Sacrificing Accuracy for Transparency in Recidivism Risk Assessment: The Impact of Classification Method on Predictive Performance. Issue 3 (2nd July 2016)

• Record Type: Journal Article
• Authors: Duwe, Grant; Kim, KiDeuk
• Is Part Of: Corrections. Volume 1:Issue 3(2016)
• Page Start: 155
• Page End: 176
• Publication Date: 2016-07-02
• Subjects: Risk assessment -- recidivism -- prediction -- machine learning -- Burgess; Punishment -- United States -- Periodicals; Criminal justice, Administration of -- United States -- Periodicals; 364.60973
• DOI: 10.1080/23774657.2016.1178083
• Languages: English
• ISSNs: 2377-4657

Abstract

ABSTRACT: Recent studies have compared the performance of machine learning algorithms versus logistic regression models in predicting recidivism. Existing research, however, has not compared their performance to Burgess methodology—a transparent, simplistic and summative classification technique used to develop some of the most widely risk and needs assessment instruments used within corrections today. Using multiple performance metrics and measures of recidivism on 40, 740 Minnesota offenders released from prison between 2006 and 2011, we evaluate the performance of prediction models developed with both Burgess methodology and supervised learning algorithms (i.e., statistical and machine learning algorithms). The results show that, compared to the best supervised learning classifiers, use of Burgess methodology yielded inferior performance in terms of predictive discrimination, accuracy, and calibration.