On Informative Bayesian Multilevel Model Priors for Variance Components (Poster 20)

• In Event: AERA Poster Session Eleven
  In Poster Session: Propensity Score, Bayesian Techniques, and Applications in Multilevel Modeling

Abstract

In multilevel models it is often difficult to estimate parameters well when the number of units at Level 2 (clusters) is small. Compared to a frequentist approach, Bayesian estimation of such models holds great promise precisely for small-sample scenarios. Nevertheless, when estimating models within the Bayesian framework, researchers must make assumptions about the prior distribution of all model parameters, including random effects (aka variance components). The present study extends earlier research by investigating potential informative prior distributions for random effects using Monte Carlo simulation of 2-level data. Results show that inverse gamma and normal distribution priors can both be helpful for obtaining precise Level 2 variance estimates in small- and large-sample scenarios.

Authors

• Liu Liu, University of Georgia
  Presenting Author

• Elizabeth A. Sanders, University of Washington
  Non-Presenting Author