Re: Help with Post-hoc analysis for repeated measures one-way anova!

On 18 Mar 2006 16:20:11 -0800, "drammt" <dramirez@xxxxxxxxxxxxxxxx>
wrote:

I am conducting a series of Repeated Measures one-way ANOVA. I have
three groups two independend variables and one dependent variable. Whe
I do the analysis the Mauchly's Test of sphericity shows significance,
therefore I am using Greenhouse-Geisser. I have read in different
textbooks that if Spericity is not assumed, Tukey HSD post-hoc tests
are not the best option, and I should do t-tests correcting either with
Bonferroni or Dunn-Sidak. However I got into an argument with a
colleague becuse she thinks that The use of Tukey in this case is
valid. What is the correct way of performing the Post-hoc tests?

Not Tukey.

Also consider: Do you really have multiple hypotheses which
deserve "corrections"? Should you (say) be testing the overall
hypothesis using one of the variables, or a composite score?

Tukey HSD testing is defined for oneway ANOVA, where there
is a pooled variance and multiple groups. I do not think that
there is much acceptance of whatever you may have found -- if
you found something -- that pretends to adapt it to repeated
measures.

The most common advice these days about followup testing
in repeated measures is that you should use paired t-tests
for the contrasts. Bonferroni-type correction is possible.

Mauchly's Test of sphericity is extremely weak at detecting
or warning of problems with contrasts that have vastly
different errors. The relevant advice that I recall is that
if you want to use Mauchly's test, use the 0.50 (not the
0.05) p-value as a warning.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.