February 16, 2018
On February 9, I learned that Allegheny County Department of Human Services had responded with concern to the excerpt from Automating Inequality that was published in Wired magazine on January 15, 2018, calling key aspects of the piece inaccurate. Below is a response to their unsigned statement, released by the office of Allegheny County’s Chief Executive, Rich Fitzgerald, and available on the County’s website.
***
I deeply appreciated, and have in fact widely praised, the willingness of the Allegheny County Department of Human Services (ACDHS) to speak with me and provide key information during my year of reporting on the Allegheny Family Screening Tool (AFST). As I make clear in my book, I believe Marc Cherna, Erin Dalton, and the other members of the agency with whom I spoke are intelligent, committed individuals with the best of intentions to protect Allegheny County’s children.
Their response to Automating Inequality, and the excerpt that appears in Wired, however, mischaracterizes my work and raises crucial additional questions about the AFST.
The writer argues that I unfairly describe the AFST as targeting poor families. This is false. The agency readily admits that the AFST oversamples households that rely on public assistance programs. In an interview with Erin Dalton, Director of Allegheny County’s Office of Data Analysis, Research and Evaluation, she offered: “We definitely oversample the poor. All of the data systems we have are biased. We still think this data can be helpful in protecting kids.”
My larger point is not that we must wait to entirely eradicate poverty before we try to improve Children, Youth and Family Services, but that the AFST only has access to data collected on families using public services, not on those that access private resources for parenting support. Because this will result in higher risk scores and more scrutiny of poor and working-class families, I believe the system is unjust and discriminatory. Further, the County’s January 31 statement suggests that for 55% of families – the majority – receipt of public services does in fact raise their AFST score, leaving them disproportionately vulnerable to child welfare investigation.
ACDHS has been rightly praised for releasing the list of predictive variables that make up the AFST. However, they have only been partially transparent about the model. ACDHS and its contractors, Drs. Rhema Vaithianathan and Emily Putnam-Hornstein, have refused to release the weights of the variables despite my direct request for this information. This makes it difficult for an independent observer to ascertain how each variable in the model may impact a family’s AFST score. That is, without the weights, it is impossible to determine if each variable is seen as a positive/supportive factor or a negative/risk-producing factor. I ask that, in the name of full transparency, the ACDHS release the variable weights of the AFST model to the public.
The writer is also incorrect in his or her claim that my statement “if a family’s AFST score is high enough, the system automatically triggers an investigation” is false. A system has been automatically triggered if it will move forward without being overridden by a supervisor.
But more importantly, the County acknowledges in its statement that 30% of the cases where children have been classified at highest risk by the AFST are thrown out as baseless. This suggests an unacceptably high rate of error at the most consequential decision point in the system – where an AFST score can trigger an automatic investigation. This is especially troubling in light of Dan Hurley’s assertion in the New York Times Magazine that the AFST is 90% accurate and his statement that “[C]all screeners and their supervisors will now be given less discretion to override the tool’s recommendations.” I urge ACDHS to release more complete, accessible information about the AFST’s error rate and any agency plans to restrict call screener and supervisor discretion over the model.
Allegheny County residents deserve to know how this tool is affecting their children and their futures.
For families,
Dr. Virginia Eubanks
Author, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor
Virginia Eubanks 
I enjoyed the story on NPR about your new book. I worked for 38 years in Unemployment Insurance and watched profiling and efforts to automate adjudication of claims help lower the recipiency rate of UI benefits to extreme lows. That and the drastic reductions to eligibility by legislatures accross the country following the Great Recession will make the UI system less able to replace lost wages in the next recession. I’m curious if your studies have looked at UI eligibility and the use of profiling in the national system operated by the individual states.