I'd like to make a correction on my statement below. FDR is not biased by having dependence in the data (though it may make it more variable). Thanks to Mike Harms and Tom Nichols for pointing this out.
doug
Douglas Greve wrote:
Hi Petra, if the data are independent, then large amounts of smoothing does not affect the validity of the statistics (it helps cluster-wise correction). If the data are not independent, then the problem is that the voxel-wise p-values will be inaccurate (causing clusters to be too large or too small, and messing up FDR). If you don't know something about the dependence, then I don't know of how to fix it. If you know something about it, then you might be able to incorporate it into the GLM (as is done with time series analysis of fMRI) or include it in a simulation. Maybe someone else on the list has a better idea?
doug
On 3/29/11 8:18 AM, Habets P (SP) wrote:
Hi,
This is probably a question for Doug Greve,
Which is to most preferably correction for multiple comparisons when investigating surface based cortical thickness changes between groups in smoothed data. Furthermore subjects between groups are related in other words not independent from each other.
It is not clear for me whether FDR or cluster wise correction is feasible in highly smoothed and data that does not fulfill the criteria of independence.
Best wishes,
Petra Habets
**
*Petra Habets * /PhD-Student School for Mental Health and Neuroscience /_p.habets@maastrichtuniversity.nl mailto:p.habets@maastrichtuniversity.nl _www.maastrichtuniversity.nl http://www.maastrichtuniversity.nl
Vijverdalseweg 1, 6226 NB Maastricht P.O. Box 616, 6200 MD Maastricht, The Netherlands T + 31 43 368 8659 F + 31 43 368 8689
*Note: the email addresses have changed!*
*Please consider your environmental responsibility before printing this e-mail.*
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer