Hi Lesley,
> Should I still put in all the other conditions? There are four conditions in total.
> In the AvsB.1D I used only 0,1. because I did not expect an interaction with baseline.
Yes, you should include all of the original regressors in your typical regression model in addition to the two new regressors: seed time series plus the interaction regressor. Understandably you're not interested in all of those except for the interaction, but that doesn't mean you should not account for the variability in the BOLD signal.
If you get a low correlation coefficient (or insignificant beta) at the seed voxel for the interaction, it simply means that there is no interaction between the contrast (or condition) effect and the seed region ON the seed region, which does make sense. Again, what you're looking for is the interaction between the contrast (or condition) effect and the seed region, not the correlation between a target region and seed time series.
HTH,
Gang