Dear Torsten,
I'm afraid TFCE was never intended to be used with COPEs. The specific values of H & E are motivated specifically by the behavior of test statistics that (under Ho) are mean 0 and variance 1, and I can't predict how our H&E values will work with your COPE images (depends on how their scaled). I suppose you could estimate the image-wide standard deviation of the COPE image and scale the image (as a whole, one scalefactor) to unit variance, but it's probably not going to work. In particular, we know that the variance is going to vary over the image. If you standardize the variance at each voxel you've just created a zstat image.
In short, I'd recommend sticking with zstats.
-Tom
On Tue, Dec 6, 2011 at 11:39 AM, Torsten Ruest
<[log in to unmask]> wrote:
Hi there,
we completed 1st level analyses of a fMRI activation study and also applied randomise tfce on the resulting 1st level copes. Inspired by the technical report on TFCE and some suggestions on the mailing list, we trialled several parameter combinations of tfce height and extent. We are aware of the recommendations (H=2 E=0.5) and the discouragement to change these settings, but may I ask if corrected for multiple comparisions, the results would be valid? So far we seem to get the best results at the lower end of H end E, that means, I assume, that the input signal is not much enhanced at all (H=0.1 E=0.1, see figure 10 in technical report)? Likewise, we are using cope images not raw tstat or zstat images, so has anyone made experience with suitable values?
Thanks very much in advance,
Torsten
--
__________________________________________________________
Thomas Nichols, PhD
Principal Research Fellow, Head of Neuroimaging Statistics
Department of Statistics & Warwick Manufacturing Group
University of Warwick, Coventry CV4 7AL, United Kingdom