Developing and Validating a Risk Assessment Scale to Predict Inmate Placements in Administrative Segregation in the Correctional Service of Canada

Abstract

Concern over the use of administrative segregation has motivated efforts to reduce segregation placements. The purpose of this study was to develop and validate an actuarial risk assessment scale to predict admissions to administrative segregation (for reasons of jeopardizing security or the inmate’s own safety) for at least six consecutive days within two years of admission to the Correctional Service of Canada (CSC). The sample (N = 16,701) included all male offenders admitted to CSC from fiscal years 2007/2008 through 2009/2010 and all female offenders admitted from 1999/2000 through 2009/2010. Offenders were randomly divided into a development sample (N = 11,110) and a validation sample (N = 5,591). Analyses were separated by reason for administrative segregation, gender, and Aboriginal ancestry. Overall, 413 potential predictor variables were examined, including items from assessment scales, demographic information, current offence information, flags/alerts/needs, and information from previous federal sentences. Approximately 24% of offenders were placed in administrative segregation. Of the 413 variables examined, 86% significantly predicted segregation placements. The item pool was reduced using Principal Components Analysis, tests of unique contributions within the measured components, and considerations of general utility and face validity. Several scales were developed and validated. Considering both accuracy and efficiency, the optimal scale had six static items (age, prior convictions, prior segregation placement, sentence length, criminal versatility, and prior violence) – this scale was called the Risk of Administrative Segregation Tool (RAST). Attempts to develop scales unique for men and women and those of Aboriginal ancestry did not yield meaningfully higher accuracy than the overall RAST. The RAST generalized well to the validation sample (AUC = .80) with high discrimination, suitable calibration (mostly non-significant E/O indexes), and superior performance to other risk scales used by CSC. Normative data (absolute segregation rates, percentiles, and risk ratios) were presented for the RAST. The RAST is an appropriate scale to use in practice for identifying risk of administrative segregation placements among CSC inmates and may serve as a first step in future efforts to divert offenders from segregation. Limitations of the current study and suggestions for future research are discussed.