In Support of the Use of Significance Tests and of Unplanned Pairwise Multiple Comparisons

Previous reviews have criticized the use of significance tests in general and of unplanned pairwise multiple comparisons in particular. In this paper, it is argued that these tests are nonetheless useful. Significance tests compare treatment variation to error variation, which complements interval estimation. Unplanned pairwise multiple comparisons play a unique and valid role in separating treatment means. These tests are in no way invalidated by the previous authors' examples of misuse, nor wholly replaceable by the methods that they suggest as alternatives, which often have different purposes. Additional research questions are suggested.