Predicting Homeless Shelter Entry

In partnership with the NYU Furman Center, the Center for Innovation through Data Intelligence (CIDI) analyzed administrative data on human services and neighborhoods to predict families' risk of homelessness and to identify buildings that are likely to house at risk families.

To do so, the project examined families that received cash assistance or Medicaid benefits at some point from 2006 to 2015. The researcher team then linked these data to information on homeless shelter applications and stays, building characteristics, and neighborhood characteristics. With this sample, temporal cross-validation in combination with machine learning algorithms were used to predict (1) family shelter applications and (2) buildings with families at risk of entering shelter in 2013, 2014, and 2015. The research additionally explored whether these algorithm-driven predictions could improve the ability of homeless prevention programs to find and assist those most at risk of homelessness.

Findings

Predictions of family shelter applications on out-of-sample data were 20 times better than random guessing and 1.5 times better than predictions based on currently seeking homeless prevention assistance. The most important predictors of these shelter applications were previous application to shelter, living in a building that previously housed a homeless family, and receipt of family assistance.

Predictions of buildings with families at risk of entering shelter were 20 times better than random guessing and 1.3 times better than predictions based on current information used for building outreach. The most important predictors were neighborhood rates of shelter entry, housing code violations, and emergency repairs.

This work suggests that use of prediction analysis of homelessness can help prevention programs find families most likely to benefit from assistance as well as specific dwellings mostly likely to house those most at risk of homelessness and so most likely to benefit from outreach.

Documents

Predicting Homelessness for Better Prevention Research Brief (full)
Predicting Homelessness for Better Prevention Research Brief (short)

Partners

NYU Furman Center
NYC Department of Homeless Services
New York City Department of Social Services
NYC Human Resources Administration
NYC Department of Housing Preservation and Development
New York State Office of Court Administration