Opinion: Should algorithm decide who gets criminal bail?

Jail.

Credit: David Silverman

Credit: David Silverman

Jail.

Civil rights groups signed a statement in late July calling for states to ditch pretrial risk assessment tools as a means of evaluating whether an individual accused of a crime should be detained pretrial, contending such data-driven tools do little to remedy racial disparities in the criminal justice system.

Conservatives, meanwhile, have expressed deep concerns over releasing individuals based on screenings that often don’t account for likelihood of risk to family members or the individual’s possession of a firearm.

Others contest these tools have significantly reduced the number of individuals detained pretrial in the states where they are used, potentially saving states billions of dollars, though that has often not been the case in practice.

Lucas County, Ohio, which includes Toledo, for example, has seen costs rise 23 percent in the three years since it first implemented its program. (The county began using a pretrial risk assessment tool from the Laura and John Arnold Foundation in 2015.)

Using the same tool in New Jersey, however, the pretrial detention population has dropped by 38 percent over the last two years.

Pretrial risk assessment tools have been implemented in some states as a replacement for money bail methods of detaining or releasing individuals pretrial.

The algorithmic tools use predictive analytics and various data points — like past criminal or re-arrest history, whether the individual failed to appear in court before, and socioeconomic factors like income and employment stability — to determine whether an individual is likely to show up for his or her court date and whether he or she is a threat to public safety.

Depending on the tools’ conclusions, judges may decide an individual should be released instead of held in jail pretrial.

The tool implemented in many of these criminal justice systems was funded by the Laura and John Arnold Foundation. The Arnolds, left-leaning Texas billionaires who made their fortune from defunct Enron and a hedge fund launched by Mr. Arnold, have pushed for their tool to be adopted in communities across the country.

In some states, the Arnold pretrial risk assessment tool has raised concerns as violent crime and rising costs have garnered the attention of lawmakers and residents.

In New Mexico, for example, Republican Gov. Susana Martinez said rising violent crime in the state makes the tool unusable. As violent crime has been on the rise, she said, releasing the accused pretrial is a serious threat to public safety.

According to a report issued by the Cuyahoga County (Ohio) Bail Task Force in Cleveland, current pretrial risk assessment tools like the Arnold Foundation’s do not take into account whether an individual released pretrial is a threat to spouses or family members. In some cases, evaluating an individual’s threat to public safety isn’t enough.

Perhaps the most incriminating case against pretrial risk assessment tools is the story of a 71-year-old man in San Francisco who was murdered by an individual who was on felony probation, had been re-arrested, then released pretrial based on a judge’s discernment after evaluating the individual’s threat to public safety via the Arnold pretrial risk assessment tool.

San Francisco has been a telling case study in the risks of implementing such a program. Of those released as a result of the assessment, more than a third were either booked on a new offense or failed to appear, according to a study.

While pretrial risk assessment tools can be modified to take into account many concerns — including racial bias — a wide variation in efficacy in early stages of implementation suggests that not all states and communities may have the conditions necessary for the tools to be successful. Further complicating the debate is the role of human discernment: ultimately a judge makes the final call, and judges make mistakes all the time.

The more common alternative, money bail, faces criticism, as well. Under the money bail method, individuals could pay their way out of pretrial detention. This method has long received criticism because it tends to favor the wealthy, leaving poorer individuals stuck in jail pretrial despite not yet being convicted.

“There’s a lot of criticisms about money bond and the fact that most people who are in jail are quite low income and might not get out of jail at all, and end up locked up longer because they’re poor,” said Jesse Jannetta, senior policy fellow at the left-leaning Urban Institute. “You can also have the criticism on the other hand which is that if you have access to a lot of resources you could get out even if the bond amount is quite high, so there’s a lot of pressure on money bond as a release mechanism because it’s unfair.”

Furthermore, being locked up pretrial could have a significant effect on a person’s quality of life and prospects, even if that person is found to be innocent in court.

“Jail incarceration, even if it’s short, has a lot of really bad consequences,” Jannetta said. “A day or two in jail could cause you to lose a job in at-will employment situations.”

Those seeking to reform the criminal justice system have targeted pretrial detention as a way to reduce prison populations and spending. The Prison Policy Initiative, which seeks to “expose the broader harm of mass criminalization,” found in 2017 that pretrial detention costs local governments $13.6 billion each year.

