The effects of low-intensity supervision for lower-risk probationers: updated results from a randomized controlled trial

This paper explores the effects of reduced supervision intensity for probationers who were identified, using a random forest forecasting model, as presenting a low risk of committing new serious offenses. It expands on previously reported results of the Philadelphia Low Intensity Community Supervision Experiment, a randomized controlled trial performed from 2007 through 2008. We update our previous one-year recidivism results to include 18 months of follow-up data, and assess additional measures that were not available in earlier analyses, including drug-testing results, officer contact compliance, probation violations, and absconding from supervision. The updated analysis affirms previous findings, showing that reduced supervision intensity does not increase the prevalence or frequency of new offending by low-risk probationers, and does not appear to result in any additional threats to public safety. We conclude that low-intensity supervision, when used in concert with valid and reliable risk forecasting, offers community supervision agencies a powerful tool for managing large offender populations, allowing the agencies to focus scarce resources on higher-risk offenders and perhaps reduce administrative costs. Further research is needed to quantify the exact cost reductions, and to determine the best means of supervising offenders whose risk level makes them ineligible for low-intensity supervision. Keywords: probationparolerandomized controlled trialsrisk forecastingrandom forestsrisk-based supervisioncompliancerecidivism Notes 1. It should be noted that this rearrest rate seems rather high for offenders who had been classified as low risk. It is roughly equivalent, for example, to the 90-day rearrest rate for the highest-risk offenders in Philadelphia, which currently stands at 11.1% (Barnes 2011). 2. At the time when the sample was selected, one offender appeared in the data under two different identifier values, and was randomly assigned into the experimental group twice. One of this offender's two experimental cases has now been deleted from the data, which has reduced the size of the experimental group from 800 to 799, and the size of the total sample from 1559 to 1558. 3. Gill (2010) examined recidivism over a two-year follow-up period, and found no significant difference between the experimental and control groups. 4. The agency's internal evaluation of the experiment (Ahlman & Kurtz 2008) did examine absconders, finding that the experimental group experienced a significant reduction, during the first year following random assignment, in the issuance of warrants for absconding from supervision. 5. In total, 42,475 offenders were listed in the agency's caseload on 1 October 2007. Some of these offenders, however, had never been photographed or fingerprinted by the Philadelphia Police Department, and therefore lacked the identifier values needed for a criminal records search. Other offenders had been placed into purely administrative caseloads within the agency, and were not being actively supervised or sought for absconding. 6. Although not relevant to the Low Intensity Supervision Experiment, the third risk classification designated offenders as high risk when they were forecasted to commit at least one serious offense over the subsequent two years. These offenders were placed into separate Anti-Violence (AV) units, with vastly increased supervision requirements. 7. Those who compare these results with our earlier findings (Barnes et al. 2010) may notice that these first-year contact counts are somewhat higher than appeared before. More inclusive searching rules were used here, resulting in more contacts being pulled from the officers' case notes. 8. To be fair, the wait times outside the agency's headquarters became much more endurable once the caseloads were stratified by risk level. Instead of requiring nearly every offender to visit once per month, 43% of the agency's active offenders were stepped down to AS, with just two office visits per year. The result was a modest, but still important agency-wide decrease in office visits. During the time between the first year prior to random assignment (2006–2007) and the third year of follow-up (2009–2010), the number of office visits dropped by 3.4%, despite the fact that the number of unique offenders reporting for such visits rose by 10.6%.