Multiple Comparisons, Selection, and Applications in Biometry. (Reannouncementwith New Availability Information)

摘要

The preceding chapters do much to cleanup conventional multiple comparisons: allsimple comparisons and the Dunnett case. It is high time to go further. Here I will discuss, more or less briefly, four ways in which we should go. Ab initio, we will have to recognize that null hypotheses are convenient limiting cases and not something that ever happens. Non-significant means that we do not know the direction. And the complicated nature of multiple comparison results urges stress on more and better kinds of graphical presentations. First, we should recognize that, as measurement in a field evolves, we move from hoping for a few confident directions toward trying to get the sign of every possible simple comparison (every difference) straight. The corresponding sequence of multiple comparison techniques is probably not yet complete. Second, we need to face the study of two-