Dear Stephan,
> I hope this is not too elementary a question, but after absorbing the
> whole SPM video, the notes, and combing the archives nothing has turned
> up (I have learned after 8 hours of watching that the SPM folks are sharp
> dressers, though), so here goes:
It sounds as though you may have missed the talk by John Ashburner, which is
a pity since it is really very interesting.
>
> We have three conditions, A, B, and a baseline condition C, and two
> groups. Group one has a particular response to A (for example, fear), but
> has no particular response to B. Group two has the same response (fear) to
> B but not to A.
>
> The analyses of interest are:
> 1) characterize the (fear) response in each group separately, taking the
> A - C contrast for group one and the B - C contrast for group two and
> performing a random effects analysis separately for each group.
>
> 2) Compare the neural correlates of the (fear) response between the
> groups. If both groups were responsive to the same condition (say, both
> had fear of A), this would be a simple matter of entering the A - C
> contrasts for both groups into a 2-sample t-test at the second level. My
> guess was that since the A - C contrast is psychologically the counterpart
> to the B - C contrast in the example, it would be valid to enter A-C for
> group one, A-B for group two, and do the 2-sample t-test. This would not
> control for activations related to specific material content in A and B,
> but this could be assessed with a separate contrast.
>
> The only alternative I can see is to qualitatively compare the two
> patterns found in each group (vs. performing a statistical test).
>
> So, in general the question is: when 2 groups have the same type of
> psychological response but in different conditions, how should one go
> about analyzing this?
>
As far as I understand you have two groups in which you have elicited the
same response (fear) but as a consequence of different stimuli. You now wish
to see if the neural correlates of fear is different in the two groups, but
your problem is that your parameter of intestest is completely colinear with
state (A v.s B) which in your case is a confound.
My understanding is that your only chance of disentangling the two is if you
have an independent record of fear for each scan (if it is PET) or
epoch/event if it is fMRI. This could be a subjective self rating or some
"objective" measure as skin conductance or hear rate. In that case you could
create a regressor from these responses, which you need to orthogonalise with
respect to the (two or three) categorical condition regressors. This
regressor would then give you the neural correlate of fear, after having
controlled for condition. It is very likely that the "rating regressor" is
almost in the space of your condition regressors, and your test will be very
insensitive but as far as I can see kosher.
Note that in this particular case you really need to explicitly orthogonalise
your regressor since you wish to enter your parameter estimates into a second
level analysis. If you had only wished to test the "rating regressor" within
your fixed effects model that would not have been necessary.
Your suggestion of comparing the A-C contrast from one group with the B-C
contrast from another sounds rather iffy to me.
>
> thanks,
> Stephan
Good luck
Jesper
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|