JUNEAU — Alaska's recently passed justice reform bill asks state prison officials to do something new: use a computer model to predict whether an arrested suspect is likely to be rearrested if released before trial.
The corrections department says it is studying the legislation to determine a response. The bill hasn't been signed by Gov. Bill Walker, and unless he does, the department won't proceed.
But with expensive pretrial incarceration rates on the rise in Alaska, the bill attempts to help judges determine who can be let out of jail without endangering the public. The legislation also looks for a way to bypass traditional bail — its cash bond to assure appearance at trial has been insurmountable for some low-income defendants.
Predicting future behavior, though, has always been tricky. Back in the late 1800s, phrenology suggested that brain size and shape could forecast criminal behavior, but it was later debunked as a pseudo-science. More recently, criminologists have suggested DNA, pain sensitivity and mental deficiency were factors.
Now computers are being called into service. Authorities in several states have used a software program from a Traverse City, Michigan, company called Northpointe, which says its assessments are objective and scientifically validated.
But a report this week by investigative journalism site ProPublica found that two years of Northpointe's projections incorrectly predicted future criminal behavior by black defendants at nearly twice the rate as for white people. In some cases examined in the two-year study by the website, African-Americans judged by the software to be seriously risky were in fact not arrested again, while white defendants said to have only mild risks went on to commit new felonies.
Alaska's criminal justice reform bill, sponsored by Sen. John Coghill, R-North Pole, and approved by the Legislature earlier this month, would require Walker's administration to start using a risk assessment tool.
Northpointe's is one of dozens used across the country, which ProPublica said are rarely subjected to independent validation.
In an interview, Coghill pointed to the language of his legislation, which requires the implementation of a risk assessment tool that's "objective, standardized, and developed based on analysis of empirical data and risk factors."
ProPublica's analysis, he added, addresses "a single company and a single product." Alaska's system, he said, will be subject to tight scrutiny and review.
"There's going to be a lot of eyes watching anything new," Coghill said.
And, he pointed out, the bill is aimed at reducing the number of people in Alaska's prison system, where Alaska Natives are over-represented.
The state's risk assessment software will be aimed at combating a sharp rise in the state's pretrial population — the number of people in jail who are awaiting trial, which rose from 900 in 2005 to 1,700 in 2014, when they comprised about one-third of the state's total prison or jail population.
One way the reform legislation, Senate Bill 91, would reduce that population is by changing how Alaska's justice system sets bail and other conditions of pretrial release.
Under the current system, the majority of defendants are ordered to pay some amount of bail — which means that pretrial release is often linked to a defendant's ability to pay instead of his or her likelihood to commit a new crime or miss a hearing, according to a report by the state's Criminal Justice Commission, which made many of the recommendations included in SB 91.
The reform bill would change the state's pretrial release law to make bail conditions more reliant on the results of an assessment designed to show a defendant's risk of rearrest and failure to appear in court.
Under the new pretrial law, defendants charged with nonviolent misdemeanors, with some exceptions, cannot be forced to post cash bail unless they're deemed by the assessment as high-risk.
According to the legislation, Dean Williams, Walker's corrections commissioner, will have to approve the risk assessment tool.
Northpointe, whose risk assessment software was the subject of the ProPublica analysis, calls its products scientifically validated.
But the analysis showed that just 20 percent of the people in a Florida county who the company's software predicted would commit violent crimes in the future actually did so. And it also showed racial bias: Even when ProPublica accounted for defendants' criminal history, age and gender, blacks were 77 percent more likely to be labeled as being at higher risk of committing a violent crime in the future.
Northpointe's risk assessments come from more than 100 questions that don't include race. They do include factors like a defendant's education, family members' criminal history, and the number of friends or acquaintances who are illegally using drugs.
Northpointe is not the only creator of risk assessment tools; there are others created by for-profit and nonprofit corporations, and by states themselves.
The system used in Kentucky, Arizona, and New Jersey, created by the Laura and John Arnold Foundation, excludes what what the foundation calls potentially discriminatory factors like race, gender, education and socioeconomic status.
Asked how the corrections department would choose its tool, a spokesman for Williams, Corey Allen-Young, said that the department has been studying other states' reform efforts.
"We are certain that we will end up with an assessment tool that will be specific to Alaska's unique population and meet the needs of our offenders," Allen-Young wrote in an email. "There will be several agencies involved in structuring the pretrial assessment tool, and racial disparities will not be tolerated in our system."
Asked how the assessment tool could be built with a guarantee of no racial bias, Allen-Young responded simply that the corrections department has "been given time to build the right tool," and said the state will "make sure it's accurate and fair to reduce pre-trial incarceration for Alaska's unique and diverse population."
The Pew Charitable Trusts, which supported the Alaska Criminal Justice Commission as it developed its reform recommendations, referred questions about risk assessment and the ProPublica analysis to an expert, Alex Holsinger, a criminal justice professor at University of Missouri-Kansas City.
He said in a phone interview Tuesday that the ProPublica analysis creates "reason for concern," and is a good reminder for researchers to test and correct for racial bias in their risk assessment tools. But, he said, it's still clear that using those tools to guide pretrial release conditions is preferable to "gut-level instinct."
"My preferred method would be the judge and the decision-maker using a tool as at least part of their decision-making rubric," he said.