Baltimore uses a little-known risk assessment tool to help make bail decisions. It’s supposed to be an objective way to keep non-violent defendants out of jail, but some fear it might be reinforcing racial bias.
Clarence Barnham is a soft-spoken 53-year-old man with a wheezy voice and bright, twinkling eyes. He’s struggled with drug addiction and mental illness for over a decade, and has a long arrest record because of it. Nearly all of his fourteen convictions have been for drug charges; he’s never been accused of a violent crime and only missed a court date once.
It’s not the kind of a criminal history that would seem to indicate that Barnham, who’s lived in West Baltimore his whole life, would be a public safety or flight risk. But after an August arrest for distribution and possession of heroin (charges he contests), he sat in jail for three months, unable to pay his $25,000 bail. In October, having found an in-house drug rehabilitation program through his public defender, Barnham was granted another bail review. At the rehearing, both Barnham's public defender and the prosecutor agreed he should be released without bail, citing the drug treatment program, his steady employment and family support network. But the pretrial service agent insisted on $5,000 bail—reversing his own agency's position in the first hearing and even interjecting after the prosecutor spoke to insist on bail being given.
Why would pretrial services—which is supposed to be a neutral party that provides background information on defendants to judges—be tougher on a defendant than the prosecutor? Part of the answer may lie in the secret pretrial risk score the agency generated for Barnham.
Baltimore’s Pretrial Release Services, like many agencies nationwide, uses a risk assessment tool to give defendants proceeding through the court system scores based upon statistical likelihoods of failure to appear or rearrest. These scores are supposed to help pretrial service agents recommend bail decisions to judges based on objective, standardized criteria. But no one else involved in the case, including the defendant and their attorney, gets to see or even hear about their score, much less the impact it has on their bail recommendation.
The use of risk assessment tools has rapidly spread to many different parts of the criminal justice system, from bail to parole, in jurisdictions across the country. Less than 10 percent of U.S. jurisdictions use pretrial risk assessments today, but support for their introduction is growing in cities and states nationwide, financed by foundations interested in finding ways to release more “low-risk” defendants—and cutting down on the billions of dollars spent to hold them.
But in Baltimore, even some defense attorneys are unaware of the use of this tool in the bail recommendation process. Little is known about the factors selected to calculate risk—and the potentially disparate impacts these calculations may have on bail recommendations for defendants from highly policed communities. Transparency on how these risk scores are being interpreted—and how judges are weighing the recommendations based on them—is especially important, as calls for bail reform through the expansion of these tools are growing in Maryland.
Interpreting the risk assessment algorithm
Baltimore’s Pretrial Release Services refused multiple interview requests from CityLab to discuss risk assessments. But a source outside the agency provided CityLab with a copy of one such risk assessment, shedding light on how Baltimore’s risk assessment tool is designed. Gerard Shields, a public information officer for the State of Maryland, confirmed the validity of this copy by releasing a blank risk score template with the same categories in response to a public records request.
In the form above, the defendant was arrested on a nonviolent drug charge, Controlled Dangerous Substance With Intent to Distribute, like Barnham. Unlike Barnham, the defendant has no failures to appear to court. But other factors combined to give this individual a total score of ten points—numerous past arrests, early age of first arrest, current drug charge, and current alleged drug use. That adds up to a final recommended “high” level of risk.
Going by these measures, Barnham would probably have racked up at least 14 points. He might have been moved down to “moderate” because of “mitigating” factors such as his stable employment, age, and family support. But such a decision, unlike the drug charge mark-ups, would be left up to the discretion of a pretrial service agent. This discretion may have played a role in the $5,000 bail recommendation Barnham received from pretrial services in his second hearing.
Several of the criteria the tool weighs—such as age of first arrest and number of arrests—mean that those like Barnham who live in in high-poverty and non-white areas are more likely to be considered high risk by pretrial services because of greater, and often discriminatory, contact with law enforcement. Such assessments could accurately reflect historical data showing that residents of highly-policed black neighborhoods in Baltimore are more likely to be rearrested pretrial or miss court dates. The key question then becomes: How is this information used?
