BTW, Is the error message regarding insufficient data for estimating gone?
OK, suppose you have 3 conditions: C1, C2, and C3, plus fixation (F).
The regression analysis would fit your original signal at each voxel with a baseline with a constant pluse trend (e.g., a + b*t) during those time points you specify as fixation (F), and with a linear combination of the three regressors based on the time series you provided for each condition, plus white noise.
If you make one of the three conditions, i.e. C2, as baseline, what are you going to do with the fixation period F? Treat F as another condition in the original place of C2, or combine F with C2? Either way, you would come up with distorting the fitting becaues C2 would be fitted with a straight line (-polort 1) instead of hemodynamic response function. Besides, the beta estimates for C1 and C3 and their statistical values would also be affected due to the change of baseline fitting.
Gang