Latent Class Analysis in Social Science Research
This 5-day workshop begins with an introduction to latent variable modeling (LVM), a comprehensive applied statistical methodology that includes latent class analysis (LCA) as a special case. The connection of LCA to the closely related statistical frameworks of factor analysis, item response modeling, and latent profile analysis is thereby underscored. A brief introduction to the highly popular LVM software Mplus is then provided. This workshop then attends to issues pertaining to handling of missing data, which may be seen as a rule rather than exception in applications of latent class analysis.
After this introductory part of the course, the quantitative foundations of LCA are focused on. In particular, the law of total probability will be illustrated with examples. The fundamental assumption of local independence is then attended to, and the formal LCA model is introduced. The parameters underlying an LCA model are discussed subsequently in detail, and their estimation using maximum likelihood is covered. Model testing against empirical data is then focused on, including in particular the increasingly popular bootstrap likelihood ratio test. The involved process of deciding on the number of latent classes is next dealt with in detail and demonstrated through empirical examples. Quality of classification to classes is then discussed and similarly illustrated empirically. LCA with covariates is subsequently dealt with. A discussion of technical issues including differentiation between local and global solutions, which could be involved in an application of LCA, is similarly provided.
Throughout the workshop multiple data examples are used for method and discussion illustration purposes. Thereby, the popular LVM program Mplus is utilized on numerous occasions. The command files needed then are discussed in detail, as are the output files furnished by these LCA and Mplus applications.
Fee: Members = $1700; Non-members = $3200