Nodewise Predictability in Cross-Sectional Data Does Not Outperform Mechanical Totals in Predicting Sexual...

Area of Science:

• Forensic Psychology
• Criminology
• Risk Assessment

Background:

• Predicting sexual reoffending is crucial for public safety and offender management.
• Dynamic risk factors are commonly used in risk assessment tools like STABLE-2007.
• Understanding the interrelationships between risk factors may improve predictive accuracy.

Purpose of the Study:

• To compare the predictive accuracy of two methods for assessing sexual reoffending risk: mechanical totals (sum score) and nodewise predictability.
• To evaluate the effectiveness of accounting for interrelationships among dynamic risk factors in predicting recidivism.

Main Methods:

• Utilized a dataset of 5,315 North American men assessed with STABLE-2007 dynamic risk factors.
• Employed a cross-validation approach with 300 iterations, splitting data into 20:80 training and testing sets.
• Calculated the area under the curve (AUC) to measure predictive accuracy for both methods.

Main Results:

• Mechanical totals demonstrated significantly higher predictive accuracy (AUC=0.67, SD=0.04) compared to nodewise predictability (AUC=0.50, SD=0.03).
• The difference in AUC was statistically significant (t(299)=80.2, p < .001), with a large effect size (Cohen's d=4.63).

Conclusions:

• The conventional method of summing dynamic risk factors (mechanical totals) is superior to nodewise predictability for assessing sexual reoffending risk at a group level.
• Future research should investigate incorporating temporal effects, individual variances, and network centrality into nodewise models to potentially enhance their accuracy.