COMPAS vs HCR-20 (Historical Clinical Risk Management-20)

COMPAS excels at providing rapid, statistically-derived risk scores by analyzing a large set of static and dynamic factors. Its strength lies in algorithmic consistency and scalability, processing a standardized 137-item questionnaire to generate decile-based risk scores for general recidivism, violent recidivism, and pretrial misconduct. For example, a 2016 ProPublica analysis highlighted its 61% accuracy for violent recidivism prediction, but also revealed significant racial disparities in false positive rates, underscoring the tool's predictive efficiency and its ethical vulnerabilities.

HCR-20 takes a fundamentally different approach by using a Structured Professional Judgment (SPJ) framework. Rather than replacing the clinician, it guides them through a systematic evaluation of 20 key Historical, Clinical, and Risk Management items. This results in a final risk formulation—Low, Moderate, or High—that integrates actuarial data with professional expertise. The trade-off is a more nuanced, context-aware assessment that excels in dynamic risk management and violence prevention planning, but requires extensive training and time, with inter-rater reliability heavily dependent on the evaluator's skill.

The key trade-off: If your priority is rapid, scalable, and statistically consistent triage for high-volume court processing, choose COMPAS. If you prioritize a comprehensive, formulation-based analysis that directly informs individualized treatment and risk management strategies in forensic clinical settings, choose HCR-20. The decision hinges on whether the system requires an efficient sorting mechanism or a deep, defensible clinical workup for high-stakes decisions.

COMPAS excels at providing rapid, consistent, and low-cost risk scores because it automates the entire assessment process using a fixed algorithm. For example, in a high-volume pretrial context, COMPAS can process a defendant's criminal history and generate a recidivism score in minutes, requiring no clinical training to administer. This algorithmic consistency eliminates inter-rater reliability concerns, ensuring that two defendants with identical records receive the same risk classification. However, this strength is also its primary weakness: the tool is a 'black box' that cannot account for dynamic, contextual factors like a recent change in medication compliance or a specific, credible threat made against a victim.

HCR-20 takes a fundamentally different approach by structuring, rather than replacing, professional judgment. It guides a trained clinician to evaluate 20 key historical, clinical, and risk management items, resulting in a final risk formulation of Low, Moderate, or High. This process excels in forensic settings where dynamic risk management is paramount, such as formulating a violence prevention plan for a psychiatric patient being discharged into the community. The trade-off is significant: HCR-20 assessments are time-intensive, require specialized training, and are subject to inter-rater variability, with studies showing intraclass correlation coefficients for the final summary risk rating typically ranging from 0.65 to 0.85, compared to COMPAS's perfect algorithmic consistency.

The key trade-off: If your priority is algorithmic consistency, rapid triage, and cost-effective screening for a large population, choose COMPAS. If you prioritize dynamic risk formulation, violence prevention planning, and a defensible integration of clinical context for a high-stakes individual case, choose HCR-20. For many justice systems, the optimal architecture is a sequential model: use COMPAS as an initial screener to identify high-risk cases, and then deploy HCR-20 for a deeper, formulation-driven assessment of that smaller, high-risk cohort.