Encoding Inequality: A Pluralistic Fairness Evaluation of Pretrial Risk Assessment in the U.S. Criminal Justice System

Abstract

Pretrial risk assessment tools are increasingly used to inform detention and release decisions, yet debates over their fairness often rely on limited validation metrics. This thesis compares COMPAS and the Ohio Risk Assessment System-Pretrial Assessment Tool (ORAS-PAT) and argues that existing aggregate validation obscures structurally unequal impacts across racial groups. In the absence of publicly available ORAS data, the study reconstructs ORAS-style scoring and validates it using Broward County defendant data published by ProPublica in 2016. By examining error rates, risk classifications, and base-rate differences, the analysis demonstrates how race-neutral algorithms can reproduce racial disparities through structural bias, emphasizing the need for more equity-oriented validation standards in pretrial risk assessments.