Dear Henrik,
At 19:02 26/11/99 +0100, Henrik Walter wrote:
| when I estimate a model, Spm will create a xCon.mat file and beta
| images.
The xCon.mat file contains details of the contrasts defined for the
analysis in this directory. Default F-contrasts (for effects of interest)
are generated by the model setup programs, and are used for selecting
"interesting" voxels at which to save raw data for future plotting, and are
saved in the xCon.mat file written out during the estimation stage.
The beta_????.img images are images of the parameter estimates for the
model. There will be one image for each column of the design matrix.
The model error variance is also saved as ResMS.img.
See the help for spm_spm.m for further datails of the image and MAT-files
written out during the estimation stage.
| When I compute contrasts, SPM generates ess*.img files for the f-
| contrasts and con*.img for the t-contrasts.
|
| The extension number of ess*.img and con*.img files equals the
| number of the contrast in the contrast manager.
A simple contrast is an estimable linear combination of the parameters,
estimated by the corresponding linear combination of the parameter
estimates. When you specify a contrast for an SPM{t} (by specifying the
contrast weights vector in the contrast manager), SPM computes the linear
combination of the parameter estimates and saves it as con_????.img, with
the ???? matching the number of the contrast (as you note). Thus, the a
con_????.img image is just a weighted sum of the beta_????.img images, with
the weights for the sum being those of the contrast weights vector.
The SPM{t} is computed by dividing the contrast image (con_????.img) by its
standard error (a multiple of the square root of ResMS.img). The SPM{t} is
saved as spmT_????.img.
To test multiple linear hypotheses, using an SPM{F}, you specify an
F-contrast in the contrast manager. The SPM{F} is essentially an
extra-sum-of squares test (See Draper & Smith "Applied Regression
Analysis"), where the additional variance explained by a section of the
model is compared with the error variance. The ess_????.img files are
images of the extra sum of squares for the corresponding F-contrasts. The
SPM{F} is spmF_????.img, computed by dividing the ess_????.img by the
ResMS.img error variance estimate image, and scaling appropriately.
See spm_getSPM.m for details of SPM{t} & SPM{F} computation.
| Now, if I later define a new Contrast manager which I copy in this
| directory the computation of contrasts will create new ess*.img
| files and new con*.img files.
The xCon.mat contains design specific information, so you should not copy
xCon.mat files from one analysis to another unless the designs are
identical. (For fMRI, the xCon.mat contains data which also depends on any
temporal smoothing and intrinsic temporal autocorrelation options.)
However, I'm not sure this is what you mean by "creating a new contrast
manager"?
In general, SPM results section is robust to the deletion of contrast
(con_???.img), ESS (ess_????.img) and SPM images (spm?_????.img). It will
re-calculate them when they are needed, hence the "reappearance" of these
files.
| (By the way: There is now warning message which may cause problems
| if one has not controlled one's actual pwd)
Not following you here... Can you elaborate?
| Now here ares the question: The beta-images do not change. Is that
| how the world should be? What do the beta-images tell me?
Hopefully this is clear now!
Hope this helps,
-andrew
PS: I see Michael Erb has posted an answer while I was composing this one!
Two replies for the price of one...
+ - Dr Andrew Holmes mailto:[log in to unmask]
| Robertson Centre for Biostatistics ( ,) / _)( ,)
| Boyd Orr Building, University Ave., ) \( (_ ) ,\
| Glasgow. G12 8QQ Scotland, UK. (_)\_)\__)(___/
+ - http://www.rcb.gla.ac.uk/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|