How predictive analytics keeps corrections staff, inmates safe

Corrections departments are using predictive analytics to curb a growing problem of violence against correctional officers.

Within the Indiana Department of Correction, “more and more of our staff were being severely injured,” said Sarah Schelle, the department’s executive director of data science and analytics. Adult facilities were averaging 320 violent assaults against correctional officers per month. Although IDOC uses the Indiana Risk Assessment System (IRAS) to predict the likelihood of recidivism for offenders, “we didn’t have a way to predict violence,” Schelle said.

In South Carolina, the Department of Corrections experienced about 127 assaults per month between January 2019 and July 2020. “They were tracked just on a spreadsheet,” said SCDC Director Bryan Stirling. “It was very difficult to get this information. They were definitely not predictive. It was definitely reactive.”

For help, the agencies turned to SAS predictive analytics and machine learning solutions to aggregate data and help predict offender violence.

At IDOC, that meant collecting data from various sources, including IRAS and the offender information management system. Corrections staff helped refine the variables that would inform a predictive analytics model -- more than 100 risk variables were tested as modeling inputs.

IDOC uses a decision tree to calculate risk scores of how likely someone is to become violent and that can inform how staff members interact with either an individual or the population at large, which numbers about 24,000 across 18 adult facilities.

“Once we score the population -- and we rescore the population once a week -- we deliver that information via a dashboard, and it is up to the facility as far as how far down they share that information,” Schelle said.

IDOC uses SAS Visual Analytics for the dashboards, which show staff a weekly hotlist of which inmates were recently recategorized as high risk. The data staff to determine their next steps, such as intervening with mental health or assigning an extra guard.

“The dashboard is one-stop visual analytics,” Schelle said. “It includes a color-coding system that allows individuals to see the categorization of offenders so far as if they are very high, high, moderate or low for that particular week. It allows them to triage their concerns, and it includes variables such as why they believe they scored high this week.”

The dashboard provides both a high-level view across facilities and a more granular look at specific buildings.

Some locations use the analytics more than others, Schelle said. “Those facilities are much more liberal with who gets it,” she added. “I don’t think that’s a bad thing because that then communicates to the line-level staff that particular individuals need to be treated with care. Again, that’s part of our training…. We don’t use this for things like takedown. We don’t use this to punish anyone. We use this merely as a way of informing ourselves on how to interact with these individuals this week.”

SCDC sees benefits

SCDC also pulls information from about 100 factors such as inmates’ educational level when they enter, the number of disciplinary actions taken against them in the past month and any suicide attempts. It looks at about 15,300 inmates across 21 facilities.

“I just compare those that actually committed assault to those that did not commit assault and come up with odds ratios that way,” said Cheryl Bolchoz, a biostatistician and owner of Alpha Datalytics who is working with SCDC on the effort. “Then we use those odds ratios to create a prediction on a weekly basis.”

Because the data changes every week, so does the model, she added. She creates a weekly spreadsheet that goes to department administrators.

“I’ve also been able to incorporate a link such that on the spreadsheet, when the warden gets this information, he can actually [click] a link and it will show the inmate’s face, it will show all kinds of information about their past and what they’ve been incarcerated for,” Bolchoz said. “It’s a lot of information to be able to pull in one single spreadsheet.”

The wardens use that information to determine what might be bothering inmates on the hotlist, Stirling said. For instance, if the issue is with a cellmate, the warden can make reassignments, which de-escalates the tension between inmates.

Using this data and taking those actions, SCDC was able to predict about 44% of assaults week after week, and the average number of monthly assaults fell to 102 between August 2020 and July 2021 -- about a 20% decrease.

“In prison, violence begets violence, so if we have 20% less violence just this year – I’m looking for this to go down again next year – that helps us make the prison safer and helps us save tax dollars, which is vitally important,” Stirling said.

For IDOC, staff assaults fell by 50% during a six-month test period and inmate-on-inmate assaults decreased by 20%.

How it works

SAS coordinates with customers based on their IT environment – whether it’s on-premises or in the cloud, what tools they have and what data they’re bringing to the application, said Mary Beth Carroll, director of Justice Involved Populations at SAS. The company helps with data resolution and cleansing to ensure it’s in a usable format despite the types of systems correction departments use and pull from.

The firm also helps protect personally identifiable and other confidential information by applying the appropriate protocols based on the data going into the system, Carroll said.

“It also gives the customer the ability to set up within their environment their level of security for access,” she said. “Not every facility needs to see what’s going on in every facility. Not every position within the correctional environment needs to have access to certain information.”

In Indiana, Schelle said that IDOC won’t use this system in any other way, but it is investigating doing modeling around suicide prevention.

”We use models that are meant to predict certain outcomes for only those purposes,” she said.

Stirling said SCDC is also interested in using predictive analytics for suicide prevention and to match offenders based on their classifications, while Bolchoz added that she’s looking to move into SAS’ new Viya analytics platform to take advantage of the dashboards. She also plans to use analytics for staff retention -- to better understand what factors predict that employees will stay longer.

“Correctional organizations are data-rich, but just like with anything else, the data gets siloed,” Carroll said.  With day-to-day decisions that could mean the difference between life and death, “having the ability to act proactively, rather than reactively, changes the game in their operation,” she said.  “It also allows them to start looking in the future and using data to anticipate what their strategic planning may be for the future or even assist how they should change policy or procedure,” she added.