> Do you think it is methodologically kosher to just modify the p and
> q parameters in GAM(p,q) so that I get the right peak time (i.e., to
> account for the delayed response in left PFC), rather than using
> waver? I think that waver could give me more fined-grained control
> over my model parameters, but I don't think I need all that much
> control.
Well there is no easy answer to this. It depends on what kind of experiment design (block vs. event-related) you have. Varying p and q in GAM(p, q) does give you some room of flexibility in terms of HDR shape. However you have to keep in mind that this is still a one-shape-fits-all approach, and it may or may not work well depending on the scenario. If the GAM option gives what you were looking for, you don't have to mess around with waver.
> Putting aside the problem of negative beta values for now, is there
> any reason to look at differences in HDR shape other than increased
> sensitivity, given that I lack such predictions and interpretative
> ability at this point?
There might have different HDR shapes across regions in the brain, across groups of subjects, across conditions, etc., that you may fail to detect with a one-shape-fits-all approach.
> you say "You could still get a sort of measure of percent signal
> change with piece-wise curve fitting strategy by summing over
> the beta's for all the tents. However you have to be very careful
> when negative betas occur among those beta's because they
> would cancel each other." Could you say a little more about what
> you have in mind here? Perhaps you could give an example?
Oh, that was the situation almost two years ago. I was referring to the situation in which some people sum up some beta's as sort of percent signal change by ignoring the one at both ends. This is not an optimal solution from the current perspective.
Gang