Robustness of Latent Class Measurement Error Models

Brian J. Meekins and Daniell Toth


The technique of latent class analysis relies on a number of model assumptions which might be violated by the underlying process being investigated. This study is to determine the reliability of the analysis done on four stage Markov Latent Class models containing the classification of individuals in one of two indicator categories. The estimation is done using the EM algorithm on simulated data under specified model assumptions where those assumptions are violated to varying degrees.