Dear SPM experts,
I would like to hear your opinion about F-contrasts / factors with more than two conditions. Basically one would set up an F-contrast and then run post-hoc t-tests inside significant clusters (?) to find out whether the effect is due to A vs. B, A vs. C, B vs. C.
I got the impression that F contrasts are quite uncommon though. ANOVAs are reported regularly for the behavioral data, but the very same papers do not care anymore when it comes to the fMRI part. Rather one seems to run a bunch of planned (well...) t-tests right from the start. Others set up an ANOVA but run t-tests across the whole brain inside this model (to incorrectly increase the df?).
Now I wonder:
1) Where does this attitude against F-contrasts come from?
2) Do you know any papers that might serve as a guideline how to correctly treat factors with more than two levels and use a reasonable approach, e.g. F-contrast and then t-tests?
Finally some general problem specific to F-contrasts:
3) How to define a signficant cluster which could be tested with post-hoc t-tests? There is nothing like a p-value on cluster-level.
Best,
Gabor
|