Date: Feb 17, 2013 9:21 AM
Author: Luis A. Afonso
Subject: No-Null HST . . .
No-Null HST . . .

My insistence in this point is only justified because the controversy, sometimes turned sour, has any chance .to get an end. The anti-NHST are right only at one point: when dislikes the revues to publish as relevant papers simply because it was found that the Null Hypotheses was not credible to hold. Some ones advise authors to show confidence intervals instead p-values in order to evaluate what´s the extension the null cannot be rejected. It is an idea, but it must be said is not free of danger. In fact the well-known No-Null HST (my acronym) can be straightly invoke: it suffices to add to the null an positive quantity c and test H0: Teta<=c against H1: Teta>c. Then if H0 is not rejected the conclusion is that we do not know if the parameter is larger than c, otherwise if the test falls at the rejection region we are sure (at the set significance level) that Teta is at least as large as c. Exclusively NHST are, for example, those that compare the two Population difference of means using the Fisher´s Shuffling (exact test) or by Bootstrap Technique as well. Originally they were so: to transform them in a No-Null HST it suffices to start to add a quantity c to one sample items.

Luis A. Afonso