Hi Jim,
Other than using user-specified expressions (-EXPR), such as tent functions, as your basis, there are other options as well in waver, such as -WAV (Cox special), -GAM (gamma function). You would probably have to try each of them to see which would give you a better fit for your data.
The reason why 0, 4, and 8 were used in the example in the education material is because the whole duration of the hemodynamic response was about 12 seconds. Since 3 basis functions were adopted with the first one being only the descending slope, it would come up with diving the duration (12 s) into 3 sections, [0, 4], [0,8], and [4,12] for the three basis functions respectively. That the first basis function is only the right side of a tent is because the response right after the stimulus is presumably big instead of being 0.
If you decide to go with three tent functions, there is no point using TR value (3.6) as a denominator in your basis functions. Since the response peaks around 5.5 s, you could break the response duration (for instance, 12 s) into three intervals, just like the one in the above example, and set up three basis functions for waver. See if this setup would give you a satisfactory fitting.
Gang Chen