A nested expectation-maximization algorithm for latent class models with covariates

We propose a novel nested expectation-maximization algorithm for latent class models with covariates which allows maximization of the full-model log-likelihood and, differently from current methods, is characterized by a monotone log-likelihood sequence.