CSDA child welfare data tool piloted in LA County, California
Child welfare staff in three regional offices in Los Angeles (LA) County, California, are trialling a data analytics model, built with the help of the Centre for Social Data Analytics. Designed to enhance the quality of supervision during child maltreatment investigations, the model uses machine learning techniques to extract insights from existing child welfare data, which are then utilised by staff and leadership via three applications.
The model was developed as part of a risk stratification project led by the Children’s Data Network through funding from philanthropic partners. The project aims to help emergency response social workers and their supervisors align workforce resources during open investigations.
Specifically, the goal was to help supervisors better prioritise the small but critical share of child maltreatment investigations where children and their families have complex histories that place them at heightened risk of adverse outcomes, while also laying a foundation for strategies that could help divert more low-complexity investigations to community service pathways. A third critical component was to use the model to identify potential unwarranted variation in the treatment and delivery of services to Black families.
The CSDA team used child welfare data from LA to build the Risk Stratification Model, which is a predictive risk model (PRM) trained to assess a child’s risk of long-arc adverse outcomes. Once the PRM was built and validated, the CSDA and Children’s Data Network teams worked closely with the Los Angeles County Department of Children and Family Services (DCFS) to deploy the PRM within the DCFS system.
The model generates scores daily and makes them available for various applications in the pilot offices, including:
- complex-risk protocols
- investigation overview report
- racial equity feedback loop
The complex-risk protocol isolates the most complex referrals in the investigation caseload and prompts supervising social workers with suggested actions to aid the investigation, like consulting specialists within their office, including the Assistant Regional Administrator, and engaging community partners in early service discussions.
Supervisors are also provided information from an investigation overview report, which provides a data visualisation of related child welfare history as a reference point for social workers and supervisors as they progress an investigation. This report was co-designed with staff in the three offices.
Finally, the racial equity feedback loop has been set-up as a quality assurance strategy rather than to manage individual investigations. With work led by the Office of Equity, it will help DCFS identify and address hotline screening practices and community reporting patterns that may result in racial or socioeconomic bias.
The Risk Stratification Model applications were implemented as a pilot in three DCFS regional offices - Belvedere, Lancaster, and Santa Fe Springs – in early August. After the 90-day trial DCFS plans to incorporate feedback gathered from supervisors and staff during the piloting process to improve applications that draw upon information from the model and refine supervision practices that might be scaled in other offices.
The CSDA team that delivered the model includes Dr Diana Benavides-Prado (data science), Megh Mayur and Athena Ning (deployment), Emily Kulick (business process) and Larissa Lorimer (project management) with support from Rhema Vaithianathan. The Children’s Data Network is led by Emily Putnam-Hornstein and Jacquelyn McCroskey.