How an algorithm will help keep families together
As the director of the Northampton County Department of Human Services, I would like to respond to a recent essay published in The Morning Call: “Algorithms don’t counter child welfare bias. They magnify it” by Richard Wexler.
“Find where the poor people are,” he wrote. “Send money.”
It would be nice if we lived in the world Mr. Wexler described, but we do not. We live in this world, and Northampton County DHS will do everything in our power and use all available resources, both human and data, to keep our children and families safe, while reducing bias.
One of the primary goals of the Decision Aid Tool is to do exactly what Mr. Wexler would like — keep families together and reduce the number of children in out-of-home placement.
The DAT will both decrease trauma for families who do not require investigations by the child welfare system and save the taxpayers money by keeping children in their own homes with the earlier delivery of preventive services. This tool is simply one part of the decision-making process; human caseworkers and supervisors will always make the final determination on when to investigate or screen out a referral.
While those impacted by poverty are disproportionately involved in the child welfare system, there are a couple of explanations for this.
First, poverty increases stress and therefore child welfare issues. Second, people with pre-existing challenges such as addiction or mental health issues, and who have challenges keeping children safe, are also more likely to be living in poverty.
While poverty is acknowledged as a common factor in referrals to the child welfare system, the DAT will not introduce any additional focus on poverty in the screening or investigation stage.
Child welfare history is the major data source currently considered by call screeners. Once the DAT is implemented, referrals that do not have sufficient child welfare history for a score to be generated will be considered using the existing processes already in place in Northampton County.
It is, therefore, incorrect to suggest that the tool will only “target” families who are more likely to be poor. An ethics report commissioned in New Zealand argued that DATs can bring more attention to the unmet needs in communities, allowing officials to direct limited resources in a more equitable manner.
Northampton County already funds emergency hotel stays and rental assistance, as well as food and other essential items, in an effort to keep families together and children out of placement due to poverty. Families are also connected with extended kin and community resources and supports.
When children are placed in out-ofhome care, it is the direct result of unmitigated safety concerns, never the result of a risk score or poverty.
Bias already exists in the nation’s child welfare system. The question is: What can be done about it?
Contrary to Mr. Wexler’s unsubstantiated claims, there is no evidence the Decision Aid Tool increases racial bias. In fact, an independent evaluation demonstrated it decreased racial bias. In addition, the algorithm will not utilize race as a metric in the data, as it has not shown to improve the accuracy of the tool.
Concerns about bias will be addressed by two specific steps:
1) The model will be tested before, during and after introduction to ensure the tool performs equally well for all subgroups including ethnicity and race.
2) The accuracy of the tool will be validated before it is introduced, against outcomes that are external to the child welfare system, but related to child maltreatment, such as hospitalizations or egregious, near-fatal or fatal incidents.
In addition, there was evidence from Allegheny County’s model that demonstrated a decline in the rates at which Black children were being screened in for investigation.
Mr. Wexler wrote that a “scientific” algorithm is sending investigators to families’ homes, but this is simply false. The DAT is intended to support child welfare workers to make decisions using their own professional judgment, not to dictate screening decisions.
This means investigators have no reason to proceed with an investigation on the basis of “the tool said yes” — since in other jurisdictions where the tool is in use there are frequent situations where referrals with low risk scores are screened in, and cases where higher risk scores are screened out.
Investigators will be aware of how the tool is used and will understand that a case has reached them as a result of human judgment plus the risk score, but in what combination they cannot be sure.
Northampton County will co-design policy and process decisions for use of the tool with front-line staff before the DAT is introduced, taking account of input from the Centre for Social Data Analytics, the community and stakeholders. We are eliciting feedback at every step of this process to ensure the tool performs as planned and all concerns are addressed prior to implementation.
Northampton County is proud to use all available means at our disposal to keep children safe and families together. And we believe ongoing testing and analysis of the DAT will demonstrate an improved ability to do both.