The sections below review separately the analyses for each of the inmate population sub-groups: adult males, adult females, and youthful offender males. Each section follows the same pattern, beginning with base recidivism rate estimates for each treatment-control group comparison. It is instructive to compare the recidivism rates for public and private adult male inmates within the six methods of identifying the two groups without control variables.
Then differences between the treatment and control groups on recidivism-related factors are reported. Although six different methods of identifying public versus private inmates were examined, the differences in the adult male private and public group on these descriptive statistics are, with some exceptions, reasonably consistent across the six measures. For brevity only one of the treatment-control group measures is reported for the descriptive analysis of factor differences. The purpose of examining basic differences in the public and private inmates on variables known to influence recidivism rates, regardless of the types of facilities they were housed in, is to provide a general sense of the characteristics of the inmates in both groups and to demonstrate that controlling for these factors is important. To best achieve this, factor differences for the purest measure (definition B1) of treatment and control groups are reported.
The proportional hazard regressions controlled for all continuous variables (e.g., age, number of disciplinary reports, months in prison, etc.) as such. The descriptive statistics tables display these factors in categories only to facilitate reporting. The proportional hazard model results displayed are those from the treatment and control comparison groups when all factor effects, including a dichotomous variable for the treatment-control group effect, were entered. These models for the adult male and adult female sub-groups yielded effect sizes and significance levels for the factors similar to those reported for by the FDOC. The department reports no data on these effects for youthful offenders, so a comparison of these effects on this sub-group is not possible. These models were also run using a stepwise selection method (entry p<.05) to verify that the treatment effect variable does not enter the models.