Title: Model averaging weighted estimators for latent variable models in a longitudinal data setting
Authors: Vassilis Vasdekis - Athens University of Economics and Business/Research Center (Greece) [presenting]
Kostas Florios - Athens University of Economics and Business (Greece)
Dimitris Rizopoulos - Erasmus University Rotterdam (Netherlands)
Irini Moustaki - London School of Economics (United Kingdom)
Abstract: In the latent variable setting for longitudinal data, different sets of weighted average estimators for model parameters are proposed. All of them are based on ideas coming from the composite likelihood. The first two are produced by minimizing the total variance of the resulting estimator. The weights are matrices either unrestricted or diagonal. The third estimator uses univariate weights. For their construction we propose a sequence of models each of which provides estimates for model parameters and a composite likelihood information criterion. The latter is used to evaluate the weights. The estimator seems to perform better than the other two estimators providing variance components with high coverage, especially when the number of time points at which measurements are obtained is large, regardless of the size of the cluster size.