Predictive validity performance indicators in violence risk assessment: a methodological primer.

Area of Science:

• Forensic Psychology
• Risk Assessment Methodology
• Clinical Criminology

Background:

• Predictive validity of violence risk assessments is crucial for informed decision-making.
• Commonly used metrics, such as the area under the receiver operating characteristic curve (AUC), incompletely assess predictive validity.
• AUC primarily measures discrimination, neglecting the calibration component of risk assessment performance.

Purpose of the Study:

• To provide an overview of calibration and discrimination performance indicators for violence risk assessments.
• To highlight the limitations of the AUC in fully portraying predictive validity.
• To recommend a comprehensive approach to evaluating risk assessment tools.

Main Methods:

• Review of existing literature on violence risk assessment performance indicators.
• Explanation of calibration and discrimination as components of predictive validity.
• Discussion of various performance metrics for global, high-risk, and low-risk group assessment.

Main Results:

• The AUC is a limited measure of predictive validity as it does not assess calibration.
• Calibration indicators are essential for understanding how well predicted risks align with observed risks.
• A multifaceted approach using various performance indicators is necessary for a complete evaluation.

Conclusions:

• Future research should employ multiple performance indicators to comprehensively assess violence risk assessment tools.
• The limitations of reported indicators must be clearly stated in research.
• A thorough evaluation requires assessing both calibration and discrimination for accurate risk prediction.