Hi Andrea,
> significant but also very similar pattern of activation for all three
> conditions (A, B, and C). This is also confirmed by the lack of robust
> results when subtracting C from A and B. (1 0 -1; 0 1 -1).
[...]
> However, when trying to visualize the A (1 0 0) and B (0 1 0) contrasts
> exclusively masked with the C contrast (0 0 1) I obtain very robust and
> interesting results.
I don't think this is valid, since regions which are just below
significance (at your masking level) in C could be only just above
significance in e.g. A; these could show up in this analysis, without
actually being significantly different. There are some related
arguments in Jernigan et al's "More mapping..." paper:
http://doi.wiley.com/10.1002/hbm.10108
Hopefully others will chip in if they disagree.
What happens if you try your approach but using A or B to exclusively
mask C? Presumably if you still find similar patterns, that would make
you less confident in your results for A or B masked with ~C... If you
don't, well, maybe you could argue something... but I don't think it
would be statistically convincing.
Hope that helps,
Ged
|