COMBINED SIGNIFICANCE OF NON-INDEPENDENT TESTS FOR REPEATED MEASUREMENTS

A procedure is given for assessing the strength of evidence for difference between two treatments in experiments with repeated measurement. Significance levels of correlated univariate tests, each performed on data from a different period of measurement, are combined in a way that utilizes the intra-treatment correlations between data from all pairs of periods. The accuracy of assessing significance is greater than that achieved by standard univariate analysis of variance or by combining significance levels as if they were independent. The procedure is not dependent on an assumption of homogeneous variance from treatment to treatment, or from period to period. An example of use of the procedure in [animal] reproductive physiology and endocrinology studies are given.