Predicting Sexual Assault Perpetration in the U.S. Army Using Administrative Data

Area of Science:

• Military health research
• Criminology
• Public health

Background:

• The Department of Defense employs a universal sexual assault prevention framework across military branches.
• Targeted interventions are cost-effective for high-risk populations, necessitating accurate identification methods.
• This study focused on developing predictive models for male U.S. Army soldiers at risk of sexual assault perpetration.

Purpose of the Study:

• To develop actuarial models for identifying male U.S. Army soldiers at high risk of administratively recorded sexual assault perpetration.
• To operationalize predictors using administrative data for risk assessment.
• To inform targeted prevention strategies within the military.

Main Methods:

• Utilized administrative data from 821,807 male U.S. Army soldiers (2004-2009).
• Employed penalized discrete-time survival analysis to identify stable predictors for risk stratification.
• Developed separate models for assaults against non-family and intra-family individuals (adults and minors).

Main Results:

• Identified 4,640 perpetrators against non-family adults, 1,384 against non-family minors, 380 against intra-family adults, and 335 against intra-family minors.
• Risk prediction models achieved top-ventile concentration of 16.2%-20.2% for non-family assaults and 34.2%-65.1% for intra-family assaults.
• Key predictors included prior criminal involvement and history of mental disorder diagnosis and treatment.

Conclusions:

• Administrative data are effective for developing actuarial models to identify a significant proportion of sexual assault perpetrators.
• A consolidated administrative data system could enable periodic risk predictions for targeted preventive interventions.
• The cost-effectiveness of such interventions depends on intervention costs, efficacy, and competing risk factors.