Advocates of risk assessments say that, in addition to weeding out low-risk defendants, these tools can be used to delineate what social supports, such as drug rehabilitation and court date reminders, certain high-risk defendants need—and identify the fraction of them that must be detained for public safety. Critics contend that the intended use of the tool is beside the point, if pretrial service agents do not have resources or requirements to support defendants, and are instead using risk classifications to promote harsh money bail decisions against defendants from poor, highly-policed neighborhoods.
Todd Oppenheim, Clarence Barnham’s public defender, says in his experience pretrial agents mostly interpret high-risk scores to mean people are dangerous and need to be given high bail amounts to keep them in jail. “Pretrial never finds these services like we sometimes do for our clients,” says Oppenheim. “Sometimes, when pretrial is late, I’ll make a sarcastic comment to the judge and say, ‘We know they are going to ask for a bail increase, so can we just go ahead and make our arguments already?’”
Zina Makar, co-director of the Pretrial Justice Clinic at the University of Baltimore School of Law, also argues that the courts’ continuing practice of setting inaccessible bail amounts for poor defendants suggests the tool is not an effective means of pushing for the release of defendants based on their flight or public safety risk. “There’s no rational relationship between one’s ability to make bail and one’s threat to public safety,” says Makar. “If there is continued reliance on money bail without regard to the defendant’s ability to pay, any benefits a risk assessment tool may have cannot be optimized while operating in an arbitrary system.”
If such experiences are not anomalous, pretrial services’ recommendations, based on these risk scores, could be further intensifying racial disparities already present in the Baltimore bail system.
Black residents in Baltimore, for example, would seem to be far more likely to receive high risk scores under this risk assessment algorithm, which takes into account number of arrests (not convictions), earliest age of arrest, and drug arrests. According to city arrest data over January 2013 to December 2016, black residents accounted for 81.5 percent of all arrests (107,319 of 131,543 arrests), despite making up 63.7 percent of the population according to Census estimates. Of black arrests in this period, 36.2 percent were of individuals twenty-five years or younger. Of white arrests, only 21.3 percent were of individuals twenty-five years or younger. According to city drug arrest data from January 2013 to November 2016, black residents accounted for 84.6 percent of all controlled dangerous substance and narcotics arrests.
Geographic data from Baltimore drug arrests shows how drug policing is concentrated in black neighborhoods. Over January 2013 to October 2016, 80.1 percent of all drug arrests took place in majority black census tracts, as can be seen in the time lapse map below. Scroll over census tracts to see specific racial breakdowns:
How the tool was envisioned
It is unclear how long Baltimore’s Pretrial Release Services has used risk assessments, but the most recent version was formulated in 2010, according to James Austin, president of the Washington, D.C.-based JFA Institute, which works with governments to evaluate criminal justice practices. Austin was a co-creator of Baltimore’s current pretrial risk assessment. He says that the tool, in theory, is supposed to help, not punish, vulnerable residents, by identifying who will have more difficulty getting back to court and staying out of jail.
”In pretrial just about everyone is a good risk to be released because the chances of a defendant’s failure to appear or be arrested is in only in the 10 to 20 percent range,” says Austin, who argues such tools improve the accuracy of courts’ subjective decision-making processes by using historical data to identify those more likely to fail to appear or be rearrested. “I can’t speak for judges in Baltimore, but it should be used to release low-risk people on their own recognizance and create conditions of supervision for higher risk people.”
But the record suggests these risk score-informed bail recommendations do not appear to be encouraging judges in Baltimore to release significant proportions of defendants either on their own recognizance or with social supports and supervision, as Austin intended. A 2012 study from the Justice Policy Institute, a D.C. think tank that focuses on reducing incarceration rates, found that 57 percent of Baltimore City’s jail population were held without having not been offered any bail on one or more charges. And a new study from the Maryland Office of Public Defender concludes that bail decisions disproportionately hurt black defendants financially: Over 2011 to 2015, two of Baltimore’s poorest, majority black zip codes alone (Park Heights and Sandtown-Winchester) accounted for 20 percent of the $113 million paid out in premiums to bail bondsmen across Baltimore city. (Eboni Pearson, an analyst for the Baltimore City Circuit Court, did not respond to CityLab’s requests for more updated data on the percentage of defendants given bail, no bail, or released without any bail.)
