SAEM-MCMC: some criteria
Résumé
The SAEM-MCMC is a powerful algorithm used to estimate maximum likelihood in the wide class of exponential non-linear mixed effects models. The main problem of this method is that several parameters of simulation need to be calibrated. In this paper we propose some criteria to fix these parameters and we show on a real data set and by simulations that we need to run long markov chains in the Metropolis-Hastings algorithm to obtain an accurate estimator, which is relatively time consuming. In a second part, we applie our method to a model that does not belong to the exponential class, and we show on a simulated data set that we obtain the same results as the exact SAS NLMIXED procedure based on Gaussian quadrature. Our method seems to be appropriate for estimation in this class of non-linear models also.
Domaines
Calcul [stat.CO]
Origine : Fichiers produits par l'(les) auteur(s)
Loading...