View PDF Version - RePub - Erasmus Universiteit Rotterdam
View PDF Version - RePub - Erasmus Universiteit Rotterdam
View PDF Version - RePub - Erasmus Universiteit Rotterdam
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Chapter 3.2<br />
158<br />
Table 2. Dutch PBC Study. Posterior summary statistics for standard deviations of random effects (square<br />
roots of diagonal elements of the matrix D) and error terms (σ1 , σ2 , σ3 ) in models with K = 2.<br />
Group 0 Group 1<br />
Posterior<br />
Median<br />
95% HPD<br />
Interval<br />
Posterior<br />
Median<br />
95% HPD<br />
Interval<br />
Bilirubin<br />
intercept 0.39 (0.30, 0.51) 1.24 (0.66, 2.12)<br />
time (months) 0.0050 (0.0034, 0.0071) 0.17 (0.09, 0.27)<br />
Error 0.24 (0.24, 0.25) 1.34 (1.24, 1.44)<br />
Albumin<br />
intercept 0.12 (0.10, 0.14) 0.18 (0.12, 0.28)<br />
time (months) 0.0010 (0.0009, 0.0012) 0.0045 (0.0026, 0.0071)<br />
Error 0.069 (0.067, 0.071) 0.10 (0.09, 0.10)<br />
Alkal. phosph.<br />
intercept 1.18 (1.03, 1.36) 1.82 (1.32, 2.40)<br />
time (months) 0.0070 (0.0059, 0.0084) 0.046 (0.026, 0.077)<br />
Error 0.62 (0.61, 0.64) 0.97 (0.89, 1.04)<br />
the standard deviations of the random effects given in Table 2 which are higher in<br />
Group 1 than in Group 0. Especially for bilirubin, the random effects are clearly<br />
necessary to model the between-patients variation of the longitudinal evolution.<br />
Further, we computed the posterior predictive density of the random effects in all<br />
considered models and explored in more detail the univariate and pairwise bivariate<br />
marginal densities. We conclude that in models with K = 2, neither of these densities<br />
is clearly bimodal. Nevertheless, in many cases, the two components mixture<br />
helped to capture skewness in the distribution of the random effects. As an illustration,<br />
Figure 2 shows the estimated pairwise marginal densities for two selected pairs<br />
in models with K = 1 and K = 2 in Group 1.<br />
Let 0 ≤ ti,1 < ···