Study 1
Hyatt and Barnes (2017) evaluated the impact of the Intensive Supervision Program (ISP) on probationer recidivism in Philadelphia, Pennsylvania. During the period between May 2010 and April 2011, a randomized controlled trial assigned 447 cases to the ISP treatment group and 385 cases to the control condition, which consisted of regular probation. Only those who were newly categorized as “high-risk” for committing a serious offense within 2 years were randomly assigned. Individuals who were already on high-risk supervision or who had a previous high-risk forecast in the previous year were excluded.
The ISP sample was all male, and 72 percent Black, 21 percent white, and 1 percent other ethnicity. The control group was all male, and 71 percent Black, 22 percent white, and 1 percent other ethnicity. The mean age at assignment for both groups was 29. At intake, a majority in both the ISP group and control group had experienced prior incarceration (94 percent and 96 percent, respectively) or probation (63 percent and 67 percent, respectively). There were no statistically significant differences between the two groups at baseline.
People were classified into the “high-risk” category using a statistical procedure known as random forest forecasting. The random forest model used 48 predictors of future criminal behavior, including demographics, prior criminal history, prior sentences, and neighborhood characteristics. The screenings were conducted by an automated computer program integrated into the Adult Probation and Parole Department’s (APPD’s) case-management system. The automated risk forecasting allowed for the experiment to be double blind, as supervising officers and treatment and control group members were not aware they were participating in a randomized trial.
Outcome measures collected were offense and criminal history data for 12 months following enrollment in the study. Recidivism was quantified as any charge for a new offense, not including violations of probation conditions. The charges were grouped categorically: violent, serious, nonviolent drug, property, and sexual offenses. Information on new criminal offenses was collected from the unified, computerized databases used by the police, courts, and correctional agencies in Philadelphia. Data on absconding was taken from the APPD’s case-management system. An individual was deemed to have potentially absconded after missing (without excuse or justification) two consecutive scheduled contacts. The study also looked at incarceration rates and violations of probation. Incarceration rates included examination of people who had been incarcerated in the local jail system for any reason (including pretrial detention, short sentences for technical violations, and any new judicial sentences of up to 24 months). Violations of probation included individuals not complying with the requirements of supervision, including failing to report a positive drug screening, missing treatment, or not paying fines or court costs.
The analysis used an intent-to-treat design. Inferential statistical tests conducted on the collected data included a two-sample t-test and Kaplan-Meier survival analysis. The study authors did not conduct subgroup analyses.
