[VoxBo] PSC in FIR models

[Daniel Y Kimberg]

Just a few random points to add to Charan's reply.

Russell Epstein wrote:
> But, when I go back and look at results from an earlier study (rapid
> event-related, in case that matters) I find that FIR peak is about
> 0.7%, but the beta values for the normal model translate into 3-4%.
> How is is possible?

The two will generally give you different answers, although I'd be
less surprised if it were the other way around.  If you aren't
mean-centering your data, VoxBo can still scale your data as a percent
signal change, it just divides by the intercept.  In either case, you
also need to make sure your covariates are scaled appropriately, so
that the change you're interested in has a magnitude of 1 in your
covariate.  Depending on how you set up your covariates, you might
already have it that way, or you may need to click the "unit
excursion" button.

With rapid event-related designs, it's easy to have some kind of
mis-modeling mess things up.  I would double-check the collinearities
to see who's possibly stealing who's variance.  You might also, with
just the FIR models, have the G matrix designer create your covariates
at the final resolution (i.e., set the "upsampling rate" to be the
same as the TR).  That way you'll be seeing exactly the final
covariates in the G matrix designer.

dan