Date of Award

Spring 1-1-2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Tim Wadsworth

Second Advisor

Jason Boardman

Third Advisor

David Pyrooz

Fourth Advisor

Zachary Hamilton

Fifth Advisor

James Dykes

Abstract

The actuarial assessment of recidivism risk is the foremost way that criminal justice systems balance managing caseloads, prioritizing rehabilitative services, and protecting public safety in the modern era of punishment. As such, many researchers have worked to develop new and better methods for more accurately classifying offender risk, but this has often happened at the expense of making theoretical contributions to the field of criminology. Furthermore, researchers have ignored some innovative methods that hold great promise for improving the accuracy of risk predictions. In this dissertation, I use Washington State Department of Corrections data to develop and validate the first recidivism risk assessment instrument to use Bayesian statistics, which I argue are particularly well suited to the task of prediction on a theoretical level. By comparing my results to those of a similar instrument developed using more common, frequentist regression models, I demonstrate the utility of Bayesian statistics for reducing model uncertainty and classification error. Findings also show that repeat property and drug offenders exhibit patterns of behavior that suggest a great deal of specialization in offending, whereas violent criminals display more versatile criminal behaviors, and highlight a troublesome tendency for risk assessments of all types to systematically overclassify recidivism risk among black men. Implications for life course, racial formation, and intersectional theories are discussed, along with recommendations for practice.

Share

COinS