits ≥ 30 days), and time to first (overnight) return to shelter postexit (through November 30, 2012). We used univariate general linear method analyses to assess the effect of intervention group on the prevalence of exiting shelter into subsidized housing and on total days spent in shelter during the study period (April 26, 2010–November 30, 2012). We conducted all statistical analyses with SPSS version 19 (IBM, Armonk, NY).

RESULTS

We found no significant differences between the intervention groups on 7 of 9 baseline variables (Table 1). We did find significant differences in the mix of housing subsidy certification types in the subpopulation of recent-entry families who entered shelter fewer than 180 days before the pilot start date (P ≤ .01) and a near-significant difference in the full sample (P = .06). We found no significant differences between the intervention groups in the subpopulation of older-entry families who entered shelter at least 180 days before the pilot start date. Because the fixed-income–based and child-based housing subsidy certifications were mutually exclusive and strongly correlated (r = −.472; P < .001) and work-based subsidy certification was much more prevalent, we included work-based certification as a covariate in all further analyses of differences between the intervention groups. The study sample consisted of 49% recent-entry and 51% older-entry families, the same ratio found in the New York City DHS family shelter system as a whole in April 2010 (New York City DHS, unpublished data, 2010). Of 138 families randomized to the experimental group, 117 (85%) chose to enroll in Home to Stay services.

TABLE 1—

Tests of Intervention Group Comparability: Home to Stay Pilot, New York, NY, April 26, 2010–November 30, 2012

	All Clients				Older-Entry Clients				Recent-Entry Clients
Characteristics	Experimental Group (n = 138), No. (%) or Mean ±SD	Control Group (n = 192), No. (%) or Mean ±SD	χ2 or t (df)	P	Experimental Group (n = 74), No. (%) or Mean ±SD	Control Group (n = 88), No. (%) or Mean ±SD	χ2 or t (df)	P	Experimental Group (n = 64), No. (%) or Mean ±SD	Control Group (n = 104), No. (%) or Mean ±SD	χ2 or t (df)	P
Episodic	28 (20)	42 (22)	0.12a (1)	.73	17 (23)	20 (23)	0.00a (1)	.97	11 (17)	22 (21)	0.40b (1)	.53
Recidivist	126 (91)	171 (89)	0.45a (1)	.5	66 (89)	77 (88)	0.11a (1)	.74	60 (94)	94 (90)	0.59b (1)	.44
Certification type			5.66a (2)	.06			0.59a (2)	.75			12.29b (2)	< .001
Work	77 (56)	129 (67)			48 (65)	60 (68)			29 (45)	66 (66)
Fixed	22 (16)	17 (9)			9 (12)	12 (14)			13 (20)	5 (5)
Child	39 (28)	46 (24)			17 (23)	16 (18)			22 (34)	30 (29)
Pilot start from shelter entry, d	245 ±200	220 ±206	1.07b (328)	.28	385 ±171	370 ±217	0.47b (16)	.64	83 ±61	94 ±59	−1.14b (166)	.25
Adults	1.36 ±0.56	1.45 ±0.64	−1.45b (328)	.15	1.32 ±0.60	1.44 ±0.64	−1.21b (160)	.23	1.39 ±0.52	1.46 ±0.64	−0.75b (166)	.46
Children	2.19 ±1.18	2.24 ±1.34	−0.39b (327)	.7	2.15 ±1.24	2.26 ±1.33	−0.54b (159)	.59	2.23 ±1.11	2.23 ±1.36	0.02b (166)	.99
Age at pilot start, y	33.5 ±8.7	33.9 ±7.2	−0.46b (328)	.65	34.5 ±8.4	34.5 ±7.0	−0.04b (160)	.97	32.4 ±9.0	33.4 ±7.3	−0.79b (166)	.43

^aχ² test.

^bt test.

Time to Exit and Exit Subsidy

Even after we controlled for work-based subsidy, Home to Stay services significantly outperformed standard services in a survival analysis using Cox regression of time to first 30-day exit from shelter (events = 298, censored = 32, total = 330: model χ²₂ = 38.231, P < .001; work subsidy Wald χ²₁ = 26.737, 95% confidence interval [CI] = 0.417, 0.674; intervention group Wald χ²₁ = 6.068, 95% CI = 0.589, 0.942; proportional hazards assumption not violated). Visual examination of the data suggested a differential effect of intervention group on clients entering shelter long before the Home to Stay pilot and those entering shelter only shortly before the pilot. To further explore this effect, we split the sample into recent-entry clients and older-entry clients and ran additional Cox regressions on intervention group effect for these subpopulations (Figure 1).

FIGURE 1—

Cox regressions of time to exit on Home to Stay or standard services for (a) older-entry clients and (b) recent-entry clients: Home to Stay Pilot, New York, NY, April 26, 2010–November 30, 2012.

