Florida Hopes Predictive Analytics Can Improve JJ System

Print

Big businesses have come to rely on a budding technology called predictive analytics to help them guess the likely future actions of their customer base. What customers are most likely to upgrade their service? Which insurance claims are the most likely to be used by fraud perpetrators?

One of the largest producers of predictive analysis software (PASW) is SPSS, a Chicago-based company that was bought last summer by IBM for $1.2 billion.

And why should you care about any of this?

Last month, SPSS issued a press release introducing one of its most recent clients, the Florida Department of Juvenile Justice (DJJ). Generally, such a press release is business’s way of putting like entities on notice that something is happening, and there’s a very good chance that many a juvenile justice agency was made aware about the news of the relationship.

People in Florida certainly took notice, because DJJ”s Chief of Research and Planning Mark Greenwald fielded plenty of angst-laden phone calls asking if the agency was using the SPSS software to replace Florida’s respected and long-used risk assessment instrument, Positive Achievement Change Tool Assessment (PACT), which is used to assess youths at intake and monitor their progress while involved with the department.

It’s easy to see why he got those calls. SPSS’s press release says that DJJ would use predictive analytics:

“To reduce recidivism by determining which juveniles are likely to reoffend. Identified at-risk youth can then be placed in programs specific to the best course of treatment to ensure offenders do not re-enter the juvenile justice system.”

Makes it sound like each youth’s case would be plugged into a predictor machine at intake, Minority Report style, no? But after a discussion with Greenwald, the company announcement appears to be pretty misleading.

“Some people are under the impression that we were getting rid of our screening tool; that is not the case,” Greenwald said. “I humbly think we have the best screening tool going.”

The real deal on Florida’s use of PASW seems to be a fairly low-cost, long-term investment in quality assurance, so agencies probably should take notice.

“It’s an interesting and cool endeavor,” said Greenwald.

Here’s the story:

Greenwald’s predecessor, Bob Dale, was the one who first expressed interest in predictive analytics back in 2008. And actually, the department agreed to buy the SPSS software in 2009, but only recently did Greenwald’s staff finish training by SPSS on its use.      

The SPSS brand of predictive analytics cost Florida $12,000, which Greenwald said buys the use of the software indefinitely. An annual maintenance fee buys access to SPSS product support and updates to the ­­software (Florida isn’t currently paying for the maintenance because of budget issues).

Greenwald’s research and planning staff will take data from the PACT risk assessments, such as current and past offenses and diagnosed mental health needs, and enter that into the SPSS software. Other pieces of data, not collected during risk assessment, will also be included, including information about what specific treatments and services a youth received while in DJJ custody and how frequently he received them.

DJJ has two central objectives. First, and Greenwald said his staff have already started some experimental attempts at this, is to predict future trends in juvenile delinquency.

Predicting any changes in the nature of juvenile delinquency, Greenwald said, can be help DJJ and the governor’s office determine use of resources, which is, of course, a major issue in fiscally-challenged Florida. So if the inputs predict that there will be more female offenders in the coming year, for example, the state theoretically could put more resources towards programs that serve girls. That sounds a tad more fluid than the real-life reaction of systems to trends.

The other objective, which is more evaluative than it is predictive, is to use the software to determine whether the department is making the right decisions for certain youths.

“If we find that kids with certain criteria were sent to one type of program, and the same types of kids were sent to a different program and did better,” Greenwald said, the SPSS software allows them to identify that and start sending more youth to the program that yields a better outcome.

It could drill down farther than a placement or program into derivatives. A hypothetical: how predictive of non-recidivism is multi-systemic therapy (regardless of which program conducts it) for juveniles who are found delinquent of violent offenses?

The department could start using the software to forecast delinquency trends soon, but using it to do macro-analysis of placement decisions is a few years off, according to Greenwald. After a few years of data have been input, his department will start conducting evaluations on a “semi-annual or annual” basis.

“There is nothing wrong with our screening process,” he said. “We’re looking at the aggregate level to see if we can tweak some decisions.

At $12,000 plus the man hours needed to do analysis, “you don’t have to redirect many kids before you see a return on investments.”