Bayesian Estimation of a Longitudinal Mediation Model for Three-Level Clustered Data

• In Event: Bayesian Estimation and Propensity Score Analysis in Educational Statistics

Abstract

In this study we use the hierarchical linear modeling framework to parameterize a three-level latent variable regression model for testing longitudinal multiple-mediation hypotheses (both 3->2->2 and 3->3->3) using a fully Bayesian approach. Interpretation of the model’s parameters is demonstrated using ECLS-K data with a dichotomous, cluster-level treatment variable (full- versus part-time kindergarten) modeled to affect growth in an outcome (math achievement) mediated by growth in a mediator (earlier reading achievement) including both cluster- and cross-level mediated effects. A small-scale simulation study assessing parameter recovery is also underway and will be completed by the fall. The final paper will include full details about the model’s parameterization as well as guidelines for its estimation and directions for future research.

Authors

• Anita Israni, The University of Texas - Austin

• Susan Natasha Beretvas, The University of Texas - Austin