A case study of algorithm-assisted decision making in child maltreatment hotline screening decisions
; Proceedings of the 1st Conference on Fairness, Accountability and Transparency, PMLR 81:134-148, 2018.
Every year there are more than 3.6 million referrals made to child protection agencies across the US. The practice of screening calls is left to each jurisdiction to follow local practices and policies, potentially leading to large variation in the way in which referrals are treated across the country. Whilst increasing access to linked administrative data is available, it is difficult for welfare workers to make systematic use of historical information about all the children and adults on a single referral call. Risk prediction models that use routinely collected administrative data can help call workers to better identify cases that are likely to result in adverse outcomes. However, the use of predictive analytics in the area of child welfare is contentious. There is a possibility that some communities—such as those in poverty or from particular racial and ethnic groups—will be disadvantaged by the reliance on government administrative data. On the other hand, these analytics tools can augment or replace human judgments, which themselves are biased and imperfect. In this paper we describe our work on developing, validating, fairness auditing, and deploying a risk prediction model in Allegheny County, Pennsylvania, USA. We discuss the results of our analysis to-date, and also highlight key problems and data bias issues that present challenges for model evaluation and deployment.