Barbara,

I am not so sure what exact experiment design you have, but usually there should not have any columns of all zeros in the design matrix. This is something you have to be very careful, and you can exam your design matrix, and try to figure out why those columns of all zeros occur. Sounds like you didn't browse the AFNI site carefully enough. :) See Bob's note here:

[afni.nimh.nih.gov]

Yes, you can normalize your data before 3dDeconvolve. See a previous discussion here:

[afni.nimh.nih.gov]

Should I run my ttests on the ‘coef’ subbrick of my contrasts of interest? Is that the right thing to do? (statistically speaking) or is there another strategy to adopt? (for example, I am interested in regions activated in the group for the contrast stimulus A – stimulus B, and I am also interested in comparing activations in response to that contrast between two conditions or two groups of subjects).

For one group, you can directly run simple t test on those contrast coefficients, or run paired two-sample t test on those regressor coefficients. The two should be equivalent.

For the comparison of the two groups, you can run two-sample t test on those contrast coefficients, or run 3-way ANOVA with those regressor coefficients. The former is more straightforward in this case.

Gang