Abstract
Linear mixed model with covariance structures have become increasingly popular in longitudinal data analysis because of its wide applications. There is a problem of dealing with missing data in longitudinal analysis which need careful attention. In missing data analysis there are two different mechanisms referred to as missing completely at random and missing at random. In this paper estimation of parameters with distinct covariance structures using Maximum Likelihood and Restricted Maximum Likelihood methods are considered for the evaluation of models using six different information criteria with data missing at random. The study involves both the nested and non-nested covariance structure for comparison based on model selection information criteria. Also, evaluation using bootstrap method to identify the most plausible covariance structure for longitudinal models are studied.
Author: J. Mohanraj and M. R. Srinivasan
Received on: November, 2015
Accepted on: February, 2019