Orange County Turns to Big Data in Effort to Remake Child Support Services

Getting parents to pay child support is tough work. Part counseling, part debt collecting, it often means dealing with struggling kids and struggling adults and people who want to succeed but fail and then fail some more. It’s an enormous responsibility. The U.S. Census Bureau in 2008 estimated that child support programs serve a quarter of all the children in the country and half of all the children born to poor parents—some 18 million kids then, and millions more now.

Orange County, California, is honing a program that brings advanced analytics developed over a decade to bear on the work. The county is mining data to assign parents predictive “iScores” that estimate longer-term earning capacity and likelihood to make payments.

Steve Eldred, director of the county’s Department of Child Support Services, says effective operations hinge on seeking efficiencies and that’s what the new program is all about.

“People think about human resources and images of offices and desks flash through their head,” he told Route Fifty. “But it’s more like that UN truck you see on the news parked before fields of hungry people. There’s never enough resources to do the work. So this new program is about allocating resources in a scientific, analytical way to do the best you can with what you got.”

Eldred’s child support office handles 69,000 active cases and opens 12,000 new ones each year. It’s a ton of work. And that work generates a ton of data. The Orange County office figured it could use the data to better find patterns that would make case-worker predictions about who could pay child support and how much they should be paying more reliable. What’s more, the predictions, as well as the plans of action drawn up by case workers that the predictions informed, could be made increasingly more reliable for being evidence based.

In other words, Big Data could take the office beyond experienced guess work; it could help staffers get better at managing cases by more accurately gaming out not weeks or months but years or a decade into a parent’s life.

An added benefit of the approach, in theory, is that punitive measures—court orders, garnished wages, jail time—would be trained only on true “deadbeat” parents, not the vast majority of “deadbroke” parents who miss payments.

“We wanted to find a way to get beyond the typical focus on wringing out a payment this month then a payment next month,” said Eldred. “Instead of saying ‘Drop out of school and get a low-paying job,’ we can say, even at the risk of sacrificing short-term payments, ‘In your case, staying in school is a much better bet, or getting rehabilitation, or expunging a criminal conviction.’”   

How does Orange County determine a parent’s iScore?

Eldred’s staff built the project in-house with the help of SAS Enterprise Miner software to work at first with 400 possible variables and winnow them down to find the “most likely to predict scenario” for each parent.

“What we found was interesting,” Eldred wrote in an email. “Some factors confirmed our industry best practices. [But] some factors were eliminated, destroying decades of widely accepted folk knowledge. The most indicative factors included frequency of payments (any payment effort), monthly income, driver’s license status, amount of support ordered compared to income, criminal history, stability of housing, and education level. Some of the factors are unique to the individual; some factors are derived from census, crime pattern records, municipal records and other non-individual-linked information.”

One of the benefits of the program is that, done right, it will make rookie caseworkers nearly as effective as veteran caseworkers. Prescriptions would be spelled out and limited on a case by case basis. It would be much more difficult to get things terribly wrong.

Which all sounds great, except that any storyline that features personal data points scrolling across screens in a government office where computers are assigning something like future competence scores to parents has an admittedly dystopian sci-fi ring to it.

Eldred says iScores are “as secret as a person’s social security or medical information, and are intended to be shared or discussed only with that person.”

He acknowledges there is no secure data in the era of great hacks, security breaches, firewall foul-ups and government snooping, foreign and domestic, but he adds that the iScore is based on personal-finance credit scores, so they’re not static. You can correct them, if they’re somehow inaccurate, and they improve fast when you make payments, achieve housing stability, finish school, save money and so on, depending on how factors are weighted on a person’s particular score.

The program has been in use in Orange County since January of this year. Sixty case managers have been putting it to work and assessing its effectiveness. The office held weekly meetings for the first few months after it went live, Eldred said, and two of the in-house research team that put the program together walk the floor to problem solve. Eldred now holds bi-weekly meetings on the program to track caseworker experience and “customer acceptance of the model.”

“It’s an approach that I think takes a while to assess,” he said. “Take the compliance factor, for example. We don’t know right away who is following our advice. We won’t know right away whether someone has gone back to get a high school degree or the equivalent…

“We’re anxious to share a completed, fully vetted product with the rest of the child support community,” he said. “But for now we’re holding onto it pretty tight.”

Eldred expects he’ll be ready to take iScore on the road to interested child services departments next summer.