Allegheny Family Screening Tool

The Allegheny Family Screening Tool (AFST) is a risk modelling tool used by Allegheny County, Pennsylvania, that aims to improve child safety by predicting the likelihood that a child referred for abuse or neglect will later experience a foster care placement in the two years after they are investigated.

The ASFT analyses personal data including birth, Medicaid, substance abuse, mental health, jail and probation records for each person reputedly involved in an allegation of child maltreatment and generates a ‘Family Screening Score’ that predicts the long-term likelihood of a child requiring protection.

The ASFT was developed by a team from Auckland University of Technology, University of Auckland, University of Southern California, and the University of California. ASFT Version 1 was released in 2016, followed by a second version in December 2018.

Transparency and accountability 🙈

Allegheny County has been praised for making a significant effort to be transparent and open with the public and other stakeholders about the Allegheny Family Screening Tool, notably concerning its implementation and the release of the list of predictive variables that make up the model.

However, the ASFT is seen to suffer from a number of important transparency and accountability limitations.

• Algorithmic black box. The way the tool decides if a family needs help has not been made fully public, nor has it been made available to judges.

• Data sources. The information the tool uses to make decisions is unclear.

• Third-party assessments and audits. Outside experts are unable to assess how the tool works to make sure it is fair and effective.

• Human overview. It is unclear how the tool's advice influences decisions made by social workers.

• Complaints and appeals. If a family thinks the tool made a mistake about them, there's no straightforward way to challenge it.

• Privacy. It is unclear how the ASFT protects the privacy of the families it is assessing.

Risks and harms 🛑

The Allegheny Family Screening Tool has been criticised for perpetuating historical biases and inequalities due to its reliance on potentially skewed data.

Research, advocacy 🧮

• Stapleton L., Hao-Fei Cheng H-F., Kawakami A., Sivaraman V., Cheng Y., Qing D., Perer A., Holstein K., Steven Wu Z., Zhu H. (2022). Extended Analysis of “How Child Welfare Workers Reduce Racial Disparities in Algorithmic Decisions” (pdf) 

• ACLU (2021). Family Surveillance by Algorithm: The Rapidly Spreading Tools Few Have Heard Of

• Oh S.,S. (2020). An Ethical and Technical Evaluation of the Use of Machine Learning Models in Health and Human Services: A Case Study of the Allegheny Family Screening Tool (pdf)

Investigations, assessments, audits 🧐

• AP (2023). Not magic: Opaque AI tool may flag parents with disabilities

• AP (2022). Oregon dropping AI tool used in child abuse cases

• AP (2022). An algorithm that screens for child neglect raises concerns

• Goldhaber-Fiebert J.D., Prince L. (2019). Impact Evaluation of a Predictive Risk Modeling Tool for Allegheny County’s Child Welfare Office (pdf) 

• Hornby Zeller Associates (2018). Allegheny County Predictive Risk Modeling Tool Implementation Process Evaluation (pdf)