Multi-level modelling of longitudinal child growth data: a comparison of growth models in the generation XXI birth cohort

Abstract

Several methods have been used by different authors to estimate growth curves. In this paper, five growth models are compared using weight and height data from 6668 children from the Generation XXI birth cohort. The goal was to determine the model(s) that better describe the growth pattern from birth to 10 years of age using mixed-effect modelling. The study compared the fitness of four structural (Jenss-Bayley, adapted Jenss-Bayley, Berkey-Reed 1st order and Berkey-Reed 2nd order) and one non structural (cubic spline based) model. The goodness of t of the models was examined using standard deviation of the residuals, Akaike Information Criterion and Bayesian Criterion. The adapted Jenss-Bayley and the spline based model had the better tting for weight while for height the better models were Berkey-Reed 2nd order and the spline.