Modelling the Covariance Structure of Repeated Measurements

Models for repeated measurements, used by Wilson, Hebel, and Sherwin (1981, Biometrics 37, 553565) and Mansour, Nordheim, and Rutledge (1985, Biometrics 41, 287-294) are considered. These two models are generalized by the (a, P, e , K) model, obtained by introducing an extra parameter K. A more general class of models described by Kenward (unpublished Ph.D. thesis, University of Reading, 1981) is briefly discussed. Simulation results are presented in which model parameters are estimated by maximum likelihood, and likelihood ratios are used as criteria for choosing between models. A set of blood pressure data is used to illustrate model fitting.