Setting bail is a difficult task for judges. They must try to foretell whether the defendant is likely to commit another crime, hurt someone or skip out on the next court date.
Now comes help in a distinctly modern form: an algorithm.
After two years of testing, the formula, developed at a cost of $1.2 million by the Laura and John Arnold Foundation, is being rolled out to 21 more jurisdictions, including states like Arizona and New Jersey and cities like Chicago and Pittsburgh, the foundation announced on Friday. The algorithm 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 most of the country, there is little science behind the bail decisions made thousands of times a day by magistrates, commissioners and judges. In some places, bail is based on the charges alone; in others, court officials may weigh a host of factors like criminal record, employment status and substance-abuse history. Hidden biases against the poor and minorities can easily creep into the decision-making. And a growing body of evidence indicates that the nation’s bail system keeps many low-risk defendants incarcerated before trial, while those who may pose a higher risk are released because they have the money to make bail.
Many law enforcement groups and defense lawyers have supported the use of scientifically validated risk assessments, but fewer than 10 percent of jurisdictions use them, partly because of cost.
The Arnold Foundation eventually plans to make the tool, called the Public Safety Assessment, available to any jurisdiction.
The effort comes amid new scrutiny of municipal courts after a federal report on the predatory, racially biased practices of the court system in Ferguson, Mo., and a growing bipartisan consensus that reducing the nation’s heavy reliance on jails and prisons should be a priority. In developing the risk assessment, the Arnold Foundation went beyond the typical philanthropic functions of conducting studies and awarding grants in the hope of achieving concrete, widespread change. Its assessment is designed to be more economical than existing risk assessments and effective regardless of location.
The foundation is based in Houston and is dedicated to using evidence-based practices in the public sector.
The Arnold assessment has been met with some skepticism because it does not take into account characteristics that judges and prosecutors normally consider relevant: the defendant’s employment status, community ties or history of drug and alcohol abuse. Instead, after crunching data on one and a half million criminal cases, researchers found that fewer than 10 objective factors — basically age, the criminal record and previous failures to appear in court, with more recent offenses given greater weight — were the best predictors of a defendant’s behavior. Factoring in other considerations did not improve accuracy.
Some initial skeptics, including R. Andrew Murray, the district attorney of Mecklenburg County, N.C., which includes Charlotte, have slowly warmed to the assessment. Charlotte was one of the few jurisdictions in the country that already used a risk assessment tool, but it included a face-to-face interview. The Arnold assessment eliminates the interview.
“I’m expected to do everything I can to keep the public safe,” Mr. Murray said. “If we’re letting more people out earlier in the proceeding, based on more limited information, I’m going to be concerned.”
But, he said, after a yearlong trial, Charlotte’s jail population is down almost 20 percent. Crime has not increased, he said, and many poor defendants have been spared the damaging effects of incarceration, including unemployment and homelessness.
“It’s saved the community a lot of tax dollars, there’s not been an ill effect, and we’ve kept a lot of individuals from going through that turmoil,” Mr. Murray said.
A defendant’s risk scores are given to the judge before bail conditions are set. At times the scores have bolstered prosecutors’ arguments that youthful, baby-faced defendants can be riskier than they appear, Mr. Murray said.
Scott Bales, the chief justice of the Arizona Supreme Court, said the state was expanding use of the assessment from four counties and one city to all 15 counties after judges had clamored for change.
“We heard from judges that defendants were held, pending the resolutions of their charges, for longer than the sentence would have been, and that seemed fundamentally unfair,” Chief Justice Bales said. “They didn’t have information to make an assessment, and were relying on rule-of-thumb or prior practices without really knowing whether those were useful guides or not” when setting bail.
Chief Justice Bales said the assessment tool could combat implicit bias, the invisible set of assumptions based on race, class and other factors that can come into play. Some studies have shown that black defendants are given higher bail amounts than similar white defendants.
Fewer than 10 percent of all jurisdictions — including the District of Columbia, Charlotte and the state of Kentucky — use formal risk assessments, said Anne Milgram, the vice president for criminal justice for the Arnold Foundation and the former attorney general of New Jersey. Such assessments are costly to conduct and are sometimes discounted by judges because they depend on information reported by defendants. Moreover, she said, the foundation’s researchers found that many jurisdictions had made tweaks to their assessments, adding or removing questions without testing the effect on accuracy.
A defendant’s risk scores are given to the judge before bail conditions are set, and are meant to augment, not replace, the judge’s discretion, Ms. Milgram said.
Kentucky has used a risk-assessment tool for decades, and is a leader among states when it comes to court appearance rates and low recidivism. Two years ago, it switched to using the Arnold assessment. Tara Klute, the head of Pretrial Services, said that although verifying criminal records was time consuming, particularly for out-of-state offenses, the Arnold assessment required less staffing and eliminated factors that were subjective, irrelevant or even discriminatory. Historically, she said, defendants had been given credit for such things as having a telephone or being married.
Ms. Klute said that after 20 years in the pretrial field, she was not surprised to learn that certain factors that were generally considered highly relevant to defendant behavior, such as drug abuse, did not turn out to be particularly predictive. According to informal, internal data analysis of drug test results in her own office, she said, “the people who were testing dirty actually had a better appearance rate than the people who weren’t using.”
A version of this article appears in print on June 28, 2015, on page A18 of the New York edition with the headline: Judges Replacing Conjecture With Formula for Bail.