Bonnie,
In answer to your second question of "is this a problem?": it depends. The scaling of the reference function will affect the scaling of your fit coefficients after 3dDeconvolve. For example, if your reference function went from 0 to 100, then fit coefficients will be 100x larger than if the reference function went from 0 to 1. As long as you are using the same reference function for all your subjects, then this constant scaling factor won't make a difference when you get to group analysis. If, however, you want to compare fit coefficients between conditions whose reference functions are scaled differently, there will be a problem. The main place where this causes trouble, actually, is when people are making "subject-specific" reference functions based on subjects' trial-by-trial behavior. Then, if the scaling of each person's reference functions is different it will throw off the group analysis. When I make subject-specific reference functions, I normalize each wavered reference function to go from 0 to 1 by dividing by the max value. I do this in Excel, but I know others have developed programs to do it, and for all I know, there's now an AFNI program to help with this...
Hope this helps.
Lisa Eyler