We chose a standard parametric approach to model longitudinal growth. This has the advantage of natural biological interpretability of the parameters obtained from the fitted models [9], and appeared to fit our data well. There are a number of alternative approaches, for instance smoothing or regression cubic splines; these are easy to fit but the interpretation of parameters poses challenges, as does the selection of the degree of smoothness to be enforced. We attempted to fit models based on cubic smoothing splines [16] to these data, but found the results difficult to interpret and sensitive to the number and location of knots selected, and therefore present only the results for the parametric growth models.
