Testing Applications of Bayesian Statistics With Convenience Priors in Multilevel Analysis of Small Data Sets With Complex Structures: A Monte Carlo Study

• In Event: 29.036 - Marvelous Multilevel Modeling

Abstract

This study evaluates the behavior of a popular classical estimator and an alternative Bayesian estimator for multilevel models in the context of partially nested data structures, with a primary focus upon estimation of variance-components. Its primary purpose is to investigate whether the Bayesian estimator outperforms the classical estimator, given problematic conditions in a multilevel modeling context, such as unbalanced designs and small sample sizes. The study begins with an examination of several proposed methods for setting convenience priors in Bayesian analysis and then compares their behaviors through a Monte Carlo simulation. Findings from this study evidences some advantageous of using Bayesian estimators with mixture priors over traditional approaches to setting convenience priors in certain multilevel modeling contexts.

Authors

• Tyler Hicks, University of South Florida

• Eun Sook Kim, University of South Florida

• George T. MacDonald, University of South Florida

• Jeffrey D. Kromrey, University of South Florida

• Jeanine Romano, American Board of Pathology

• Harold Holmes, University of South Florida

• Sandra Archer, Archer Analytics, LLC.