Linear Mixed Model for Analyzing Longitudinal Data: A Simulation Study of Children Growth Differences

Growth developmental research is one example of the application of longitudinal data that have correlated value over time. Linear Mixed Model (LMM) is an extension of classic statistical procedures that provides flexibility analysis in correlated longitudinal data and allows researcher to model the covariance structures that represent its random effects. This paper briefly describes growth curves model as a single LMM that represent two levels of observation, which focused on modeling its covariance structure to capture correlated information over time of individual performance. We apply LMM and model different types of its covariance structure in the simulation study of children's growth differences based on the feeding methods. We perform simulation scenario using MIXED procedure in SAS system, based on three fit indices (-2RLL, AIC and SBC) and p-value significance level, we obtain Unstructured (UN) covariance is always be the best fit in presenting the characteristic of data but not the best choice considering inefficient numbers of parameters while Heterogeneous First-order Autoregressive (ARH(1)) is a proper alternative covariance structure with ease of data interpretation from fewer numbers of estimated parameters.