Dear Tugan and Roman,
The output from a FIRST vertex analysis is always an F-statistic.
It runs a multivariate GLM and this is the output of that MV-GLM
procedure (using Pillai's Trace). I'm not sure where you found the
sentence about "strength of correlation" but this was not meant to
indicate that it was a correlation coefficient - it was meant in a
non-technical sense, such that the F-statistic reflect the residual
correlation of the variable of interest after factoring out any other
confounding effects.
The short answer is that you will be seeing F-statistics, and you
can run FDR on this without problem.
You might also find the newly published paper on FIRST helpful
for clarifying how the multivariate GLM works. The paper details
are:
Patenaude, B., Smith, S.M., Kennedy, D., and Jenkinson M.
A Bayesian Model of Shape and Appearance for Subcortical Brain
NeuroImage, 56(3):907-922, 2011.
http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6WNP-527PRMC-2&_user=10&_coverDate=06%2F01%2F2011&_rdoc=8&_fmt=high&_orig=browse&_origin=browse&_zone=rslt_list_item&_srch=doc-info(%23toc%236968%232011%23999439996%233153734%23FLA%23display%23Volume)&_cdi=6968&_sort=d&_docanchor=&_ct=101&_acct=C000050221&_version=1&_urlVersion=0&_userid=10&md5=0880a49ea45171f2e3298d396fc83dc1&searchtype=a
All the best,
Mark
On 2 May 2011, at 21:31, Roman M wrote:
> re-post.
>
> Since I would like to hear the answer to that too..
>
> thank you
>
> Roman
>
> > Date: Wed, 20 Apr 2011 17:36:04 +0100
> > From: [log in to unmask]
> > Subject: [FSL] FIRST shape analysis, correlation and statistical interpretation
> > To: [log in to unmask]
> >
> > Hi,
> >
> > I wanted to clarify if I am interpreting the output of shape analysis correctly when I used correlation and how I can derive meaningful statistics.
> >
> > I am studying age associated changes and I get great looking results. But when it came to interpretation, I went and read the FIRST manual, the thesis, the paper and numerous email discussions. What I understand is, the color bar represents F-stats only if I did an analysis looking at group differences. If I am exploring shape changes that are correlated with age, the colors "represent the strength of correlation". But it does not look like any correlation coefficient and I presume an F-statistic was not derived from the CC, either. Is there a way to derive a meaningful (publishable) statistics off of this? It seems that I cannot run FDR on it, either, since the outcome is not a statistic image.
> >
> > I will be grateful if someone can shed a light on this.
> >
> > best,
> >
> > Tugan
|