Machine Learning Forecasts of Risk to Inform Sentencing Decisions

Abstract

Subjective judgment, sometimes called "clinical judgment," is an approach that relies on intuition guided by experience. The resulting risk assessments are often wildly inaccurate and their rationale opaque. "Actuarial" methods depend on data that allow one to link "risk factors" to various outcomes of interest. The associations found can then be used to forecast those outcomes when they are not known. Over the past several decades, regression statistical procedures have dominated the actuarial determination of empirically based risk factors. By and large, this enterprise has been a success. But the increasing availability of very large datasets coupled with new data analysis tools promise dramatically better success in the future. Machine learning will be a dominant statistical driver. Here, Berk and Hyatt discuss machine learning statistical procedures that will forecast risk to inform sentencing decisions.