OpenJustice shows striking disparities in California’s criminal justice system.


NYPD will document use-of-force incidents, but who can see the data?

Criminal Justice

Pennsylvania will vary jail terms for the same crime, based on where you live.

Pennsylvania's new approach to sentencing plans to look beyond past crimes, to factors like age, gender, and where an offender calls home.

Pennsylvania will soon begin using factors other than a convict’s criminal history — such as the person’s age, gender, or county of residence — in its sentencing guidelines, as The Marshall Project and the numbers-oriented blog FiveThirtyEight jointly report.

This move crosses a subtle but important dividing line. Until now, sentencing for a crime has been based on criminal justice information: the current offense, past offenses, and specific factors relevant to the crime such as whether it took place in a school zone. Pennsylvania’s new plan will use factors unrelated to the crime, and unrelated to the culpable choices of the person being sentenced, to increase or decrease proposed sentences in state-wide guidelines.

Moral concerns about the potential unfairness of statistical methods — which are a species of profiling — are similar across the investigation, sentencing and probation/parole phases of the criminal justice system. But the legal posture across these three contexts is quite different. The Constitution strongly protects fairness in investigation of crimes, and in sentencing. Probation and parole, which are largely discretionary, have used demographic predictive factors for many decades.

Discussions of a statistical tool's accuracy or transparency too often skip past questions of fairness.

What’s new now in Pennsylvania — and what the Department of Justice (DOJ) has previously described as a potential constitutional problem — is that these demographic factors are starting to be used to determine sentence length.

Last summer, Attorney General Holder warned that “[b]y basing sentencing decisions on static factors and immutable characteristics — like the defendant’s education level, socioeconomic background, or neighborhood,” such policies “may exacerbate unwarranted and unjust disparities that are already far too common in our criminal justice system and in our society.”

Similarly, the policy office of the DOJ’s Criminal Division raised a red flag about the use of demographics in sentencing as part of its 2014 annual letter to the U.S. Sentencing Commission:

We think basing criminal sentences . . .  primarily on [demographic characteristics] — rather than the crime committed and surrounding circumstances — is a dangerous concept . . . This phenomenon ultimately raises constitutional questions because of the use of group-based characteristics and suspect classifications in the analytics. Criminal accountability should be primarily about prior bad acts proven by the government before a court of law and not some future bad behavior predicted to occur by a risk assessment instrument.

These assessment tools first entered the U.S. justice system to help in determining probation and parole decisions, both of which courts and the Constitution do not strictly supervise. As the Vera Institute of Justice notes, until the 1970s, “assessment of [inmate] risk was a matter of professional judgment: correctional and clinical staff decided which offenders presented the greatest danger to public safety.” Then, the first actuarial, evidence-based risk assessments for probation and parole emerged. Those assessments used static factors that had been shown to predict the risk of reoffending — like criminal history or a history of substance abuse.

These statistical tools entered the criminal justice system in places where legal protections are fairly lax, such as in setting bail, level of parole supervision, or type of confinement within a prison.

Now, as FiveThirtyEight and The Marshall Project report, there are “more than 60 risk assessment tools in use across the U.S.” These statistical tools have entered the criminal justice system in places where legal protections for fairness are relatively lax, such as setting bail, determining the intensiveness of parole supervision, or moving inmates among different categories of confinement within a prison. “An overwhelming majority of corrections agencies nationwide routinely utilize assessment tools to some degree,” reports the Vera institute. The practice is so widespread that Pew argues that “[r]isk/needs assessments have become a cornerstone of good correctional practice.”

Proponents of risk assessment tools in probation and parole argue that data-driven systems can help counter the subjective, subconscious biases present in judges or parole officers. And when it comes to actually measuring risk, many of these tools do appear to be predictive. As the Congressional Research Service found, “research indicates that most commonly used risk and needs assessment instruments can, with a moderate level of accuracy, predict who is at risk for recidivism.” These tools can also, if properly designed and configured, create a clear record of why a decision was reached.

But discussions of accuracy and transparency too often skip past questions of fairness. Considering an offender’s geography in setting sentence length is a recipe for punishing black and brown offenders more heavily when they commit the same crimes, relative to whites. As FiveThirtyEight and The Marshall Project observed, “Pennsylvania’s proposed tool … will also include a question on where offenders live and, in some cases, penalize residents of urban areas, who are far more likely to be black.”

Indeed, in its most recent report on the project, Pennsylvania’s sentencing commission writes that “even though removing gender and county” from its recidivism predictions would have a very limited effect on the accuracy of the predictions, these factors should still be considered because they can make predictions slightly more accurate.

The “Public Safety Assessment” tool, developed by the Laura and John Arnold Foundation for pretrial bail and detention assessments, wrestles with fairness issues more directly. As the New York Times recently reported, their pretrial risk assessment tool “gives defendants two scores — one for their likelihood of committing a crime and one for their risk of failing to appear in court — and flags those with an elevated risk of violence.” In creating their tool, the researchers excluded potentially discriminatory variables — like a person’s education or where they live — from consideration. Instead, they decided to consider only factors “related to a defendant’s criminal history and current charge,” which are more strongly predictive in any case.

Pennsylvania’s step puts the difference between two competing mindsets into stark contrast. When the goal is to manage cost or anticipate future behavior, statistics can help. But when our primary obligation is to be fair to an individual person on trial, the facts and law — not the person’s demographic odds — should control what happens to the person.

External Reading

"Should Prison Sentences Be Based On Crimes That Haven’t Been Committed Yet?"

Read the joint story by The Marshall Project and FiveThirtyEight on Pennsylvania's use of a risk assessment tool in criminal sentencing. Read

"Risk/Needs Assessment 101"

Read an issue brief regarding risk/needs assessments prepared by The Pew Center on the States. Read

Vera Institute of Justice's Report

Read the Vera Institute of Justice's report regarding risk/needs assessments to Delaware's Justice Reinvestment Task Force. Read

The Department of Justice's Annual Letter to the U.S. Sentencing Commission

Read the DOJ's letter to the U.S. Sentencing Commission regarding the potential use of demographic factors in criminal sentencing.Read

We'd love to hear from you. Send us an email: