Maximum Likelihood Estimation for Incomplete Repeated-Measures Experiments under an ARMA Covariance Structure

A stochastic model is presented for the analysis of incomplete repeated-measures experiments. The general linear model is used to relate the response measures to other variables which are thought to account for inherent variation; an autoregressive moving average (ARMA) time series representation is used to model disturbance terms. Maximum likelihood estimation procedures are considered, and the properties of these estimators are derived. It is concluded that while the assumptions underpinning the ARMA covariance models may be somewhat restrictive, they provide a useful inferential vehicle, particularly in the presence of missing values.