The civil rights groups’ statement argues that the criminal justice system and data-collecting methods are riddled with racial bias, so these tools could exacerbate racial disparities in prison populations. According to Census Bureau and Bureau of Justice statistics, more blacks are incarcerated than whites, respective to the U.S. black population and white population.

The civil rights groups concerned about the assessments argue that “pretrial risk assessment instruments, if used at all, should only identify groups of people to be released immediately,” and calls for the elimination of money bail in all states.

“We believe that jurisdictions should not use risk assessment instruments in pretrial decision making, and instead move to end secured money bail and decarcerate most accused people pretrial,” the statement reads.

The Laura and John Arnold Foundation — which provided New Jersey with its pretrial risk assessment tool — responded to the statement released by civil rights groups, affirming the need to make these data-driven tools “transparent; designed and adopted in ways that reduce racial disparity; implemented with community input; and validated regularly by independent researchers.”

But Daniel Nagin, professor of public policy and statistics at Carnegie Mellon’s Heinz College, thinks the racial disparity issue is far more complicated than it’s often made out to be by proponents and opponents of pretrial risk assessment tools.

“There have been criticisms of these instruments for a variety of reasons. One is that in some cases (the tools are) black boxes and lack transparency for how they work and how reliable they are in these predictions,” Nagin said.

“The other set of criticisms, which I think needs more scrutiny, is that because the instruments do use information of a person’s prior record and some … argue that prior arrests are contaminated by racial bias, therefore these instruments are perpetuating that bias.”

Nagin thinks the criminal justice system is “contaminated” by racial bias, but he also thinks pretrial risk assessment tools may be the best current option for sifting out that bias.

“It is devilishly difficult to measure the degree to which data are contaminated by racial bias,” Nagin said. “We really don’t have good answers to that question. Some civil rights groups are very persuaded that criminal justice statistics are contaminated very heavily by racial bias. I would say they are undoubtedly to some degree contaminated, but we don’t know how much. The issue here is, how do these instruments perform to the alternative?”

The alternative, Nagin said, is money bail or a judge deciding on his own whether an individual should be detained pretrial or not. While the former may discriminate against poorer individuals, the latter is subject to human error and potentially racial bias.

With the pretrial risk assessment tools, local governments can choose what kind of data points to evaluate and can prevent racial bias from creeping into the results.

“There is, just more generally, a lot of evidence in many different settings that very simple kinds of statistical tools do a better job at predicting how well human beings will perform than people’s judgments, and that’s been used in a whole variety of contexts,” Nagin said.

According to Jannetta, “There’s always a question here about what risk assessment is supplementing and replacing. What you’re using it to guide and make more uniform is judicial discretion. I think in most places pretrial risk assessment will be more consistent and fair than pure judicial discretion, but that’s a testable proposition.”

Specific tools may show strong biases, create risk to the population, and be costly to taxpayers, but there is the possibility that a better tool can be identified.

The success of such tools really depends on what data points are being evaluated, how the tools are designed, and who is using them and to what ends.

“Like with any tool, a lot depends on what you’re trying to do,” Jannetta said. “There are lot of people who are in jail right now who the risk assessment tool will tell you are very low risk and should not be in detention at all.”

While the New Jersey data show that the number of pretrial detainees has dropped by 38 percent over the last two years (the pretrial risk assessment tool was deployed in 2017) the Pretrial Justice Institute (which advocates for pretrial reform), noted in its 2017 State of Pretrial Justice report that there is still insufficient data on the efficacy of pretrial risk assessment tools because they are so new and haven’t been used for very long.

For example, there’s no data from New Jersey yet detailing whether more whites or blacks have been released pretrial, so there’s no way to quantify whether the tools are exacerbating, eliminating, or leaving any instances of racial disparity unchanged.

“(The civil rights groups’ statement) is an interesting argument, it gets to why are we detaining people pretrial,” Nagin said. “I think the argument they’re trying to advance is one of values, which says in a free democratic society, pretrial detention should be the exception, not the rule, you only use it if you have a really good reason that this person is a flight risk, but it has to be a very compelling argument or a compelling reason for thinking they’re really dangerous roaming free. But that’s a value judgment on which people can disagree.”

Because there still isn’t much data on the efficacy and efficiency of the tools due to still being in the early stages of implementation. There’s potential for the tools to improve safety and justice in some states, but depending on the different crime and cultural climates in communities, the tools could exacerbate existing problems if they aren’t modified to evaluate those conditions and properly weigh the risk of releasing violent individuals.