Risk Assessment: What is Being Predicted by Actuarial Prediction Instruments?

Abstract

There is an ongoing debate between actuarial and clinical approaches to assessment of risk for violent recidivism where previous studies on predictive validity are said to have yielded results favouring actuarial assessments. However, these studies have all used a dichotomous outcome (recidivism “yes” or “no”) as the criterion variable. In order to not only predict but also manage recidivism risk and prevent future violence, there is a need to elaborate on the criterion variable in actuarial assessments and also consider the imminence, frequency, nature, and severity of potential reoffending. We revisited a database of actuarial risk assessments referring to a nationwide 5-year cohort of all adult men released from prison after having served a prison sentence for sexual offending. Sexually recidivistic behaviors were coded for individuals sexually reconvicted during an average follow-up of 6 years. Actuarial procedures had overall areas under the receiver operating characteristic curve (AUC) of .73 and .75 concerning any sexual reoffending. However, when the procedures were reevaluated using various outcome measures, the predictive validity ranged from AUC of .40 to .94. Actuarial risk assessment schemes worked well for imminent and less severe reoffending but were less accurate concerning repeated and injurious sexual reoffenses, and were of no value at all to discriminate intrafamilial recidivists from nonrecidivists. Replication of these preliminary findings would suggest that whereas actuarial approaches are optimized to predict the common (but less severe) occurrences of sexual reoffending, they might be of less value to inform on the assessment of risk for the more clinically worrying, severe (but less common) sexual recidivism.