Violence prediction 16 years later: An updated and extended network meta-analytic comparison of nine risk assessment instruments

Risk assessment remains a contentious yet clinically and legally relevant issue in the sentencing and management of offenders within the criminal justice system. The meta-analysis by Yang et al. (2010) synthesized evidence on the predictive validity of nine risk instruments-the General Statistical Information on Recidivism, Historical Clinical and Risk Management Scheme, Level of Service Inventory-Revised, Offender Group Reconviction Scale, Psychopathy Checklist-Revised, Psychopathy Checklist: Screening Version, Risk Matrix 2000 for Violence, Violence Risk Appraisal Guide, and Violence Risk Scale-based on 28 studies involving 12,178 individuals published between 1999 and 2009, yielding a moderate pooled effect size (Cohen's d = 0.66) and concluding that all nine tools demonstrated similar predictive validity for future violence.

Sixteen years later, we asked whether the accumulating body of research had altered these conclusions. We hypothesized that extended evidence and more refined methodology might increase predictive validity estimates and uncover meaningful differences among the nine instruments.

We extended and updated the evidence base using network meta-analysis and a multimodel moderator analysis. The updated network meta-analysis included more than twice as many studies (k = 70) involving 19,537 individuals, published between 1998 and 2024. A random-effects model was applied, and heterogeneity was assessed using Cochran's Q and I².

The network meta-analysis indicated moderate predictive validity across all nine instruments (Cohen's d = 0.76). No statistically significant differences in predictive validity were found among the tools, except for Psychopathy Checklist-Revised Factor 1, which exhibited significantly lower predictive value than most other instruments. Moderator analysis identified age as a significant predictor, with higher predictive validity observed in younger compared to older samples.

The present findings corroborate and extend Yang et al.'s (2010) seminal conclusions, confirming that all nine instruments demonstrate statistically equivalent moderate predictive validity for future violence. The persistence of moderate effect sizes over 16 years suggests limited transformative progress in risk-assessment development, underscoring the need for novel, qualitatively different methodological approaches to advance the field. (PsycInfo Database Record (c) 2026 APA, all rights reserved).