Growth Mixture Modeling: Identifying and Predicting Unobserved Subpopulations with Longitudinal Data

Abstract

An important limitation of conventional latent-growth modeling (LGM) is that it assumes that all individuals are drawn from one or more observed populations. However, in many applied-research situations, unobserved subpopulations may exist, and their different latent trajectories may be the focus of research to test theory or to resolve inconsistent prior research findings. Conventional LGM does not help to identify and predict these unobserved subpopulations. This article introduces the growth-mixture modeling (GMM) method for these purposes. Given that GMM handles longitudinal data (i.e., nesting of time observations within individuals) and identifies unobserved subpopulations (i.e., the nesting of individuals within latent classes), GMM can be construed as a multilevel modeling technique. The modeling procedure of GMM is illustrated on a simulated data set. Steps in the modeling process are highlighted and limitations, cautions, recommendations, and extensions of using GMM are discussed. Technical references for additional information are noted throughout.