Dear Danilo,
> I have some questions about FDR theory
> - are there any assumption on the test distribution in order to apply
> FDR procedure to control for multiple comparisons?
There is an assumption about the spatial autocovariance of the
statistics (in this case the residuals of the GLM). If the correction is
based on the Benjamini-Hochberg procedure (such as in spm_P_FDR.m) then
this assumes the statistics are independent, or positive regression
dependent within each subset (PRDS). Quite a theoretical thing, but it's
safe to assume this is true for smoothed gaussian random fields.
see:
http://www.sph.umich.edu/~nichols/FDR/spm_P_FDR.m
> - In case there is some dependancy between FDR threshold estimate and
> the test distribution from which p-values are derived, what happen when
> the test distribution to threshold is very skewed or super/subgaussian ?
FDR correction works on the p-values, so if you use a test for a
gaussian distribution when the noise does not actually follow a gaussian
distribution, the p-values will be biased as well (if your distribution
has a heavy righthand tail there will be more false positives).
Best wishes
Alle Meije Wink
--
Brain Mapping Unit
Dept. of Psychiatry
University of Cambridge
Downing Site
Cambridge CB2 3EB
tel: (+44) (0)1223 764676
fax: (+44) (0)1223 336581
[log in to unmask]
|