Association for Public Policy Analysis & Management

Panel: Using Data to Combat Homelessness: National and State Level Evidence

(Housing and Community Development)

Homeless assistance in the United States is increasingly emphasizing prevention oriented approaches, which aim to help households at risk of homelessness maintain their housing or help currently homeless households regain housing as quickly as possible. The nascence of this approach means that program implementation has progressed faster than the evidence to support it; this panel addresses that gap by evaluating homelessness prevention and rapid rehousing (HPRP) across a range of housing and healthcare outcomes and assessing a method of improving targeting and program efficiency. The paper by Rodriguez and Eidelman evaluates the impact of homelessness prevention and rapid re-housing on subsequent homelessness in Georgia. The study uses administrative data to compare incidence of shelter use among households exiting HPRP to a control group of households exiting transitional housing. Findings that HPRP reduces homelessness rates for families with children and single adults provide evidence for program effects across populations and support for the policy. Byrne evaluates the impact of homelessness prevention and rapid rehousing on healthcare utilization among a national sample of Veterans accessing the Supportive Services for Veteran Families (SSVF) program. A key aim of SSVF and similar programs is to assist individuals and families in accessing healthcare and other mainstream benefits that may affect their housing stability, but no study has examined the extent to which they are successful in achieving this aim. This study compares the extent and volume of VA health and behavioral services utilization between a cohort of Veterans who received SSVF services and a comparison group of at-risk and currently homeless Veterans produced through propensity score matching. Byrne finds that program impacts vary between physical and behavioral health services and across modalities, and suggests program responses. Treglia’s paper evaluates a tool addressing a known weakness of homelessness prevention: targeting. Prevention providers target households who will become homeless without their services to ensure that savings from reduced shelter use exceed program costs, but predicting which households will become homeless is extremely difficult. This study uses a machine learning algorithm – random forest – to forecast subsequent homelessness among Veterans accessing VA outpatient care who are at risk of homelessness, and compares predictive accuracy to more commonly used logistic regression. Random forest is significantly more accurate than the logistic model, correctly predicting more truly homeless individuals and reducing the rate of false positives. Using machine learning to screen for services can positively impact homeless service providers, improving program efficiency and the ability to ensure that services are provided to the most appropriate populations.