Dear Will and other SPM experts,
Thank you for you reply to my question about the scaling of contrast
weight in F-contrast last week. I posted a follow-up question about
some practical implications of this scaling, but I got no answer. So,
I thought of trying it again ...
Here is the follow-up question again that I posted a week ago.
On 2005-01-18 (Tue) at 09:49:54 +0100, Jan Gläscher <[log in to unmask]> wrote:
Dear Will,
thank you very much for your quick reply to my email. Can I ask you
a follow-up question?
The background of my question was, that I quite often use an
F-contrast that spans only the regressors of real interest (specified
via 'columns for reduced design') and use it to plot parameter
estimates for these regressors at a particular voxel (using the
Plotting Util of spm_graph). However, as it turns out that the
contrast weights of this 'effect of real interest' F-contrast are also
scaled (and not equal to 1). Since spm_graph plots c'*beta, I do not
obtain the true contrast estimate from the con-images, but rather
a somewhat "distorted" (scaled) value and the plot does not reflect
the underlying data.
Can you comment on this? Is it still OK to use the Plotting Util with
these kinds of F-contrast? How do you think about using these plots in
publications? Or would it be more appropriate to extract the data
directly from the beta-image (e.g. with spm_get_data), which would be
equivalent to SPM.xCon.c'*beta if the contrast weight matrix contains
only 1's and 0's, and the proceed with plotting e.g. the mean
parameter estimate for the group?
On a more theoretical level I wonder why this scaling of contrast
weights is necessary at all. I always thought that the ReML algorithm
calculates the pre-whitening matrix W, which is then premultiplied with
all components of the GLM (data, design matrix, and error) prior to the
ML estimation of the betas. This premultiplication rescales the design
matrix and compensates for non-sphericity. Why is it then necessary to
scale the contrast weights of F-contrasts as well?
Thanks for your insights.
Jan
--
Jan Gläscher Neuroimage Nord
+49-40-42803-7890 (office) Dept. of Neurology, Bldg S10
+49-40-42803-9955 (fax) University Hospital Hamburg-Eppendorf
[log in to unmask] Martinistr. 52
20246 Hamburg
Germany
http://www.uke.uni-hamburg.de/zentren/neuro/neurologie/mitarbeiter/glaescher_jan.html
--------------------------------------------------------------------------------------
GnuPG/PGP key id: FEC4B55C
fingerprint: 5A36 1EF6 8472 117E 805A F240 3146 A410 FEC4 B55C
--------------------------------------------------------------------------------------
|
