 |
 |
Thanks for your reply, David. The thing is that the *tfce_corrp_tstat* files don't show any significant effect. Any thoughts on to what extent the uncorrected but TFCE based p-maps would indicate that there is indeed an effect, even if the formally corrected p-maps don't show any? I mean, the uncorrected maps show clustered effects at logical locations and therefore look very plausible. Or is TFCE never a surrogate for real multiple comparisons correction? After reading Steve's TFCE paper I think it's not just a cosmetical operation but adds real sensitivity... Best regards, René On Thu, Apr 26, 2012 at 3:56 PM, David V. Smith <[log in to unmask]>wrote: > Hi Rene, > > Those are indeed the uncorrected images. You want to use the ones that end > in *_tfce_corrp_tstat. Take a look at the randomise webpage for more > information (http://www.fmrib.ox.ac.uk/fsl/randomise/index.html). > > Cheers, > David > > > > On Apr 26, 2012, at 9:28 AM, René Besseling wrote: > > > Dear FSL users, > > > > I've performed group ICA and group comparison using dual regression. The > stage 3 statistics I get include *_tfce_p_tstat*.nii and > *tfce_corrp_tstat*.nii maps of (1-p). I'm now a bit at a loss how to > interpret the *_tfce_p_tstat*.nii maps; their naming suggests they are > uncorrected, however dual regression makes use of permutation testing which > would provide multiple comparison correction. Furthermore, the results are > threshold-free cluster enhanced (tfce), which suppresses stray (and > therefore probably false-positive) significant voxels. I guess this also > provides (maybe somewhat informal) control over the Type 1 error. Should I > therefore consider the *_tfce_p_tstat*.nii outputs as truly uncorrected or > do they deserve more credit? > > > > Best regards, > > > > Rene Besseling >