Course Text: Pages 25-26
The multilevel modeling option can be used to extend LC models to account for additional correlation in nested data, such as employees nested within departments, pupils nested within schools, clients nested within stores, patients nested within hospitals, citizens nested within regions, and repeated measurements nested within individuals. Simultaneously with the identification of latent classes at the individual level (level 1), 2 or more latent group level classes (GClasses) may also be identified. The basic idea of a multilevel LC analysis is that one or more parameters of the model of interest is allowed to vary across groups via the GClasses.
The variant of the multilevel LC model that we will focus on involves including group-level random effects in the model for the latent classes, which is a way to take into account that groups differ with respect to the distribution of their members across latent classes. Not only the intercept, but also the covariate effects may have a random part. Such models are especially useful when there are many groups.
To introduce this type of model extension (see Exercise C1), we will utilize the first example in Vermunt (2003) where there are 886 employees nested within 88 teams (groups).