On Thu, 15 Sep 2005 17:50:46 +0100, Amit Etkin <[log in to unmask]> wrote:
>Hi all,
>
>I am using "con" images for a "second level analysis" using a multivariate
>method. Based on my early results, I feel that the original data was
>inadequately smoothed for this purpose, and I would like to smooth the
data
>more. There are two options for this...either go back to the beginning,
>resmooth the data with a larger kernel and estimate all of the models over
>again, or simply smooth the con images by a small kernal. Certainly
>smoothing existing con images would be faster and would allow me to
>experiment with several additional levels of smoothing
>
>I am wondering, therefore, whether these two methods equivalent (except of
>course for a rim of "edge" voxels near the brain mask border, whose
>neighbors are NaN)?
>
>Also, how would you calculate how much extra smoothing to add to con
images
>to achieve a given level of "original smoothing" had I smoothed the raw
data
>by the desired amount in the first place?
Roughly speaking, if you smooth then run stats, the resulting con images
will be the same as con images smoothed with the same kernel (with no
smoothing done before stats).
I say "roughly" because it's not exactly true in SPM2: see this post of
2005-09-15:
http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0509&L=spm&P=16379
>thanks!
>Amit
>=========================================================================
|