JERSEY CITY, N.J. ─ On the ground floor of a deteriorating county courthouse, in a room outfitted with temporary office furniture and tangles of electrical wires, a cornerstone of America’s criminal justice system is crumbling.
A 20-year-old man in a green jail jumpsuit appears on a video monitor that faces a judge. It is early June, and he has been arrested for driving a car with a gun locked in the glove compartment.
If he were in almost any other courtroom in the country, he’d be ordered to stay behind bars until he posted bail — if he could afford it. This is what millions of people charged with crimes from shoplifting to shootings have done for more than two centuries. The bail system, enshrined in the Bill of Rights, is meant to ensure that all defendants, presumed innocent before trial, get a shot at freedom and return to court.
But allowing people to pay for their release has proved unfair to people who don’t have much money. The poor are far more likely to get stuck in jail, which makes them far more likely to get fired from jobs, lose custody of children, plead guilty to something they didn’t do, serve time in prison and suffer the lifelong consequences of a criminal conviction. Those who borrow from a bail bondsman often fall into crippling debt.
At the same time, the wealthy can buy their way out of pretrial detention on just about any offense, including murder.
The bald inequity of this system has triggered a national movement to eliminate bail altogether.
But what to replace it with?
In New Jersey, the answer is an algorithm, a mathematical formula to determine whether someone is likely to return to court for trial or get arrested again.
Back in the courtroom, the formula has run the criminal history of the man caught with the gun. The young father ─ who said he didn’t know the weapon was in the car ─ has nothing on his adult record, but he pleaded guilty to sex assault as a juvenile. That record has been given a numerical weight and then boiled down to two numbers that appear on the judge’s computer screen. They say the man is a low risk of skipping court and committing a new crime.
“Release On Own Recognizance,” the formula tells the judge.
But prosecutors think he’s too dangerous. They want him locked up. Their disagreement forces a second hearing in which the government must persuade a judge to override the algorithm’s recommendation.
This is what the new vision of American justice looks like.
Created by data scientists and criminal-justice researchers, the algorithm — one of dozens of “risk assessment tools” being used around the country — promises to use data to scrub the system of bias by keeping only the most dangerous defendants behind bars, regardless of their socioeconomic status.
Six months into this venture, New Jersey jails are already starting to empty, and the number of people locked up while awaiting trial has dropped.
But it’s also become clear that data is no wonder drug.
The new system — driven by years of research involving hundreds of thousands of cases and requiring multimillion-dollar technology upgrades and the hiring of more judges, prosecutors and court workers — still produces contentious decisions about who deserves freedom and who does not.
Police officials and prosecutors have complained about the release of people charged with gun crimes, fleeing police, attacking an officer, sex offenses and domestic violence — and of those who keep getting re-arrested. In at least two cases, people have been killed by men who’d been released on earlier charges. The bail bond industry, facing extinction, has backed two federal lawsuits seeking to end the algorithm’s use.
Defense lawyers and civil rights advocates, meanwhile, say people who pose little risk have been ordered detained, only to be given plea deals or had their charges dropped — a sign, they fear, that authorities are exploiting the new system to generate convictions.
And it remains unclear whether the new approach will reduce racial disparities, drive down crime rates or be fiscally sustainable.
Still, it is the best alternative anyone’s come up with.
And if this grand experiment works in New Jersey, it could become a model for the rest of the country.
Jail as last resort
Modern algorithms promise to objectively weigh whether someone will behave a certain way. But they fall short in one key aspect: they can never reflect the mystery and uncertainty of everyday life.
Algorithms need humans — flaws and all — to oversee them.
That is why New Jersey’s algorithm, called the Public Safety Assessment, doesn’t get the last word.
The ultimate decision lies with the judge, but only after hearing arguments from defense lawyers and prosecutors, who must ask that someone be held before trial.
In the case of the man charged with gun possession, prosecutors challenged the algorithm’s recommendation, arguing that his purported gang ties made him too risky to release. A judge agreed, and ordered him to stay behind bars.
He joined the relatively small segment of New Jersey defendants who’ve been held before trial with no option for release.
With bail all but eliminated, about a third of all defendants have been subject to detention requests since New Jersey adopted its new pretrial justice system on Jan. 1. Far fewer, about 14 percent, have been ordered held behind bars until their cases are adjudicated.
