All Pairwise Comparisons among Means: All t-tests




If there were no need to be concerned about the EER, then, instead of computing an analysis of variance, one could simply compute t-tests among all pairs of means. However, the effect on the EER is not trivial. For example, consider an experiment conducted with eight subjects in each of six treatment conditions. If the null hypothesis were true and all 15 t-tests were conducted using the 0.01 significance level, then the probability that at least one of the 15 tests would result in a Type I error is 0.10. Thus, the EER of 0.10 would be 10 times as high as the PCER of 0.01. Because computing t-tests on all pairs of means results in such a high EER, it is generally not considered an acceptable approach.