Algorithmic risk assessment tools have become a popular element of criminal justice reforms, often with the explicit goal of reducing incarceration rates. The hope is that these data-driven tools will standardize decisions about pretrial detention and sentencing, and ensure that only the most high-risk offenders are incarcerated. However, the effects of these tools depend crucially on how judges use them.
This brief considers judicial reforms in Kentucky and Virginia as case studies of the effects of algorithmic risk assessment tools in practice. We show that, in both states, reforms aimed at reducing incarceration for low-risk offenders had little to no impact on incarceration rates. While these tools clearly recommended less incarceration for a large share of defendants, they had little effect on judges’ incarceration decisions. However, there is tremendous variation across judges in how closely they follow the risk assessment recommendations. We discuss what these results mean in terms of how to affect meaningful criminal justice reform, including ways citizens and policymakers can align judicial incentives so that reforms have their intended effects.