Matthias Kaeding discusses Bayesian methods for analyzing discrete and continuous failure times
where the effect of time and or covariates is modeled via P-splines and additional basic
function expansions allowing the replacement of linear effects by more general functions. The
MCMC methodology for these models is presented in a unified framework and applied on data sets.
Among others existing algorithms for the grouped Cox and the piecewise exponential model under
interval censoring are combined with a data augmentation step for the applications. The author
shows that the resulting Gibbs sampler works well for the grouped Cox and is merely adequate
for the piecewise exponential model.