Court officials say the early numbers show the new process is already working: people who aren’t dangerous are not being jailed solely because they can’t afford bail, and dangerous people aren’t being released even though they can afford to pay.
“You can disagree with an individual decision, but you can’t deny the systemic change,” said Glenn Grant, who manages the reforms as acting administrative director of the New Jersey court system.
There will always be cases in which people who get released go on to commit new crimes ─ as was the case under money bail, Grant said. Exactly how many times this has happened under the new system is not yet clear, because his office has not yet released data on such cases.
“If you want a 100 percent guarantee that any individual will not commit a crime while out on release, the only way to do that is to lock everybody up,” Grant said.
The case against bail
Across the country, reformers are chipping away at money bail, arguing that it discriminates against the poor, ruins innocent people’s lives, fuels mass incarceration and contributes to wrongful convictions. The movement is part of a much broader effort to end abuses across the criminal justice system, from biased policing to burdensome court fees to mandatory minimum prison sentences.
Philanthropic organizations are funding projects in more than three-dozen states to eliminate bail and adopt algorithm-based risk-assessment tools. Judges are pushing similar efforts in Maryland, Arizona, and Indiana. Lawmakers are the driving force in a handful of other states, including Illinois and California.
Civil rights organizations have challenged bail-based systems in a dozen local jurisdictions, arguing that they violate constitutional protections of equal treatment and due process. The targets range from tiny Calhoun, Georgia, where the U.S. Justice Department has argued that the town’s use of cash bail was unconstitutional, to Harris County, Texas, where in April a judge ordered authorities to stop detaining people charged of low-level crimes.
In July, a team of Republican and Democratic U.S. senators introduced a bill that would encourage more states to replace money bail with risk assessment tools.
The strategies vary, but the goal is the same: improve public safety and reverse an unprecedented growth in incarceration rates.
The number of people behind bars in the United States has jumped more than threefold, to 2.1 million, since 1980, according to the Bureau of Justice Statistics. Of the 693,300 prisoners held in local jails in December 2015, 62 percent were waiting for trial, at an estimated cost of $14 billion a year. The Prison Policy Initiative projects that 70 percent of those pretrial detainees are charged with non-violent crimes — mostly involving drugs or property offenses.
In theory, bail — and the network of bail-bond companies that provides loans to those who can’t pay — gives just about anyone the opportunity to remain free while awaiting their fate. Judges typically set bail amounts based on the charge, a look at a defendant’s record and home life, and their own experience and intuition. Bondsmen and bounty hunters say they provide an essential service, and help make sure that defendants show up for their court dates — and pay the price if they don’t.
But many Americans can’t afford the smallest of payments. Some who manage to scratch together the money remain in hock for years.
Deep-pocketed defendants, meanwhile, face no such challenges.
From there, the inequities pile up. People sitting in jail while awaiting trial on low-level charges are more likely to be black and Hispanic. And the longer they sit behind bars, the more likely they are to commit a crime once they get out, researchers have found. That could be because jail can wreak havoc on a person’s ability to maintain a stable life.
Being locked up also makes it harder to fight a case. Many end up pleading guilty just to go home — including some who are innocent.
“It’s an inherently discriminatory system,” said Roseanne Scotti, director of the New Jersey chapter of the Drug Policy Alliance, which led the campaign to do away with bail. “People who have the resources can get out, and the people who can’t are punished by sitting in jail. They lose their jobs, housing, the connection with their families.”
It happened to Mustafa Willis1 seven years ago. He still isn’t over it.
Mustafa Willis was arrested in June 2010 by police who said they saw him holding a gun on a Newark street. Willis, 24 at the time, was given a $50,000 bond, meaning that he would have to pay 10 percent, or $5,000, to secure his release. But Willis worked part-time for a liquor distributor and didn’t have the money.
While locked up, Willis heard that someone had shot a video that appeared to show police planting the gun. But he wasn’t able to get a copy until nearly four months later, when a judge cut his bail to $30,000 and relatives co-signed a bondsman’s loan, he said. After prosecutors saw the video, they dropped the charges.
By then, Willis had lost his job, making it hard to repay the bondsman. He got arrested on a second gun charge, which also was dismissed. But it left him deeper in debt.
