A simulation study of a proposed graphical diagnostic for assessing goodness-of-fit

In this article, we investigate for various noninformative improper priors the propriety of the resulting posterior distributions of variance components in a mixed normal linear model. We have considered the usual Jeffreysʹ prior, reference priors and Jeffreysʹ prior based on residual likelihood for variance components. The latter prior is recently advocated as a standard default prior for Bayesian analysis using the PROC MIXED procedure of SAS.