Maximum likelihood estimators in multivariate linear normal models

A unified approach of treating multivariate linear normal models is presented. The results of the paper are based on a useful extension of the growth curve model. In particular, the finding of maximum likelihood estimators when linear restrictions exist on the parameters describing the mean in the growth curve model is considered. The problem with missing observations is also discussed and the EM algorithm is applied. Furthermore, a multivariate covariance model is generalized.