The first posterior results we obtained under the previous prior presented some label switching because the groups were not identified in the proper order required by the prior on μ. In order to reduce label switching, we modified our initial prior information and augmented the asymmetric informative prior on μj with an asymmetric prior information onView the MathML source, selecting the prior expectations equal to (0.25, 0.50, 1.50) with prior degrees of freedom equal to 50. This new prior manages to reduce label switching because now the groups do correspond to their prior order. In order to appraise the convergence of the Gibbs sampler, we use CUMSUM graphs, as displayed in Fig. 1, Fig. 2, Fig. 3 and Fig. 4. CUMSUM graphs were first proposed by Yu and Mykland (1998) as a simple diagnostic method in order to assess the convergence of MCMC chains. The idea is to plot the evolution of a partial moment. If xj is the jth MCMC draw, then View the MathML source represents the MCMC estimate of the mean and View the MathML source the MCMC estimate of the standard deviation, both over the complete MCMC set. The (standardized) partial moment computed over the first i draws is View the MathML source. The standardized CUMSUM graph is formed by plotting ci against i for i=1,…,m where m is the total number of draws. This is a standardized version of the diagnostic procedure as proposed and detailed in Bauwens and Lubrano (1998). In the graphs, a ±10% confidence band is displayed as explained in Bauwens and Lubrano (1998).