A group of graduate students and faculty member Karen Nylund-Gibson presented the symposium “An Overview of Latent Class Analysis: Applications and Issues” at the Western Psychological Association conference on Thursday, April 28.
Latent Class Analysis (LCA) is rapidly becoming a more commonly used analytic technique for psychological researchers given its capacity for informing the exploration of unobserved heterogeneity in a population. Recent developments in mixture models, including the addition of mixture components to existing statistical models, permit the specification of much more complex models. This complexity offers greater flexibility in the parameterization of population heterogeneity, which is indeed an advantage in that it allows the analytic models to more accurately reflect the complexity behavioral processes and individual differences. However, this flexibility also necessitates careful attention to model building procedures, and an increased understanding of the sensitivity of empirical results and the corresponding interpretations are to model specifications.
The symposium brought together five papers that all relate to the application and specification of LCA models. The first paper – “Introduction to Latent Class Analysis (LCA),” written by Hadar Baharav, Amber M. Gonzalez, Alma S. Boutin-Martinez, and Ani Dzhidaryan – introduced the cross-sectional LCA model, its parameters, and the modeling procedures used in its application. The next two papers provided applications of the LCA model. For instance, the second paper –“Latent Class Analysis (LCA) with Ordered Categorical Variables,” written by Shelley R. Hart –used a national dataset measuring victimization experiences with and without complex sampling weights, comparing/contrasting results and inferences that can be made from both models. The third paper – “Examining the Factors Influencing Academic Achievement: Latent Variable Approach,” written by Igor Himelfarb & Karen Nylund-Gibson – used a latent class variable as the outcome in a mediation model which examines the relationship of extracurricular activities and delinquent behavior in a sample of middle school students.
The latter two papers explore model specification issues. The fourth paper – “Including Auxiliary Variables in Latent Class Analysis Models,” written by Karen Nylund-Gibson and Katherine Masyn (Harvard Graduate School of Education) – presented a method of interpreting and graphically representing the LCA with ordered categorical outcomes (i.e., that is instead of binary). The fifth paper – “Comparing Latent Class Models With and Without Sampling Weights Using the NCVS-SCS Dataset,” written by Diane Morovati, Shelley Hart & Karen Nylund-Gibson – presented the results of a simulation study looking at the impact of misspecifed covariate effects on latent class enumeration. Together, these papers provide an overview of LCA, examples of its use in psychological research, and provide cutting edge information about the application and specification of mixture models.
[The members of the Latent Variable Group are available for interviews; contact George Yatchisin at 805 893 5789]
– end –