A Priori and A Posteriori Tests on Repeated Measurements

Abstract

Failure to meet the sphericity assumption in repeated measurements analysis of variance can have serious consequences for both omnibus and specific comparison tests. It is shown that, in educational research journals, the relevance of this assumption has hardly been recognized. The risk of an inflated Type I error rate can be minimized by calculating separate error terms and by applying conservative tests. This paper illustrates how this is done. Some notes on the use of three mainframe computer packages are also provided. It is argued that when using these packages, as specific comparisons are typically tested as planned instead of post hoc tests, the outcomes should be interpreted in a conservative sense.