Summary: By adopting Bayesian approach in connection with Tiku’s method, this paper develops robust procedures for comparing means and variances from a one-way fixed effects ANOVA model under heteroscedasticity and nonnormality.
It is shown that the posterior distribution of the ratios of variances can be approximated by a finite series involving F distributions while posterior distributions of linear combinations of means can be approximated by a scaled-t distribution. For the numerical example taken from a medical study, the approximations appear to be extremely good.