Willis has since gotten the arrests expunged from public records, but he still struggles to find long-term work. He sued the police, and the city settled for $6,000, most of which he used to reimburse his mother for helping him through the last few years.
Willis, who is 31, spent nearly four months in the summer of 2010 in jail on a gun possession charge that prosecutors dismissed after he tracked down a videotape that he said showed Newark officers planting the weapon. He lost his job and relied on family members to repay his bondsman. A second gun charge, also dismissed, added another bondsman’s loan.
Around that time, a campaign to reform New Jersey’s justice system began gaining momentum, driven by a 2013 report commissioned by the Drug Policy Alliance that found 38 percent of inmates in New Jersey jails were there solely because they couldn’t afford bail.
That effort coincided with a growing movement, sparked by the shooting death of Michael Brown in Ferguson, Missouri and the post-jail suicide of Kalief Browder in New York, that focused the nation on the harms of fines, fees and bail in state and local courts.
New Jersey reform advocates found Willis and recruited him to lend his voice. In July 2014, he testified before the state Senate’s Law and Public Safety Committee, and a month later the new system became law, with a January 2017 start date.
“Next time somebody else get a gun put on them, they might not have to sit more than three months like I did,” Willis said recently. “They might be going back home to their family. They might be going back to their job.”
The prediction problem
America has struggled with the fairness of pretrial detention since the earliest days of the republic, when the protection against excessive bail was written into the Eighth Amendment of the Constitution. The U.S. system, borrowed from England’s, has been through three waves of reform as authorities sought to balance equal justice and public safety.
The first came in the 1960s, when anti-poverty crusaders led a charge for courts to release low-level offenders without conditions. Rising crime rates in the 1970s and 1980s fueled a second round of changes aimed at keeping dangerous defendants off the streets: judges were required to consider a person’s risk of committing a crime while out on bail.
That led to a new golden age of bail, as the commercial bail-bond industry grew in size and political strength and the number of pretrial detainees rose nearly fourfold, from 113,984 in 1983 to 434,600 in 2015.
The current wave of reform seeks to reverse the swing toward detention, and correct the system’s frequent failure to determine who poses the greatest risk.
That is where the algorithms come in.
Predicting crime is notoriously difficult, and highly divisive. The earliest prediction tools, first developed in the early 1900s, explicitly used nationality and race as factors, along with subjective descriptions of a person’s personality — a practice that gave way after the civil rights era to a reliance on more statistical methods. Even now, there is a fierce debate among researchers about the balance between prediction and bias.
Critics argue that even the most basic factors, such as someone’s criminal history, can exacerbate discriminatory treatment, because that history can be profoundly affected by race, class, neighborhood and the ability to afford a lawyer.
“One goal is clearly to eliminate the racial and class bias,” said Bernard Harcourt, a Columbia University law professor whose book, “Against Prediction,” charts the rise of actuarial methods in targeting criminals and meting out punishment. “What I’m saying is these methods actually make them worse.”
Criminal history, Harcourt said, is a reflection of how often someone has been caught breaking the law. “That’s very different from the likelihood of offending.”
There are dozens of risk assessment tools in use today, developed by universities, governments, private companies and nonprofit agencies. They are used at various points of the criminal justice system, from pretrial to sentencing to parole. (Apart from bail reform, New Jersey recently began using a risk assessment tool that measures whether someone accused of domestic violence is likely to assault an intimate partner in the future.)
The Public Safety Assessment, created by the Laura and John Arnold Foundation, has been adopted by Kentucky, Arizona and more than two dozen local jurisdictions, but none to the extent of New Jersey.
Proponents of the tools argue that relying on empirical evidence is far better than leaving it exclusively to judges. They acknowledge that poorly crafted tools can perpetuate bias, a problem that can be overcome through rigorous testing.
“Both models are imperfect, but which is less imperfect?” asked John Pfaff, who teaches criminal law at Fordham University. “You might be able to fix the model faster than you can fix a judge.”
The use of risk assessment tools in sentencing — which has expanded to more than 20 states — has raised the most concerns. In 2014, then-Attorney General Eric Holder warned against relying on data-based prediction methods to determine jail time, saying they could inadvertently undermine justice by focusing on factors outside of a defendant’s control. In Wisconsin, a man sentenced to prison based on the result of an algorithm argued that his inability to challenge the secret formula violated his due-process rights; the state Supreme Court ruled against him, and the U.S. Supreme Court declined in June to hear the case, allowing the state ruling to stand.
