Hi Kara,
> I think you thought I was talking about the baseline and
> linear drift estimation during 3dDeconvolve, but I was
> actually talking about the baselines for each run during the
> normalization process, before they are concatenated and
> 3dDeconvolved.
Yes, I indeed misunderstood what you meant by "baseline" in the previous messages. With that clarified, I am still confused with what you referred as "negative" percent signal changes. Normalization is a scaling process with the signal at each voxel divided by some number, either mean value across time or baseline at that voxel. Ideally the percent signal change would be more accurate if the scaling is done with the real baseline instead of mean intensity for each run. However, no matter which number being used, the difference would be really small unless the BOLD signal is very strong, and, more importantly, it would only affect the relative magnitude of those percent signal changes, not their sign, positive or negative.
> I am basically just wondering if it is possible to combine
> runs that have different control conditions due to the fact
> that the baselines (as calculated above - mean intensity
> of voxel) are different.
This points to two different approaches to obtain percent signal changes. One way is what you did (and also what it is done in HowTo #5), and the other is converting those regression coefficients from 3dDeconvolve by scaling them with the real baseline values. The first approach is a little more straightforward while the second is slightly more accurate. Most of the time they should differ much.
Gang