Machine Learning Risk Assessment at Preliminary Arraignments for Domestic Violence

WP 2016-2.0

Richard A. Berk

Risk assessments have a long history in criminal justice settings. Since the 1920s, parole release decisions in many jurisdictions have been informed by projections of how inmates would perform under supervision once released. Over time, these assessments have relied more heavily on actuarial methods and increasingly sophisticated statistical procedures. Sentencing decisions are undergoing a similar evolution from lengthy pre-sentencing reports to legislatively mandated statistical forecasts of risk. Beginning in the 1980s, pre-trial decisions have in many jurisdictions been shaped by actuarial forecasts of flight risk and of new crimes committed while an offender is awaiting trial.

PDF icon WP2016-02_Berk_DVReport_05.16.2016.pdf