Using Predictive Analytics to Inform Policy and Practice: An Introduction to Predictive Modeling and Recent Applications Predicting Superutilization of Child Welfare, Health and Other Services

The use of administrative data in social work research is not new, but there is a growing interest in leveraging its potential to inform policy and practice. Researchers and practitioners are particularly interested in knowing the best way to use available administrative data to inform decision-making to better serve program participants, target services more efficiently, and ultimately improve outcomes. This workshop provides an introduction to predictive modeling using administrative data and presents applications from a recent study which identified factors that can predict child and families most at-risk of using high levels (“superutilization”) of child welfare and health services.

This workshop will start with an overview of predictive modeling—including a description of different methods and approaches—along with key thing to consider when applying predictive methods, such as data quality; statistical power; and use of a training data set, a hold-out sample, and different metrics to assess predictive performance.

Workshop presenters will share insights and examples of applications of predictive analytics from a recent collaborative study conducted by Mathematica Policy Research and Casey Family Programs. This study applied predictive modeling techniques using administrative data to identify and predict superutilization of child welfare, health, and other services among children and families in the child welfare system. The study was designed to answer several research questions for program administrators who wanted to understand which children and families experience superutilization of child welfare and other services, what types of services they receive, and what factors can help predict superutilization. To identify predictors of superutilization, researchers linked and analyzed five years of administrative data from child welfare, Medicaid, and other services from two sites. To identify types of superutilization based on patterns and combinations of service use, we first used descriptive and latent class analysis to assess service use by children and parents. Next, we applied predictive modeling techniques to identify key factors that predict the potential for superutilization of services. Implications for the use of predictive results were identified with input from child welfare and Medicaid agency partners on how results might inform their policy and practice priorities.

The presenters will also lead a discussion of the potential benefits and challenges of using predictive modeling to inform decision-making for service provision. The workshop will include time for questions and discussion so audience members can share their experiences, discuss potential concerns of using predictive modeling, and consider possible applications in the field of social work.