Predicting domestic violence: A meta-analysis on the predictive validity of risk assessment tools

Abstract

Risk assessment tools are increasingly being used to guide decisions about supervision and treatment of domestic violence perpetrators. However, earlier review studies showed that the predictive validity of most of these tools is limited, and is reflected in small average effect sizes. The present study aimed to meta-analytically examine the predictive validity of domestic violence risk assessment tools, and to identify tool characteristics that positively moderate the predictive validity. A literature search yielded 50 independent studies (N = 68,855) examining the predictive validity of 39 different tools, of which 205 effect sizes could be extracted. Overall, a significant discriminative accuracy was found (AUC = 0.647), indicating a moderate predictive accuracy. Tools specifically developed for assessing the risk of domestic violence performed as well as risk predictions based on victim ratings and tools designed for predicting general/violent criminal recidivism. Actuarial instruments (AUC = 0.657) outperformed Structured Clinical Judgment (SCJ) tools (AUC = 0.580) in predicting domestic violence. The onset of domestic violence (AUC = 0.744) could be better predicted than recurrence of domestic violence (AUC = 0.643), which is a promising finding for early detection and prevention of domestic violence. Suggestions for the improvement of risk assessment strategies are presented.