The name is absent

22

Figure 2.2: Illustration of the distribution of the latent variable v_il in the model
described in (2.2)-(2.4) when the covariate takes values —1 and 1. Here we consider:
J = I type of toxicity, no patient-specific random effect r_i, G = 2 components in the
mixture of normals and three, K = 2, possible ordinal outcomes. In both mixtures,
the darkly shaded, lightly shaded and white areas correspond to the probabilities πι⅛
of the ordinal outcome taking the values 0, 1 and 2, respectively.

On the other, large values of G may overparameterize the model leading to poorly
mixing Markov chains. Alternatively, using reversible jump MCMC (Green, 1995),
G could be included in the parameter vector and estimated as part of the inference.
But since the parameters of interest are the cell probabilities π_j⅛, and inference on
mixture-specific parameters is not of interest we prefer the approach with fixed large
values of G. Formally implied inference on the parameters of the mixture model,
including μj_g and pj_g, should not be interpreted. Problems related to label switching
(arbitrary permutation of the terms in the mixture) and node duplication (replicating

<<   31   32   33   34   35   36   37   >>

More intriguing information