Dear Michael,
> I am attempting to calculate signal intensity changes at a given voxel
> using the corresponding beta coefficients and contrast weight vectors.
> In doing so, I discovered that an old analysis (I think with SPM99b)
> yielded effect sizes (beta coefficients) which are different than a
> repeat analysis (with SPM99) with exactly the same data and model (at
> least everything in the model appears to be identical unless I am
> missing something very subtle.). This is the case except for the
> constant beta- which is identical across analyses. {I am running t-tests
>
> with 2 conditions modelled as events and looking at [-1 1] and [1 -1]
> contrasts}. I think my problem relates to a previous message that Karl
> addressed, but did not resolve:
>
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0003&L=spm&P=R5027&m=2612
>
> Note that I have carefully checked that I am tracking the same voxel and
>
> using the same scaling (global) between analyses. Furthermore, the
> resulting SPMs, Beta.imgs, and models appear indistiguishable between
> analyses.
>
> What might cause such a discrepancy in resulting beta coefficients
> (excepting the constant)?
I can't think of a good reason, why your parameter estimates would
differ, if you used exactly the same model for both analyses.
One way to check, whether the data and the model are really the same is
to load the resulting SPM.mat files for both analyses and check
carefully the stored design matrices in xX by plotting them. The
estimated global values are (I think) stored in VY.pinfo. You could also
check the temporal filter used in xX.K{:}. Any difference in all these
structs would explain your observed differences in parameter estimates.
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
