Replicating the violence risk appraisal guide: a total forensic cohort study

Area of Science:

• Forensic Psychology
• Psychometrics

Background:

• Assessing violence risk is crucial in forensic settings.
• Psychometric properties like discrimination and calibration are key to evaluating risk assessment tools.
• The Violence Risk Appraisal Guide (VRAG) is widely used, but its calibration requires further investigation.

Purpose of the Study:

• To replicate the VRAG's development study in a European context.
• To evaluate both the discrimination and calibration of the VRAG.
• To assess the accuracy of VRAG's probabilistic risk estimates.

Main Methods:

• Prospective study of 206 violent offenders in Switzerland.
• VRAG administered by trained assessors prior to community release.
• Follow-up for 7 years, recording subsequent violent offenses.
• Discrimination assessed via ROC analysis; calibration via Sanders' decomposition and Bayesian credible intervals.

Main Results:

• VRAG demonstrated modest discrimination (AUC = 0.72).
• Calibration was poor, with an average error of 21% in probabilistic estimates.
• Bayesian credible intervals indicated that expected risk rates were not met in 5 out of 9 risk bins.

Conclusions:

• There is a need to improve the measurement of calibration validity for risk assessment instruments.
• Replication studies focusing on the calibration of actuarial tools like the VRAG are essential.
• Caution is advised when applying VRAG's probabilistic risk estimates in clinical practice.