On the analysis of mixed longitudinal growth data

Mixed longitudinal growth data consists of several observations on a characteristic over a limited age range for each individual in a study. This data is then combined to model growth over the total age range of all individuals in the study. The limited data collected on each individual precludes a subject-specific approach to modelling so that a population-based approach must be used. Here we propose a method for the analysis of mixed longitudinal data using linear models constructed from cubic splines to model both the mean and variance curves of an observed characteristic. The method is illustrated on growth in height and head circumference for children in a recently collected mixed longitudinal data set concerning the growth of Victorian schoolchildren, where particular concern was the timing of growth spurts and increases in variability.