Latent Class Analysis
Veronika Karnowski · The International Encyclopedia of Communication Research Methods · 2017
Latent class analysis is a statistical procedure to identify underlying and unobservable categorical latent variables based on categorical observed indicators. By attributing cases to the value of the categorical latent variable it most probably belongs to, latent class analysis can be used for clustering. Latent class analysis does have a series of advantages compared to conventional clustering techniques: (i) The choice of a cluster criterion is less arbitrary due to the underlying statistical model. (ii) Being a model‐based approach it provides several rigorous statistical tests to assess model fit. (iii) It also provides formal criteria to make decisions about the appropriate number of clusters. (iv) Latent class clustering can include observed variables of different scaling and measurement levels. (v) The probabilistic nature of cluster membership in latent class cluster analysis leads to less biased estimations of class‐specific means, as each case only contributes to this mean weighted by its class membership probability. (vi) The misclassification error can be calculated and reported. Despite these advantages latent class analysis has only scarcely been used in communication research thus far, a fact that is likely to change with the ever‐increasing uptake of comprehensive solutions for data analysis (e.g., the software environment R).