> I designed the matrix so those other stimuli would
> be dumped into the baseline, instead of accounted
> for with their own parameter, whereas in the
> individual-item case every stimulus was accounted
> for by a parameter.

Well, believe it or not, whether you treat a regressor as a normal one or banish it to be part of the baseline (a regressor of no interest such as motion parameters) with -stim_base, it really does not matter at all in terms of the regressors you are interested in. The ONLY effect is that you would get a different Full F. The reason is that, either way, you have exactly the same model, and the Full F is calculated as

Reduced model with [Baseline (including regressors of no interest) + Noise) versus (Baseline (including regressors of no interest) + Regressors of interest + Noise)

Therefore the difference is how you group those regressors of no interest in the same model. Try the different approaches and verify whether this is true.

Cheers,
Gang