Abstract
Reference centile charts are widely used to monitor child growth, and yet they are often used inappropriately. Charts derived from cross-sectional data ought not to be used to monitor longitudinal data. Such monitoring is only valid if two distinct sets of centiles are available, one for height distance and one for height velocity. There is also the problem of regression to the mean, in that expected height velocity is negatively correlated with initial height during infancy and puberty, and this requires a regression-based conditional standard. Furthermore, such conditional standards usually assume that the measurement of interest is normally distributed, which may not be appropriate. The paper describes modifications to the conventional growth chart which address these issues, illustrated using height data from the French Longitudinal Study.