NCJ Number
85371
Date Published
1981
Length
19 pages
Annotation
This study briefly presents the assumptions of the analytic techniques most frequently used to measure recidivism in correctional program evaluations -- the binomial model and the failure-rate model -ultimately leading to the mixed exponential approach, the major focus of the study.
Abstract
The binomial model is used most frequently in correctional program evaluations, but there has been a recent strong interest in the modeling of recidivism as a failure process over time, similar to device reliability analysis and mortality or morbidity calculations. The work of Maltz and McCleary (1977) has provided a key next step in the process of measuring recidivism. Their approach is to assume that a population is subject to a split into two subpopulations; one is prone to failure according to a constant hazard rate while the second never fails. The mixed-exponential model brings the Maltz-McCleary model closer to the statistical mainstream by assuming that each member of the second subpopulation is no longer deemed a sure success but may be subject to a hazard allowing early failure only with a very small probability. One reason for this extension is the realization that a failure rate of zero for one subpopulation may be incorrect. In applying the models to two sets of recidivism data, it appears that the mixed-exponential model compared to the models most frequently used redefines what is meant by the recidivism rate. Whereas in the past it was defined as the percentage of failures within a fixed time, now it is defined as the percentage of failures within a fixed time, now it is defined as both the fraction ultimately expected to fail and the time rates of failure. Graphic and tabular data and 16 references are provided.