Shantanu,
There are a couple of issues here which are not so clear to me:
(1)
Want to look at data generated when no stimulus, so stim.1D is all zero.
If I understand it correctly, you want to model the baseline by setting up your stim.1D as an input in 3dDeconvolve. However I fail to understand why you set the file with all zeros. In fact you are NOT supposed to tell 3dDeconvolve at which TR the rest (baseline) period is. Instead all those TR's at which their values are set to be 0 at all stimulus files are automatically interpreted as rest period.
Or you don't have any stimulus with this dataset?
(2)
I created a bunch of HEAD/BRIK's with different TR's, and all have the same fit coefficient.
When you created those HEAD/BRIK files, did you also adjust the length (number of TR's) of those stimulus files to reflect the TR change?
Theoretically speaking, 3dDeconvolve would generate a proportional change of c(1) with c(0) being constant with such a TR change, which is why I thought you would NOT need any time scaling. In other words, 3dDeconvolve does not know what TR is; Instead it interprets the time in the unit of TR's based on the number of subbricks in the input dataset. Thus the time unit in the baseline model (and also the regression model)
c(0)*t^0 + c(1)*t^1
should be in the scale of TR, but not seconds. While c(0) is constant regardless of TR, c(1) should be proportional to TR: the bigger TR would correspond to bigger c(1). You can convert this TR-variant slope into one in the unit of seconds by dividing c(1) by its corresponding TR.
However, with your original stim.1D of all zero, I am not so sure what would happen. Any further clarification would be appreciated.
Gang