The tool cited in the Wisconsin case, called COMPAS, is based on dozens of pieces of data about defendants, collected from a questionnaire that includes their own moral judgments and the criminal activity of family and friends. COMPAS is used not only for sentencing, but also in pretrial cases, and was the subject of a 2016 ProPublica investigation that found it treated black defendants more harshly than whites. Those findings, however, have been challenged by researchers who said they found no evidence of racial bias.
Unlike COMPAS, the Public Safety Assessment is used exclusively in pretrial settings and is designed to be shared publicly and used anywhere.
The Public Safety Assessment, or PSA, is among a new generation of tools that claim to do away with factors that exacerbate biased treatment of minorities. Its makers say it has been calibrated to weigh aspects of someone’s criminal history that have the best predictive qualities in order to minimize bias.
For example, the PSA gives more weight to jail sentences of longer than 14 days, because researchers found that shorter periods of incarceration were typically associated with people who pleaded guilty because of their inability to post bond. The PSA also assigns more value to a person’s history of skipping court before their cases have been adjudicated. That decision is backed by research that showed people who failed to appear in court after disposition generally did so because they couldn’t pay their fines.
Conversely, the PSA’s developers excluded factors that were predictive but also likely served as proxies for race, such as a person’s arrest history and number of misdemeanor convictions.
The PSA and other modern pretrial tools have not been tested thoroughly enough to determine whether they affect people differently based on their race; the Arnold Foundation says it has asked three independent research agencies to evaluate the PSA. But proponents argue that they are at least an improvement over the current state of affairs. The Arnold Foundation’s own analysis of the PSA in Lucas County, Ohio, found that the arrest rate of pretrial defendants dropped from 20 percent to 10 percent, and that black and white defendants were released at similar rates.
New Jersey’s new law bars any system that results in discriminatory release patterns, and the courts plan to produce a report at the end of the year analyzing the results. The law also created a commission to review that report, look at what other states are doing, and recommend changes.
“I respect the concerns about racial bias, but we also need to look at the system that was in place before, which was phenomenally race- and social-status based,” said Marie VanNostrand2, the architect of the PSA and author of the 2013 Drug Policy Alliance report that drove New Jersey’s bail-reform campaign. “We can’t eliminate bias, but we can disrupt the cycle of bias and that’s what this tool is intended to do.”
Marie VanNostrand is among the top developers of algorithms aimed at making the pretrial justice system more fair. After decades toiling on the outskirts of criminal justice reform, she has found herself in high demand.
Her interest began as a probation and parole officer, when she saw large numbers of people with jobs, families and no criminal records stuck behind bars because they couldn’t afford bail. She also noticed people with long records who got released quickly because they could pay.
“The disparity and discrimination I saw based on race and socioeconomic status outraged me and fueled my desire to dedicate the rest of my life to changing that,” she said.
One problem: few were paying attention. This was in the mid-1990s, when little was being done to improve the pretrial system.
While in graduate school, VanNostrand volunteered to help Virginia create one of the country’s first pretrial “risk assessment tools.” She became an expert in the field, and in 2003 started her own business, Luminosity. Her first clients were jails being sued for overcrowding.
That led her to New Jersey, where she authored a study on jail populations that became a centerpiece of the campaign to end bail.
Reform and backlash
The New Jersey system, written into law in 2014, includes not only the PSA but also a new pretrial services unit and a speedy-trial mandate. The reforms also required changes to the state constitution, approved by voters, that eliminated the public’s right to bail and allowed for high-risk people to be held before trial.
The PSA, built from an analysis of 750,000 criminal cases across the country, reduces risk factors to a defendant’s age, criminal past, history of skipping court and whether the pending charge is violent. It crunches that data and produces two scores, both on a scale of one to six: one for failure to appear, another for committing a new crime. There is also a flag for those with a high risk of future violence.
