Self-Reported Inmate Conduct

Abstract

This article applies Walters’s criteria of dynamic factors in prison assessment to assess the predictive capabilities of the Prison Inmate Inventory (PII) in a sample of offenders from a southeastern state in the United States. Nested negative binomial regression was used, as well as ROC/AUC (receiver operating characteristics/area under the curve) to evaluate accuracy. The results show that beyond the effects of demographic and static variables, dynamic factors incrementally improved model fit and estimated counts of female inmates, non-White inmates, and inmates reporting more disciplinary actions and violent behavior. All PII scales demonstrated accuracy above .50; however, only two were within the range considered acceptable for criminal justice instruments. Practical implications have been discussed.