Hi,
On 20 Feb 2008, at 23:49, Jessica Kirkland wrote:
> Hello,
>
> I have a few questions about HLA with covariates of interest, and I
> am a
> novice at doing this in FSL. My design is this: 1 group, several
> first-level
> contrasts of interest, and several questionnaire measures. What I
> would like
> to do is see how individual subject activation on the contrasts
> correlates
> with the questionnaire measures. I am switching over to FSL from
> AFNI...so
> previously I could do two things: correlate whole brain activation
> for all
> subjects on a contrast with questionnaire values, OR extract mean
> subject
> values for a contrast of interest using clusters from the group
> contrast as
> an ROI and then correlate those values with the questionnaire. I am
> really
> interested in setting up this latter case in FSL.
>
> I have followed the help list, and so far have set up my analysis so
> that my
> EVs are my group mean and my demeaned questionnaire scores. I am
> still not
> clear on how to set up the contrasts if what I am interested in is the
> correlation of the First-Level contrast and the questionnaire data.
> I am
> also not sure if masking with a thresholded first-level contrast is
> the way
> to get the results I'm hoping for.
Hi - it's certainly straightforward to put your cross-subject
covariates in a second-level voxelwise design, but it sounds like you
want something different: you want to look at each subject's first-
level analysis, and extract a mean COPE (contrast of parameter
estimates or betas) for a first-level significant cluster, summarising
each subject's first-level results by a single number. This is easy to
achieve by looking at the cluster tables; see the COPE MEAN column.
Alternatively, if you want more flexibility in extracting this
information, Featquery should be able to help.
Once you no longer have image data (because you have just one number
per subject) I presume you would then want to do your group-level
modelling in SPSS, matlab, etc?
> Finally, if I include multiple
> questionnaires, will I be changing the correlations of contrast-each
> individual score?
In general, yes, unless they are orthogonal to each other. if they
partially correlate with each other you need to decide whether the
question you want to ask is best addressed by a) only running one at a
time, b) putting them all in and leaving them as is, and c) putting
them all in and orthogonalising some wrt others.
Cheers, Steve.
>
>
> Group EV1 EV2 EV3 EV4
> 1 1 3.5 4 2
> 1 1 -2 3 -1
> etc...?
>
>
> Thank you for the help--I'm having difficulty putting what I need into
> words, and I realize this isn't the clearest explanation, so please
> let me
> know if there is more information you need.
>
> Jessica
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|