I tried with 10 runs from one session. Still I see the negative beta values but the no of voxels with negative beta values is alot lower than when I did the analysis across all runs i had.
Still, one thing not clear to me is when I just compute a t-score between baseline condition and motion present condition i get alot of voxels with positive t-scores especially in areas expected to respond to stimulus. But when i do the regression and look at beta values for presence of motion regressor there are way too many voxels with negative beta values. Is there something I am missing in the regressing part?