 |
 |
On Mon, 13 Jun 2005 16:11:56 -0500, Li, Yu <[log in to unmask]> wrote: > >Dear SPMers: >In fMRI box-car group study, we want to see correlation between each >subject's activation map with 2 clinical scores A or B. A and B are >suppose negative correlated. I did following 2 tests: >1. I chose the "simple regression" and enter each subject's "con*.img" >and corresponding "A" as regressor. >2. I chosed the "multiple regression" and enter corresponding "A" & "B" >as regressors. >The result from 1. with contrast "-1(A)" is different than from 2 with >contrast " -1(A), 0(B)". Could someone please explain? If A and B are negatively correlated, the design "2." has multicollinearity, i.e., A and B are partly confounded with each other. This will make the results differ as you've noticed. A commonly cited discussion of this in the context of neuroimaging is "Ambiguous Results in Functional Neuroimaging Data Analysis Due to Covariate Correlation," Alexandre Andrade, Anne-Lise Paradis, Stephanie Rouquette, and Jean-Baptiste Poline, NeuroImage 10, 483–486 (1999). >Many Thanks!! > >Julia > >