Graphical models with marginals in the exponential family

Graphical models encode conditional independence statements of a multivariate distribution via a graph. Traditionally, the marginal distributions in a graphical model are assumed to be Gaussian. In this paper, we propose a three-level hierarchical model that functions as the hyper-Markov law that enables a graphical model with marginals in the exponential family with quadratic variance function. Inference on the model parameters is made using a Bayesian approach. We perform a simulation study and real data analyses to illustrate the usefulness of our models.