Title: Some issues on Bayesian analysis of binary bidirected graphs
Authors: Claudia Tarantola - University of Pavia (Italy) [presenting]
Abstract: Bayesian analysis of binary bidirected graphs has not been developed as much as traditional methods. No conjugate analysis is available and MCMC methods must be employed. The likelihood of the model cannot be analytically expressed as a function of the marginal log-linear interactions, but only in terms of the probability parameters. Hence, at each step of the MCMC an iterative procedure needs to be applied in order to calculate the cell probabilities and consequently the model likelihood. Finally, in order to have a well-defined model of marginal independence, the considered MCMC algorithm should generate parameter values leading to a joint probability distribution with compatible marginals. We will present a novel MCMC strategies that handles the previously discussed problems. A simulation study will be discussed.