Predictive Power at What Cost? Economic and Racial Justice of Data-Driven Algorithms

Abstract

This paper studies how algorithms use variables to maximize predictive power at the cost of group equity. Group inequity arises if variables enlarge disparities in risk scores across groups. I develop a framework to examine a recidivism risk assessment tool using risk score and novel pretrial defendant case data from 2013-2016 in Broward County, Florida. I find that defendants' neighborhood data only negligibly improve predictive power, but substantially widen disparities in defendant risk scores and false positive rates across race and economic status. Higher risk scores may lead to longer pretrial incarceration and downstream consequences, by impacting labor market outcomes. These findings underscore that machine learning objectives tuned to maximize predictive power can be in conflict with racial and economic justice.