Dear Sukhi,
Dear All
I am analysing a fMRI data set in 8 subjects performing a language task at
3 different rates (60 times a minute, 30 times a minute and 15 times a
minute) within one block of scanning; each task performed 3 times for 30
seconds at a time. So effectively ABCCBACAB.
I am interested in looking at regions which respond in line with the
increased rate. Using the parametric modulation and linear model reveals
little of interest. I assume that a) this is a true negative finding - no
regions with associated activity OR b) that the regions associated with
increasing rate do not respond in a linear fashion..
One little warning here. What you test when you test a specific regressor (in
this case the regressor that
contains the presentation rates) is the variance that is explained by that
regressor when all variance
explicable by other regressors have been removed. Hence, if you have included
the categorical regressors in
the model, your "rate" regressor will be almost (due to finite length) exactly
a linear combination of the other
regressors, and it will explain virtually nothing over and above those. My
excuses if you haven't made this
little mistake.
My question is where to go from here?
Is there a scientific rationale for using different models - further order
polynomials as per the menu, or is there another smarter way of defining an
optimal model.
Lets say you have included the categorical regressors, i.e. you have one
column in the design matrix for each
condition A, B and C. If you now specify an F-contrast, testing these three
together, that will be equivalent
to testing any polynomial (or other function) of your rate parameter.
If that F-contrast doesn't show much, then it is probably not worth searching
further for any effects of
presentation rate.
Many thanks
Sukhi
Good luck Jesper
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|