The Predictive Performance of Risk Assessment in Real Life: An External Validation of the MnSTARR

Abstract

ABSTRACT Using multiple performance metrics, this study externally validates the Minnesota Screening Tool Assessing Recidivism Risk (MnSTARR) among a sample of 3,985 inmates released from Minnesota prisons in 2014. While the Minnesota Department of Corrections implemented a fully-automated risk assessment (MnSTARR 2.0) in 2016, the original MnSTARR was a manually-scored, gender-specific recidivism risk assessment that predicted multiple types of recidivism – felony, nonviolent, violent, and both first-time and repeat sexual offending (only for males). The results show the MnSTARR achieved adequate predictive performance. The average area under the curve (AUC) was 0.73 for males and 0.77 for females. Nonetheless, the MnSTARR would have achieved better predictive performance had it used an automated scoring process. Further, the findings showed the MnSTARR performed better for Whites than Nonwhites, and the magnitude of this difference would have been minimized using automated scoring. In sum, while the MnSTARR had adequate validity, performance is likely to be improved with automated systems.