Title: Divisive latent class models in R.
 
Author: Dani‘l W. van der Palm, L. Andries van der Ark, Jeroen K. Vermunt
 
Presenting author: L. Andries van der Ark

Affiliation: University of Amsterdam
 
Abstract: Traditionally, latent class models (LCMs) are used to identify
substantively meaningful clusters.  More recently, LCMs have also been used
as a density estimation method for categorical variables; for example in
missing value analysis, reliability analysis, pattern-recognition, and
smoothing sparse data.  The LCM describing the multivariate density should
fit the data well, and a considerable number of increasingly large LCMs may
have to be estimated before sufficient model-fit has been achieved, which
may be a slow and cumbersome process.  The divisive latent class model
(DLCM) is an attractive alternative over the LCM as a density estimation
tool.  The DLCM estimates the density of categorical variables at least as
well as the LCM. However, the DLCM can be estimated relatively fast because
it consists of a sequence of small LCMs (1 and 2 latent classes only), and
because it can easily utilize multiple processor cores.  We will explain
density estimation using the LCM and the DLCM, and demonstrate an R package
for estimating the DLCM using an applications in missing value analysis and
reliability analysis.