Toward Understanding Homelessness: The 2007 National Symposium on Homelessness Research. Accountability, Cost-Effectiveness, and Program Performance: Progress Since 1998.. Supportive Housing


Published studies of various housing interventions have demonstrated that housing does indeed resolve homelessness, though to varying degrees depending on the nature of the intervention, and with varying cost offsets. An experimental study in Boston (Dickey et al., 1997) of people who were homeless with severe mental illness compared people who were placed in an “evolving consumer” group living arrangement with similar persons placed in independent subsidized housing. The group living model had staff supervision, which was expected to decline over time as consumers developed a mutually supportive community. Housing outcomes were positive in both conditions, with retention rates at 18 months of 83 percent in independent housing and 92 percent in the group settings. No changes in clinical or functional outcomes were associated with either type of placement. The cost of the group living condition was much greater than the cost of the independent living condition, mostly attributable to the staffing costs. There was not a “no housing” condition in the study, so it is not possible to estimate the degree to which housing placement in either condition was associated with reduced inpatient days or other services use. Rosenheck notes that inpatient services use at baseline in both conditions was modest, more in line with the national samples. This would suggest that the intervention included a broadly representative sample of people who were homeless with mental illness and did not target heavy service users. Thus, it is unlikely that an offsetting cost effect would have been found.

Several studies have included study designs that enable some inferences about the impact of housing on services utilization and costs. A quasi-experimental study (Culhane, Metraux, & Hadley, 2002) with a pre/post design and a matched control group evaluated the New York-New York Agreement, a joint state- and city-funded initiative to develop supportive housing for people who were homeless and had a severe mental illness in New York City. The study analyzed administrative records from seven service systems to estimate the impact of supportive housing placement on services utilization for two years post placement as compared to two years prior to placement. The study found that supportive housing placement was associated with declines in hospitalizations, incarcerations, and shelter stays. Ninety-five percent of the costs of the supportive housing were offset by service reductions ($17,200 per unit per year), resulting in an estimated net annual cost of the supportive housing programs of approximately $1,000 per unit per year. While an advantage of the study was its inclusion of multiple systems to measure impacts on services use and costs, it did not include all potential costs, including police and court costs, emergency medical transport, and emergency room costs. The primary limitation of the study is that it did not involve randomization. It is possible that there was sample selection bias, and that people were selected for housing only if they were sufficiently stable or had received sufficient treatment before program entry. Indeed, as noted in previously, the pre-intervention average cost of $40,500 per person per year suggests that a heavy service-using population was targeted for the intervention and that such high utilization made it possible to achieve such a high degree of offsetting costs.

Rosenheck, Kasprow, and Frisma (2003) conducted an experimental study of a housing intervention for veterans who were homeless and had a severe mental illness. The housing condition included a Section 8 housing voucher coupled with intensive case management. It was compared to a condition of intensive case management alone and a standard care condition. The results estimated potential cost offsets through administrative data from the Veterans Administration that track health services use and through participant self-report in a quarterly interview for non-VA health and other services use (including shelter and jail stays). The results found generally positive housing outcomes regardless of the study condition, including standard care. The housing voucher and intensive case management condition was associated with greater cost offsets, and the intervention had a net annual cost of approximately $2,000 per unit per year. According to the authors, a limitation of the study is that attrition was high in the “standard care” condition. It is possible that persons with recurrent homelessness or in other unstable circumstances were differentially lost to follow-up, biasing the sample in the standard care condition to more favorable outcomes, and thereby lessening the observed differences with the experimental conditions. In any case, although the study was based in part on self-report and included randomization, the findings are roughly consistent with the $1,000 net cost found in the NY-NY evaluation.

A supportive housing intervention in San Francisco (Martinez & Burt, 2006) tracked people who were homeless and who had at least two qualifying conditions (an axis I or II mental disorder, a substance use disorder, or HIV/AIDS). A list of eligible persons was generated through a solicitation of applicants recruited at shelter and feeding programs. Just over 200 people were randomly selected from the list to enter one of two supportive housing programs. Their health service use at the San Francisco General Hospital for two years prior to program entry was compared with service use for two years following placement. A small control group from the waiting list was also selected for comparison purposes. The results show that housing placement is associated with declines in emergency room use, hospitalizations, and inpatient days. The authors estimate that the service reductions offset approximately $1,300 of the cost of the intervention, or 10 percent. The authors attribute the comparatively small cost offsets to their limited access to administrative records, having included only visits to the San Francisco General Hospital, and not including other health or social welfare systems. 

