NCJ Number
85372
Date Published
1981
Length
21 pages
Annotation
The importance of the Cox model and the grouped-data adaptations (failure-rate regression techniques) is their nonparametric nature and their allowance for time-varying covariates and censored data, so they can reveal harbingers of postrelease failure.
Abstract
A nonparametric failure-rate regression model with a proportional hazard-rate assumption was developed by Cox (1972), exact likelihoods based on grouped-data failure-rate regression models have been proposed by several authors. In such models, failures are assumed to occur in intervals. Since recidivism data are often grouped, the grouped data models seem more appropriate than the Cox model. The techniques were applied to a Connecticut parole study of a sample of 108 exoffenders. The techniques permit the measurement of the effect of time-varying factors such as income, drug use, and contact with criminal associates on postrelease failure rate. This permits the realistic assumption that an individual's propensity to recidivate varies over time, depending on circumstances. Many criminological studies are necessarily observational and require some way of accounting for the influence of factors that cannot be eliminated through randomization or blocking. Failure-rate regression techniques provide an accounting for these factors and can be applied whenever the program outcome is a time-dependent threshold event. The threshold event might denote a failure, such as rearrest, a return to drug use, or violation of parole. On the other hand, the event may be associated with success. A job-training program evaluation might use a weekly income of $200 as the threshold event. In this case, a program that leads to early events is preferred. A discussion of the numerical maximization of the likelihood function is appended. Tabular and graphic data and 32 references are provided. (Author summary modified)