Uses of structural equation mixture models in psychological research
11/5/2012, 1:00 pm - 2:30 pm
Psychological Sciences Quantitative Methods Colloquium Series presents Daniel J. Bauer. Associate Professor, Quantitative Psychology, University of North Carolina at Chapel Hill
Structural Equation Mixture Models (SEMMs) represent a hybridization of traditional structural equation models with latent class analysis. To date, most applications of SEMMs have aimed to identify hidden population subgroups, such as latent classes of individuals following distinct longitudinal trajectories, or latent classes that are more or less responsive to treatment. In these applications the classes are interpreted in a literal sense as representing distinguishable groups of people. I review research that shows this interpretation is seldom accurate and that, more often, the latent classes serve other roles. Although this fact may stymie attempts to identify actual population subgroups, it also provides an opportunity for other types of SEMM applications. Specifically, the flexibility of the latent classes can be exploited to gain traction on a variety of difficult problems. I describe several such applications of SEMMs: density estimation for latent variables, nonlinear effect estimation, longitudinal data analysis with non-random missing data, and multivariate survival analysis.