Accuracy of Risk Assessment in Corrections Population Management: Where’s the Value Added?

The history of risk assessment shows a trend towards increasing complexity in both instruments and analytic methods, reflecting trends in both research on the causes of criminal behavior and the development of statistical methods, particularly as influenced by the availability of inexpensive computing power. A recent innovation is the introduction of classification and prediction methods based on machine learning algorithms developed outside of the traditional statistics milieu by computer scientists; these are claimed to improve accuracy (Berk, Sherman, Barnes, Kurtz, & Ahlman, 2009; Breiman, 2001a) while removing the dependence on unrealistic parametric assumptions and capturing complex interactions in the data (Breiman, 2001a; Svetnik, Wang, & Aliferis, 2003).