Fig. 6.

Simulation setup with equal weights: MCMC run of a single data set, $K=15$, standard prior. Trace plot of the number of observations allocated to the different components, burn-in included (left-hand side) and point process representation of the posterior component mean draws where $K_0^{(m)}=4$ and which were non-empty, ${\mathit{\mu }}_{k1}^{(m)}$ is plotted against ${\mathit{\mu }}_{k2}^{(m)}$ (right-hand side)