Estimating Distributions and Ranks of Cluster-Specific Effect Parameters: New Bayesian Tools for Value-Added Modeling

JoonHo Lee

Education policymakers and researchers often rely on standard random-effects models and empirical Bayes shrinkage to measure the effectiveness of schools or teachers. However, these methods can be misaligned with specific inferential goals‚ such as estimating the shape of effect distributions, identifying outliers, or establishing reliable ranks‚ particularly when effect sizes and their standard errors are correlated, or when latent parameter distributions deviate from normality. This project develops new Bayesian tools to overcome these challenges, motivated by the evaluation of value-added effects for 1,524 Alabama pre-K classrooms. First, it proposes flexible deconvolution methods that capture non-normal, precision-dependent (i.e., correlated effect sizes and standard errors) effect distributions, offering robust tail-area estimates in large-scale, noisy data. Second, it tailors posterior predictions to ranking goals, creating context-sensitive "report card" measures aligned with policymakers' loss functions (e.g., tolerance for misclassification). Third, it extends these techniques to a multivariate framework, enabling joint modeling of multiple academic and nonacademic outcomes‚ thereby improving precision in smaller clusters and unveiling cross-domain relationships. While grounded in the Alabama pre-K context, these methods can be widely applied to any education setting aiming to identify high- or low-performing schools, detect within-site disparities, or explore the interplay of multiple outcomes. By bridging advanced Bayesian theory with practical policy demands, this project aims to equip researchers and decision-makers with more precise, transparent, and goal-oriented tools for value-added modeling.

About JoonHo Lee

JoonHo Lee is an assistant professor of education research at the University of Alabama. His methodological research develops advanced statistical tools‚ particularly Bayesian hierarchical models and modern causal inference techniques‚ to tackle challenges such as treatment effect heterogeneity, value-added assessment, and missing data in education. His applied research centers on understanding why increased funding often fails to narrow achievement gaps or promote greater equity, focusing on school finance reform, teacher and school quality, and early childhood interventions. Lee's work bridges methodological advancement and real-world policy application. Currently, he leads a project funded by the Institute of Education Sciences to refine estimation methods for site-specific effects in multisite trials‚ critical for identifying which contexts or schools benefit most from specific interventions. In parallel, he collaborates with federal and state agencies, including the Administration for Children and Families and the Alabama Department of Early Childhood Education, leveraging large-scale administrative data to develop actionable, evidence-based policy solutions. Lee's scholarship has appeared in journals such as the American Educational Research Journal, Journal of Educational and Behavioral Statistics, and Educational Policy. He received his Ph.D. in Quantitative Methods and Evaluation from the University of California, Berkeley, where he was an AERA-NSF Dissertation Grantee, and holds bachelor's and master's degrees from Seoul National University.