The Effect of Reentry Court Participation on Post-Release Supervision Outcomes and Re-Arrest

ABSTRACT

This research examines the impact of participation in a federal reentry court program on supervision violations, revocations, and re-arrest. We evaluate a federal reentry court that utilized random selection and random assignment of participants to treatment groups and random selection for control groups. Outcomes examined include: number of supervision violations, non-technical violations, and re-arrests; and relative risk of supervision violation, non-technical violation, supervision revocation, and re-arrest. Negative binomial and Cox regression models suggest that participation in reentry court had little impact on supervision outcomes or re-arrest. Specifically, although the reentry court participants had more supervision violations compared to the control group, on average, there were no other differences based on reentry court participation. The survival analysis revealed no statistically significant differences in supervision outcomes or re-arrest for reentry court participants compared to other study groups. This study contributes to the scant and conflicting literature on reentry court programs.

KEYWORDS:

• Reentry court
• problem-solving courts
• recidivism
• revocation
• supervision violation
• offender reentry

Acknowledgments

The authors wish to thank the Honorable M. Casey Rodgers, Judge, United States District Court, Northern District of Florida, for her interest in offender reentry, and commitment to quality research. We are also grateful to the women and men with whom we worked in the offices of the Clerk of Court, United States Probation and Pretrial Services, United States Attorney, Federal Public Defender, and community partners. Undergraduate and graduate students in criminology and criminal justice at the University of West Florida provided important logistical support.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. The NDFL reentry court is formally known as the Robert A. Dennis, Jr. (“RAD”) Reentry Court Program. Mr. Dennis, a former federal public defender, was instrumental in the establishment of the initial pilot program of reentry court.

2. Features of the PCRA are beyond the scope of this study and are discussed elsewhere (Johnson et al., 2011; Lowenkamp et al., 2013). PCRA is based on and emphasizes supervision practices consistent with the risk, needs, responsivity principle (Andrews, Bonta, & Hoge, 1990), which is also true of the reentry court model.

3. The risk principle “states that the level of supervision and treatment should be commensurate with the offender’s level of risk” (Lowenkamp et al., 2006, p. 77).

4. The reentry court team (individuals holding voting rights) consisted of the Judge, two U.S. probation officers, the NDFL Deputy Chief Probation Officer, Assistant U.S. Attorney, and Assistant Federal Public Defender. The reentry committee consisted of the team, therapist, two representatives from the Florida Division of Vocational Rehabilitation, and a clergyman.

5. Although some researchers suggest a rule of thumb that Cox regression models should have a minimum of 10 events per predictor variable (EPV) (Concato, Peduzzi, Holford, & Feinstein, 1995; Peduzzi, Concato, Feinstein, & Holford, 1995), more recent literature suggests that this “rule” should be relaxed (Vittinghoff & McCulloch, 2007). Both Vittinghoff and McCulloch (2007) and Peduzzi et al. (1995) recommend that when the EPV is low, results should be interpreted with caution and compared with reduced models with higher EPV. Although the EPVs for the full Cox models presented in the current study are low (5.8 in Table 3; 5.1 in Table 4; 3.2 in Table 5; 4.6 in Table 6), the substantive findings regarding study group assignment are, as noted above, similar compared to the reduced models, for which the EPVs are greater than the rule of thumb suggested by Concato et al. (1995) and Peduzzi et al. (1995).

6. We also analyzed all of the multivariate models controlling only for the predictors that differed significantly across study groups in the bivariate analyses. There were few substantive differences in the effects of study group assignment or in statistical power in these reduced models (e.g. non-significant effects). Model fit statistics for several of the models were improved in the full models; therefore we present the full models here.

7. Some researchers argue that post-hoc power analysis should be used to examine null findings to determine the likelihood of Type II error (see Onwuegbuzie & Leech, 2004); however, Levine and Ensom (2001) provide a persuasive argument that post-hoc power analysis should not be used, as it underestimates power and can mislead readers. Given this debate, we do not report the statistical power for non-significant findings within the main text, but note non-significant relationships for which there was low power within table footnotes.

8. We cannot rule out the possibility of Type II error for several of the non-significant findings due to low power associated with those hazard ratios; however, see Onwuegbuzie and Leech (2004); and Levine and Ensom (2001) for discussion regarding the appropriateness of post-hoc power analysis.