Hi Doug,
Thank you for your quick response. Concerning question3, do you mean that if
there is an interaction with age, it is all right to use DODS (without
stratification)?
Thanks, Maaike
Date: Wed, 24 Dec 2014 15:31:53 -0500
From: gr...@nmr.mgh.harvard.edu
To: freesurfer@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] DODS design and demeaning
On 12/24/14 5:21 AM, maaike rive wrote:
Dear Freesurfer experts,
I would be very greatful if you could help me with
three (probably basic) questions regarding an mri_glmfit
design.
For the design and contrasts I used FSGDF. I have a
2x2 factorial design with one covariate (age). I specified all
four groups (A1B1, A1B2, A2B1, A2B2) and the covariate in one
model, because I would like to test for the interaction
between A and B (regressing out the effects of age). Age was
demeaned using the mean of all subjects (i.e. of all 4
groups).
My first question regards the contrasts:
If, besides the interaction and main affects of A
and B, I would like to assess the difference between
individual groups, for example A1B1 - A1B2, is it okay to use
the same model (1 -1 0 0 + further 0's for the age
regressors); or should I make a new FSGDF containing only the
subjects of group A1B1 and A1B2?
It is a tricky question. You will get the same group means
regardless of whether you combine all subjects or use a 2nd FSGD.
The difference will be that you will have a much higher DOF if you
combine them all. This means that the noise in the other subjects
informs you about the noise in those two groups. If you can justify
this, then you can leave them combined. If the 2nd model gives you
the results you want, then reporting that would be a stronger
scientific result.
My second question is associated with the first:
If it is allowed to use the same model, how does
freesurfer deal withdemeaning? Since I used the mean of all
subjects to demean and not just of the subjects of for example
A1B1 and A1B2.
The demeaning is also tricky. I would actually test whether there is
an interaction between group and age. If there is no interaction,
then I would re-run using a DOSS model in which case the demeaning
won't make a difference.
My third question regards the DODS/DOSS issue.
I used DODS, because I do not believe the slopes of
age by volume (or age by area/thickness) will be exactly
parallel between groups. I tested this also be using contrasts
of the age regressors, (for example using the contrast 0 0 0
0 1 -1 -1 1 for a AxBxage interaction) and indeed there are
interactions with age in some brain areas. For thickness, they
were FDR corrected still significant, for other measeurs the
were not significant after correction, but that seems natural
given the sample sizes. No I wondered whether it would suffice
to use a DODS design (since slopes differ between groups) or
whether I should stratify groups further for age (for example
use two models, one for young subjects, on for old subjects).
If the latter is the case, I would end up with very small
samples, I am afraid.
If there is no interaction in the areas that show up in your
contrast of interest, then it is safe to use DOSS. If there are,
then I don't think that stratifying them solves the problem.
doug
Thanking you in advance,
Maaike
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail. _______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
