Right-sizing growth mixture models as multi-group growth and confirmatory factor models

Multi-group growth curve models with variant factor structure across classes can be used as the basis for growth mixture models. Such models identify qualitatively different patterns of growth/decline within class. Initial assessment of the dimensionality and patterning of growth factor loadings, prior to determining the functional form of growth, prevents the identification of artifactual latent classes as well as the use of overly complex or overly simple models. Simulated data sets illustrate the estimation of such loading variant mixture models and the comparison of candidate models. The ability of a variety of fit indices to correctly identify the correct model is explored under several sample size conditions. Analysis of a real-world data set is considered in which a two-factor growth model provides both superior fit and conceptually different classes than the "cat's cradle" pattern identified by linear growth or quadratic growth mixture models.