Choosing a timepoint for 3dttest++ when using CSPLIN model

I ran 3dDeconvolve on a CSPLIN model with the values (0, 22, 11) and a TR of 2.2 for an event-related design. When I pull up the results to check the activation maps, I have 11 timepoints to choose from. I’m wondering which of those timepoints I should choose to model in 3dttest++. For instance, does it make sense to use the average activity for each of the two tasks and/or the peak of the HRF?

Check out the following paper:

Chen, G., Saad, Z.S., Adleman, N.E., Leibenluft, E., Cox, R.W. (2015). Detecting the subtle shape differences in hemodynamic responses at the group level. Front. Neurosci., 26 October 2015.