A simple and fast alternative to the EM algorithm for incomplete categorical data and latent class models

摘要: Abstract Incomplete categorical data and latent class models play an important role in biostatistical medical literature. The most common maximum likelihood procedure for accommodating these types of is the EM algorithm. We present a faster alternative to approaches that improves upon recently introduced likelihood-based by Molenberghs Goetghebeur (1997. J. Roy. Statist. Soc. Ser. B 59, 401–414) two ways: higher-dimensional problems via more time points longitudinal employing less tedious iteratively reweighted least-squares (IRLS) approach than Newton–Raphson used MG. This IRLS also will facilitate potential extension with random effects context complete incomplete classes. illustrate our method application. As MG approach, we maximize observed instead under multivariate generalized logistic model composite link function. results convergence rate algorithm, allowing easily obtainable variance estimates. proposed estimation using from HIV study involving four dichotomous tests measured on each individual, assuming disease variable levels.