Constructing actuarial devices for predicting recidivism a comparison of methods

A recent contribution to the prediction literature by Steadman et al. features a novel "iterative classification" procedure for constructing risk screening devices. In this article, the authors apply the iterative classification procedure to a large recidivism data set, across a range of recidivism outcomes and cross-validation conditions. The purpose of this study is to assess the generalizability of the iterative classification procedure and to draw comparisons with more traditional methods of device construction. Results show the iterative classification procedure to outperform other standard device construction procedures in terms of the percentage of cases classified as high or low risk but not to outperform more traditional device construction procedures on a variety of other performance measures. Implications for future research on the construction and evaluation of risk screening devices are discussed.