First, we used a robust locally weighted regression method25 of S-plus software (MathSoft Inc, Seattle, Wash) to conduct empirical smoothing growth curves for heights, weights, and BMIs against the chronological age and the time from menarche, which is the time difference between chronological age and menarcheal age, respectively. Growth velocity of height and weight across ages of the onset of menarche was assessed with slope change. The chronological age, age at menarche, and school were adjusted as covariates in the fitted model. In this longitudinal study, each schoolgirl had 4 to 8 height and weight measures, which are correlated. Furthermore, individuals' birth heights, weights, and growth rates may be varied owing to some unmeasured factors, including genetics, environment, nutrition, etc. Therefore, the mixed-effects linear regression models, which are a generalization of the standard linear model, are used to handle the correlation of height and weight measures within each girl as well as the natural heterogeneity of birth height, weight, and growth rates among girls. Specifically, the coefficients representing birth height, weight, and growth rate in the mixed-effects model are assumed to be random effects, which are normally distributed with zero means and unknown variances. The adjusted slope changes of the height and weight on the different groups of menarcheal ages were estimated by using the procedure of the mixed-effects models in the SAS software (version 6.12) (SAS Institute Inc, Cary, NC).