Association for Public Policy Analysis & Management

The U.S. Federal Government aims to end homelessness for families by 2020. Meeting this goal will require communities to design and implement programs that effectively alleviate housing insecurity among vulnerable families.  New York City’s Department of Social Services (DSS) is currently developing and testing new homelessness prevention programs, using predictive analytics to guide initial program design, and process and impact evaluation to inform ongoing planning and contribute to the evidence base.  DSS’s Office of Evaluation and Research (OER) developed a high risk family profile based on a 2012 cohort of adult cash assistance clients, utilizing survival modeling to predict homeless shelter application in the subsequent two years.  The first pilot effort, implemented in 2015, identified approximately 2,000 cash assistance clients with children who appeared at risk of becoming homeless based on factors such as a prior homelessness spell, lack of subsidized housing, a short current spell on cash assistance, history of noncompliance with cash assistance program rules, and indications of substance use or disability.  Given the program leadership’s decision to roll out the pilot by ZIP code, we used a quasi-experimental evaluation design that split the high risk individuals into a treatment and a comparison group, carefully matched by demographic and economic characteristics to ensure similarity across the two groups.  Treatment group clients were contacted by DSS’s homelessness prevention office, asked a series of housing questions to assess their current housing insecurity, and referred to an array of services for which they might be eligible that could help mitigate their housing insecurity (e.g., help paying rent arrears for families behind on rent, or legal assistance for those having disputes with their landlord) and address other potential sources of household instability.  Evaluation results show that the pilot achieved a good contact rate and succeeded in identifying cash assistance clients at risk of applying to stay at a homeless shelter, with clients’ geographic location playing a key role.  Despite the successful outreach, service receipt and shelter application rates were statistically similar between the treatment and comparison groups, suggesting minimal impact of the pilot on reducing homeless shelter applications.  Recognizing that a highly targeted intervention among a small group of those most at risk would likely have a small impact on the overall homeless shelter census, DSS also designed a second “lower-touch” intervention in 2016 with contact via mail, targeted to a larger group of 10,000 clients with a history of shelter use.  For this intervention, HRA consulted with a behavioral economics research team to enhance outreach effectiveness, using a random assignment design to test the relative effectiveness of different communications approaches.  Clients were randomly assigned to one of two treatment groups, each receiving a different mailer.  OER will evaluate this second pilot using matched HRA-Department of Homeless Services administrative data, examining key process measures (including assessment rate, categories of self-reported issues, and referral type), and impact on service receipt and subsequent homeless shelter application.  This paper presents our predictive analytics, program development for both pilots, and evaluation findings for the first pilot.