Association for Public Policy Analysis & Management

There has been growing evidence that permanent supportive housing is an effective prevention policy for long-term homelessness. Since housing resources are limited, one of the key challenges of the intervention is how to identify and target the “right” people to be efficient.  Predictive models are a very efficient and effective way of matching people with the highest costs with the available supply of housing and generate cost offsets equal or higher than the cost of the interventions.

Santa Clara County Triage Tool was developed using an integrated database built by linking eleven agencies’ administrative records over a six-year period, which provide demographic, clinical and service use information. The model with 38 predictors was validated and showed a very high accuracy and performance in predicting high-cost users.

Having 6 years of data, a dynamic analysis of false positives and false negatives was conducted based on pre and post prediction period actual costs. The analysis estimated significant cost-offsets and showed that the model was particularly effective in separating cases with one-year cost spikes from cases with steady high costs. The model is expected to assist Santa Clara County as an efficient resource allocation tool to improve their targeting with significant cost offsets.