A0840
Title: On the estimation of multilevel cross-classified latent class models
Authors: Silvia Columbu - University of Cagliari (Italy) [presenting]
Nicola Piras - University of Cagliari (Italy)
Jeroen Vermunt - Tilburg University (Netherlands)
Abstract: An extension of latent-class models is presented for dealing with the clustering of multilevel cross-classified data. The model is formulated to allow two levels of clustering, one for lower-level units and one for cross-classified units, i.e. observations simultaneously nested within two or more groups. Given the dependency structure in the data, maximum likelihood estimation cannot be directly performed using a standard EM algorithm. A variation including a stochastic step is proposed. Global model selection criteria are also provided to determine the number of latent classes at both levels of the multilevel structure. The performances of the estimation algorithm and the selection criteria are verified through simulation studies. An application to educational data is finally discussed.
