Can an Algorithm Improve Parole Decisions?
Wayne Winston · 2020
This chapter uses logistic regression to construct a simple example that illustrates the “classic” approach to calculating risk scores. It discusses the widely cited ProPublica searing criticism of risk scores. The chapter discusses how Jennifer Skeem and Christopher Lowenkamp showed that Post Conviction Risk Assessment does not exhibit significant test bias. It introduces the important concept of machine learning and discuss Professor Berk's use of machine learning to create risk scores for prospective parolees. The worksheet Base of the workbook Parole.xlsx contains fictitious data that combined with logistic regression to create risk scores. ProPublica analyzed risk scores developed by Northpointe that were used at time of trial to forecast the likelihood of recidivism. The Correctional Offender Management Profiling for Alternative Sanctions algorithm used the variables to calculate for each defendant a 1–10 score, where a higher score means the defendant was more likely to commit another crime.