Latent Class Modeling with Covariates: Two
Improved Three-Step Approaches
Jeroen K. Vermunt
Latent Class methods are used to construct a typology or clustering based on a set of observed variables; that is, to classify observational units into a preferably small set of LCs. There are two ways to deal with covariates in LC analysis: a one-step and a three-step approach. To overcome the disadvantages of the two methods, the author proposed a modified BCH procedure that removes several limitations of the original BCH approach and a new three-step maximum likelihood (ML) procedure
A simulation study was conducted which used to show when the various three-step methods work and when they do not.
The results show that the new three-step ML method is more efficient—yields smaller SEs for the covariate effects—than the BCH approach.
Comments：
In table2, the performance of one-step ML seems more steady among the conditions and not less than new method the author proposed.
In the empirical study，the LC solution has similar estimates as PCA, we may PCA as an external criteria to name the latent class, or can we use the results of PCA to set some cut-points which may be used for grouping when the structure is simple?