Big Problems, Big Solutions, Big Data: A Defense of the Use of Predictive Analytics in Child Welfare
by Matthew F. Katz
As of August 1, 2018, 46 states, the District of Columbia, and Puerto Rico were operating or developing Comprehensive Child Welfare Information Systems (CCWIS) [Children's Bureau]. This next generation of information system, adopted by the Department of Health and Human Services in 2016, offers states the opportunity to create integrated data systems that connect human services data across multiple agencies to help inform and track child welfare case management decisions. Despite the widespread adoption of CCWIS, jurisdictions are not using these big data systems to their full capabilities: predictive analytics. Utilizing CCWIS data to drive predictive analytics offers states a cheap and efficient way to reduce bias; inform best-practice decision making; and aid in accomplishing child welfare goals of safety, permanence, and family preservation.
Despite the widespread adoption of CCWIS, child welfare decisions remain highly subjective. Caseworkers are often tasked with making quick decisions based on minimal investigation. This subjectivity has created a biased system highlighted by misrepresentation along racial lines. An overwhelming majority of parents, 81%, admit to the use of corporal punishment [Brookings]. Despite nearly all families participating in corporal punishment, we see over-representation of impoverished and black-identifying children in the child welfare system. Black families comprise almost double the proportion of families involved in child welfare, 24%, as they do in America, 13% [Child Welfare Information Gateway].
Allegheny County, Pennsylvania, presents a prime example of how integrated data systems like CCWIS can be used to improve the biggest challenges in child welfare. In 2016, Allegheny County became the first jurisdiction in the world to begin using predictive analytics to inform child welfare decisions after a string of child deaths related to the county’s Office of Child, Youth, and Families’ (C.Y.F.) missteps [New York Times]. Before unveiling the data system, 48% of low-risk calls to the county’s C.Y.F. were screened in and investigated. Twenty-eight percent of high-risk cases were screened out by a caseworker based on judgment, opinion, bias, belief, and interpretation of legal precedent. Before the initiation of Allegheny County’s predictive analytic system, 44% of the 18 child death cases in Allegheny County had been screened out during their initial report to county offices [Child Welfare Information Gateway]. Nationally, the challenges in the reporting process mimic Allegheny County. Forty-two percent of the allegations received in the United States in 2015 were screened out. In 2015, 1,670 children died as a result of abuse and neglect [New York Times]. Consistently, child welfare agencies are spending resources investigating and providing interventions to the wrong families.
Allegheny County’s data system links information from more than 100 categories in eight databases related to criminal justice, social services, welfare, drug and alcohol treatment, and more. When a call is received by C.Y.F., the databases are scanned and produce an immediate report on the predicted threat posed to a child based on risk factors present in the child's household. The high-risk cases are given further attention, and low-risk calls can be screened out.
Within the first 16 months after implementation of the county's predictive screening tool, the rates of racial over-representation declined, and there was a significant drop in low-risk cases being investigated [New York Times]. This resulted in more time, money, and resources for caseworkers to spend providing interventions and support to high-risk families.
Allegheny County shows the utility of CCWIS, big data, and predictive analytics in making child welfare decisions. However, artificial intelligence can do more for child welfare than just screen calls. It can be used to make decisions related to investigations, substantiation of reports, interventions for children and families, and out-of-home placements. It can also inform best practices that increase child safety, permanence, and family preservation. Additionally, big data can reduce caseload size, limit hours spent on investigations, and allow child welfare staff to spend their hours providing resources and management that improve family functioning. Furthermore, the use of predictive analytics reduces bias against marginalized groups and protects well-functioning, non-abusive homes from being subjected to the intrusiveness of an investigation.
Critics of big data’s infiltration into the child welfare system have identified challenges with using analytics in child welfare [Chronicle of Social Change]. The system has to be “fed” information to allow it to make decisions. If the system is fed biased information, the entire system is now being controlled by biased analytics. However, artificial intelligence adapts as it is used. The system learns as new information is fed into the algorithms, mistakes are identified, and consequently bias is pushed out [CMS Wire].
Additionally, people question the ethics surrounding computers making decisions affecting families and child safety. Why is a computer worse than a liberal arts major with limited training and an over-capacity caseload? Computers didn’t screen out 44% of the child death cases in Allegheny County - people did. We continue to rely on a system that has proven to create ineffective results. Predictive analytics does not need to make all of our child welfare decisions. Instead, it can be used as a tool to help inform, back up, and defend caseworker decisions. States should be striving to find the data that power predictive analytics to make decisions that have a positive impact on child safety and minimize human error.
The child welfare system has been flawed since its conception. A system tasked with keeping children safe is overwhelmed, biased, and allows too many cases to fall through the cracks. Although artificial intelligence is not yet a perfect answer, it offers us a tool to solve systemic problems within child welfare across the country. More jurisdictions should take advantage of federal funding to develop CCWIS and follow Allegheny County into using their data systems to power predictive analytics. CCWIS and big data offer us a tool to create a modern, impactful, and trusted child welfare system.
Matthew F. Katz is a current Master of Social Work student in the University of Pennsylvania's School of Social Policy and Practice. Previously, Matthew attained a Bachelor of Science in Human Development and Family Science from the University of Georgia.