Hi Daniel,
I'm not too sure, but I think something like this sounds okay, though
I think maybe you should drop the constant, as G1 and G2 model the
group means separately, and the simpler single-group correlation has
no concept of a shared grand-mean term.
[...]
> G1 G2 V1 V2 C (G=group, V=variable, c=constant)
> 1 0 2 0 1
> 1 0 4 0 1
> 1 0 3 0 1
> 1 0 5 0 1
> 1 0 8 0 1
> 0 1 0 6 1
> 0 1 0 1 1
> 0 1 0 3 1
> 0 1 0 2 1
> 0 1 0 4 1
[...]
> In my head, the contrast [0 0 1 0 0] from the multiple regression
> analysis and the contrast [1 0] from the simple regression analysis should
> be (somewhat) identical, barring estimation/statistical differences that I
> am not aware of.
I would guess so too, but might be wrong. Anyway, I'd suggest that you
compare the con and spmT images for these tests (e.g. in check reg,
and right-clicking to compare exact values) rather than looking at the
final results, as this will clarify whether it's something to do with
degrees of freedom or multiple comparisons (though I wouldn't expect
that). As suggested above, I'd try this without the constant term,
though it might not change things.
[...]
"Glabus, Michael" <[log in to unmask]> 1/10/2007 5:54 PM >>>
> There's also a procedure that someone developed that converts the t-map
> to an r-map that I'll send when I find it!
This is in Volkmar's volumes toolbox and in Christian Gaser's VBM2
toolbox:
http://sourceforge.net/projects/spmtools
http://dbm.neuro.uni-jena.de/vbm/vbm2-for-spm2/threshold-and-transform-spmt-maps/
Or you could implement the formula yourself
% correlation coefficient:
% --------------------------------
% sign(t)
% r = ------------------
% df
% sqrt(------ + 1)
% t*t
But perhaps you don't want to do this conversion anyway... hopefully
you'll get some more replies from statisticians!
Best,
Ged.
|