Luis A. Afonso
Posts:
4,613
From:
LIsbon (Portugal)
Registered:
2/16/05


Two nickpicks concerning Significance Tests
Posted:
Jun 18, 2012 4:21 AM


Two nickpicks concerning Significance Tests
As I?d showed by simulation the test statistics pvalues are uniformly distributed only when the respective null hypotheses is EXACTLY TRUE. By other words, it is not ENOUGH at all the norejection decision to assure uniform pvalues: if we not hit the *bull?s eye* they tend strongly to be smaller than for true H0, therefore one does find *anomalous* higher proportion of the p< alpha condition. ****** By other hand the usual *technical jargon* there is no difference (on means, on variations, on correlation coefficients, etc.) is advisable to be banned for didactic/teaching purposes, exactly because it fatally suggests (wrongly) that one can ever to state equality by means of significance tests. This interpretive error I was able to found here.
Luis A. Afonso

