Just as factor analysis is commonly used to infer the presence of underlying continuous latent variables, a related modeling technique – mixture modeling – can be used to inform researchers about underlying categorical latent variables. Often referred to as classification and conceptually similar to traditional clustering, latent class analysis (LCA) and latent profile analysis (LPA) use measured characteristics of individuals to identify latent classes, or phenotypes, through mixture modeling. For instance, mixture modeling has been used to identify types of families and children that are most receptive to interventions and to detect subgroups with similar developmental trajectories (i.e., growth mixture models). The use of such person-centered approaches is gaining popularity in a number of research contexts, including early childhood and education research. This presentation will (a) discuss traditional and modern methods of classification, and (b) provide examples of empirical identification and theoretical validation of latent subgroups within a population.

Kevin Kupzyk received his master’s degree in Quantitative Psychology from the University of Kansas in 2005. He is now a methodological consultant for the CYFS Statistics and Research Methodology Unit and a doctoral student in Quantitative, Qualitative and Psychometric Methods in Educational Psychology at UNL. His research interests include power analysis and optimal design of experiments, educational measurement, multilevel modeling and latent variable growth models.