Dear Richard and Jesper,
Thanks for your comments. I've got a better sense of the relevant issues
now. In retrospect, I could have stated the question of interest a bit
more clearly. Although I think both of you got the gist the following
will hopefully clarify:
Briefly, the idea was to characterize the brain
activations associated with the same nominal psychological state in two
different groups. For example, how do the activations related to fear or
some other emotion compare between males and females? For some emotional
states we can use the same stimuli to elicit the states in males and
females and use the standard random effects analysis to characterize
differential patterns of activation. But for some emotions (for various
reasons) we are forced to use different (albeit visually similar) visual
stimuli to evoke the same emotions.
Here are the possible solutions so far:
1. The simplest solution appears to be to qualitatively compare the two
patterns of activation. So, taking fear as one example, we would have
something like: males activated areas X,Y, and Z in the A -C (A and B are
target emotion conditions, C is neutral) contrast between fear and
baseline whereas females activated areas W,X, and Y in the B - C contrast.
Disadvantage: the material differences are confounded with the emotion
contrast. However, if we data on what activations were related to the
material differences (for example, the regions associated with processing
spider pictures vs. toad pictures, when fear was absent) and we knew or
had good reason to believe that these areas did not interact with gender,
then one could use this information to help interpret the qualitative
analysis.
2. Jesper's suggestion to use a regression approach is good except that as
he noted, it will be highly insensitive unless we have enough subjective
variation in the "fearsomeness" of the stimuli within conditions. Perhaps
it would be better to completely do away with the conditions and simply
have an event related design in which various visual stimuli are
presented in a pseudorandom order -- to continue the example, toads,
spiders, snakes, miscellaneous vermin, as well as neutral stimuli. For
each stimulus the subject's subjective and physiological response would be
recorded. Then, these responses would be entered as regressors for each of
the groups (would one still need to orthogonalize here between the two
regressors?), and the parameter estimates would be taken to the second
level analysis to compare between groups.
3. Alternately, run a contrast for the group x condition interaction ,
taking A-B into a random effects between-group test (same for B-A). This
is definitely valid, although it does not address (4) below
4. Finally there is the interesting concluding comment from Richard:
>> However,
if you want to demonstrate that there is a DIFFERENCE in the pattern
of the fear response of arachnophobes to spiders compared with the
fear response of bufunophobes to toads, then you are interested in a
group by condition by brain region interaction, which needs more
thought (unless you are happy to leave it as a 'qualitative'
observation).
>>
Yes, this is the type of analysis I had in mind, as well as the group x
condition interaction in (3).
>>To convince us that brain regions X and Y show
significantly different response patterns, in that area X responds
more in A - B and area Y responds more in B - A, you should strictly
speaking use a statistical test in which data from one these regions
is directly compared with data from the other, not something that is
routinely done in SPM.
>>
Hmm, do you mean identifying ROIs and counting the number of voxels
activated at a certain threshold and comparing between groups? Could you
elaborate a bit more on this? Or, if there is a good reference that'd be
fine.
--Stephan
Stephan Hamann
Dept. of Psychology
Emory University
Atlanta, GA 30322
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|