Hello,
> I have a question about parametric designs. Suppose I had a design which
> included a high intensity heat stimulus intended to provoke pain and a low
> intensity stimulus, also intended to provoke pain, but lower. The typical
> responses to each of these stimuli vary quite a bit but the bimodal
> distributions do not necessarily overlap in all participants. The most
> straight forward way to model this would be as two discreet regressors and
> run a contrast of the two levels. But this would seem to ignore a
> substantial amount of variability in experience of pain. For that reason I
> was planning on modeling all events as a single regressor and including the
> trial by trial ratings to obtain a parametric result. However I was told
> that this is not necessarily appropriate because I technically have only two
> levels of my variable. Does anyone have an opinion on this? Are there
> assumptions of doing parametric analyses concerning the distribution of the
> data?
Although it's called a parametric modulator, I don't think that there
are assumptions about the distribution of values in the regressor—what
you are doing is adding an additional column to your design matrix.
So, you could have a parametric regressor that had only two values
(coding for two conditions). Or, multiple levels, regardless of their
actual distribution. I think that the only caution would be to be
somewhat cautious in your interpretation; if the ratings are basically
reflecting high/low, then even if you get a difference for your
"parametric" modulator it may not indicate a continuous scaling of
brain response and pain rating. But this seems like a minor issue.
I'm sure someone else will chime in if I'm wrong about characteristics
of parametric modulators… :)
Hope this helps!
Best regards,
Jonathan
--
Jonathan Peelle, PhD
Assistant Professor of Otolaryngology
Washington University in St. Louis
Office: (314) 362-9044
http://peellelab.org || http://jonathanpeelle.net
|