A Randomization Test for Comparison of Groups of Growth Curves with Different Polynomial Design Matrices

Abstract

Well defined methods for comparing mean polynomial growth curves exist when subjects are measured at identical times, modeled with polynomials of identical degree and when multivariate normality of the measurements can be assumed. Relaxing these conditions creates problems solved here for the special case of a completely randomized design by developing a randomization test which is both practical and interval specific. The procedure is applied to longitudinal data from a study of human growth.