Abstract:
|
In this talk, we consider Bayesian analysis for the general hypothesis testing problem in linear regression models with spherically symmetric errors. The use of these error distributions generally reduce the influence of outliers and can thus make the statistical inference more robust. Specifically, we consider modified mixtures of g-priors for the regression coefficients under some general linear constraints and derive a Bayes factor which is shown to be independent of the considered class of spherically symmetric error distributions and is also just a simple function of the classical F-tests. The proposed approach works very well in the context of general linear models with spherically symmetric error distributions and do not require the deigns matrix to be of full rank. The presented results extend some existing ones of the Bayesian procedures in the literature. Finally, some numerical examples are presented for illustrative purposes.
|