This work is an extension of the analyses performed under the authors' previous grant and reported in "Predicting Recidivism Using Survival Models" (Schmidt and Witte, 1989). The current research investigated the extent to which the previous models of Schmidt and Witte could be improved by greater attention to the ways in which explanatory variables are entered into the models. These models included proportional hazards models and also parametric models based on the lognormal distribution. The explanatory models did provide a more complete profile of the way in which explanatory variables affect recidivism. In particular, age and number of prior incarcerations were found to have strong nonlinear effects that the previous models did not reveal. The increase in explanatory power was not matched by a commensurate increase in the quality of out-of-sample predictions. Predictions for the entire 1978 and 1980 validation samples were improved only slightly by using expanded models. Expanded specifications resulted in greater improvements in predictive ability for individuals instead of for groups. 61 tables
Improving Predictions of Offender Recidivism and Patterns of Offender Crime: Final Report
NCJ Number
135637
Date Published
April 2025
Length
108 pages
Annotation
This report presents results from tests of the abilities of a proportional hazards model and parametric (lognormal) models to improve predictions of offender recidivism and patterns of offender crime.
Abstract
Date Published: April 30, 2025