Duke University
presents:
Julia Varshavsky
Eli Lilly and Company
"Marginal Distributions in Nonnormal Linear Model"
Abstract: In this talk I will present the result providing a simple closed form expression for the marginal distribution of a minimal data set arising from a nonnormal linear model with reference noninformative priors. Specifically, I will show that this marginal distribution is independent of the distribution of the error when the latter satisfies a certain symmetry condition. I will demonstrate how this result can be generalized to a class of nonlinear models. A link between the symmetry condition and the distribution of a maximal invariant under the appropriate group of transformations will be established. Finally, I will mention some implications of these results and discuss their applications to Bayesian model selection, in particular to the Intrinsic Bayes Factor, a model selection criterion recently proposed by Berger and Pericchi.
Friday, December 1, 1995
11:45 - 12:45
116 Old Chemistry Building