[R-sig-ME] power calculations in mixed models

[Mark Kimpel]

One thing I forgot to mention.... We have a reasonable estimate of the
bounds of the biologic variance and have subtracted that out to get
the bounds of the technical variance. For estimation purposes, I am
assuming that the within sample replication variance will be the same
as the between sample technical variance. So for power calculations
the between sample fixed effects variance is the same as within-sample
random effects variance.

Mark

On Sun, Jun 15, 2008 at 3:33 PM, Mark Kimpel <mwkimpel at gmail.com> wrote:
> I have data from a micro-array experiment that has turned out to be
> underpowered. Classical power calculations based on observed variance
> of presumably non-differentially expressed genes and the delta that we
> would like to measure have led to this conclusion. We have enough of
> each sample to re-run the assays, thus obtaining technical replication
> within sample, which would be a random-effect. As a simplistic
> approach to predicting the power obtained when using technical
> replication, I have performed a simulation using averages for each
> sample. This demonstrates that variance is cut in half with
> replication.
>
> I assume that power would be further increased in a real experiment if
> all measurements are included in a mixed-effects model. Is that likely
> to be true? If it is, is there any way to predict, based on the
> knowledge we already have, the power of the data-set expanded with
> technical replication?
>
> Thanks,
> Mark
>
> --
> Mark W. Kimpel MD ** Neuroinformatics ** Dept. of Psychiatry
> Indiana University School of Medicine
>
> 15032 Hunter Court, Westfield, IN 46074
>
> (317) 490-5129 Work, & Mobile & VoiceMail
> (317) 663-0513 Home (no voice mail please)
>
> ******************************************************************
>



-- 
Mark W. Kimpel MD ** Neuroinformatics ** Dept. of Psychiatry
Indiana University School of Medicine

15032 Hunter Court, Westfield, IN 46074

(317) 490-5129 Work, & Mobile & VoiceMail
(317) 663-0513 Home (no voice mail please)