Partial likelihood analysis of within-unit variances in repeated measurement experiments

Abstract

The objectives of some experiments are to compare the variances of two or more treatments, products, or techniques. If the investigator is more concerned about within-unit variances rather than between-unit variances, then a repeated measurement design is needed. We invoke a random effects model with heterogeneous within-unit variances for certain repeated measurement designs. We do not impose any distributional assumptions for the random effects, whereas we assume either a normal or multivariate t distribution for the random errors. We propose a partial likelihood analysis for population-based inference and individual-based inference. We illustrate the methodology with an example from a trial comparing serum cholesterol measurements from a routine laboratory analyzer to those of a standardized method.