> can the -stim_times option in the new version of 3dDeconvolve
> be used in conjuction with deconvolution, or only with multiple
> regression?

Only regression.

It seems that some confusions exist here. For an event-related design, the real question about modeling at individual subject analysis level is not really about regression or deconvolution+regression although it is portrayed that way most of the time. Instead the issue is: Do you want to model the impulse response functions yourself or just use some standard impulse response function?

With TR-locked timing, you can run deconvolution+regression to model impulse response function, or run regression analysis on pre-convoluted stimulus files with waver.

With irregular timing, deconvolution is difficult to implement. Instead 3dDeconvolve uses different kinds of basis functions to model impulse response functions. Those basis functions still give the user different options: If you are not interested in modeling the impulse response functions yourself, simply use any of standard impulse functions such as gamma variates; If you want to model voxel-based response functions , use more subtle basis functions such as tent.

So for irregular timing, the modeling decision of impulse response functions hinges on the type of basis functions you select, not about the choice of regression/deconvolution.

> How would one specificy the specific irregular stimulus timings in
> the context of a deconvolution analysis? For example, if the stim
> was always delivered at the midpoint of the TR, how could that
> be indicated?

For timing files, the user simply provides all the moments when a stimulus was given. If your stimulus was always delivered at the midpoint of the TR (TR = 2 sec), an example of stimulus timing file for condition 1 is something like

11
25
37
43
59

Gang