[VoxBo] vbstatmap bug

[Daniel Y Kimberg]

Geoffrey K Aguirre wrote:
> If the G matrix of the GLM is constructed so that covariates of no-
> interest precede covariates of interest, then the contrast vector
> specified is improperly applied to the covariates of no-interest,
> yielding weird results.

Hi Geoff.  It looks like this kind of thing can happen if you use
old-style contrasts (e.g., 0 1 -1 0) as opposed to the new format
(e.g., foo t vec 0 1 -1 0).  I took a look to see if there are bugs in
the handling of the old style, and it looks like the code may
mistakenly use the numbers you provide to weight the first n kept
covariates instead of the first n covariates of interest.  Is that
possibly consistent with what you saw?  I will patch that momentarily
and put the fix out with the imminent new release.

I've really been trying to push the use of the contrasts.txt and the
format used there, instead of just strings of numbers.  I wasn't
trying to discourage it by making it especially buggy, but that can
happen too.

dan