Hierarchical baxes predictive means and variances with application to sample survey inference (Q3978036)

In a simple and clear way the possibilities of the Bayesian approach to statistical inference by the predictive distributions of unobserved random variable values are presented. Between the predictive and the estimative point of view, the author takes the former as framework. The results of posterior predictive means and variances of sample averages of unobserved values taken from \(k\) populations based on samples from these populations modelled according to the hierarchical Bayesian structure are applied to the finite population situation in order to obtain the predictive means and variances of \(k\) stratum means and predictive means and variances for totals, means and proportions. An application to small area estimation is done too. This paper also contains results on the Bayesian analysis of natural exponential families, posterior means and variances using conjugate priors, posterior mean and covariance matrix of the mean vector, applications to normal univariate exponential families and to binomial data, and the selection of a second-stage prior. A complete section for Bayesian finite population theory from a predictive point of view is developed. Twenty-six references are mentioned. This paper also contains new results for the superpopulation approach to sample survey inference with a predictive point of view.