5. In Example 13.8 include a two-component discrete model varying on the Weibull slope as well as the regression intercept. Sample replicate times from this model to ascertain whether the 95% intervals of replicate data ti,rep contain the actual times (observed failures only).

Also include code to obtain Monte Carlo estimates of CPOs and assess any subjects not well fitted by the model. Finally consider the predictive criterion C of Ibrahim et al. (2001), adapted to allow for latent failure times tcens of censored cases, as well as observed failure times tobs. Let D = (tobs, t∗) where t∗ are the censoring times. This is best implemented by obtaining posterior means νi = E(ti,rep|D) and ξi = E(t(2)

i,rep

|D) for the replicate data ti,rep from an initial run. Then a second run is made sampling t(r )

cens for iterations r = 1, . . . , R and obtaining

with k > 0 defining the balance between precision and bias in C.

n C=(v)+ = (i - v) + - i=1 k (k+1) (3-4)? + (1) ti observed (vi ti censored r=1