R help - Apr 2007 - Power analysis and mixed model

Un texte encapsul? et encod? dans un jeu de caract?res inconnu a ?t? nettoy?...
Nom : non disponible
Url :
https://stat.ethz.ch/pipermail/r-help/attachments/20070404/0f61f54a/attachment.pl

One option is to use lmer in a monte carlo simulation. I just did this last
week. Check out the article published in the American Statistician and can be
found at http://maven.smith.edu/~nhorton/R/r.pdf.

The article is not about power per se, but is about R as a toolbox for
mathematical statistics. But, it has a section illustrating how to do power
calculations via simulation using the nlme package in R, which is the function
for multilevel models.

I simulate data for clustered designs differently than the authors and I also
use lmer2() in the lme4 package and not nlme as it is much, much faster.

We needed to know how many clusters to have to obtain sufficient power. I ran
power calculations for a decently complex model with cluster sizes of 10 to 200
in increments of 10 (i.e., 10, 20, ..., 200) with 1000 iterations at each
cluster size. This is a total of 20,000 iterations (20 * 1000) and took about
1.5 hours on my desktop machine.

There are power calculators, like Optim, but simulation gives you much more
flexibility and the option to estimate power in situations where formulae do not
exist.

Harold
> -----Original Message-----
> From: r-help-bounces at stat.math.ethz.ch 
> [mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Julien
> Sent: Wednesday, April 04, 2007 10:40 AM
> To: r-help at stat.math.ethz.ch
> Subject: [R] Power analysis and mixed model
> 
> Hi 
> 
> Is there any way to compute power analysis for a mixed model ?
> 
> Julien
> 
>  
> 
> Julien Martin
> 
> Candidat au Doctorat
> 
> Laboratoire de Marco festa-Bianchet
> 
> Universit? de Sherbrooke
> 
> Sherbrooke, Qu?bec
> 
> J1K 2R1 Canada
> 
> 819 821 8000 poste 62059
> 
>  
> 
> 
> -- 
> 
> Checked by AVG Free Edition.
> 
> 12:48
>  
> 
> 	[[alternative HTML version deleted]]
> 
>

-----Original Message-----
From: Jean G. Orelien [mailto:jorelien at scimetrika.com] 
Sent: Wednesday, April 04, 2007 11:22 AM
To: 'Julien'
Subject: RE: [R] Power analysis and mixed model



Julien,

You could write your Linear Mixed Model as a General Linear Multivariate
Model (GLMM) and use the power software develooped by Muller, Lavange, Ramey
and Ramey (JASA, 1992).  

The program is written is SAS IML. If you can't find it by "googling
it",
let me know.

Jean Orelien
-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Julien
Sent: Wednesday, April 04, 2007 10:40 AM
To: r-help at stat.math.ethz.ch
Subject: [R] Power analysis and mixed model

Hi 

Is there any way to compute power analysis for a mixed model ?

Julien

 

Julien Martin

Candidat au Doctorat

Laboratoire de Marco festa-Bianchet

Universiti de Sherbrooke

Sherbrooke, Quibec

J1K 2R1 Canada

819 821 8000 poste 62059

 


-- 



12:48
 

	[[alternative HTML version deleted]]

Like Harold, I have also used simulation-based methods for power analyses, which
I discuss (briefly) in the following article (with R code in an appendix):

Atkins, D. C. (2005).  Using multilevel models to analyze marital and family 
treatment data: Basic and advanced issues.  Journal of Family Psychology, 19, 
98-110.

However, depending on the nature of your design, it could be far easier to use 
Steve Raudenbush's Optimal Design software with menus and GUI, downloadable
at:

http://sitemaker.umich.edu/group-based/optimal_design_software

cheers, Dave
-- 
Dave Atkins, PhD
datkins at u.washington.edu


From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Julien
Sent: Wednesday, April 04, 2007 10:40 AM
To: r-help at stat.math.ethz.ch
Subject: [R] Power analysis and mixed model

Hi

Is there any way to compute power analysis for a mixed model ?

Julien



Julien Martin

Candidat au Doctorat

Laboratoire de Marco festa-Bianchet

Universiti de Sherbrooke

Sherbrooke, Quibec

J1K 2R1 Canada

819 821 8000 poste 62059