The large T-value for the cluster may be due to the underestimated
smoothness. Check the smoothness FWHM at the bottom of your SPM output
and see if the numbers are small (<0.2 voxels). If they are indeed very
small, you might want to check RPV.img (resel-per-voxel image) in your
analysis directory and see if there are any unusually bright areas. Such
bright areas are likely to be causing your error.
Your problem is actually quite common in VBM analyses, and there have
been some solutions suggested in the SPM mailing list. For example, see
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Ashley Conway
Sent: Friday, August 08, 2008 1:01 PM
To: [log in to unmask]
Subject: Re: [SPM] Paired t test error
Thank you. I did use an absolute threshold of 0.1 originally. I tried
masking with the brainmask image, and I also tried an absolute threshold
0.2. Both changes resulted in the same error for FWE-corr. I looked more
carefully at the results when I enter p<.001 uncorr, and noticed that
getting warnings in the Matlab window:
Warning: Returning NaN for out of range arguments
Additionally, my set level and cluster level p values are NaN, and the
for a few GIANT clusters are in the 5*10^15 range!
This is a subset of my subjects, and when I perform the same paired t
n=14, I do not have any of these problems.
What's going wrong?