On 25 Mar 2003 08:05:45 -0800, [EMAIL PROTECTED] (dennis roberts) wrote: [ ... ] > > 1. how useful REALLY ... is null hypothesis testing? ... as we generally > practice it > [snip #2.] > > as to the first item ... take correlations ... when we form the typical > null hypothesis of rho being 0 in the target population ... and if we are > interested in the relationship between ability and achievement (or lots of > other cases) ... aren't we wasting our time testing such a null?
Yes, we would, if we ever bothered to do that. Mostly, we don't, and that's why. If it is a total "waste of time," I am pretty sure that there is a good reason for a good hypothesis tester to avoid it. As I recall, we did this before -- Dennis put forth a couple of "straw men" cases, where he thought it was worthwhile to do a study even though he could not come up with a Null; and I re-wrote what he had, so show how to frame the relevant hypothesis. For correlations "reliability" studies, there is practically no interest in the r=0 null. Unless the sample is so small that the r will be large, anyway. You do not understand that there > ... i find > most of the nulls that we routinely test ... to be rather silly and again, > we know they are not true > Do you want to try it again? - I will show you what you are *supposed* to be asking, or saying, instead of what you have that is silly. Practicality -- I stated in another note how I use p-levels in combination with effect sizes, and also, knowledge of design, in my practice of Reading the Newspaper, to come to interpretations of several concrete instances. Meeting a 5% level is helpful. Having an Odds Ratio of 1.50 is not enough to be impressive for a trial that is not randomized and control.ed. Dennis, since I have to compute the p-levels myself, from scarce numbers in the text, I know that they are not an accident, or something forced upon me by an editor. In what fashion am I fooling myself or misleading myself, when I use the p-level to elevate my distrust? - Herman only suggests that my strategy might properly make me throw out *everything* because of too much multiplicity. You don't have the p-levels -- Do you have any guidelines to offer, that can serve *any* purpose? -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
