![IconGems-Print](images/IconGems-Print.png)
421 – Nonparametric Methods for Longitudinal Data to Promote Healthy Birth, Growth, and Development
Characterizing Child Growth Trajectories
Craig Anderson
University of Technology Sydney
Ryan Hafen
Purdue University
Louise Ryan
University of Technology
There is a wide and growing literature on growth curve modeling based on longitudinal data, including parametric and semi-parametric (spline-based) random effects models, functional data analysis methods and latent growth curve models. We compare and contrast these various methods in terms of how well they do in terms of predicting individual child growth trajectories, based on data from the HBDGki project. Prediction accuracy is assessed using a "leave one out" strategy for fitting and then comparing predicted values with the observed values of those left out. Methods are also presented for extracting key growth features such as faltering and catch-up. We conclude with recommendations about how growth data can be most effectively modeled in epidemiological cohort studies such as the ones encountered in the HBGDki project.