Evaluating Bias, Shrinkage and the Home-Field Advantage: Results from a Revalidation of the MnSTARR 2.0

Abstract

ABSTRACT In 2016, the Minnesota Department of Corrections implemented the Minnesota Screening Tool Assessing Recidivism Risk 2.0 (MnSTARR 2.0), a fully-automated, gender-specific instrument that predicts multiple types of recidivism. Using multiple performance metrics, this study externally validates the MnSTARR 2.0 on a sample of 8,997 people released from Minnesota prisons. The results show that while the MnSTARR 2.0 overestimated risk for felony and nonviolent recidivism, it achieved adequate predictive discrimination, with an average area under the curve (AUC) of 0.73 for males and 0.74 for females. The MnSTARR 2.0 produced predictions that were generally consistent with the observed recidivism outcomes by race/ethnicity, and the amount of “shrinkage” from the internal validation sample to the external validation sample was minimal. The MnSTARR 2.0 significantly outperformed the LS/CMI on a sub-sample of individuals who received both assessments, which provides additional evidence that customized instruments tend to perform better than off-the-shelf tools that are designed to be used with a variety of correctional populations.