The apparent failure of Baltimore’s pretrial risk assessments to transform the city’s bail practices cannot be pinned on pretrial services alone, given the role of judges and prosecutors in bail hearings. But public defenders contend pretrial services is often hampered by institutional and ideological factors that get in the way of its ability to use risk assessments to push for the release and support of more defendants. “Their ultimate fear is danger of the worst-case scenario, and they have very little guidance beyond the thing that says ‘high risk’ score,” says Oppenheim. “But also these agents are boxed in a system where they don’t have options. They need social workers, drug treatment experts, and mental health experts in their offices.”
Makar often finds services for her clients on top of her legal work. “In a perfect world, it should be pretrial connecting people to these pretrial release services,” she says.
“I wouldn’t have talked to them, if I’d known what you just said.”
Baltimore’s risk assessment categories raise other questions about defendants’ rights. For example, how does a pretrial service agent determine and mark up an individual for active use of crack, cocaine, or heroin? Unless the individual’s arrest report documented evidence of drug use around the time of arrest, the information must be extracted in an interview in jail without a lawyer present.
Baltimore’s Pretrial Services declined to discuss its risk assessment practices, but defendants recently out of jail described how they were questioned by agents. Jeffrey Thomas, who is currently facing allegations of assault, claims that pretrial service agents gave him leading questions about his drug use and did not inform him such information could be used for a risk assessment. “They asked me, ‘What drugs do you like?’ so I told them what my drug of choice was,” says Thomas, who says he is currently battling a heroin addiction. He says he thought he was being asked questions to provide him with rehabilitation opportunities. “They didn’t inform me of that score. Anytime anybody asks you anything that affects your case, you should be able to speak to an attorney first.”
In his interview with pretrial services, Clarence Barnham also reports that pretrial services did not tell him that his answers about his residency and drug use could be fed into a risk assessment algorithm, a tool he had never heard of until our interview. “They asked me how long had I been using,” Barnham says. “I’d told them I was snorting heroin off and on for about twenty years. That was the last question… I wouldn’t have talked to them, if I’d known what you just said.”
Austin suggests such lines of questioning are important for ensuring accurate risk score data, assuming individuals have been told they are being screened for risk. “Typically, if you want to get a good answer on it, you would ask, ‘What drugs are you using at the time of arrest?’,” says Austin. “If you ask people, ‘Are you using drugs?,’ some people will deny that, even though they are using drugs. The objective is try to get them to be as truthful as they can because people who say yes to that have higher rates of Failure to Appear and getting arrested again.”
Defendants can always deny the right to answer questions, Austin adds, but both Barnham and Thomas say they were not informed of that right—and public defenders point out such a denial would not look good in court. “If someone has a client that doesn’t want to talk, pretrial services person announces it at court,” says Natalie Finegar, Baltimore’s Deputy District Public Defender. “It’s seen as a big red flag.”
It is difficult to discern whether these alleged interview practices violated departmental protocol, but they get at the heart of the controversy over pretrial risk tools, which largely boils down to how risk assessment information is selected and put to use. ”We don’t have these risk assessment calculations because pretrial has refused to give it to this,” says Finegar. “How can we explain our client if we can’t put them in the context of what’s forming the basis of the recommendation decision specifically?”
Makar echoes this point. “It is not discussed openly,” she says. “If tools are to be used, we need to make sure the process is transparent, so defense counsel knows why the tool produced the results it did, in what context it pertains to the crime at hand, and how much weight the court gives to it.”
As long as the relationship between risk scores, bail recommendations, and bail decisions remains opaque, it’s difficult to definitively say whether the implementation of Baltimore’s risk assessment tool is helping defendants get out or further justifying the jailing of many more. What little evidence is available on pretrial services’ institutional capacities and courts’ bail setting records suggest the non-incarceration route is often passed by.
Which route is taken can have major consequences. Thanks to a judge’s October decision to ignore the pretrial service agency’s recommendation and release him without bail, Barnham is now out of jail and in a drug rehabilitation program at a Johns Hopkins Hospital facility. He’s also getting treatment for mental illness issues he had been dealing with unknowingly for years.
“I used stay up four, five days at a time without sleeping, worrying. You hear voices, think people are talking about you,” says Barnham, who has been in the program for over a month now. “Twelve years of using, using, using. So God landed me here.”