The impact of algorithmic risk assessment tool legislation on racial disparities in criminal sentencing

Abstract

The prevailing presence of racial disparities in criminal sentencing motivated the introduction of algorithmic risk assessment tools (RATs) in the U.S. judicial system. These tools provide judges with an algorithm-generated risk score and sentencing recommendation to consider in their decisions. Although this technology is well-intentioned, researchers find that RATs produce racial disparities in their outputs. My research examines the impact of state laws regulating the use of RATs on racial disparities in sentence length and likelihood of receiving probation. Utilizing Gardner's (2021) two-stage differences-in-differences methodology, I exploit the natural experiment arising from 29 states passing some form of RAT law at different times. I find that the impact of RAT laws depends on the components of the state's RAT law, and that the effect varies by racial group. My results suggest that RAT laws significantly decrease the racial sentencing disparity for Hispanics, but increase the disparity for Blacks. Although my results are somewhat sensitive to specification, they still bear critical policy implications regarding the use of RATs in the judicial system.