Hi Christopher,
I would argue very much that you should do (a) and model with four
conditions.
You want to remove the effects of incorrect responses, but whatever
brain response occurred on those trials would still be best modeled as
an hrf. Modeling as a simple onset without convolution would just have
the effect of removing the single timepoint of the incorrect response
from further analysis, and would not be correct. You would be leaving
the rest of each hrf that occurred due to the incorrect response to be
folded into baseline. One would use a regressor such as this in
situations where one wants to remove a "bad" volume from the imaging
analysis.
Modeling as an hrf allows you to estimate what is occurring during
incorrect responses separately from the correct responses, rather than
leaving them to be folded into baseline, which would make comparisons
against baseline more difficult to interpret,
Joe
On Aug 11, 2008, at 10:27 PM, Christopher Benjamin wrote:
Dear SPMers
I've haven't received a response re. this issue, apologies if seems v
simple but would
greatly appreciate input.
I want to model response accuracy in an efMRI design (I'm assuming a
canonical HRF). I
have conditions A & B, which, in some but not all cases, can be
subdivided into correct &
incorrect responses. I want to compare correct-A to correct-B. My
question is, is it
MORE valid to –
(a) Specify a model with four canonical-HRF-convolved regressors (four
conditions), or
(b) Two canonical-HRF-convolved regressors (Conditions A, B) and one
binary accuracy
regressor for each condition?
----
Joe Moran, Ph.D.
Department of Brain & Cognitive Sciences
46-5081, MIT
Cambridge, MA 02139
tel: 617.324.5124
fax: 617.324.5311
email: [log in to unmask]
http://web.mit.edu/gabrieli-lab/People/moran.htm
|