Biased by Design? The Structural and Technical Limitations of Pretrial Risk Assessment Tools in Criminal Courts

Pretrial risk-assessment tools are increasingly being deployed in U.S. criminal courts to assist judges with release and detention decisions. This policy brief, prepared for the Pennsylvania House Judiciary Committee, examines the structural and technical limitations of these systems. Drawing on the algorithmic fairness literature, the brief identifies two distinct sources of disparate outcomes: structural drivers, where tools inherit biases from policing data that reflects where police patrol rather than where crime occurs; and technical constraints, where, when base reoffending rates differ across demographic groups, no algorithm can satisfy all fairness metrics simultaneously.

Achieving predictive parity produces unequal error rates across groups; equalizing error rates violates predictive parity. The choice between fairness metrics is therefore a value judgment, not a technical optimization. The brief outlines three policy levers available to legislators: addressing the upstream resource constraints that drove courts to adopt these tools, mandating equal-error-rate fairness over predictive parity, and requiring input disclosure and third-party audits to enable independent oversight.