Validation of a Virginia Work Release Risk Prediction Model: A Methodology For the Improvement of the Reliability of Correctional Decision Makers

• Main Author: Osborne, William Nathan, Jr.
• Format: Others
• Published: VCU Scholars Compass 1994
• Subjects: Public Administration
• Online Access: https://scholarscompass.vcu.edu/etd/5239; https://scholarscompass.vcu.edu/cgi/viewcontent.cgi?article=6323&context=etd

This study identifies and validates variables which are significant predictors of work release success on 439 Virginia work release participants. The variables were selected on the basis of whether they would exert internal or external control over the inmate, with a view toward offering empirical support to control theory. A retrospective longitudinal research design was employed by randomly selecting inmates who had participated in either of three work release centers from 1987 to 1991. Two of the programs housed male inmates while the other housed female inmates. Data were collected from inmate files on thirty-one variables over a six month period. Analysis employed logistic regression using work release success or failure as a dichotomous dependent variable. A prediction model was developed using a construction sample of 416 cases. The resultant model was then used to predict and classify inmates using a randomly selected validation sample of 226 cases. Of the thirty-one variables under study, four individual factors (previous commitments, age of offense, time on the street, and prior misdemeanor convictions), two program factors (time in work release and year of work release), and one institutional adjustment factor (no institutional drug or alcohol violations), emerged as significant predictors. The study revealed that the work release staff has been successful in identifying low risk inmates, with a success rate of 86% and a failure rate of 14%. Of the failures, only six had new charges (1.4% of the total population), and three escaped or absconded (0.7% of the total population). The remaining forty-nine failures (11.1% of the total population) failed urine screens or failed due to poor work performance. The prediction model was able to classify 88% of the validation sample correctly which is a minimal improvement over the department of corrections selection procedures.