Date: Oct 31, 2012 6:47 PM
Author: Luis A. Afonso
Subject: Problems with NHST? None at all . . .
Problems with NHST? None at all . . .
JEOFFREY A. GLINER
NANCY L. LEECH
GEORGE A. MORGAN
Problems With Null Hypothesis
Significance Testing (NHST):
What DO the Textbooks Say?
. . . Kirk (1996) went on to explain that NHST was a trivial exercise because the null hypotheses is always false, and rejecting it is merely a matter of having enough power.
What?s an amazing absurdity for me are exactly supposedly learned people, as are Psychologists and others Researchers (kirk) had started many decades ago, an endless discussion with no solution in view. The difficulty will disappear in an eye clink if they returned to their Handbooks and take into account that beyond the Null text there are a No-Null one. Explicitly, if we read, for example (difference of two means, large normal samples, variances unknown):
_____Z = (xbar - ybar - 0) /sqrt (sx^2/nX + sy2/nY)
not to paying attention to the 0, we are tried to perform a NHST even that we do not recognize the intent as so. The z value if larger than 1.645, for example, we conclude that the X mean Population exceeds the Y Population mean with 95% confidence, but stays unknown how much the difference is.
However, on contrary, if we write, with a specific D,
_____ Z = (xbar - ybar - D) /sqrt (sx^2/nX + sy2/nY)
and Z is 1.645 or larger we can properly assert that at least the difference is D (with 95% confidence). So we made a NNHST (my own acronym, no-null hypotheses statistical test). This procedure is perfectly sound and can be used always a parameter test is in view. No need of further *tools*.
It seems that the only *problem* is that some insufficient knowledge among statistics users . . .
Luis A. Afonso