Abstract: A Latent Profile Analysis of Foster Youth Well-Being: Translating Complex Phenomena to Inform Child Welfare Programs and Services (Society for Prevention Research 27th Annual Meeting)

345 A Latent Profile Analysis of Foster Youth Well-Being: Translating Complex Phenomena to Inform Child Welfare Programs and Services

Schedule:
Thursday, May 30, 2019
Pacific N/O (Hyatt Regency San Francisco)
* noted as presenting author
Jeffrey Waid, Ph.D., Assistant Professor, University of Minnesota-Twin Cities, St. Paul, MN
Background

Child welfare systems collect vast amounts of data, yet utilization is typically restricted to descriptive analysis and periodic reports to state and federal stakeholders. This under-utilization limits the ability of practitioners and administrators to make data informed decisions about the programs and services they provide to at-risk children and families.

Purpose

The purpose of this study was to (1) pilot a brief assessment tool designed to measure the well-being of youth in foster care, and (2) characterize child well-being using a latent profile analysis (LPA) approach. The goal of this research is to support the development of new approaches to assessment, analysis and interpretation of data which can direct policy and inform the delivery of services to children and families.

Methods

Legal guardians for 359 youth in four U.S. states completed a brief electronic survey which asked respondents to assess multiple dimensions of child well-being. Constructs included a five-item measure of difficult behavior (alpha=.88) and three, six-item measures which assessed youths’ social relationships (alpha=.87), sibling relationships (alpha=.88), and emotional well-being (alpha=.91). Confirmatory factor analysis supported construct items loading on their single common factors.

Analysis

Latent well-being profiles were estimated in R using the tidyLPA package. Models were fit by calculating Loglikelihood, Bayesian Information Criteria, and Entropy statistics across a series of models where variances and covariances were equal or varied. Models with lower BIC and higher entropy indicated a preferred solution. After determining a preferred fitting model, constructs were standardized and profiles were plotted with grouped bar charts to support visual interpretation of the data.

Results

A 4-profile model with equal variances and covariances fixed to zero best fit the data (LL=1317.98, BIC=2771.28, Entropy=.90). Well-being profiles included youth who were “thriving” (high ratings across all constructs; n=50), “surviving” (near-zero rating across all constructs; n=196), “concerning” (moderately low ratings across all constructs), and “alarming” (significant behavior problems and moderately low ratings across other constructs; n=29).

Conclusion

Child welfare systems are in need assessment tools and analytic techniques that provide substantive and interpretable feedback which can inform the delivery of services to children and families. This study provides preliminary support for the use of web-based data capturing and LPA techniques to establish easily interpretable, multidimensional graphical representations of foster youth well-being. Future research should examine the potential of automated information systems to support real-time feedback for latent modeling approaches (e.g., LCA, multilevel analysis, latent transition analysis, covariate modeling) to better explain the complex phenomena which define child welfare practice.