Dear SPM experts,
I have a question concerning analysis and interpretation of fMRI-data in a multiple regression design.
I present participants with words, that vary on 4 psycholinguistic variables and am interested in the brain regions that are sensitive to these variables.
Thus at the 1st Level my GLM contains all 4 parametric variables as predictors. The 4 variables are moderately correlated between each other, so the design is not orthogonal per se. I do not orthogonalize the design, because I am interested in the unique variance explained by each and there is no clear order in which they should be entered in the model.
Here are my questions:
- In a whole-brain analysis I find that brain regions A,B,C,D are selectively correlated with variables 1-4 respectively (beta >0). Can I conclude that region A processes variable 1, but not variable 2 based on this information only? Or do I have to look at the contrast directly comparing variable 1 and 2 (that is [1 -1 0 0]) on the second level?
- If the latter is the case, what is the correct way to do that? I find it problematic to do the usual paired t-test, since the two variables are based on the same trials, thus nested.
When I try to transfer the logic of analysis for behavioral data, I find that this is a comparison never one. One would do a multiple regression analysis of, say, RT, and be happy to say that variable 1 but not variable 2 was significantly correlated with RT. One would not compare the correlation between RT and Variable 1, to the correlation between RT and variable 2, since they are dependent.
Thank you for your thoughts,
Yulia Oganian
|