Our lab is seeing an unexpected result when we rerun our data using -acf for clustsim (compared to the -fwhm method). Our cluster sizes are smaller, much smaller, going from 17.2 voxels at .001 with fwhm, to 3.7 voxels at .001 with acf. This is when looking at the full grid - we now go with a whole brain mask, but in the interest of comparing the fwhm and acf we did the exact same clustsim, just changing to the acf. Because our understanding was that the cluster thresholding should get more conservative we were puzzled and wanted to do a sanity check, is there a reason we’d be getting smaller thresholding cluster sizes?
If the strangeness is when running without a mask, that is
less worrisome, since the non-brain voxels, and even step
from brain to non-brain, may throw off the estimates a lot.
Does it work as expected using a brain mask?
Using a mask does change things and actually we realized yesterday that we were looking at data what had not been blurred (we will be running MVPA down the road), so I think that also explains the reduction in the thresholding values - because you take a hit when you blur, right?
That is right. The more blur is applied, the higher the estimates will
(or at least should) be, and so the resulting cluster sizes will be bigger.
With the typical/old model of FWHM, the parameters might show just
above one voxel size. The ACF parameters should be a little different,
but still small.
Awesome, that is very much in line with what we are seeing!