Hi,
There are various valid options here. If you don’t care about small negative values in B, then you can just mask the A-B contrast with A>0 on its own. That is probably what I would do, just in terms of simplicity. However, your alternative of forming the A+B contrast and masking with A+B>0 is also fine. The main difference between them would be in cases where A was moderately large and positive but B was quite large and negative. In that case A>0 would still show the result but A+B>0 would mask it out. So it depends on what you care about.
All the best,
Mark
> On 5 Nov 2014, at 23:39, John Coetzee <[log in to unmask]> wrote:
>
> Hi Eugene,
>
> thank you for the link; but could you please clarify one issue for us:
> In the practical it says that to ensure no "reverse activations" one should use Z>0 masking on BOTH A and B. So, for the contrast A-B, you would only include voxels if A>0 AND B>0 (and of course to be significant A-B>0). Doesn't this mean that if a voxel had a value of 1000 in A, and a value of -1 in B it would get masked out, leading us to miss it (declaring it a reverse activation)?
>
> Would an alternative be to create a mask where you force "A + B > 0" ? With this mask you would not be requiring A and B to be jointly > 0 but the more B is negative, the more A has to be positive to "compensate" therefore avoiding reverse activations.
>
> Just a thought..
>
> cheers
|