Latent Class Analysis

The term latent class (LC) analysis refers to a class of statistical analyses that use the LC model to explain the associations among a set of observed variables. The LC models are advantageous generally because they bring in unobserved (latent) categorical variables, each category of which is defined as a subgroup. Thus, in LC models, the associations among observed variables are explained by the relationships between the latent categorical variables and each observed variable. LC models have been used in many applications in statistical analysis, such as clustering, diagnostic classification, density estimation, and dealing with unobserved heterogeneity. Further, LC models assume that observations within each subgroup are generated from an independent random process, and therefore, the distribution of overall observations can be seen as a mixture of the distribution of observations from each subgroup. In this sense, LC models are more generally referred to as finite mixture models.

This entry begins with a brief review of the history of LC analysis. Then, it introduces two approaches (i.e., probabilistic and log-linear) in parameterizing LC models. Further, the entry focuses on the basic methods for model estimation and model evaluation.