In the race to dominate the child abuse prediction market, the world’s largest data analytics firm has its eye on what it calls “perpetrator” networks.
SAS, with a global workforce of 14,000 and $3.16 billion in revenue in 2015, delivered Florida’s Department of Children and Families a lengthy technical report in August of last year. The report claimed that the firm had developed the strongest child abuse prediction algorithm to date by focusing on the many adults in a child’s life who could be a threat.
By mining these perpetrator networks, SAS says it was able to predict which adults were destined to become what it calls “chronic perpetrators.”
The firm found that 42 percent of the 291,499 adults in its study group who had one report for child maltreatment would be reported again by the end of the eight-to-ten-year follow-up period. Roughly 10 percent fell into what the SAS researchers call the “chronic maltreatment group,” those who had five or more maltreatment reports in the study period.
SAS says this development warrants “a radically different approach to child welfare;” one that flips the focus from a child’s risk of being abused to the adults in a child’s life who present the greatest threat.
“This involves advocating policy and practice changes at the national level, as well as legislative changes at the state and local level, and to change some practices to focus on the adults around the children from re-perpetrating rather than forming policies and practices strictly based on protecting current children from relatively imminent maltreatment,” the report reads.
SAS’ work for Florida’s child welfare agency is one of the most robust child maltreatment focused data-linkage projects ever undertaken. It linked child abuse records with birth and death records, alongside information on public assistance and mothers who had been involved in the state’s home visiting program. The firm’s data scientists looked at every child born in 2004 and 2005, and followed them until 2014, providing as much as 10 years of data.
They also looked at adults in those children’s lives, as ascertained by data gleaned from other data sets, to better understand the threats lurking out of sight to a child abuse investigator who may not look further than the immediate family.
Because it is very hard to identify the same individuals across these various data sets, SAS employed a computer science technique dubbed “entity resolution” to increase the probability that they were identifying the right adults.
Will Jones is the firm’s child wellbeing industry consultant, and is leading its work in Florida. Prior to taking that job, Jones was the chief of programs at Eckerd Youth Alternatives, a national service provider that contracts with the state’s Department of Children and Families to provide foster care. During Jones’ tenure there, Eckerd developed a predictive analytics tool called Rapid Safety Feedback that is employed in Florida’s Hillsborough County and a number of other jurisdictions throughout the country.
He says that what makes SAS’ model competitive over other predictive analytics tools and more commonplace actuarial risk assessments is the overlay of “social network analysis.”
“We are going deep down into the data to identify relationships that impact risk,” Jones said. “Nobody currently has that in production that I am aware of.”
In December, Jones wrote a blog post describing how mining a child’s adult networks could improve child abuse prediction. He used the example of a four-month-old who died of Sudden Infant Death Syndrome.
By going back 10 years and looking at data on the adults in the baby’s family network, SAS was able to find 127 reports “demonstrating historic, generational abuse going back to the child’s grandparents.”
The SAS report mirrored earlier linked-data projects showing that the children of parents who were victims of maltreatment themselves are much more likely to be substantiated victims than other children. In a slide show accompanying the technical report, SAS wrote: “many of the highest risk perpetrators are young mothers with young children, a history of victimization, and a large number of networked reports in the past.”
Given the intergenerational abuse present in the case of the four-month-old, and the sheer number of reports in that baby’s family network, SAS scored the child in its 99.6 risk percentile. If the SAS approach were used, such a high score would warrant heightened scrutiny from child abuse investigators
Jones said that any concerns about whether it was ethical to focus on a child’s family networks were misplaced.
“SAS is mining data and integrating data that is already currently available to the work force,” he said. “This is not data they could not acquire as is. We are not necessarily tapping into data sources they don’t have access to in some shape or form. We are simply making it more automated.”
But Tim Dare, a professor of philosophy at the University of Auckland in New Zealand, who has written a number of papers on the ethics of predictive analytics in child protection, says that Jones’ argument is not entirely satisfactory.
“The tools being developed allow those interrogating the data to spot correlations between data points, and to examine larger sets of data, in ways which were previously impossible,” Dare said in an email. “The tools allow us to know things we could not [have] previously known, even if the data was in some sense available to us. Given that, it’s reasonable to think the tools create new ethical questions, even if they are looking only at preexisting information.”