Classifying Risk Development and Predicting Parolee Recidivism with Growth Mixture Models

Using two data sets, containing 582 total cases, this study investigates whether classifying offenders on trajectories of risk scores helps predict parolee recidivism. One data set has 4 years of risk scores and another has three. Both data sets contain control variables measuring released inmates’ characteristics. The dependent variable measures arrest or return to prison over a 2-year span. A growth mixture model, classifies offenders into three classes, a stable and high trajectory group, a group with a high but declining risk trajectory, and a small, low-risk group with little change. Trajectory class membership correlates with recidivism in both data sets. Supplementary analyses show that assigned classes are better predictors of recidivism than last risk scores or simple change scores. Discussion centers on the appeal and relevance of trajectories of risk, as opposed to static measures, for predicting offender misconduct and other outcomes.