Add option to neglect missing values for higher-order terms in M-step

[jdrefs]

Currently, M-steps use data estimates in case of missing values. Accuracy of higher order terms (i.e., variance parameter) may be improved when ignoring missing values (see, e.g., here, here) and considering corrections for missing values (e.g., here).