> Does this mean one should NOT model fixation with
> a separate regressor? I have run 3dDeconvolve with
> and without modeling fixation, and this seems to
> change my data quite a bit.
Indeed, people generally do not model fixation with a separate regressor, but rather consider the fixation as an implicit baseline to which the other conditions of interest are being compared.
Note that by modeling fixation as well, you might run into near-collinearity issues, as the fixation regressor will be anti-correlated with the sum of the regressors of your experimental condition. It is quite possible that 3dDeconvolve gives at least some warnings when you try to include a fixation regressor together with the other ones.
best,
Nick