Most items in structured risk assessment instruments do not predict violence

Abstract

The predictive ability of static risk assessment instruments may be explained by a limited number of their items. This study examined the independent predictive accuracy of individual items in the Psychopathy Checklist-Revised (PCL-R), Violence Risk Appraisal Guide (VRAG) and Historical-Clinical-Risk Management-20 (HCR-20) for violent reconvictions following release among 1353 male prisoners in England and Wales. The study found most items in the three instruments were not independently predictive. Items not independently predictive were removed and all significant items in the original three instruments were combined, resulting in negligible gains in predictive accuracy for the VRAG and HCR-20, but a small improvement in the PCL-R. The study demonstrated that the predictive power of the PCL-R, VRAG and HCR-20 are based on a small number of their items. This may partly explain the ‘glass-ceiling’ effect beyond which further improvement cannot be achieved. Instruments lack outcome-specificity for violence, and independently predictive items include measures of general criminality.