Date of Award
Fall 11-24-2023
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Mathematics, Statistics and Computer Science
First Advisor
Michael Zimmer
Second Advisor
Darren Wheelock
Third Advisor
Walter Bialkowski
Abstract
Tools for predicting an individual’s risk of recidivism are implemented throughout the criminal legal system, from informing decisions about release pretrial to identifying appropriate treatment plans for successful reentry. Data-driven risk assessment is increasingly being incorporated during sentencing proceedings across the United States. There is significant variation across jurisdictions in how these tools are constructed and what they aim to predict. The use of algorithmic risk assessment at sentencing is hotly contested due to concerns over the opacity of algorithms developed by private corporations, the potential for perpetuating disparities present in the criminal legal system, and violations of the constitutional rights of defendants. This thesis explores the debate surrounding the ethical use of risk assessment algorithms in the context of sentencing. Using risk assessment algorithms at sentencing is additionally challenged by the behavior of the humans engaging with them. The predictive capabilities of these tools may be misinterpreted by decision-makers due to insufficient reporting of different accuracy measures, as well as the disconnect between predicted outcomes and the analytic strategies used to generate them. This work concludes with recommendations for improving the interactions between human decision-makers and algorithmic predictions of risk.