Bayesian analysis of finite mixture models of distributions from exponential families (Q2463672)

A general approach to address Bayesian mixtures of distributions from natural exponential families with quadratic variance function (NEF-QVF) is proposed. The authors consider the case with a known number of components. The Gibbs sampling procedure and the relabeling procedure of \textit{M. Stephens} [J. R. Stat. Soc., Ser. B, Stat. Methodol. 62, No. 4, 795--809 (2000; Zbl 0957.62020)] are adapted to explicitly handle mixtures of distributions from NEF-QVFs. The authors propose a simulation-based algorithm to choose the parameters in the prior distribution. An illustrative example for a finite mixture of Poisson distributions is presented.