What Can We Learn About the Incidence of Foster Care Placement from Birth Records? Findings from the Cross Jurisdiction Model Replication Project

Publication Date: October 19, 2022
replication brief cjmr project cover

Download Brief

Download PDF (1,492.25 KB)
  • File Size: 1,492.25 KB
  • Pages: 14
  • Published: 2022

Introduction

Research Questions

  1. Can information recorded at birth consistently predict the risk of foster care placement?
  2. Can a model developed (trained) using data from one jurisdiction be used by other jurisdictions to estimate the share of children who might have a heightened likelihood of foster care placement?
  3. What opportunities exist for using birth record models to help inform the ongoing and accurate national surveillance of foster care placement? That is, can these models be used to understand differences in foster care placement across counties and states?

This brief describes the Cross Jurisdictional Model Replication (CJMR) project. The CJMR project sought to understand the degree to which a risk prediction model built from population-level and anonymized birth records in one state could be used to differentiate the risk of foster care placement in other jurisdictions. Specifically, this project estimated population-level differences in the risk of being placed in foster care by applying a single risk prediction model (from California) to anonymized birth records from Alaska and Kentucky. The risk prediction model was designed to explore whether birth record data could help jurisdictions understand geographic variation in risk, improve planning for services, and focus limited resources on the most at-risk communities.

Purpose

This brief describes the development, validation, and cross jurisdiction replication of a risk prediction model designed to predict foster care placement.

Key Findings and Highlights

  • Findings highlight the potential to leverage birth records to understand geographic differences in foster care placement risk. The risk prediction model created from California data generalized well to two other jurisdictions: Kentucky and Alaska. The model generalized to these other jurisdictions despite differences in governance, child welfare agency structure, and definitions and policies from the jurisdiction used to develop the model.
  • Findings suggest that birth records can be used to model and differentiate risk of foster care placement. The risk prediction model created from 2012-2015 California birth data was able to accurately predict foster care placement within the first three years of life.
  • Using this risk prediction model, other jurisdictions could use only birth records (even if that data is not linked to anything else) to predict risk of foster care placement. This risk prediction model may be useful for jurisdictions that do not have the analytic capacity or data accessibility to develop their own risk prediction models.
  • Implementation of the risk prediction model may inform efforts to plan and focus community-based resources and help identify populations experiencing disparities.

Methods

The risk prediction model used birth records to explore whether a uniform set of characteristics could consistently differentiate risk of foster care placement. The model was not designed to identify individual children who would experience maltreatment. Instead, it aimed to use prediction methodologies to understand population-based risk differences in foster care placement.

The CJMR project took place in three stages: development, validation, and replication. A model was developed using recent birth record data from California. This model was then validated using two datasets unrelated to the project sample: (1) records for children born in California in the year 2000 and (2) records for children born in Allegheny County, Pennsylvania from 2012-2015. To examine the generalizability of the model and the ease with which the statistical program could be used to code a common set of birth record predictor variables, state agency staff from Alaska and Kentucky carried out cross-jurisdictional replication.

Appendix

Appendix

File TypeFile NameFile Size
PDFreplication brief cjmr project appendix sept 2022649.46 KB

Citation

Emily Putnam-Hornstein, Eunhye Ahn, Rhema Vaithianathan, Jared Parrish, Robyn Husa, Matthew Walton, Claire Smither Wulsin, M. C. Bradley, and Beth Varley (2022). What can we learn about the incidence of foster care placement from birth records? Findings from the Cross Jurisdiction Model Replication Project. OPRE Report #2022-255, Washington, DC: Office of Planning, Research and Evaluation, Administration for Children and Families, U.S. Department of Health and Human Services.

Glossary

CJMR:
Cross Jurisdiction Model Replication