The results are then entered automatically into a “decision-making framework,” written by Arnold with customizations by New Jersey, that produces a recommendation of whether the defendant should be released, and, if so, what conditions should apply. (Those conditions can include calling a newly formed pretrial services unit once a month, making weekly visits, wearing an electronic monitoring device, and home detention.) The most serious charges, such as murder, robbery and sex assault, carry a recommendation of detention no matter what the scores. But prosecutors still need to ask for it.
The judge considers the PSA results, along with arguments from prosecutors and defense lawyers. If the judge goes against the automated recommendation, the judge has to explain why. If prosecutors ask for detention, then the case moves to a mini-hearing, in which a second judge decides whether the risk is high enough to warrant pre-emptive incarceration.
In the vast majority of cases, mostly involving low-level crimes, defendants appear briefly before a judge the day after their arrest, hear what their PSA score is, and are quickly assigned conditions for release. Prosecutors usually do not contest those decisions.
But a number of cases have sparked legal challenges. One, prompted by the outcry against the release of a man accused of soliciting sex from a girl, led to an appellate court ruling on the handling of sex offenders and defendants’ juvenile histories. Authorities also complained about the release of serial criminals and people accused of fleeing police. But the strongest backlash came in response to the release of people charged with gun crimes.
The most dramatic example was the death of Christian Rodgers, shot in South Jersey in April, allegedly by a man who’d been released a few days earlier on a gun possession charge. Rodgers’ mother has filed a wrongful death lawsuit against the state, saying the PSA did not do enough to protect him.
A similar scenario unfolded in June, when a man awaiting trial on charges he threatened his ex-girlfriend shot her to death outside his Newark home, then killed himself, authorities said.
Other cases publicized by critics included a young man charged with shooting someone in the leg, a father accused of pulling a pistol on his step-daughter’s boyfriend, and several instances of street gunfire. But in several of those instances, prosecutors did not ask for the defendants to be detained.
As it turns out, a charge of gun possession doesn’t raise a defendant’s PSA risk score. That’s because the researchers who created it didn’t have data showing whether such a charge had predictive value.
But the new system allows officials to tweak how the automated recommendation is calculated by adding gun possession to the list of charges that call for detention. That is what the state attorney general’s office asked the courts to do, and the courts acquiesced. The attorney general also ordered prosecutors to be more aggressive in requests for detention. Since then, more gun defendants are being held before trial.
“We were seeing cases of firearms defendants, plainly dangerous, not being detained, and we saw people offending, being released and re-offending, sometimes within days, sometimes three or four times,” said Elie Honig, director of the state Division of Criminal Justice. “I’ve seen these and gone, ‘How can that possibly be?’”
Civil rights advocates and public defenders opposed the tweaks, saying they were based on anecdotal cases and would undermine the algorithm’s objectivity. They pointed out that there have always been people who commit crimes after being released from jail. A 2014 report on the bail industry by the state Commission of Investigation outlined a litany of abuses, including cutting deals with hardened criminals who went on to commit more violent crimes.
“Anyone can pick out a non-random selection of cases and come to the conclusion that the risk-based system is flawed,” said Joseph J. Russo, who runs the public defender’s office in Hudson County. “That would be a fundamentally erroneous conclusion to reach.”
He added: “You also have to realize that individuals are presumed innocent. Sometimes we forget that.”
Just ask Bruna Toovey3. The single mother’s home was burglarized twice by a serial thief who’d already been arrested and released under the new system.
In April, Bruna Toovey’s neighbor called her at work with an emergency: She and Toovey’s boyfriend had caught a burglar inside her Newark home. It was the second time in less than two months that Toovey’s place had been ransacked, and it turned out that the same man did it. He had a dozen felony convictions on his record. He’d also been arrested on burglary charges weeks earlier, but was released pending trial.
Police officers told Toovey all about it.
“I said, ‘How? Why?’ and they said, ‘We put them in, and they come out,’” Toovey recalled.
Prosecutors sought to keep the man locked up, but he was released again, only to be picked up for more burglaries. Only then did a judge agree to detain him.
Toovey assumed the burglar was stealing to support a drug habit, and she sympathized. But she lived alone with her 6-year-old daughter. What if they’d been home during the break-ins?
“I know these people need help, but that has nothing to do with us,” Toovey said. “I’m a good person. I go to work. I live with my daughter. He robbed my house and he came back for more. It’s just not safe.”
