We would like to thank Guochen Song for pointing out an error in Follmann & Albert 1999 who developed a Bayesian approach to the monitoring of discrete survival data. The basic formulation was to regard the discrete failure times as coming from a multinomial distribution, for which the conjugate prior is Dirichlet. With censoring, the posterior distribution is a mixture of Dirichlet distributions. Guochen Song realized that the mixing probabilities for the mixed posterior Dirichlet distribution are incorrect.

The theorem in Follmann & Albert (1999) incorrectly stated that the mixing probability for inline image was proportional to

  • display math

where inline image if inline image and

  • display math

if inline image.

The correct mixing probabilities have

  • display math

if inline image and inline image if inline image.


  1. Top of page
  2. References