Veteran Homelessness 2000-2010: Examining a Comprehensive Model of Outcome Predictors

In this presentation, we examine a multi-domain model of predictors of housing intervention outcomes in a large VA-funded, transitional, community-based housing program. Data for this study were provided by the VA Northeast Program Evaluation Center and consisted of records of individuals receiving services in community-based housing intervention programs supported by the VA Grant and Per Diem program during 2000-2010. This program defrays the cost of housing support/supportive services for up to two years and is designed as a transitional program leading to permanent housing. The dataset was derived from comprehensive intake, admission, discharge, and facility descriptor data. Analyses were based on a dataset consisting of 100,167 records for 68,921 veterans who were admitted to one of 435 provider programs. We examined a comprehensive model of predictors for several intervention outcomes (e.g., program completion, employment on discharge, improvement in substance abuse). This model included predictors in the following domains: demographic characteristics, pre-admission status (e.g., employment, housing), substance use (e.g., recent use, hospitalizations), mental health status and history, health status (e.g., chronic illnesses, disease index), intervention components (e.g., length of stay, number of supportive services), and facility characteristics (e.g., program philosophy). Hierarchical logistic regression was used to examine outcome predictors. In order to estimate the amount of variance in outcomes (e.g., program completion) explained by each domain, we calculated the McKelvy-Zavoina index, which serves as a measure of effect size. We will present results with a focus on domains of predictors that have the most explanatory power in estimating housing intervention outcomes.