Main content

A Rejoinder to Dressel and Farid: New Study Finds Computer Algorithm Is More Accurate Than Humans at Predicting Arrest and as Good as a Group of 20 Lay Experts

Volume 82 Number 2
Date: 
September 2018
Abstract: 

The authors respond to a recent article by Dressel & Farid that called into question the accuracy and fairness of the COMPAS risk assessment tool specifically and all statistically based prediction tools more generally. Dressel and Farid argue that laypeople are at least as accurate and fair in predicting reoffending as statistically based risk assessment instruments empirically designed to predict reoffending. The authors closely examine premises, methodology, and conclusions, focusing on some omissions and incorrect assumptions; in addition, while Dressel and Farid focus on the binary decision of "future crime" (yes vs. no), the authors argue that risk assessment has important justice-related objectives beyond predicting new criminal conduct.