Hi,
I have a main effect A and main effect B. When I look at the activation maps of these main effects I see reasonable areas of activation.
Now, when I do a contrast, say A-B, I get activation not only in the areas associated with the main effect maps, but also in other areas that were not active in either main effect maps.

I realize the nature of a subtraction is just to find the difference between correlation or intensity values, so it mathematically makes sense. Theoretically I am wondering if there is a problem with looking at contrast activations that are not associated with main effect activations. This problem would be “what does this mean?”

So is it statistically sound to create a combined mask at a predefined threshold from the main effects A and B, and limit a contrast to only those voxels? This would ensure the subtraction only left voxels that were significant in the main effects and would make interpreting the results easier in a theoretical sense.
I am just attempting to see if this type of ROI analysis is statistically and theoretically reasonable, and if there is an error in my thinking.
The answer may be one of opinion, but any advice would be much appreciated.

As always, thank you so much for your time!
Jeremy