The ANOVA models are common in analyzing data from a wide range of
areas such as biology, psychology, and sociology. Two versions exist
in literature, namely the fixed effect model and the random effect
model. The main difference of these two types of models lies in whether
or not the effects of the factor levels are treated as random variables.
In practice, however, there are many cases where it is not clear whether
or not the factor effects should be treated as fixed or random. Moreover,
from the Bayesian point of view, all parameters are considered as random
variables, making the distinction between fixed effect model and random
effect model rather obscure. The primary goal of this talk is to seek for
a unified Bayesian approach to deal with one-way ANOVA models with fixed
effects and random effects. We propose a modification of the Zellner-Siow
prior, and show that the proposed prior will result in good frequentist
properties in terms of model selection in the settings of either fixed
or random effect.
Meet the speaker in Room 212 Cockins Hall at 4:30
p.m. Refreshments will be served.