Schumacher, et al. (2002) focused only on people with substance abuse disorders (particularly addiction to crack cocaine), rather than mental illness, and randomly assigned people to either abstinence- and work-contingent housing and treatment or day treatment alone. The results show that the enhanced services model (housing, work, and treatment) cost more per person (approximately $7,700 versus $3,300 per year). The authors did not investigate any potential cost offsets from other service systems that might be associated with the enrollment in the study, but it is possible that shelter, criminal justice, and other health system costs could have reduced the net cost of the intervention. Given the modest success of the program (abstinence was higher at 6 months in the enhanced condition but there was no significant difference at 12 months) and given the modest cost, the authors argue that investments in programs providing housing and treatment to adults who are homeless with addictions are comparable in net cost to other common social interventions of comparable value.

A quasi-experimental study by Clark and Rich (2003) compared people who were homeless with a severe mental illness and who were placed in a comprehensive housing program, including rental subsidies, support services, and case management, to similar people who were placed in a case management-only condition. The results again indicated that housing outcomes were positive irrespective of the study condition. However, persons who had high psychiatric symptom severity and high substance abuse had better outcomes in the comprehensive housing than in the case management alone condition. Although the authors did not include a cost component in their analysis, they conclude that “the effectiveness, and ultimately the cost, of homelessness services can be improved by matching the type of service to the consumer’s level of psychiatric impairment and substance use, rather than by treating mentally ill homeless persons as a homogeneous group” (p. 78).

The fact that several of the studies reviewed here have found positive housing outcomes over time, irrespective of treatment conditions, reinforces the idea that narrower program targeting might be considered as a means of improving the efficiency and effectiveness of programs, as suggested by Clark and Rich (2003). As has been previously noted, most people who experience homelessness, including most people who have a severe mental illness, exit homelessness quickly and do not return within three years; most do so without formal exit support from the homelessness service system (Kuhn & Culhane, 1998). The subpopulation with histories of heavy services use, who experience chronic homelessness or who are otherwise unable to exit homelessness without added supports, should be considered the priority target of the more costly interventions. They not only are the people most likely to need them, but they also are the group for whom the opportunity for demonstrating cost-effectiveness is much greater.

Finally, as with “cost studies,” there has been a recent surge in interest in cost offset studies, particularly for interventions targeting people who are chronically homeless. Nineteen such studies have recently been identified by the U.S. Interagency Council on Homelessness (ICH) — more than have been conducted in the entire period preceding 2003. The Appendix exhibit provides ICH’s summary of 14 studies. The studies are being led by a variety of people, including consultants, academics, planners, and advocates. The interventions are primarily permanent supportive housing programs, some of which include an ACT component. The samples are typically enrollees in the various interventions. They do not appear to have been randomly assigned, so, the opportunity for selection bias exists. Because the people targeted by these projects have experienced chronic homelessness, they are likely to be relatively higher service users, as compared to the homeless population in general. Therefore, they are among the populations for whom significant cost offsets are likely to be achieved following a housing intervention. Most of the studies do not include comparison groups and appear to be pre-post designs. The studies vary in the degree to which they included multiple service systems and in the comprehensiveness of their data, but most have included at least some health services data and some criminal justice systems data. Of course, data coverage determines the ability to find costs as well as cost offsets, and variability in coverage and in other study design issues make the studies not truly comparable.

Results of the completed studies indicate substantial reductions in services use associated with both the ACT and supportive housing interventions (half of the studies are ongoing and do not yet have results). Perhaps not surprisingly, the largest study, based on nearly 5,000 formerly homeless people with mental illness in California, reports the most modest cost offset, at $5,614 in hospitalization, incarceration and emergency room costs. The size of the sample suggests that the intervention did not target people who were chronically homeless, but may have been more broadly representative of people with mental illness who experienced some type of homelessness. Several of the studies report annual cost reductions per person in the range of $13,000–$18,000, which would be roughly consistent with cost offset found in the NY-NY evaluation. Intervention costs are not shown in the Appendix exhibit summarizing the studies, so it is not possible to discern here the degree to which these service reductions offset the costs of the interventions. But such average reductions would appear to be fairly substantial, comparable to the costs of supportive housing in many jurisdictions.

The amount of interest and activity in this area suggests that there is a great appetite for research of this nature. The 10-year planning processes and the resulting housing efforts targeting people who are chronically homeless have no doubt inspired communities to evaluate their progress. But this is an area where federal leadership and support could make a valuable contribution, both by helping to establish standards and comparability in the research and by providing funding that would engage academic researchers as partners with local planners and implementers. The use of administrative records and the lack of randomization would seem to preclude NIMH as a source of funding for these efforts. Many of these communities are seeking more timely feedback than is likely in most prospective, longitudinal studies based on primary data collection. While research of that nature continues to be needed, mechanisms are also needed for supporting qualified research based on administrative databases and research that is more evaluation than experimental in nature.

View full report


"report.pdf" (pdf, 561.34Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®