Evaluating Algorithmic Risk Assessment

Algorithmic risk assessment is hailed as offering criminal justice officials a science-led system to triage offender populations better manage low- versus high-risk individuals. Risk algorithms have reached the pretrial world best practices method aid in reforms reduce reliance upon money bail and moderate detention’s material contribution mass incarceration. Still, these promises are elusive if algorithmic tools unable achieve sufficient accurate rates predicting failure. This article presents an empirical study of most popular tool used United States. Developers promote Public Safety Assessment (PSA) national tool. Little information known about PSA’s developmental methodologies or performance statistics. The dearth intelligence alarming being high-stakes decisions whether detain individuals who not yet been convicted any crime. uncovers evidence accuracy using variety validity metrics and, novel contribution, investigates use three diverse jurisdictions evaluate how well generalizes real-world settings. Policy implications findings may be enlightening officials, practitioners, other stakeholders interested across decision points.