Accuracy and impact of supervision risk assessment tools across racial groups

Abstract Objectives Estimates the predictive validity of the Ohio Risk Assessment System Community Supervision Tool (ORAS-CST) and racial differences in outcomes both across assessed risk levels and against a comparison group of unassessed individuals. Methods Using a dataset of individuals under supervision across four Pennsylvania counties in 2017 ( n = 6, 339 ), we perform Cox proportional hazard modelling to test for racial differences in the association between risk level and recidivism (up to 5 years). We implement a propensity-score matching model to estimate the potential effects of risk assessment on future recidivism across race. Results Risk assessment shows good predictability across racial subgroups with no statistically significant difference when risk level is interacted with race. We estimate that assessment reduces one-year recidivism rates by 8% points ( p < 0.001). Conclusions While we find similar effects for racial subgroups separately at one year, estimated treatment effects of assessment fall to zero for Black individuals by year five.