Dear Heike,
This sounds like you are talking about a fixed effects group analysis as you
appear to build a huge design matrix containing all your subjects in order
to do your group statistics.
The good news is that you have a good number of subjects to base your group
analysis on. Rather than doing the above type of group analysis, which can
be biased/driven by a few subjects with a large effect in a particular area,
a different type of group analysis might be appropriate (of course, this
depends on your paricular study and purpose which is not further specified):
conjunction, mixed effects, random effects... .
If you look at Friston 1999 "How many subjects constitute a study", this
should get you started.
Briefly, e.g. if you wanted to perform a random effects analysis (see above
paper), then you could analyse every subject individually and specify the
single subject [T-] contrast according to the inference you want to make
(both at the group and the single subject level).
Every contrast will be assigned a number (you'll find it to the left of the
contrast name
in the contrast manager, and also on the results page). In the analysis
directory, you will find so called con-images according to that contrast,
like
con[contrast-ID].img and
con[contrast-ID].hdr
example: con002.img; con002.hdr
Use this in a 'Basic Model' , 'One sample t-test'.
Select the con images of all subjects you want to include, leave the default
settings [in the first place], estimate and specify a +1 or -1 T-contrast.
Hope I understood your problem correctly and this is a starter! There should
be plenty more on the mailing list, though.
Best wishes,
Helmut
----- Original Message -----
From: "Heike Schmidt" <[log in to unmask]>
Sent: Friday, April 28, 2006 3:20 PM
Subject: Groupmaps
> hi,
> Just got a question that someone might be able to answer. I wanted to make
> an fMRI groupmap of 17 subjects but this didn't seem to be possible as the
> program kept on crashing at the model estimation stage. The data attached
> to the model was for 17 subjects each with scans comprising of 320
> slices. The model itself was ok as it has ran for another sample of 19.
> So my question is, can SPM handle such a large sample?
>
>
> Also I'm wondering if there's a way to combine the two seperate groupmaps
> of 8 and 9 subjects and to analyse them as if it was a group of 17 by
> grouping the two together. Does anyone know a way to do this?
> Thanks,
> Heike
>
|