Liz,

Sorry that my response was not clear enough to you. Let me try it again.

The output from 3dDeconvolve includes the statistics, beta coefficients, and the modelled signal (the original signal with noise removed). As you said, the statistics would not be affected by how you create your regressors. The same holds for the modelled signal. Only the beta coefficients would depend on the scaling of the regressors.

I am not so sure what results from 3dDeconvolve you would like to report. If you want to detect activation regions and/or compare the % change of the modelled signal among different voxels, there is no need to concern about the scaling.

As your experiments with different scaling showed, the beta coefficients are inversely proportional to the magnitudes of your regressors: peak 1 ==> beta =8.0817, peak 2 ==> beta=4.014718. As I said in my previous message, beta's are quite arbitrary because they depend on the scaling of your stimulus files. Therefore if you want to compare the relative values of beta among different voxels or across different subjects, you'd better convert the beta's into their % change instead of using their arbitrary values. The % changes of beta then become a common currency for comparison. In this sense, you don't need to worry about the magnitude of your regressors when you calculate the % change of beta's, because your irf's (basis function of gamma variates) have a peak of one ( the option -peak 1 in waver).

Gang