On Thu, 6 Dec 2007 12:09:02 -0500, Maria Densmore
<[log in to unmask]> wrote:
>Hi,
>
>I have an fMRI study to analyze covarying for age. We have 2 groups,
>control(n=20) and patient(n=14) each with 8 conditions. These conditions
>are modeled in the 1st stage fixed effects level for each subject
>(=con.img).
>Is it reasonable to model all the condition and subjects in one anCova
>model in the second stage?
>Would I state 2 groups, 1st has 20*8=160 con.img files the 2nd group
>having 14*8=112 con.img files ?
>Or would I state 16 groups?: condition1group1 condition2group1
>condition3group1.....condition1group2 condition2group2.....
>
>Is it more correct to compare each condition within a 2 group design
>seperately, so I would run 8 seperate ancova analyses.
You might want to consult "ANOVAs and SPM", by Henson and Penny, at
http://www.fil.ion.ucl.ac.uk/~wpenny/publications/rik_anova.pdf
I think it's also available in almost the same form as a book chapter in
one of the SPM books.
If you include more than one image per subject at the group level, you
have a repeated measures design (at the group level). This is a textbook
statistics issue, and is dealt with by software packages like SAS and
SPSS, but dealing with the issue at the group level in SPM is a little
complicated, though doable.
One thing you have to decide is whether you want to do "partitioned error"
or "pooled error," as the previous respondent to your post stated. The
traditional treatment in textbooks, as far as I can tell, is to use a
partitioned error, but it's more complicated in terms of setting things up
in SPM. (This topic is discussed in the Henson/Penny monograph.) On the
other hand, my impression from browsing the list is that many people in
the SPM community are comfortable with using the pooled error.
>Thanks for your time and attention.
>Maria
>=========================================================================
|