Connecting state and local government leaders
A predictive computer model is a key part of the overhaul.
With sweeping bail reforms set to go into effect in the state early next year, New Jersey Courts technology staff are putting in place a sophisticated, new computer platform, designed to help judges make decisions about when people awaiting trial should go free or remain in jail.
Ushered in by legislation and a voter-approved constitutional amendment passed in 2014, the reforms call for courts in the state to shift toward what’s known as a risk-based system when it comes to determining what happens to pretrial defendants. The idea is to prevent situations where people accused of minor offenses languish behind bars because they cannot afford bail, while others who can cover the cost avoid jail time—even though they might go on the lam, intimidate witnesses or commit crimes.
There’s also the possibility that someone who poses moderate risks might be released with special conditions, such as check-ins with authorities, curfews, or electronic monitoring.
“What we found was about 15 percent of our jail population was made up of individuals who were in jail on a $2,500-or-less bail,” Jack McCarthy III, Information Technology Office director for New Jersey Courts, said by phone in late April. “So that person might sit there for a day, or six months waiting for their day in court, and it was just because they didn’t have money.”
“We also were faced with individuals who, no matter how high the bail was posted, that individual was going to get out and they may be a problem for society,” he added.
To provide judges more information they can use to assess a defendant’s risk level, New Jersey has turned to a tool from the Laura and John Arnold Foundation.
Dubbed the Public Safety Assessment, or PSA, the pretrial risk assessment tool provides a way to generate three risk scores. These scores are meant to predict the likelihood a person will fail to appear in court, commit a new crime, or commit a violent crime, while awaiting trial. Producing the scores involves examining nine criminal risk factors, such as the defendant’s current charge, previous convictions, prior jail time and past failures to appear in court.
The assessment tool is generally viewed as one of the most thoroughly vetted frameworks for statistically sizing-up the risks defendants might pose. It was developed using a database of over 1.5 million cases drawn from over 300 U.S. jurisdictions.
But implementing the tool in New Jersey has required some lifting by the state courts IT staff.
‘Risk Assessment in About Five Minutes’
“We’re trying to automate the entire process,” McCarthy said. Rather than having someone search case management systems for records needed to come up with defendant risk scores, he explained, “we’re providing that information in a few seconds by clicking a button.”
To run a risk assessment, it’s necessary to access past criminal justice data for a defendant, and any aliases they may have used. In New Jersey, this means pulling information from roughly 40 million records, across about a dozen different case management systems used by courts and law enforcement agencies in the state, along with calling on federal databases.
“Once we have a finalized complaint,” McCarthy said, “we can run a risk assessment in about five minutes.” After an assessment is complete, the final product delivered to a judge is a report that shows the three risk scores, as well as the information used to produce them.
To develop the computer application that will be used to run the pretrial risk assessments, New Jersey Courts has relied on the Pega 7 platform, from Pegasystems Inc. “It's like a giant sandbox of Lego blocks,” Doug Averill, the company’s global public sector business line leader, said during a recent interview, as he described the platform. “What the people on site in New Jersey do is then reassemble those blocks into exactly what they need.”
‘A Tool in a Toolbox’
Once they’re in use, risk assessments are not the only factor that judges weigh when deciding whether or not to let a defendant go free.
“It’s a tool in a toolbox and, you know, there’s certainly other things that a judge could consider when setting bond,” said Robyn Withrow, an assistant district attorney in Mecklenburg County, North Carolina, which includes Charlotte, where the Arnold Foundation risk assessment system is in use.
Withrow noted that a member of a defendant’s family might speak on their behalf at a pretrial hearing for instance, or that a victim could be there in the courtroom.
She emphasized that while the assessments can be helpful in keeping low risk defendants out of jail, having a strong legal advocate early on in the judicial process is crucial as well. It is also important, Tvedt said, for defense lawyers to understand how risk assessments could be biased against a certain race or gender. “There may be circumstances where the risk assessment tool may be skewed in some ways,” she said. “Your lawyer should be able to distinguish that.”
Matt Alsdorf, vice president of criminal justice at the Arnold Foundation, was adamant that special care has been taken to avoid this kind of bias with the tool the group has developed.
“From jump, we have been looking at the issues around racial equity,” he said. “We don’t use any of the factors that most people have pointed to as potentially having a racially disparate impact. So things like neighborhood, or the age the first time someone was arrested, or employment,” Alsdorf added, “those are things that people have said: ‘uck, that makes us queasy because, man, I could see how that could be racially disparate.’”
He made another point as well.
“I think that there's a lot of evidence that the status quo system has a lot of negative ramifications for minority communities,” he said. “It’s not like risk assessment then makes the system perfect. It doesn’t solve everything. But it can help alleviate some of the real problems.”
Data analysis so far seems to show the tool is producing good results, and is avoiding the pitfalls surrounding bias concerns, according to Alsdorf. Kentucky has seen a reduction in pretrial detention, alongside a reduction in pretrial crimes, he said. So, in other words, more people are getting released, but fewer people are getting rearrested.
As for how the PSA has performed so far in Mecklenburg County, Withrow said: “I think it’s a useful tool, it’s certainly been validated, it seems to predict what it was intended to predict.”
‘A Big Sea Change’
Alsdorf said during an interview in late April that 30 jurisdictions, including the states of Arizona and Kentucky, are either using, or are in the process of implementing, the Public Safety Assessment. This year, between five and 10 more places are expected to join their ranks. And the foundation has received over 200 serious inquiries from jurisdictions interested in the tool.
“Ultimately we’re committed to making the tool available to any jurisdiction that wants it,” he said. But implementing the assessment can be a challenge for a number of reasons.
One is that it can require a significant shift in mindset for judges and others involved in the criminal justice system, as they transition to a framework for pretrial decisions that is based on risk, as opposed to criminal charges. Alsdorf noted: “That's a big sea change.”
Another hurdle is on the technical front.
“By and large, criminal justice data across the country is dismal,” Alsdorf said. “I’ll just say it.”
“It’s often really, really hard in any system to pull things that you would think would be really basic metrics,” he added. “And to then make sure that you’re able to pull the nine risk factors that go into the PSA, and that those nine risk factors are, in fact, going to be accurate.”
McCarthy pointed out that, in New Jersey, courts from the municipal level to the state Supreme Court fall under one administrative unit: the Administrative Office of the Courts. This eased the difficulties somewhat when linking together the information needed for the risk assessment tool.
Some of those records, however, are housed using 30 to 40 year old computer technology.
“While these systems have always been viewed as a burden for us,” he said. “There’s really a tremendous amount of value in having this historic information.”
All told, McCarthy said the IT costs for establishing the risk assessment tool would be few million dollars, though he could not provide an exact figure.
He voiced pride that his team completed their work in March on the algorithm that provides the foundation for the tool, long before the Jan. 1, 2017 deadline for the bail reforms to set in.
“The Arnold Foundation will come in and validate our tool over the next few months,” McCarthy explained. “By having the tool already done and in place, we're able to go through and provide six to seven months worth of testing to make sure that it’s behaving the way it should.”
He also highlighted that the assessment tool is flexible, and can easily be revised to accommodate future changes to the predictive model behind it.
“We should be able to provide more information to the judge than they’ve ever had before,” McCarthy said as he discussed the tool. “At least they’ll have a fuller picture of the individual in front of them.”
Bill Lucia is a Reporter for Government Executive’s Route Fifty.