Objective Bayesian Inference for the Half-Normal and Half-t Distributions Articles uri icon

publication date

  • December 2008

start page

  • 3165

end page

  • 3185

issue

  • 20

volume

  • 37

international standard serial number (ISSN)

  • 0361-0926

electronic international standard serial number (EISSN)

  • 1532-415X

abstract

  • In this article, Bayesian inference for the half-normal and half-t distributions using uninformative priors is considered. It is shown that exact Bayesian inference can be undertaken for the half-normal distribution without the need for Gibbs sampling. Simulation is then used to compare the sampling properties of Bayesian point and interval estimators with those of their maximum likelihood based counterparts. Inference for the half-t distribution based on the use of Gibbs sampling is outlined, and an approach to model comparison based on the use of Bayes factors is discussed. The fitting of the half-normal and half-t models is illustrated using real data on the body fat measurements of elite athletes.