On 08/23/2016 03:18 AM, maaike rive wrote:
Dear Douglas,
Sorry to bother you with this, but after reading the thread about the Eklund papers I'm getting confused about what I actually did when correcting for multiple comparisons and I'd also like to know whether I'm on the safe side.
I used pre-cashed Monte Carlo simulation with a cluster finding threshold of 0.05 (at least so I thought):
mri_glmfit -sim - -glmdir - -cache 1.3 abs - - cache-dir - - cwpvalthresh 0.01
Am I correct?
This command line will use a voxel-wise threshold of p<.05 (-log10(.05)=1.3) and will report clusters that are significant at p<.01 (rather than the usual p<.05).
And am I also correct that the FDR may be about 13% instead of 5% when doing this?
It is hard to say without knowing what the smoothness level is. For my simulations, it was aabout 13%, but this was also thickness data. As you point out the lGI is much smoother.
(I smoothed thickness and surface area data with FWMH 10mm, but not the lGI data, because they were already very smooth and pre-cashed simulation didn't work after additional smoothing).
Thanks, Maaike