Title: ABC inference for a spatio-temporal autologistic model
Authors: Shuxian Li - PrimBio Genes Bio-Tech (China)
Anne Gegout-Petit - INRIA Universite de Lorraine-IECL BIGS (France) [presenting]
Abstract: A centered autologistic model is proposed to model the occurrence of an illness on a regular lattice. It allows us to test spatial dependencies and effect of covariates. In the framework of Markov chain of Markov field, we first prove the existence of the spatial law of the field at time $t$ given the covariates and the past of the process. We show the utility of a good spatio-temporal centering for the correct modelling and interpretation of the effect of the neighbours on the occurrence. It preserves the large-scale structure of the phenomena. About inference, we show trough simulations, the good performance of the expectation-maximization (EM) pseudo-likelihood estimation. We study also the possibility to use Approximate Bayesian Computation (ABC) for model selection of the structure of neighbourhood.