Statistical Methods Course Description
Overview: An introduction to “person-centered” data analysis. Topics include latent class analysis, latent class cluster analysis, modeling predictors and outcomes of latent class membership, and select extensions. Hands-on practice with Mplus is provided.
This five-day summer institute, sponsored by IMMAP, is an intensive short course in the fundamentals of latent class analysis and finite mixture modeling.
Finite mixture models are a type of latent variable model that express the overall distribution of one or more variables as a mixture of a finite number of component distributions. In direct applications, one assumes that the overall population heterogeneity with respect to a set of manifest variables is due to the existence of two or more distinct homogeneous subgroups, or latent classes, of individuals. These approaches are often termed “person-centered” analyses in contrast to the “variable-centered” analyses of conventional factor and SEM models.
This course will introduce participants to the prevailing “best practices” for direct applications of basic finite mixture modeling to cross-sectional data, specifically latent class analysis (LCA) and latent class cluster analysis (LCCA; a.k.a. latent profile analysis), in terms of model assumptions, specification, estimation, evaluation, selection, and interpretation. Models that allow for the inclusion of correlates and predictors of latent class membership as well as distal outcomes of latent class membership will be presented. The course will also explore “hybrid” latent variable models that include both latent factors and latent classes (termed factor mixture models) and will touch briefly on some multilevel and longitudinal extensions of mixture modeling, as time allows (for a more in-depth treatment, see the Stats Camp Session 2 course on longitudinal mixture modeling). The implementation of these models in the most recent version of the Mplus software will be demonstrated throughout the course.
Instructor: Katherine Masyn Ph.D.
Katherine is an Associate Professor of Biostatistics at Georgia State University’s School of Public Health. She was previously on the faculty at the Harvard Graduate School of Education and received her doctorate in Social Research Methodology at UCLA under the mentorship of Prof. Bengt Muthén. Dr. Masyn’s research focuses on the development and application of latent variable statistical models related to: survival and event history analysis; multivariate and multifaceted longitudinal processes, e.g., latent transition growth mixture models; and, more broadly, the characterization and parameterization of both observed and unobserved population heterogeneity in cross-sectional and longitudinal settings, e.g., factor mixture models. Dr. Masyn enjoys close collaborations with colleagues from the fields of Human Development, Education, Psychology, Public Health, and Prevention Science and is known for her enthusiastic instruction on applied latent variable techniques for substantive audiences.
Software and Computer Support
Participants should bring a laptop computer. Instruction will be provided for the methods using the most current version of Mplus (base program with mixture add-on or base program with combination add-on). Mplus is available for Windows, Mac, and Linux environments (www.statmodel.com).
Participants who do not have access to software will be given temporary access to the server that contains fully functioning versions of the recommended software.
Note: We will also make use of Excel and R to do various post-processing summaries.
Participants will receive an electronic copy of all course materials, including PowerPoint slides, Mplus scripts, output files, relevant supporting documentation, and recommended readings.
If you are interested in learning “person-centered” statistical modeling techniques that can identify unobserved subgroups (latent classes) characterized by qualitative differences in observed multivariate outcome distributions, this course is for you. You should have a good working knowledge of the principles and practice of multiple regression and elementary statistical inference. You will get the most out of the course if you already have experience with binary and multinomial logistic regression. You do not need to know matrix algebra, likelihood theory, or SEM, although that knowledge would be beneficial. No previous knowledge of mixture modeling, latent class analysis, or Mplus is assumed. Participants from a variety of fields—including psychology, education, human development, public health, prevention science, sociology, marketing, business, biology, medicine, political science, and communication—will benefit from the course.