Hudson Institute

Criminal Justice Reform and the First Step Act’s Recidivism Reduction Provisions: Preliminary Issues for Policymakers

President and CEO
Senior Fellow
The Robert F. Kennedy Department of Justice building in Washington, D.C. (GETTY IMAGES)
The Robert F. Kennedy Department of Justice building in Washington, D.C. (GETTY IMAGES)

Propelled by years of public commentary and concern about the size of the federal prison population—which has receded from its peak of over 219,000 in 2013 to roughly 186,000 in 2017 but remains more than seven times larger than it was 40 years ago—Congress recently passed, and President Trump signed into law, a major piece of criminal justice reform legislation, commonly referred to as the First Step Act. The new law reduces enhanced minimum mandatory sentences for prior drug felons; expands the existing “safety valve” that allows below-statutory-minimum sentences for certain non-violent, low-level drug-crime defendants; and establishes a variety of initiatives and policies at the Bureau of Prisons (BOP) intended to ease re-entry for soon-to-be-released federal offenders. This much of the bill is fairly straightforward.

The central and most notable elements of the First Step Act, however, are contained in its Title I “Recidivism Reduction” provisions. Title I requires the Attorney General and his Justice Department colleagues (in consultation with an “Independent Review Committee” of outside experts and practitioners) to devise or select (by late July of this year) and then fully implement (by late January 2020) a comprehensive “risk and needs assessment system” that will use a combination of actuarial tools and individualized appraisals to classify (and regularly re-classify) everyone in BOP custody according to the likelihood that he will recidivate post-release. Certain offenders who achieve and maintain a “minimum” or “low” recidivism risk classification will then be eligible to earn time credits for successful participation in BOP rehabilitative programs. And these time credits may subsequently be redeemed through assignment to pre-release custody or supervised release during what would otherwise be the end of an offender’s regular prison sentence.

Recidivism is a serious issue for the federal justice system. According to an ambitious and systematic U.S. Sentencing Commission study in March 2016, nearly half of those released from BOP custody or placed on probation are rearrested for either a new crime or “conditions violation,” most of them within the first two years. Not all rearrests result in reconviction, of course, and not all reconvictions result in reincarceration. But many crimes go unreported or otherwise undetected, so even the rearrest rate, our least restrictive and largest proxy category for federal recidivism, gives a much-less-than-complete indication of the problem’s size. In the simple interest of public safety, then, if we are to shorten sentences and thus incrementally reduce the federal prison population, it stands to reason that we must do so only with the utmost care and caution, employing only the most rigorous and reliable risk-prediction methods.

And this will by no means be a straightforward or easy task. Before the effort even begins, the Justice Department officials and lawmakers responsible for undertaking and overseeing it should have a clear-eyed understanding of certain baseline complexities they will surely encounter along the way—and temper their expectations about what a recidivism-targeted BOP “risk and needs assessment system” can reasonably achieve.

There is no such thing as a “perfect” risk-assessment device. A large body of research suggests that actuarial systems based on large data pools—especially if they incorporate “dynamic” factors (like age and psychological profile) into their otherwise “linear” or “static” predictive algorithms—provide more accurate risk assessments than are generally possible using the judgments of experienced, professional individuals alone. And, in theory at least, when well designed and scrupulously validated, these statistical systems should be less prone to bias. They are not foolproof, however. Auto insurance premiums for younger drivers aren’t higher because each of those drivers will inevitably be involved in an accident, after all. And in the criminal justice context, even the most robust and exhaustively researched risk assessment instruments—like the Post Conviction Risk Assessment (PCRA) program used by the U.S. probation and pretrial services system—still produce “false positives” (offenders classified as higher risk who do not in fact violate the terms of a probationary sentence) or “false negatives” (offenders classified as lower risk who do) at a roughly 30 percent rate.

**The very particular “risk and needs assessment system” described at considerable length in Section 101 of the First Step Act does not currently exist and may take time—lots of time—to properly develop and implement.** The federal Bureau of Prisons now makes determinations about individual inmates’ security and custody needs using a relatively simple tool statistically validated only to predict future misconduct in its facilities. BOP currently makes no independent, actuarially-informed effort, as required under the new law, to determine the recidivism risk level of each prisoner at intake; identify the kind and degree of rehabilitative programming appropriate for each prisoner; or assess what might constitute the “successful completion” of such programming on which eligibility for the First Step Act’s earned time credits will depend. Designing such a reliable and legally defensible system from scratch—or even adapting and customizing something roughly analogous and already in use like PCRA—will and should involve an elaborate project of data collection, variable selection and weighting, predictive validity testing, and assessment and refinements of any proposed algorithms for their likely real-world results and general “fairness.” Six months may not be enough. Congress should be prepared, if necessary, to grant the Attorney General a measure of flexibility with his deadlines.

**Certain effects of the required risk and needs assessment system may prove to be in tension with one or more of the First Step Act’s broadest goals.** A not insignificant percentage of federal prisoners whose prior criminal histories—the most reliable statistical predictor (along with age at sentencing and release) of post-release recidivism—would generally suggest a lower risk classification may nevertheless remain ineligible for the new law’s earned time credits because they have been sentenced for one of the First Step Act’s long list of excluded offenses. How large an effect these exclusions may have on the law’s implicit prisoner population-reduction goal is impossible to estimate with any accuracy from currently public data. The Bureau of Justice Statistics’ most recent Criminal Case Processing Statistics reporting on conviction offenses for BOP inmates dates from Fiscal Year 2014. Reliable assessment of the First Step Act’s recidivism reduction and prisoner population effects may well require significantly more aggressive data collection and dissemination efforts from BJS and other relevant federal agencies. And depending on what those effects ultimately turn out to be, Congress may subsequently wish to revisit the First Step Act’s earned time credit eligibility and exclusion rules.