Helmut,
I’m curious. Does the same apply for cluster level FWE?
Regards,
Jeff
On Jul 13, 2015, at 1:39 PM, H. Nebl <[log in to unmask]> wrote:
> Dear Elena,
>
> FDR on voxel level has been criticized by Chumbley and Friston (2009, Neuroimage) and Chumbley et al. (2010, Neuroimage), so you should rely on topological FDR on cluster level or topological FDR on peak level if you want to go with FDR. This is also the reason why voxel FDR can't be chosen per default any longer, it has to be enabled by adjusting spm_defaults.
>
> For any of the two topological FDRs you first need a "feature-defining" threshold, which would be an uncorrected voxel threshold, e.g. the default .001. Don't go with an initial uncorrected voxel threshold of .01, this is much too liberal. This feature-defining threshold has to be reported as well, as it makes a difference whether you go with .001 or .99. The FDRc always corresponds to the size of the smallest, significant cluster surviving a corrected cluster q = .05 in combination with the original uncorrected voxel threshold. If you want to go with a corrected cluster q = .01 then you have to look at the different q values and determine the size of the smallest, sig. cluster with q < .01 yourself. So something like
>
> "To control for multiple testing a topological FDR on cluster level was applied at q = .05 corrected, based on an initial feature-defining voxel threshold set to p = .001 uncorrected".
>
> should do the job.
>
> Hope this helps
>
> Helmut
-----------------------------------------------------------------------
Jeff Browndyke, Ph.D.
Clinical & Research Neuropsychologist
Durham VA & Duke University Medical Centers
[log in to unmask] / [log in to unmask]
office: (919) 286-0411 ext. 4656
cell: (336) 264-4222
|