COMPAS and State v. Loomis — a Proprietary Recidivism Score Allowed at US Sentencing, With Mandatory Warnings

COMPAS is a commercial risk-assessment tool (Northpointe, now Equivant) that scores defendants' risk of re-offending from a 137-item questionnaire. Wisconsin's Department of Corrections used it in pre-sentence investigation reports.
In State v. Loomis (881 N.W.2d 749, decided 13 July 2016) the Wisconsin Supreme Court held that a sentencing court's consideration of a COMPAS score does not violate due process if used with awareness of its limits. It required every pre-sentence report containing a COMPAS score to carry written advisements: the tool's proprietary method is undisclosed, it was normed on national rather than Wisconsin data, studies suggest it may classify minority offenders as higher risk, it needs ongoing monitoring and recalibration, and it was designed for corrections, not sentencing. Scores cannot determine whether a defendant is incarcerated or how severe the sentence is.
Independent evidence on performance is unflattering. Dressel and Farid (Science Advances, 17 January 2018), using 1,000 Broward County, Florida defendants, found COMPAS 65.2% accurate versus 67% for crowd-sourced volunteers given seven features, and a two-feature model (age and prior convictions) performed about equally.
Lesson: a court-imposed disclosure-and-limits regime is a governance mitigation, not validation. The case is a cautionary reference for public bodies buying opaque scoring tools; it shows what judicial guardrails can and cannot fix.

Read the full analysis: https://law.justia.com/cases/wisconsin/supreme-court/2016/2015ap000157-cr.html