 |
 |
Hi Roman, Sorry, but you can't have everything -- The way to go is masking C3 with C2, and indeed, a voxel that has Z=-0.000001 will be lost. In principle it is possible to have a contrast that instead of significant if larger than zero, is significant if larger than some value "x", but to have this done in practice, you'd have to have a sensible and justifiable way to define "x", and would have to manually add it to the COPE, and recompute, by hand, the test statistic. If you're finding difficult to write the methods section as is, I think it will be even harder to justify not only the need for "x", but also the value used for it, and the fact that the COPEs would have been edited manually. I would leave as is. Cheers, Anderson On 16 November 2015 at 19:40, Roman M <[log in to unmask]> wrote: > Dear All, > > yes I know this topic has come up before, but as I am writing a methods > section it occurs to me that I am no longer clear on the appropriate way of > applying post-thresholding contrast masking. > > The scenario is farily strightforward: > > I've got 3 EVs: > C1=CondA v Baseline > C2=CondB v Baseline > C3=C1>C2 > > Now, to avoid reverse activations I'd like to use contrast masking on C3. > However: > if I only mask it with C1 (Z>0) then I could get voxels active because > they are Z=0.00001 in C1, but Z=-10 in C2, so no good. > If I only mask with C2 (Z>0) then I could lose voxels that might be Z=10 > in C1, but happen to be negative in C2, even by just very little, so also > no good. > If I mask with C1 AND C2 (both Z>0) then it means that I'm only looking at > voxels which have Z>0 for BOTH conditions, so if a voxel has Z=10 in C1, > but Z=-0.000001 in C2 I would lose that voxel, which is also no good! > > For this reason, I am having a hard time justifying any of these three > approaches in my methods section. > > Is there a proper way to use this tool to avoid getting reverse > activations without losing voxels? > > Cheers > > R >