“It’s so sad, the whole thing,” Toovey said. “We live in fear.”
Even with the recent tweaks, results can seem inconsistent. A woman with high risk scores who is charged with assault gets released while prosecutors seek detention for a man with lower scores and drug charges. On the same day that prosecutors sought to detain the young man caught with a gun in Jersey City, they did not oppose the release of another young man accused of stabbing someone in a fight.
Some defendants, appearing before the judge via video link, appear surprised to be let go.
Defense attorneys, meanwhile, warn that the new system could give rise to a new type of injustice: prosecutors using detention motions as cudgels to obtain guilty pleas. They point to cases in which defendants have been ordered held on serious offenses such as assault or robbery only to be offered to plead guilty to low-level charges — and sentenced to time already served in jail.
“The reason I came to bail reform was seeing clients plead guilty to stuff they couldn’t prove, just to take probation and get out of jail,” said Alex Shalom, a former public defender who now leads bail reform efforts at the American Civil Liberties Union of New Jersey. “It has changed, because now the system isn’t about poverty. But with detention you have people pleading to stuff they wouldn’t have if they were out.”
There is also the matter of cost.
While the prospect of declining jail populations could one day lead to budget reductions, bail reform in New Jersey has required a significant investment in technology and upgraded courthouses as well as the hiring of additional court personnel, forensic investigators, public defenders, prosecutors and judges. The cost to the state for the coming fiscal year is estimated at about $46 million, with additional expenses covered by county governments, who have argued they shouldn’t have to pay.
The state’s portion is entirely funded by court filing fees, a fluctuating source of revenue that judges fear will turn out to be insufficient. Grant, who heads the Administrative Office of the Courts, has predicted that, without a new source of revenue, the program would be left without enough money to operate in fiscal year 2019.
The disappearing bondsman
The biggest loser in all of this, by far, is the bail bond industry.
It became obsolete virtually overnight.
“We saw it coming, but we didn’t expect it to be this quick,” said Al McCallen, a bounty hunter and manager of Ace Bail Bonds in Jersey City.
For many years, McCallen made a good living chasing down absconders, or “skips.” But that business evaporated. The company hasn’t written a bail bond all year. It was only a matter of time before he’d have to shutter the cluttered office across from the Hudson County courthouse and start working out of his car.
Other bondsmen were fighting back, supporting the Rodgers family’s wrongful-death claim and, in a second lawsuit, asking a federal judge to force New Jersey to bring back bail.
McCallen began leafing through dusty cardboard boxes for files of people who owed Ace money on old defaulted bonds.
“Since the founding of the country there’s been bail here,” McCallen said. “It has always existed, and it always worked. And with swoop of a pen, these guys say no more bail.”
He stood up and began pacing. At 48, he looked imposing: shaved head, chest-length goatee, black T-shirt and jeans, black work boots, sidearm. But he felt powerless.
“I’ve been doing this for 16 years,” he said. “I’m proud of being a good-standing licensed bondsman and private investigator. Now I don’t know what I will do. Maybe construction?”
McCallen recalled the January morning when bail reform went into effect, sitting in a courtroom and watching people he’d bailed out before — some of whom he considered flight risks — getting released.
“The idea of putting people in jail – isn’t that the f–king point of not committing a crime?” McCallen said. “You don’t do it because there’s a penalty. Now there’s not a penalty. These smug jackoffs on the street are laughing.”
Defense lawyers and civil liberties advocates disagree.
So do people like Mustafa Willis.
Willis, an aspiring actor, sometimes wonders how different his life would have been if he’d been arrested under the new system he helped get implemented.
The PSA very likely would have given him a low risk score. The gun charge may have raised flags, however, and prosecutors may have argued for his detention.
But what if he’d been released? He wouldn’t have suffered three months in bedbug-infested cells, missed a young cousin’s funeral, or lost his job. He wouldn’t have had to borrow from family and friends to pay back his bondsman. By now, maybe, he’d be living on his own, and closer to a career in film.
“A lot of things,” he said, “would have gone better.”
Clarification (Aug. 22, 2017 9:50p.m.): A previous version of this article said that researchers had refuted a ProPublica investigation that found a Wisconsin assessment tool treated black defendants more harshly than whites. Although researchers have challenged that investigation, they have not successfully disproved it.