Hi,
I think the relevant things here are that:
- FDR is likely to be less sensitive than randomise
- the new version only looks for motion along the normal to the boundary, and so is univariate
- the old version looked for motion in any direction (including along the surface, although it is questionable how accurate that can be estimated) and was multivariate (which for only 3 dimensions is likely to lead to a loss of sensitivity too)
So it is certainly possible to see different results.
Personally, I'd trust the new results more.
It would be informative to look at the direction of the movement in the old results, to see if it is largely along the surface, rather than normal to the surface.
All the best,
Mark
On 14 Nov 2012, at 16:03, Ilya Veer <[log in to unmask]> wrote:
> Hello,
>
> I have done a between groups FIRST vertex analysis using the new as well as the old method. Randomise gives me nicely corrected results (with F cluster thresholding). In contrast, surface_fdr on the .vtk file does not give any corrected results. However, when I look at the uncorrected results of surface_fdr, there is a strong trend at a specific location. I would expect that this is the same location as came out corrected in the shape analysis based on randomise, but it wasn't. What exactly is the difference between the old and new vertex analysis? Do both answer (slightly) different questions?
>
> Best,
> Ilya Veer
|