The Efficacy of Violence Prediction: A Meta-Analytic Comparison of Nine Risk Assessment Tools

Actuarial risk assessment tools are used extensively to predict future violence, but previous studies comparing their predictive accuracies have produced inconsistent findings as a result of various methodological issues. We conducted meta-analyses of the effect sizes of 9 commonly used risk assessment tools and their subscales to compare their predictive efficacies for violence. The effect sizes were extracted from 28 original reports published between 1999 and 2008, which assessed the predictive accuracy of more than one tool. We used a within-subject design to improve statistical power and multilevel regression models to disentangle random effects of variation between studies and tools and to adjust for study features. All 9 tools and their subscales predicted violence at about the same moderate level of predictive efficacy with the exception of Psychopathy Checklist Revised (PCL-R) Factor 1, which predicted violence only at chance level among men. Approximately 25% of the total variance was due to differences between tools, whereas approximately 85% of heterogeneity between studies was explained by methodological features (age, length of follow-up, different types of violent outcome, sex, and sex-related interactions). Sex-differentiated efficacy was found for a small number of the tools. If the intention is only to predict future violence, then the 9 tools are essentially interchangeable; the selection of which tool to use in practice should depend on what other functions the tool can perform rather than on its efficacy in predicting violence. The moderate level of predictive accuracy of these tools suggests that they should not be used solely for some criminal justice decision making that requires a very high level of accuracy such as preventive detention.