Modelling variation of lower leg length growth in early life

We consider the estimation of sources of variation for panel data with repeated measurements. With no repeated measurements and known measurement error, models for variation decomposition have been proposed when there are one or more types of measurements. Estimation was performed using the EM algorithm accompanied by model augmentation that demands more computational efforts. In this article we extend previous variation models and modify the estimation methods in order to estimate various variation components after eliminating the unknown effects of measurement error. Specifically, methods that dispense with model augmentation and estimation of time-dependent covariates are considered. A set of lower leg length data from Chinese infants is analysed by using the proposed model. Interestingly, our results are consistent with the well-accepted three-phase (infancy-childhood-puberty) growth transition proposition for human growth. Moreover, gender effect is found to be time-varying.