科研成果详情

We reparametrize the marginal covariance matrix arising in longitudinal studies to model, jointly, the mean and covariance structures in terms of three polynomial function of time. We also compare model selection procedures based on regressogram estimation with those based on a direct search of the joint model space. Using a BIC-based model selection criterion to identify the optimum degree triple of the three polynomials, we show that the use of a saturated mean model is not optimal, explain why regressogram-based model estimation may mislead and give a new computational algorithm, based on a criterion involving three pairwise saturated profile likelihoods, for finding the global optimum model efficiently.