Prognosticating Offending in Early Adulthood: How Early Can We Predict?

Intoduction/Aim: Extant tests of developmental theories have largely refrained from moving past testing models of association to building models of prediction, as have other fields with an intervention focus. With this in mind, we test the prognostic capacity to predict offending outcomes in early adulthood derived from various developmental theories. Methods: Using 734 subjects from the Rochester Youth Development Study (RYDS), we use out-of-sample predictions based on 5-fold cross-validation and compare the sensitivity, specificity and positive predictive value of three different prognostic models to predict arrest and serious, persistent offending in early adulthood. The first uses only predictors measures in early adolescence, the second uses dynamic trajectories of delinquency from ages 14–22, and the third uses a combination of the two. We further consider how early in adolescence the trajectory models calibrate prediction. Results: Both the early adolescent risk factor only model and the dynamic trajectory model were poor at prognosticating both arrest and persistent offending in early adulthood, which is manifest in the large rate of false positive cases. Conculsion: Furthermore, existing developmental theories would be well served to move beyond cataloging risk factors and draw more heavily on refinements, including a greater focus on human agency in life course patterns of offending.