Discrimination in the Criminal Justice System: A Mechanism for Balancing Misclassification by Race

Abstract

This paper demonstrates a method of fairly classifying whether criminal defendants of different races will recommit crimes when released before trial. This thesis uses existing predictions of recidivism—the likelihood of recommitting a crime—from the Correctional OffenderManagement Profiling for Alternative Sanctions (COMPAS) algorithm. The motivation for this thesis comes from glaring asymmetries in error rates systematically committed by theCOMPAS algorithm, where black defendants are incorrectly labeled as high-risk (false positive) and white defendants are incorrectly labeled as low-risk (false negative). This paper creates adjusted score thresholds that balance false positive errors, false negative errors, and both false positive and negative errors at the same time. The result is a readjusting of errors that does not unfairly punish one demographic.