JoshPowell Wrote:
-------------------------------------------------------
> Hi Rick et al-
>
> I've had a similar problem with 3dFWHMx
> underestimating smoothness. In my case, it seems
> to happen only when I use the 'detrend' option.
> I'm not sure whether this is a bug with detrend,
> or perhaps detrend is inappropriate when using
> data that doesn't have anatomical detail (i.e.
> residuals.)
Interesting finding!
I tried this on random gaussian data (99 volumes generated in matlab) that was subsequently smoothed with 8 mm FWHM; the smoothness estimates are very similar with and without detrend. There is absolutely no anatomical detail here, so that hypothesis does not explain the significant reduction of smoothness with -detrend:
dhcp-10-249-68-218:spm nick$ 3dFWHMx __g8_all+orig.
++ 3dFWHMx: AFNI version=AFNI_2011_12_21_1014 (Sep 6 2012) [64-bit]
++ Authored by: The Bob
8.39984 8.43527 8.43592
dhcp-10-249-68-218:spm nick$ 3dFWHMx -detrend __g8_all+orig.
++ 3dFWHMx: AFNI version=AFNI_2011_12_21_1014 (Sep 6 2012) [64-bit]
++ Authored by: The Bob
++ detrending start: 9 baseline funcs, 99 time points
++ detrending done
8.30659 8.32791 8.32377
I'm still curious what is causing the decrease in smoothness though. If voxels are detrended individually, then neighbouring voxels may have different weights for each of the detrending regressors and thus the residuals might become more independent. However the results shown above suggest that this effect is minimal (at least for real gaussian data) and does not explain the deviations noted by the others.
Nick