Modelling in repeated measures experiments
In repeated measures experiments how treatment contrasts change over time is often of prime interest and modelling these contrasts is the aim of any analysis. In long-term experiments, the number of repeated measures can be large or may exceed the degrees of freedom for error. Modelling the covariance matrix is therefore either desirable or required. We present an approach to the analysis of repeated measures data in which both the mean and the covariance matrix are modelled para- metrically. We use the correlogram and semivariogram for identification of the covariance structure and linear models for specifying the change over time of the treatment contrasts. Incomplete data are handled in the approach and estimation is based on residual maximum likelihood. Examples which motivated the work are presented to illustrate the applicability of the method.