I have a question about AM2. When I used it to covariate out the semantic distance, it gave me the brain map which are not sensitive to the semantic distance (labeled as #1). I wonder if AM2 allows separation of voxels that are active but are not detectably modulated by the semantic distance from voxels that are sensitive to semantic distance. if so, then the coeffecient of each voxel should not be changed before vs. after I used AM2. The only difference between the two maps (semantic distance sensitive or not sensitive maps) is the number of voxels. However, when I did the contrast between two conditions (e.g., A-B) using the coeffecient labeled as #1, I got a very different brain activation map (showing greater activation in condition A) compared to the brain map without regressing out the semantic distance (showing greater activation in condition B). So could you explain how the AM2 regress out the semantic distance in statistics? I know it is a complicated question. Thanks for your patience.