andek Wrote:
-------------------------------------------------------
> I have made a second level analysis through
> uber_ttest.py and would like to estimate the
> significance of a cluster.
>
> -fwhm s, is this in mm FWHM?

Yes.

> The amount of smoothing, can I use the estimated
> blur from a single subject?

You could use a single subject but it's better to take the average across subjects for a more reliable estimate.

That said, in my experience smoothness estimates don't vary that much across subjects if in the same study and everything else is equal (like scanning and preprocessing parameters).

> blur estimates : 6.50091 6.22995
> 6.48386
>
> e.g. use -fwhm 6.5 ?

Yes, approximately - but why not compute the average?

> 3dFWHMx says that a stats image should NOT be
> used

... and for good reasons.

>, but the folder for the group results only
> contains a stats file, and no file for the GLM
> errors?

You can get the residuals for individual participants by running 3dDeconvolve with -errts, or 3dREMLfit with -Rerrts. These files might already have been created during preprocessing.

A somewhat quicker but less accurate method is to use 3dFWHMx with -detrend on the preprocessed (up to and including spatial smoothing) time series.

> -mask mset, which mask should I use? For example
> mask_anat.subject1+tlrc or TT_N27+tlrc ?

Preferably a mask representative of your group of participants. For example a group mask where voxels survive in at least half of the subjects's mask. This can be achieved easily using 3dMean and 3dcalc.

best,
Nick