Journal of Official Statistics, Vol.25, No.3, 2009. pp. 405–413
Beyond Objective Priors for the Bayesian Bootstrap Analysis of Survey Data
Abstract:This article provides reasonable answers to the problems left unsolved in Aitkin (2008), a recent paper on the Bayesian bootstrap in finite population inference. These problems are essentially two: the choice of the population parameter cannot be discussed from within the Aitkin’s Bayesian bootstrap approach, which is based on a multinomial likelihood with unconstrained parameters; assumptions such as model constraints on the multinomial probabilities are difficult to implement in such a Bayesian framework. The answers are obtained by assigning suitable informative priors to the population proportions involved in the analysis.
Keywords:Survey sampling, Bayes factors, population parameter selection, post-data Dirichlet priors, constraints on the multinomial probabilities
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group