Cluster Analysis as a Method of Recovering Types of Intraindividual Growth Trajectories: A Monte Carlo Study

Abstract

There has been increased interest in and application of cluster analysis in longitudinal applications to identify distinctive developmental patterns of intraindividual change. This article used Monte Carlo experiments to evaluate the adequacy of cluster analysis to recover group membership based on simulated latent growth curve (LGC) models. The simulated LGC models were manipulated by varying growth parameters (e.g., elevation, dispersion, and shape) for subpopulation growth curves (e.g., linear and quadratic growth models). The evaluation of cluster analysis to recover individual membership in these growth curve subpopulations was completed via the Kappa statistics. Cluster analysis failed to recover adequately growth subtypes when the difference between growth curves was shape only. It was much more successful when the distance between initial mean levels was large (e.g., difference of two standard deviations), independent of difference in the shape of the growth curves. Tentative guidelines were proposed to facilitate the evaluation of the adequacy of a cluster analytic solution to recover subtype heterogeneity in longitudinal (intraindividual) growth curves.