Gravar-mail: Empirical-likelihood-based criteria for model selection on marginal analysis of longitudinal data with dropout missingness