These post hoc analyses revealed that the 2 conditions performed almost identically with older-entry clients (151 events, 11 censored, 162 total: model χ²₂ = 14.159, P = .001; work subsidy Wald χ²₁ = 13.443, 95% CI = 1.352, 2.701; intervention group Wald χ²₁ = 0.382, 95% CI = 0.802, 1.527; proportional hazards assumption not violated). For recent-entry clients, however, Home to Stay strongly outperformed standard services (147 events, 21 censored, 168 total: model χ²₂ = 26.800, P < .001; work subsidy Wald χ²₁ = 11.279, 95% CI = 1.285, 2.595; intervention group Wald χ²₁ = 6.451, 95% CI = 1.111, 2.266; proportional hazards assumption not violated).

We tested the proportion of clients exiting shelter with housing subsidies by means of a univariate general linear method analysis with placed with subsidy as the dependent variable, intervention group as a fixed factor, and work-based subsidy as a covariate. Home to Stay strongly outperformed standard services, with an adjusted 73% (95% CI = 66%, 81%) of the 138 Home to Stay clients exiting shelter with housing subsidies before December 1, 2012, compared with 56% (95% CI = 50%, 63%) of the 192 clients receiving standard services, model F (2,328) = 24.837 (P < .001); intervention group F(1,329) = 11.663 (P = .001).

We conducted post hoc univariate general linear method analyses with the same variables to explore the apparent split between older-entry and recent-entry clients mentioned earlier. The advantage of Home to Stay (76% adjusted; 95% CI = 65%, 87%) over standard services (49% adjusted; 95% CI = 41%, 58%) was even more marked in recent-entry clients; model F (2,166) = 17.516 (P < .001); intervention group F (1,167) = 13.516 (P < .001). For older-entry clients, however, the difference between Home to Stay (71% adjusted; 95% CI = 61%, 81%) and standard services (64% adjusted; 95% CI = 55%, 74%) was not statistically significant; model F (2,160) = 9.968 (P < .001); intervention group F (1,161) = 0.964 (P = .328).

The nearly identical time-to-exit curves for older-entry clients were reflected in very similar placement profiles for those families across the experimental and control groups. The time-to-exit curve for recent-entry clients remained largely unchanged for the experimental condition but was much slower for the control condition. This slower time to exit was reflected in a marginally significant change in placement profile for recent-entry clients in the control condition relative to their older-entry counterparts (χ²₃ = 7.631; P = .054). Looking at high-value housing subsidies (including New York City subsidies before September 2010 and a small number of longer term subsidies such as Section 8 and New York City public housing), we found a lower proportion of recent-entry clients exiting with high-value subsidies (39% vs 56%) and a higher proportion exiting without subsidy or remaining in shelter (53% vs 36%). The placement profile for recent-entry clients in the experimental condition, however, did not differ significantly from the older-entry profile (χ²₃ = 1.583; P = .66), with most clients exiting with high-value subsidies (70% vs 62%) and relatively few clients exiting without subsidy or remaining in shelter (20% vs 29%).

Return to Shelter

After controlling for work-based subsidy, we found that Home to Stay significantly outperformed standard services on a Cox regression of time to first return to shelter (Figure 2; events = 98, censored = 200, total = 298: model χ²₂ = 10.397, P = .006; work subsidy Wald χ²₁ = 3.073, 95% CI = 0.957, 2.196; intervention group Wald χ²₁ = 6.524, 95% CI = 0.379, 0.880; proportional hazards assumption not violated).

FIGURE 2—

Cox regressions of time to first return to shelter on Home to Stay vs standard services (a) without exiting subsidy covariate and (b) with exiting subsidy covariate: Home to Stay Pilot, New York, NY, April 26, 2010–November 30, 2012.

Given the expected potency of exiting subsidy type as a predictor of return to shelter, we conducted post hoc analyses to explore the relative impact of intervention group and exiting subsidy type on survival time to return to shelter. A Cox regression of intervention group on time to return to shelter generated a nonsignificant intervention group effect when we added exiting subsidy type to the model, although intervention group remained near significant and would likely be a significant predictor given greater statistical power (Figure 2; events = 98, censored = 200, total = 298: model χ²₄ = 69.286, P < .001; exiting subsidy type Wald χ²₃ = 49.715, P < .001; intervention group Wald χ²₁ = 2.439, 95% CI = 0.461, 1.091; proportional hazards assumption not violated).

Total Days in Shelter

After we controlled for work-based subsidy using a univariate general linear method analysis, the 138 Home to Stay client families spent an adjusted 376 total days in shelter (95% CI = 328, 424) between April 25, 2010, and December 1, 2012, compared with an adjusted 449 days (95% CI = 408, 490) for the 192 families receiving standard services; model F (2,•) = 18.676 (P < .001); intervention group F (1,•) = 5.150 (P = .024).

