Assessing violence risk in stalking cases: a regression tree approach

Area of Science:

• Forensic Psychology
• Risk Assessment
• Clinical Criminology

Background:

• Risk assessment models require validation for unique subgroups, such as stalking offenders.
• Stalking offenders pose significant safety management challenges due to potential for violence.
• Developing accurate risk assessment strategies for this population is crucial.

Purpose of the Study:

• To apply the Classification and Regression Tree (CART) approach to stalking offenders.
• To identify and understand risk assessment strategies for this specific subgroup.
• To develop and assess putative risk factors for predicting violence.

Main Methods:

• Utilized data from 204 stalking offenders undergoing psychiatric evaluation.
• Employed a series of nested Classification and Regression Tree (CART) models.
• Assessed predictive accuracy and compared CART models with logistic regression using cross-validation.

Main Results:

• Both simplified and extensive CART models demonstrated high predictive accuracy for violence.
• CART models were comparable to logistic regression in accuracy but simpler for clinical use.
• Cross-validation showed CART models had shrinkage, yet remained competitive with other actuarial instruments.

Conclusions:

• Classification and Regression Tree (CART) analysis provides a valuable and practical tool for risk assessment in stalking offenders.
• The straightforward application of CART models enhances their utility in clinical settings for safety management.
• While logistic regression showed greater cross-validation resilience, CART offers a viable alternative for predicting violence risk.