Re: st: power repeated measures anova vs mixed models

["Airey, David C" <david.airey@vanderbilt.edu>]

.

I'm confused how someone could answer your question without as many qualifications 
as assumptions you feel uncomfortable making? The question is sincere. Just trying 
to understand. Maybe all those choices in the software are the reality. What effect 
did you test? The group x time interaction? You don't say what the hypothesis was. 
You say you don't want to assume what you don't know about the covariance structure 
or variance or measurement error given the lack of pilot data, but you want to know if the 
RM ANOVA is more powerful than the "mixed model". Did you mean -xtmixed-, because 
-anova- can certainly do mixed models? You can reproduce the RM ANOVA results 
(except t versus z tests for contrasts) by assuming a specific correlation structure, 
etc. From my understanding, the split plot with sphericity assumption is a subset of 
what xtmixed can do. So I'm assuming you would get the same answer with the same 
model, using either -anova- or -xtmixed-, unless you made other assumptions that 
made the model different than -anova-. Austin's paper does mention ignoring a level
of the hierarchy, but I doubt that is relevant in this situation which is a designed
experiment.

Cheers,

-Dave


> You are missing the point. I have the sample size (n=65/group), power (80%)
> and alpha (5%), 3 groups and 6 time points. What I want to compute is the minimal
> detectable effect size. I did the power analysis using a repeated measure ANOVA and
> obtained the minimal detectable effect sizes assuming various correlations between
> the repeated measurements. What I want to know is whether the mixed model would
> have more power to detect these effect sizes? 
> 
> Ricardo Ovaldia, MS
> Statistician 
> Oklahoma City, OK


*
*   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/