first level statistics: follow up +++

Hello,

In order to compare the BETAS (ie Fisher transformed R values) of a seed-to-voxel analysis, in a single subject, between two different conditions, I just subtracted the two BETA maps, and divided by 0.198, which was the SD of the difference sqrt(1/(DOF1-1) + 1/(DOF2-3)), with DOF1 and 2=54, as provided by CONN in the "denoising" panel. the result is, I believe, a map of Z values, which I can threshold statistically (eg >3.02 for p<0.001). I then asked your advice, and you confirmed that it was OK, particularly the use of the DOF after denoising.

So far, so good. Now, I did exactly the same thing, but not with the BETAS from a seed-to-voxel analysis, but from the Global Correlation voxel-to-voxel analysis. I assumed that it was legitimate, as values are supposed to be just averages of R values with "all other voxels". It worked very well, in the sense that I get nice results. Almost too nice, ie with very high Z values. It may correct, but I have my doubts. So...

Do you think it is OK to do that, particularly
- is the same DOF still OK?
- do the options "Normalization" and "Dimensional reduction" in the computation of Global correlation change anything?
- ???

Thank you very much in advance

Laurent