By Holden Slattery
The idea of predictive analytics, where big data is crunched to better assess the likelihood of something happening, is gaining traction in the field of child protection with hot spots in Los Angeles, across the country in Pennsylvania and as far afield as New Zealand.
The U.S. jurisdiction closest to employing the potentially powerful and controversial tool is western Pennsylvania’s Allegheny County. There, child protection officials say they are as close as a year away from implementing a predictive data model to assist child welfare workers in deciding which children face the greatest risk of harm.
But in Allegheny County, like in other jurisdictions considering employing the tool, the idea of ascribing risk to certain families brings concerns that such information will be used against them and cause undue and unfair intervention of child protection workers.
Allegheny County announced in February that it has teamed up with an international research team led by a professor from the Auckland University of Technology in New Zealand to create the risk-modeling tool.
In New Zealand, the Parliament is even closer to integrating big data and risk models into its child protection practices. New Zealand is creating two information systems that will allow the public and professionals to report their concerns about vulnerable children in a database accessible by “child teams” composed of members of health, education, welfare and social services agencies.
The effort has been championed by Paula Bennett, a member of the New Zealand Parliament. A speech Bennett gave to a group of Statistical Analysis System (SAS) users in February is posted on the New Zealand National Party’s website.
“What I hope risk predictor modeling will do is transform the data government already holds and has access to, to make the picture clearer and the path of intervention more certain,” Bennett said in the speech. “This way, we will be able to go exactly where we need to be—the right family, the right child, at the right time—and we will have a better understanding of exactly what they need.”
Allegheny County’s international research team also includes University of Southern California professor Emily Putnam-Hornstein, who has led extensive research on the at-birth risk factors linked to higher rates of child maltreatment in California.
Los Angeles County committed itself to an overhaul of its child welfare system in 2013, forming a Blue Ribbon Commission on Child Protection. One of the commission’s 40-plus recommendations for the L.A. County Department of Children and Family Services (DCFS) was to adopt a predictive analytics model created by Eckerd, a private family services organization in Florida. DCFS also created its own risk prediction tool, Approach to Understanding Risk Assessment, or AURA and contracted for a pilot program with SAS, a North-Carolina-based analytics software company.
None of these tools have been implemented in LA County. L.A. County DCFS Director Phillip Browning said during a visit to USC in March that his department has not made a decision on the merits offered by any of these analytics tools. He hopes to find or develop a risk assessment tool that can assist a social worker, but said no tool can “make a decision for a social worker.”
“The three more important things I tell staff everyday: common sense, critical thinking and accountability,“ Browning said. “Common sense is pretty rare these days, so if all our workers used common sense and critical thinking, we’d be a much better organization.”
(Philip Bowning and Leslie Gilbert-Lurie, former co-chair of the county’s blue ribbon transition team, joined the Media for Policy Change class at USC’s Price School of Public Policy in March. This video includes some discussion of applying predictive analytics in L.A. County.)
A nationally used risk-assessment tool is already built into every social worker’s activities in LA County, Browning said, but is a manual process. Social workers enter certain information about a family or situation and the system categorizes them as high-risk or low-risk. The problem with a manual tool is the ability to manipulate it, he said.
An analytics tool could increase information sharing between county departments that serve families. Communication between different departments can make it clear where resources should be targeted. However, confidentiality requirements can block this communication, Browning explained.
In Allegheny County, child protective services are part of the Department of Human Services, which also includes services for mental health and substance abuse as well as homeless services.
Having an integrated department and prioritizing data collection makes birth data, school data and court records accessible, said Erin Dalton, deputy director of the Allegheny County Department of Human Services.
“We want to make use of the data we have,” Dalton said. “We have good, innovative data. It’s been 20 years or so in the making.”
Discussions about how Allegheny County’s risk model would work are ongoing, and the department is leaning toward a tool that social workers would only use once they received a call reporting child maltreatment, Dalton explained. Social workers would then access data going back to the child’s birth.
“We’d use all available data, including when dad was in jail three years before that call,” Dalton said.
While creating this tool, the department will conduct an ethical review focused on disproportionality “caused by disparity, some in society at large, and some in public child welfare,” according to Allegheny County’s February news release. The release defined disproportionality as “the over or under-representation of certain groups [e.g., racial/ethnic, gender, age] in a public child welfare agency relative to the group’s proportion in the general population.”
The idea of social workers using predictive data to decide which homes to visit and how to conduct those visits creates fears that certain populations will be targeted based on their histories and environment.
Kathryn Icenhower, CEO of Shield for Families, a nonprofit organization that provides comprehensive services for families involved in the child welfare system, said she fears predictive data will be used in a punitive way. Using analytics, a social worker can find out that a parent was arrested as a juvenile, later abused drugs, and now lives in a high-risk neighborhood. Each of those factors results in the workers checking another box and increasing a parent’s risk score, Icenhower explained.
“What a parent has done to address those things doesn’t get counted,” Icenhower said. “It doesn’t take into account families’ strengths. It only considers their challenges.”
If L.A. County DCFS adopts such a model, Icenhower fears that it will be an impetus to start removing more children from their homes again.
From 2000 to 2010, L.A. County reduced its number of foster children from 50,000 to 32,000, according to numbers cited in the Los Angeles Times. The county now has 19,899 foster children, according to Kidsdata.org.
“The data is amazing if used correctly,” Icenhower said. “If I know that these are the risk factors I can utilize services to address those risk factors.”
In Allegheny County, Dalton agrees that the data has limitations and biases. People who have private insurance and live far from their neighbors are less likely to develop a public history of child abuse than people who have public insurance and live very close to their neighbors, she explained.
“We don’t see this as binding,” Dalton said. “It gives workers a tool. People could still investigate a very low-risk case and people could still screen out a high-risk case for whatever reason.”
Dalton said she hopes the risk models will be finished by this summer. An ethical review on disproportionality will help the department decide how to adjust and implement the models, Dalton said.
“We wouldn’t hardcode bias into decision making,” Dalton said. “Let’s not pretend there isn’t already bias in the [social] workers, but we wouldn’t hardcode it. That’s why we’re taking this part of the work very seriously.”