Contact Name

Rose Runyon

Contact Phone

5738826376

Contact Email

roserunyon@missouri.edu

Speaker(s) Information

Prof. Ed Merkle, Psychological Sciences, University of Missouri

Additional Event Information

ABSTRACT:

Typical Bayesian methods for models with random effects involve directly sampling the random effects alongside the model parameters. This sampling leads researchers to specify model likelihoods that are conditional on the random effects, which in turn leads to model comparisons via conditional likelihoods (i.e., likelihoods where the random effects are considered model parameters). In non-Bayesian settings, typical model comparisons involve marginal likelihoods where the random effects are integrated out. This distinction is often overlooked despite the fact that it can influence the comparisons of interest. In the talk, we clarify and illustrate these issues, focusing on conditional and marginal Deviance Information Criteria (DICs) and Watanabe‐Akaike Information Criteria (WAICs) in multivariate normal and multivariate probit models. The conditional/marginal distinction corresponds to whether the model should be predictive for the clusters that are in the data or for new clusters (where "clusters" typically correspond to higher‐level units like people or schools). These results lead to recommendations on the general application of the criteria to models with random effects.