Differences in total days in shelter between intervention groups were also markedly different in the 2 shelter entry cohorts. For older-entry clients, we found a nonsignificant difference of 30 shelter days between the adjusted means for Home to Stay (mean = 411; 95% CI = 347, 476) and standard services (mean = 438; 95% CI = 379, 498); model F (2,•) = 8.745 (P < .001); intervention group F (1,•) = 0.363 (P = .548). For recent-entry families, we found a clearly significant difference of 124 days between Home to Stay (mean = 334; 95% CI = 260, 407) and standard services (mean = 459; 95% CI = 403, 516); model F (2,•) = 11.033 (P < .001); intervention group F (1,•) = 7.083 (P = .009).

DISCUSSION

Home to Stay significantly outperformed standard shelter services in survival analyses of client family time to exit from and first return to shelter, as well as the percentage of those exiting shelter with housing subsidies and the total days spent in shelter. Post hoc analyses suggest that the impact on time to exit was attributable to faster placement of clients who had entered shelter within 6 months of the pilot start date. This performance advantage appeared to result in a higher proportion of recent-entry clients in the experimental group than the control group exiting shelter with high-value housing subsidies. This relative advantage in high-value subsidies may have accounted for much of the intervention’s impact on time to first return to shelter.

Both Home to Stay and standard services appeared to be effective when working with older-entry clients who entered shelter more than 6 months before the pilot start date, although nonsignificant differences did favor Home to Stay for all 4 outcomes. Both interventions placed most clients in the older-entry subpopulation quickly enough to obtain high-value housing subsidies while they were still available. Home to Stay, but not standard services, was also successful in rapidly placing, with high-value vouchers, recent-entry families who entered shelter within 6 months of the pilot start date.

The study data do not reveal what drove this interaction of shelter entry cohort with intervention, but we believe the most likely explanation is the following: The Home to Stay program specifically targeted episodic–recidivist families with the intention of placing them into housing as quickly as possible. Approximately 50% of those families were part of the older-entry subpopulation, but the distinction was never explicitly noted, nor was it operationally relevant. However, episodic and recidivist clients made up only a small fraction of standard service providers’ caseloads, and clients’ episodic or recidivist status was neither noted nor relevant. For these providers, financial incentives prioritized work with older-entry clients, who made up half of shelter caseloads. This focus was likely reinforced by the awareness that many clients can exit shelter with minimal assistance, making it inefficient to expend substantial resources on recent-entry clients. From the shelter caseworkers’ perspective, a single recent-entry episodic or recidivist client family was indistinguishable from the many other recent-entry client families on a caseload. Nothing indicated that these particular families either required or would reward more intense effort, and providers followed their incentive structure and focused efforts on the older-entry subpopulation, which happened to include half of the episodic and recidivist families.

It appears likely that Home to Stay strongly outperformed standard services for recent-entry episodic and recidivist client families because standard service providers had an incentive to deprioritize these clients as recent entry and no reason to prioritize them as episodic or recidivist, whereas Home to Stay providers had no incentive to deprioritize them as recent entry and targeted them as episodic or recidivist. Older-entry episodic–recidivist clients, however, were both targeted by Home to Stay as episodic or recidivist and prioritized by standard service providers as older entry, resulting in a much smaller performance gap.

If correct, this interpretation may have important implications. First, Home to Stay services in a shelter system with robust housing subsidies might significantly outperform standard services in time to exit for recent-entry but not older-entry episodic or recidivist clients. Even for recent-entry clients, the impact on recidivism might be limited because time to exit would not be a significant determinant of exit subsidy. Shelter systems might be able to reduce time-to-exit performance gaps by incentivizing rapid placement of episodic and recidivist clients immediately into shelter. Given the low prevalence of such clients, this rapid placement would only slightly increase the overall proportion of high-priority clients and might also be a practical approach in communities with too few episodic or recidivist homeless clients to justify a Home to Stay program. It may be difficult, however, to identify a simple and effective incentive structure, and financial incentives limited to a small subset of homeless families may provide insufficient leverage to modify providers’ behavior.

This study has several significant limitations. The changes in housing subsidies and incentive structures that occurred during the course of the study greatly complicate interpretation of the data. The atypical randomization procedure, designed to comply with New York City DHS requirements while minimizing threats to validity, may have failed to eliminate some such threats. The study offers no insight into the relative importance of specific elements of the Home to Stay model. Findings based on episodic and recidivist homeless clients may not generalize to other homeless clients. Finally, outcomes obtained in the New York City shelter system may not generalize to other communities and service systems, and the study’s loosely defined and extremely heterogeneous control condition make it hard to determine which communities are most likely to benefit from Home to Stay services.

Relative to standard services, Home to Stay services can accelerate exit from shelter and reduce return to shelter and total sheltered days for episodic and recidivist homeless families. Standard shelter services may be able to narrow this performance gap by incentivizing work with all episodic and recidivist homeless families.