Hi Zarinah,
> I am trying to create inclusive masks of two different contrasts A and B
> in SPM5. I am finding that the maps are different when I mask A with B
> and when I mask B with A.. why is this
In terms of the statistic values, A-masked-by-B will have the t-values
of A, while B-masked-by-A will have those of B, so we must expect them
to be different.
But perhaps you mean that the regions reported differ? (In terms of a
binary mask of which voxels are significant and which aren't)
I assume that for the main contrast (the one being masked, rather than
the one doing the masking) you are using multiple comparison
correction? In which case the results would differ because a different
contrast would be the one being corrected.
I think if the main contrast is also uncorrected, then A-masked-by-B
and B-masked-by-A would show the same binary regions, identical to the
region that would result from the intersection of the regions found to
be significant for A and B, separately. I haven't checked this though;
perhaps you could follow up if you get the chance to check this on
your data?
> and how do I know which one is correct?
Well... Firstly, I'm not aware of much literature on contrast masking
-- perhaps someone else can help here? But I'd say that if you are
correcting the main contrast for multiple comparisons, then this
should be the one you are most interested in.
If you are interested in either/both equally, then I think you
probably want to test their conjunction, either against the global
null for "either" or the conjunction null for "both" significant.
There is a lot more literature on this (albeit with a somewhat
confusing historical development) --
http://scholar.google.com/scholar?hl=en&q=conjunction+friston+OR+nichols
I just stumbled across this post from Karl Friston, which as well as
discussing conjunctions, mentions the possibility of taking the
intersection of both A-masked-by-B and B-masked-by-A. It doesn't sound
from his post though that Karl looked into this formally, and I don't
know whether anyone has subsequently...
http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind05&L=SPM&P=R250495
I hope some of this helps, sorry it's not a very clear cut answer!
Ged.
|