Computation of maximum entropy Dirichlet for modeling lifetime data (Q1566647)

This paper presents an entropy-based procedure for Bayesian inference about the suitability of a parametric distribution as a model for the data. The procedure, referred to as the maximum entropy Dirichlet (MED), combines the maximum entropy characterization of the parametric family with the Dirichlet process prior for the unknown data-generating distribution. The MED is a computer-intensive method that generates prior and posterior distributions of an information index for assessing the suitability of the parametric family of the likelihood function in light of the data. The MED also provides prior and posterior distributions for the parameters of the model under consideration. We present the elements of the MED procedure and give a Monte Carlo algorithm for obtaining the MED prior and posterior distributions. We illustrate an application of the MED on a set of synthetic lifetime data and find that the procedure correctly identifies the data-generating distribution among the well-known families of lifetime models.