This article considers a new approximation to the log-likelihood surface in mixture models. This approximation is based on both the mean and variance of the full-data log-likelihood over imputations ...