About. The Allegheny Family Screening Tool (AFST) uses data analytics to improve child welfare services in the County. The AFST, implemented by the Allegheny County Department of Human Services in 2016, uses predictive modelling based on eight data sources to help improve child welfare call screening decisions. Allegheny County was the first globally to use a predictive algorithm to support child welfare decision-making.
Process and the Data. To develop the predictive risk model, the County first developed a dataset based on historical referrals and placements with protective services between 2010 and 2015. This data was then merged with datasets from the Allegheny County Data Warehouse (e.g., mental health services, the criminal justice system, drug and alcohol services, public benefits, census neighborhood poverty indicators and, more recently, birth records). These datasets produce more than 100 predictive factors on each child on the referral. Machine learning and predictive algorithms are then used to generate a risk score on every incoming call, in the hopes of better identifying the families and at-risk youth most in need of intervention. Call screeners use the score to help inform their decisions about investigating the allegation.
The design and implementation of the tool was a multi-year process that included careful procurement of data, community meetings, a validation study, independent and rigorous methodology and impact evaluations, and an ethical review. Monitoring the tool on real production data has been essential for revealing where continued changes/improvements need to be made.
The tool is owned by the County.
Background. Before, when an allegation of child maltreatment was received at the call centre, staff would have to manually assess a variety of different data points to decide whether or not to investigate and take action, without any structured guidance as to how these elements are connected with relevant outcomes or how they should be weighted in decision-making. An analysis found 27% of highest risk cases were screened out and 48% of lowest risk cases screened in.
Role of Philanthropy. The County was able to create this tool partly because it already had the integrated Data Warehouse, created in 1999 with the support of the Human Service Integration Fund, a collaborative funding pool of local foundations under the administrative direction of The Pittsburgh Foundation.
Development of the AFST was made possible through a public-private funding partnership that included support from the Richard King Mellon Foundation, Casey Family Programs and the Human Service Integration Fund.
Impact and scale. The system, after several rounds of improvements, has achieved 77% accuracy in determining whether a child is at risk of placement in child services or not, based on a set of previous cases from 2010-2015. This has increased the efficacy and efficiency of the call screening process: many children who would previously have been screened out even though they had a high risk of being placed in foster care within two years were now screened in, and vice versa. The tool is helping to better direct energy and resources to help more children.
This success has attracted the attention of other child welfare departments in Colorado and California. California is in the process of adapting the tool to the entire state, which has 10 million children. That tool is initially achieving 85% accuracy.
This case study is part of a larger project on the role of philanthropy in using data to solve complex problems. A global scan highlights many initiatives using data for good, the main methods, how philanthropy is engaging, and the challenges faced.
Want to read more great content like this? Sign up to our newsletter for regular updates.
For questions or comments, contact Jordan.Junge@socialinnovationexchange.org