A Bayesian Multilevel Modeling Workflow for Robust Quantitative Inference in Education Research

• In Event: AERA Roundtable Session Thursday 4:15 pm - Gold 4
  In Roundtable Session: RT6: Tutorials, Guides and Workflows in Quantitative Education Research: Putting our Quantitative Advances into Practical Use

Abstract

This proposal introduces a rigorous Bayesian multilevel modeling workflow tailored to the methodological constraints of education research. Addressing small samples, hierarchical data, non-random missingness, measurement error, non-linearity, and outliers, the workflow integrates Directed Acyclic Graphs, prior predictive checks, simulation-based design, model fitting, and diagnostics. Demonstrated with simulated and public datasets, the approach improves inference and prediction over NHST and OLS by incorporating model uncertainty, regularizing priors, and model stacking. The workflow enhances transparency, reproducibility, and interpretability, providing a practical framework for researchers and graduate training. By enabling more robust quantitative analysis under real-world constraints, this Bayesian workflow aligns methodological practice with the complexity of educational systems and improves the credibility of findings for policy and practice.

Authors

• M. Shane Tutwiler, University of Rhode Island

• Qingyu Yang, University of Rhode Island

• Yiping Zhang, University of Rhode Island