Re: one way anova overall significant, but dunnett post hoc has no single significant group?

Posted by David Marso on Oct 18, 2012; 11:23pm
URL: http://spssx-discussion.165.s1.nabble.com/one-way-anova-overall-significant-but-dunnett-post-hoc-has-no-single-significant-group-tp5715709p5715756.html

The way I recall Scheffe is that it pertains to the set of all possible linear combinations  of means, not simply the very small subset of pairwise contrasts.
http://www.itl.nist.gov/div898/handbook/prc/section4/prc472.htm

Rich Ulrich-2 wrote

I've always thought of the Scheffe procedure in the opposite
direction.

That is:  It sets a high standard for the individual contrast,
so that if there is *any* contrast that is significant, then the
omnibus test has to be.  (Basically, the contrast will account
for the full Sum of Squares needed for the overall test and d.f.)

--
Rich Ulrich

> Date: Thu, 18 Oct 2012 17:22:33 -0500
> From: [hidden email]
> Subject: Re: one way anova overall significant,              but dunnett post hoc has no single significant group?
> To: [hidden email]
>
> There is only one post-test comparison procedure that guarantees a significant result if you have a significant omnibus test and that is Scheffe. It, however, controls for all comparisons, including complex (ie. more than two means compared like (m1 + m2)/2 - m3 = 0).
>
>
 ...