I forgot to attach this to the mailing list when I replied earlier today.
Dear Andreas
Re:
>Dear SPM,
>so far nobody answered my questions, so I'll try it a second time.
>
>Using fMRI I have scanned 16 subjects from 3 different groups: 9
>subjects, 3 subjects and 3 subjects, respectively.
>The first group consists of "normal" subjects , the other two are
>composed of "atypical" subjects.
>It is rather difficult to find such "atypical" subjects. That explains
>the different group size.
>
>I have the following hypothesis:
>Although the activation pattern of the subjects belonging the "atypical"
>groups is different from the activation pattern of the "normal" group,
>they all have some cortical areas that are used by subjects from all
>groups.
>
>What sort of analysis is appropriate?
>
>I can think of the following:
>1. A conjunction analysis for each group to get the activation pattern
>that is typical for each group.
>2. Comparing the activation pattern of each group...
>
>Any comments are appreciated,
>
>Andreas
>
The conjunction analysis will give you the commonalities across groups.
If you want to compare the activation pattern between groups, you need an
interaction and a random effects analysis. You don't have enough subjects
for this.
Masking is useful when
(i) the contrasts, where you are looking for commonalities, are not
independent (non-orthoganol) thereby excluding the option of a conjunction
analysis. In this case, you could compute the overall main effect and mask
it with the effect of each group. Conjunctions are better (when you have
orthogonal contrasts) because the probability is adjusted to take into
account that you have "conjoint" effects ie, the probability that you get
an effect in contrast 1 and contrast 2 and contrast 3 is less than if you
sum over contrasts. Therefore, your results should be more significant with
a conjunction.
(ii) if you wanted to look at those voxels of the interaction that were
also activated by your main effect.
Best wishes
Cathy
|