Re: st: repeated measures analysis: random mixed models, GEE and power analysis

[David Airey <david.airey@Vanderbilt.Edu>]

On Dec 14, 2007, at 6:50 PM, Diego Bellavia wrote:

> Dear Statalisters,
>
> I am writing a grant proposal (time is an issue here)
> and by study design I will  have to analyze serial measurements
> in three predefined groups of patients (the outcome variable is  
> continous).
> The time between each measure should be uniform in all the patients.
> I should have no problems to enroll more than 40 patients per group.
I'm curious as to when GEE would be chosen instead of a mixed model?  
In Stata there are some features missing in xtmixed versus xtgee (vce:  
autoregressive, stationary, nonstationary, for example), and this may  
cause the decision to be made in favor of xtgee for Dr. Sebastian's  
design. But assuming xtmixed had the features of sas proc mixed or R  
lme, when would the choice for GEE win out? I thought I had read  
somewhere that GEE could be robust to vce misspecification versus  
mixed models, so this could be a reason even when comparable vce  
features are available. Note that with just one level of clustering,  
xtreg can be used, and xtreg does have the vce types above that are  
not available in xtmixed. One point not mentioned, and not relevant to  
this design, is xtmixed can do multilevel but xtgee and xtreg can  
handle only one cluster level. Dr. Sebastian did not say why he had  
serial measures, but the choice for population average effects or  
subject specific effects may tip the balance away from xtgee, given  
Barth's comments.

-Dave

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/