EPSA paper: Inference for multilevel models when the number of clusters is small.

In a recent article published in AJPS it is claimed that Bayesian estimators have a superior performance in the estimation of the influence of group-level covariates, especially if the number of groups/clusters is small. In the paper presented at EPSA, we show that the problems addressed by Bayesian techniques can also be adequately addressed by a frequentist technique, restricted maximum likelihood, without the problems involved in Bayesian estimation, such as the computational cost and the need to select an appropriate prior.


Coverage error of Student-t based confidence intervals of the coefficient of a group-level covariate, with different numbers of groups and two different estimation methods, maximum likelihood and restricted maximum likelihood.