Galaxy,
> I'm still wondering how to set a proper threshold for a given
> effect. In my study, if i set P< .05(corrected),the corresponding T
> threshold is about 10.5, it seems too strict.
Since the interpretation of a T value depends on the degrees of
freedom, I can't tell if 10.5 is a high threshold. For example, if
you have lots of degrees of freedom, the corrected threshold for
100,000 voxels (based on Bonferroni) is 4.9, while if you only have 10
degrees of freedom the threshold is 10.5.
(Be sure that you have all the SPM99 updates, found in
ftp://ftp.fil.ion.ucl.ac.uk/spm/spm99_updates
Since the random field thresholds are sometimes more conservative than
Bonferroni, it's important to have the Bonferroni updates posted June
2001. See
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0106&L=spm&P=R15124
But since you mention FDR, you probably have these updates.)
> I know there is also other method for correction, such as FDR,
> still, we will meet such question, say, if i set P<.05 (FDR
> corrected), then the T threshold is about 4. So my question is, can
> we just set the T threshold, if so, which value is more reasonalbe.
Since the degrees of freedom, image smoothness, and search volume will
vary from study to study, it's impossible to give a T threshold that
will always work. That is why SPM has these fancy corrected threshold
methods to deal with this challenging problem.
As for using FDR vs. the traditional FWE-controlling, it is a matter
of choosing what multiple false positive measure you're comfortable
with. With FWE there's no more than 5% chance of *any* false
positives, anywhere. With FDR, on average, no more than 5% of your
suprathreshold voxels will be false positives. FDR is more powerful,
but accepts more false postives. For more on FDR see
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0106&L=spm&P=R15237
Hope this helps.
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|