According to Kirk's (1995) Experimental design: Procedures for the
behavioral sciences, "Scheffš's procedure is always congruent with the
omnibus F test.  If the omnibus F test is significant, there is at least
one contrast among the means that is significant according to Scheffšs
test, and vice versa." (p. 155)
 
 
 
Michael T. Scoles, Ph.D.
Associate Professor of Psychology & Counseling
University of Central Arkansas
Conway, AR 72035
501-450-5418

>>> David Epstein <[EMAIL PROTECTED]> 4/3/2007 2:43 PM >>>

On Tue, 3 Apr 2007, Rick Froman went:

> How would you interpret an ANOVA result where the F-test was
> significant but none of the multiple comparisons were significant in
> an HSD comparison?

Initially, I wrote a response as follows:

"Off the top of my head, I would say: An overall effect was detected,
but the sample sizes within individual cells were not sufficiently
large to attribute the effect to any specific pairwise comparison.

"Then I would talk, with due caution, about what it LOOKED like."

Then I did a Google search and found an explanation that's more
abstruse, but still seems to invoke the problem of small sample sizes:
<http://www.ats.ucla.edu/STAT/spss/library/manglm.htm> (scroll down to
the section "My tests don't agree!").  Note that the explanation does
*not* seem to be that you're missing "some more complicated contrast
among the means."

--David Epstein
   [EMAIL PROTECTED]

---
To make changes to your subscription go to:
http://acsun.frostburg.edu/cgi-bin/lyris.pl?enter=tips&text_mode=0&lang=english




---
To make changes to your subscription go to:
http://acsun.frostburg.edu/cgi-bin/lyris.pl?enter=tips&text_mode=0&lang=english

Reply via email to