In the first eight weeks of spring, reports of child abuse in New York City fell 51 percent from the same time last year. With children out of school and confined to their homes—and not even getting regular medical checkups—few adults were on hand to see signs of abuse or neglect. As New York began opening up again over the summer, reports increased, though they remain down about 20 percent from previous years. Each new school year usually brings an uptick in reports, but with so many kids learning remotely (or not at all), hundreds, if not thousands, of children are likely in danger who have not been reported.
Child welfare has an information problem: we don’t know what goes on behind closed doors. And without violating people’s rights, it’s hard to find out. New York City’s leaders should consider adopting some new tools to combat this problem. Data technology in use in other sectors, from baseball to health care, can improve our ability to find out which children are in danger and also to locate stable, loving homes for kids who need them.
For more than four years, Pittsburgh has used big data to help screeners at its child-abuse hotline figure out which of the thousands of calls they receive need to be urgently investigated. In an interview a few years ago, the late Richard Gelles, former dean of the School of Social Policy and Practice at the University of Pennsylvania, told me: “Even the state-of-the-art assessment tools being used in New York are no better at predicting risk for a child than if you flipped a coin.” Gelles said that relying on social workers’ “clinical judgment” and “expertise” to determine which children should be removed from their homes is “simply inadequate.”
Fortunately, we now have other options. The Allegheny Family Screening Tool, which includes more than 100 factors, was rolled out in August 2016. The algorithm uses data points sourced from criminal records, educational data, and public welfare information. The system then gives the call screener a score between one and 20 that describes the likelihood that the child in question will be re-referred or removed from the home within two years. The screeners then assign an investigator to look immediately into the cases with the highest risk scores.
To avoid confirmation bias, investigators are not told what score a family received when they go to visit the home. The model predicted with 76 percent accuracy whether a child would be placed in foster care within two years, and with 73 percent accuracy whether the child would be re-referred—that is, whether child services would be alerted about that child again.
In April 2019, Allegheny County Office of Children, Youth and Families released the first independent-impact evaluation of the screening tool. The results were enlightening. First, the authors noted that “Overall, the AFST did not lead to increases in the rate of referred children screened-in for investigation. Use of the tool appears to have resulted in a different pool of children screened-in for investigation.” In other words, the data suggest that our traditional methods of evaluation are incomplete. Further, it turns out that more of the kids who were being investigated with the new tool were the ones who really needed help.
Though predictive risk modeling helps authorities identify children most in need of attention, many localities have resisted implementing it. Critics have warned that the tool will lead to greater racial disparities in child welfare. They suggest that data analytics is leading us toward a Minority Report-style dystopia in which officials will try to predict who will hurt their kids before they even do it.
In fact, as is often the case when numbers are used instead of gut feelings, the tool corrects for some of the biases that result when people go with their gut. Researchers found, for example, that “AFST led to reductions in disparities of case opening rates between black and white children. Prior to the introduction of the AFST, case-opening rates for black children were higher than for white children. During the Post-AFST period, increases in the rate of white children determined to be in need of further child welfare intervention, coupled with slight declines in the rate at which black children were screened-in for investigation, led to reductions in racial disparities.”
The impact of the pandemic on child welfare goes beyond what happens at the front end of the system. Even when we have a better idea which children are in danger, we still need better options for what happens if they need to be removed from their homes. A report from the United Hospital Fund and Boston Consulting Group found that about 4,200 children in New York State lost a parent during the pandemic; 57 percent of those deaths were concentrated in the Bronx, Queens, and Brooklyn. Almost a quarter of those cases may involve children losing their sole guardian.
During the pandemic, foster parents became even harder to come by. Many worried about their own safety or the risks in coordinating visits between biological parents and the kids living in their homes. In some places, training for foster parents ground to a halt. Finding suitable foster parents is a challenge in the best of times; research shows that between 30 percent and 50 percent of foster parents quit within the first year.
The use of data can help us do a better job finding foster families. Faithbridge, a Georgia organization that recruits, trains, and supports foster parents, has been able to use data provided by the state to learn the demographic profiles of families most likely to be successful at fostering. By partnering with such an organization, New York could do more targeted recruitment of long-term foster parents, which would result in fewer placements for foster kids.
Finally, New York should consider a partnership with an organization called Adoption Share to use its Family Match program. Instead of simply depending on families browsing profiles of children online, which can result in superficial choices, Family Match uses personality characteristics of all the parties involved to predict which pairings will work; one of the program’s developers also worked on E-Harmony’s algorithm. Among the advantages of Family Match is that it can gather information from kids and families across city and county lines, yielding a greater number of matches than having each county fend for itself.
Many problems in child welfare will be with us for the long haul. Family Court is slow and makes poor decisions. People who go into child protective services are often underqualified and undertrained for their jobs. The services we offer to families to help them when they are at risk have long waiting lists or are ineffective. We give biological families too many chances to regain custody of their children, stretching out the time that their kids spend in foster care.
But some challenges can be reduced with the aid of technology. If city officials made a smart presentation, they should be able to get private philanthropy to help fund such initiatives, which would hold bipartisan appeal. Helping figure out which kids are in danger, finding them longer-lasting foster homes, and ensuring that adoptions out of foster care are more compatible are goals that everyone should be able to get behind. If New York still needs more incentive to improve its child welfare system, it might simply remind itself: Pittsburgh is beating us.