Arthur Tabachneck <art297@NETSCAPE.NET> wrote
>Let me preface my response with the fact that I am not a statistician. I
>only wish that David still posted to the list, as he not only is a
>statistician, but a very good one at that.
We all miss David.
>I have always heard that one can do something(s) with non-significant
>findings, but I have never accepted that concept.
>Since a pvalue is only a likelihood of a null hypothesis being wrong, I
>don't see how it could validly be used as a measure of a hypothesis being
>A simple example: I hypothesize that there aren't any yellow Toyotas in the
>world. I take 10 samples of whatever size and don't find any yellow
>That doesn't allow me to confirm that my null hypothesis was correct.
It's true that you can't confirm the null, but there are things one can do with non-sig findings.
A confidence interval around a non-sig. finding means the same thing as around a sig. finding.
A nonsig. parameter estimate has just as much validity as a sig. one, it's just not statisticially significantly different from zero.
Peter L. Flom, PhD
Website: http://www DOT statisticalanalysisconsulting DOT com/