Dear Jérôme,
> We performed a multi-group analysis in SPM99 in order to test
> contrasts interactions between 2 groups (healthy vs patient) in a 6
> conditions (N,H,A,B,C,S) PET study. N and H are control conditions for
> the S activation condition
> A is control condition for B and C activation condition.
> The interactions we have done are: I1=(S-N)healthy - (S-N)patients
> I2=(C-A)healthy - (C-A)patients
>
> We would like to know if it is possible to combine a conjunction
> analysis and an analysis of the Group by Condition interaction, in
> order to answer the following question: for which brain regions is
> there a significant group by condition interaction, whether the
> conditions contrasted are S and N or C and A. In other words, can we
> perform the conjunction analysis between I1 and I2?
>
> P.S: We prefer not to use a random effect analysis because of the low
> degrees of freedom with this kind of analysis (8 suvjects / group).
Yes you can. These contrasts are orthogonal and it is very
straightforward to do a conjunction of I1 and I2. Simply select both
contrasts in 'results' and the ensuing SPM{Tmin} will tell you where
both the interactions were signficant.
However, if you really want to find brain regions in which interactions
involving S vs N OR C vs A (as opposed to S vs N AND C vs A) then a
conjunction analysis is not appropriate. In the language of set theory
a conjunction is the 'intersection'. The inclusive OR correpsonds to
the 'union'. If you want regions that show an interaction using S vs N
OR C vs A OR both, then this could be tested with a SPM{Tmax}. I am
afraid the Gaussian Feild theory for this sort of SPM has not been
derived. But do not worry, there is a much simpler approach - use the
SPM{F}. This involves putting both your T contrasts into a F-contrast
with 2 rows. The resulting SPM{F} will test the null hypothesis that
contrast 1 and contrast 2 were not significant (i.e. one or the other
or both were).
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|