In partnership with Los Angeles County Department of Children and Family Services (DCFS), the Children’s Data Network (CDN) will develop and pilot a predictive risk model and several management applications designed to support quality investigations. This project involves numerous partners, including the Centre for Social Data Analytics (CSDA) at Auckland University of Technology, the California Child Welfare Indicators Project (CCWIP) at UC Berkeley, the Carnegie Mellon School of Computer Science, and evaluation and implementation support through Mathematica.

Overall, the goal is to develop a model and associated applications that will help DCFS:

• Better align and deploy supervision and management resources to ensure children are safe and families receive requisite services during maltreatment investigations;
• Increase the use of information and data by supervisors to facilitate best practice and high-quality casework during investigations; and
• Improve the use of data to identify screening practices and community reporting patterns that may result in unnecessary investigations disproportionately burdening low-income and black families.

The Risk Stratification Model was developed and validated using Los Angeles County child welfare data. The model uses hundreds of data elements to identify differences in child maltreatment referrals that have been screened-in for investigation, but may otherwise have very different degrees of complexity and service needs. The motivation to pursue this work emerged from analyses documenting that there are a relatively small number of complex investigations and cases that are at a significantly elevated risk of future system involvement and tragic outcomes. These tend to be investigations where there are very complex assessment and service needs that require more time, attention, and resources – but can be missed due to human error, inconsistent staffing, and high caseloads.

Three applications will be developed for the pilot that draw upon data from the model. These applications include: (1) a flag provided to supervisors overseeing emergency response investigations, alerting them to a new investigation that the model has classified as “complex-risk”; (2) an investigation overview report that summarizes information that can be time consuming to assemble through the existing case management system and could be more efficiently compiled using the model; and (3) a racial-equity report and feedback loop to review and better understand low-complexity referrals that were screened in for investigation and involved Black children.

These applications emerge from feedback and convenings with numerous state and local community stakeholders dating back several years. Based on this collective feedback, several decisions were made at the outset of the project including that the model will not be used to make any decision related to whether a child should be brought into care or have a case opened for services. Likewise, it was also decided that there needed to be applications that would use the model to examine system responses that have (and continue to be) unequal due to a combination of both structural racism and individual biases, with the hope that insights gleaned could lead to changes in policies and practices that may have a disparate impact on families of color.

The Risk Stratification Model will be initially piloted in the Belvedere, Lancaster, and Santa Fe Springs Regional Offices in August 2021. During the pilot, the LA DCFS Office of Equity will work with the continuous quality improvement unit to review the service needs of low-complexity investigations, helping devise strategies for community diversion rather than investigation. Mathematica will work to gather information about how the applications are being used to support quality casework and supervision so that improvements can be made. The learning that occurs during the initial pilot phase will be shared with community stakeholders and used to inform next steps.