Douglas County Decision Aide: Methodology Document
The Centre for Social Data Analytics has published a methodology report detailing how it developed and implemented a child welfare predictive risk modelling tool commissioned by the Douglas County Department of Human Services (CO, United States).
The Centre for Social Data Analytics has published a methodology report detailing how it developed and implemented a child welfare predictive risk modelling tool commissioned by the Douglas County Department of Human Services (CO, United States).
In 2017, Douglas County asked CSDA to explore the use of predictive risk modelling to help to inform, train and improve the way County staff triaged incoming child maltreatment referrals. (Each time a referral alleging child maltreatment is received, County staff must decide whether it should be investigated.)
The CSDA team used machine learning techniques to develop the Douglas County Decision Aide (DCDA), which was deployed in early 2019. The DCDA is a tool that produces a risk score from 1 to 20 for each referral. This score is intended to help County staff to decide whether the referral should be investigated.
The Douglas County Decision Aide Methodology report explains:
- How Douglas County triaged referrals before and after the DCDA was deployed.
- How CSDA built, validated and deployed the DCDA tool; and
- How ethics concerns around the DCDA tool were identified and addressed.
Professor Chris Wildeman and Associate Professor Maria Fitzpatrick at Cornell University are conducting an independent evaluation of the DCDA, via a randomized controlled trial. This means that half of referrals are decided using the DCDA score and half and decided without the DCDA score. Data collection for the evaluation is due to finish around March 2020 with results available midyear.
Contact:
Phone:
Email: