Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: st: Re: Repeated Measures ANOVA vs. Friedman test
From
Steve <[email protected]>
To
[email protected]
Subject
RE: st: Re: Repeated Measures ANOVA vs. Friedman test
Date
Tue, 22 May 2012 14:49:01 -0500
Dave, Joesph, Rob and Nick,
Thanks for all your replies. As you might have gathered, my main
concern was whether the small sample size would be compatible with a
RM ANOVA. Since this was panel data - and in the past I've used
-xtmixed- for it, I was leaning a bit towards that but was curious
about other analytic possibilities.
Also, Joesph, thanks for the detailed post with the info on the
diagnostic plots. I had looked at those previously and there weren't
really any major outliers. If anything the data seemed a bit
uniformly distributed. However, I think the most important point you
made was the fact that since this is a pilot study, I should stop
fretting about the results and focus on the quality. :) Incidentally,
using parametric vs non-parametric or even looking at differences
between values vs factoring in time all pretty much gave the same
result with p-values ranging from 0.50 to 0.25
Once again, thanks for all the help, guys.
- Steve
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Airey,
David C
Sent: Tuesday, May 22, 2012 11:08 AM
To: [email protected]
Subject: RE: st: Re: Repeated Measures ANOVA vs. Friedman test
.
That's true; the contrast postestimation command will use a t after
regress/ANOVA but a z after xtmixed. On the other hand, the compound
symmetric covariance for regress/ANOVA may not be a good choice, and
the univariate corrections for departure from the sphericity
assumption are approximate. Also, with other than designed
experiments, missing time points are likely, and regress/ANOVA does
casewise deletion whereas xtmixed will retain cases missing one time
point.
> One more thing - if you can "get away with" the assumptions for ANOVA and everything is balanced, it should provide closer to nominal test levels than -xtmixed- because of the finite degrees of freedom for error in the former vs. asymptotic Z- statistics in the latter.
>
> Al Feiveson
>
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/