> My stimuli are presented for variable durations (RT), ranging from 1 to 4 seconds.
> By removing duration modulation, and assuming the Gamma function (even amplitude
> modulated by RT), wouldn't I be losing this information? Or is the difference small
> (I'll look at the design matrix for each).

This is why I asked before how you would like to run group analysis regarding each modulation effect. Per RT, the question is whether you want to obtain the interaction between tasks and groups while controlling RT at its average? If so, treating RT as a modulation variable seems better than the alternative (dmBLOCK).

Regarding the correlations among regressors, did you get any warnings from 3dDeconvolve directly?

> Another question - is it wrong to expect statistics for each of the modulation regressors? I don't seem to be getting them.

Yes, it's expected to obtain statistics for those modulation variables. Something definitely went wrong with your script. Without knowing the specifics of your script and stimulus timing files, I can't guess the cause of problem.

Gang