Thanks for the reply!

I was wonder about the last statement from the previous post though.


“I don't know how people practically handle this situation, but the pivotal question seems to be whether the contrast (A - B) makes sense only when either of the two main effects is statistically significant. I don't have a clear answer about this, but if it is true, then masking is probably a legitimate way to go.”

I was wondering if we could get any more clarification as to whether the non-statistically significant A or B main effect voxels can be of any use to us.
If they are not of any use, then creating a ROI main effect mask and applying further statistical analysis to that would make sense. The hesitation comes out of what we know to be the typical source of ROI’s. That being a priori information from a different data set, not so much from the same data set.
So does this issue of creating an ROI from another data set or within the same data set come up when dealing with the statistical soundness of the subsequent output?

I will appreciate any response, as I understand the nature of my inquiry is dancing precariously close to the edge of knowledge and opinion.

Thank you so much for your time!

Jeremy