Activity Number:

260
 SPEED: Topics in Bayesian Analysis

Type:

Contributed

Date/Time:

Monday, July 30, 2018 : 3:05 PM to 3:50 PM

Sponsor:

Section on Bayesian Statistical Science

Abstract #332665


Title:

A Bayesian Model Selection Approach to Multiple Comparisons

Author(s):

Javier E. Flores* and Andrew Neath and Joseph Cavanaugh

Companies:

University of Iowa and SIU Edwardsville and University of Iowa

Keywords:

Multiple Comparisons;
Model Selection;
BIC

Abstract:

Consider the setting in which independent samples from several populations are taken for the purpose of betweengroup comparisons. Specifically, interest lies in the determination of clusters where mean levels are equal. To this end, multiple comparisons testing procedures are often employed. We formulate the hypothesis testing problem of determining equalmean clusters as a Bayesian model selection problem. A comprehensive collection of models is formulated by specifying mean structures that represent all possible combinations of equal mean levels. The Bayesian information criterion is then used to approximate posterior model probabilities and posterior probabilities associated with pairwise mean equalities. Information from all competing models is combined through Bayesian model averaging in an effort to provide a more realistic accounting of uncertainty. An example illustrates how the Bayesian approach leads to a logically sound presentation of multiple comparison results.
