Florida takes aim at juvenile recidivism with predictive analytics
Connecting state and local government leaders
The Florida Department of Juvenile Justice will become the first state agency to conduct risk assessments using predictive analytics.
The Florida Department of Juvenile Justice (DJJ) will become that state's first agency to conduct risk assessments using predictive analytics.
The agency began using predictive analytics tools in 2010, but only for research and evaluation, said Mark Greenwald, the chief of the DJJ’s Office of Research and Data Integrity. With the use of new predictive tools, DJJ plans to move away from its traditional score-based format for predicting recidivism, the Positive Achievement Change Tool (PACT), he said.
The agency collects information from all over the state and from a range of different sources, such as juvenile officers, the courts and police, as well as from PACT. Demographics, arrest records and risk and needs assessment – which gauge the propensity to offend and factors that could lead to repeat offending – are some of the information that has been made available online through analytics software from Tableau since 2013.
Tableau allows the agency to instantly break down data in a variety of different ways – including type of offenses, felonies vs. misdemeanors and race or other demographics.. As a result, the agency can focus on juveniles with a history of trauma (high likelihood to reoffend) versus alcohol issues, for example. The goal of focusing efforts is to keep juveniles out of jail or residential placements.
Now DJJ has brought on Algorhythm, a company that uses data analytics to provide insights for social and public sector decision makers. In a control study using five years of DJJ data, Algorhythm put more than 140,000 cases through its system. The company built predictive models for recidivism and developed prescriptive models that can be used to tailor interventions that will reduce the odds of re-arrest, according to an article by Peter York, Algorhythm founder and CEO. The study showed predictive analytics could improve the accuracy of Florida’s current risk assessment from 61 percent to 82 percent, York wrote.
That makes the predictive model more accurate than the current PACT scoring process in predicting recidivism, according to a report in the Chronicle of Social Change.
“The difference with predictive is it’s substantially more complex,” Greenwald said. “It’s a more advanced way of doing the scoring.”
The program appears to be working. A recent study showed that a 25 percent increase in issuing juveniles with low risks of reoffending a civil citation instead of an arrest would save taxpayers as much as $61 million in Florida while keeping kids from handicapping their futures with a record.
NEXT STORY: Making sense of machine learning