The First Step Act, the federal prison reform bill that President Donald Trump signed into law on Friday, represents a bipartisan and major effort at making the criminal justice system fairer. This step will only be a baby step, however, if the engine that drives the entire piece of legislation—risk assessments of federal prisoners’ likelihood to reoffend—is not used carefully and with sound scientific and public oversight.
The statute embraced two main types of interventions designed to reduce federal reliance on incarceration. The first increases judges’ discretion to impose shorter prison sentences. Time will tell whether they use it consistently and fairly.
The second allows federal prisoners, who now number about 180,000, to earn credits toward early release based on rehabilitative programs and their risk of reoffending. The statute states that an algorithm will be used to score every prisoner as minimum, low, medium, or high risk. But the legislation does not say how this algorithm will be designed. The final version of the First Step Act, which refers to “risk” 100 times, calls for a “risk and needs assessment system” to be developed in 210 days, and then made public and administered to every federal prison within the following 180 days.
That may not be nearly enough time to carefully study all of the questions raised by creating such a massive system. Take as an example the experience in Virginia, which has been hailed as a national model and “leading innovator” by the American Law Institute for using risk assessment to divert low-risk offenders from prison. Virginia spent several years developing its risk assessment system. The Virginia Criminal Sentencing Commission carefully obtained public input, scientific evaluations, and pilot studies, before implementing it statewide.
But in a recent series of studies of the effort to divert prisoners in Virginia, John Monahan, Alexander Jakubow, Anne Metz, and I have found that there is wide variation in how courts and judges apply this risk assessment. In 2016, for example, 42 percent of low-risk offenders received alternative sentences and 58 percent did not. Most of those who were diverted from prison still received jail time.
Why did this occur? We surveyed the Virginia judges and found that many appreciate risk assessment as, in the words of one, “another tool that aids but does not supplant judicial judgment.” However, others expressed extreme distaste. One wrote: “Frankly, I pay very little attention to the [risk assessment] worksheets. … I also don’t go to psychics.” Many judges correctly noted that they cannot easily divert offenders from prison when there are not adequate resources available for rehabilitative programs.
People are not algorithms. The statute’s fairness will hinge on the discretion that prison officials exercise, informed by the scores from a risk assessment but also by their own judgment. The First Step Act’s success will similarly depend on resources for real rehabilitative programs. It calls for evidence-based evaluation of such programs, but that research will also take time.
While using an evidence-informed tool can be better than simply leaving everything to prison officials’ discretion, there needs to be more than buy-in by the decision-makers—the right tools need to be used. Michelle Alexander and others have raised concerns, for example, with risk assessments that rely on information about prior arrests or neighborhood information that can produce stark racial bias. The final version of the act speaks to the potential for bias and asks the comptroller general to conduct a review after two years to identify “unwarranted disparities.” The act also calls for an independent review body that includes researchers who have studied risk assessment and people who have implemented it. These are important steps. Involvement of scientists and the public will be needed to consider whether invidious and potentially unconstitutional discrimination results—otherwise, protracted constitutional litigation challenging these risk assessments will be a foregone conclusion.
Still, there is much that is positive about the bill’s many provisions dealing with risk. The First Step Act emphasizes not just recidivism but also programs that support rehabilitation. It is noteworthy that the legislation calls for re-evaluation of prisoners each year so that risk scores are not set in stone. All prisoners are able to reduce their classification. This should be taken seriously. The risk of any person may decline dramatically over time simply as a matter of age, as the U.S. Sentencing Commission documented in a study last year.
The statute also makes the attorney general the risk assessor in chief—with input from the independent scientific reviewers—of the risk assessment used on 180,000 prisoners each year. That scientific input is critical, and it should be solicited from the broader scientific community. It’s also worth noting that the Department of Justice has recently shut down key science advisory groups; this law hopefully takes an important first step toward bringing science back in.
The hard work that went into the First Step Act should be followed by serious scientific and public deliberation about the sound use of risk-based evaluation. The act will not do justice to the needs of our federal criminal system if the first step results in a hasty sprint to uncritically embrace risk assessment.