Increased rates of incarceration coupled with growing rates of institutional violence and major disturbances within U.S. correctional institutions have resulted in increased importance being placed on the development of accurate and efficient correctional risk classification methods. In the current study, institutional infractions were tracked from correctional intake for 17,054 male and female incarcerated offenders. In order to allow for examination of specific categories of problematic behaviors, institutional infractions were categorized according to physically aggressive, verbally aggressive/defiant, and' nonviolent infractions. Following analysis of Personality Assessment Inventory (PAl) descriptive statistics that were obtained during correctional intake, the univariate predictive utility of several static and dynamic variables in the prediction of institutional violence and misconduct was examined. Predictor variables included historical, demographic, and self-report (i.e., PAl) information. Among individual variables, ' subject age, gang affiliation, gender, and PAl Antisocial Features and Aggression scale scores were most predictive of institutional infractions after controlling for the number of days incarcerated. The majority of examined P AI scales remained significant predictors after further controlling forage, gender, ethnicity, and gang affiliation. Despite a low base rate of occurrence, the largest effect sizes were demonstrated in the prediction of physically aggressive infractions. In the final stage of data analysis, multiple regression analyses were undertaken in order to develop an institutional violence risk assessment scheme for potential use in inmate triage/classification procedures and to allow for examination of incremental predictive accuracy of predictor types (e.g., static vs. dynamic). A forward logistic regression resulted in a 9 variable model composed of historical, demographic, and sdf-report variables that was a robust predictor of adjudication for physically aggressive infractions (AUC = .715,p < .001). A violence risk classification scheme was developed that allowed for meaningful distinction between categories of relative risk based on the final model, and differences in accuracy between regression weight-based scores and a simple score method were minimal although static and historical/demographic variables were most predictive of future acts of violence, the addition of dynamic and self-report variables resulted in increased predictive accuracy, with each variable type adding unique variance to the prediction of future violent behaviors.
Files are restricted to Pacific University. Sign in to view.