A model for targeting prevention services to families at greatest risk of homelessness

One strategy for preventing homelessness involves providing services to people at risk. Such a strategy can succeed only if the services in fact prevent homelessness and they are targeted to people who are in fact at risk. Showing that recipients avoid homelessness does not demonstrate prevention – this result could reflect poor targeting rather than effective services. Services would appear even more effective if given only to millionaires. This paper focuses on how to target services better. We used intake data from 11,105 families who applied for preventive services from New York City's community-based “HomeBase” prevention program to predict shelter entry over the next three years. Intake workers judged 62.4% to be eligible for services. Others were denied for reasons including insufficient housing risk or residing in the wrong catchment area. Modeling included multiple imputation of missing data and survival analysis. No variable interacted with eligibility status in the prediction of homelessness (as might be the case if services counteracted particular risk factors). From the full model we developed a simplified screening model involving 15 variables, each scored 1, 2 or 3 points, that is 29% more effective than the current system in identifying families who enter shelter, holding constant the proportion of applicants targeted. A graph of hit rate (sensitivity) vs. false alarm rate (1-specificity) shows the effects of targeting different proportions of applicants. Comparisons between families who were and were not eligible for services suggest that prevention services were increasingly effective for families at higher levels of risk.