On Thu, 19 Nov 2009 17:26:54 -0600, Brian Essex
<[log in to unmask]> wrote:

>Hey SPM'ers,
>
>I have a design matrix for an event-related fMRI design in which there are 4
>conditions and one of the conditions has a parametric regressor associated
>with it (for this condition each trial has a different value of this
>parametric regressor). I took care of the parametric regressor by using a
>pmod. Let's call the condition that has a pmod condition 1. What I would
>like to do is compare activation in condition 1 and 2 (which can easily been
>accomplished with a subtraction), but while also taking into account the
>level of the parametric regressor that is tied to condition 1. In other
>words, I'd like to see if activation associated with the subtraction scales
>with the level of the parametric regressor (i.e. is there an interaction
>between the subtraction and the pmod). The goal is to examine this
>interaction at the single subject level and then to take those results to
>the 2nd level for further analysis.

It's not clear to me that this question is well-posed.

Off the top of my head I can't think of a conceptually meaningful way to
define an "interaction" in the way you intend. Usually interactions are
between factors of an experiment, not between levels within a factor. You're
asking for an interaction between one level of a factor and the parametric
modulation of another level.

In the model that you've set up with SPM, there's one predicted response to
condition 1, another to condition 2, and a (predicted) modulation of the
response to cond 1 by the parametric regressor. Clearly, since the
(predicted) activation during condition 1 depends on the value of the
parametric regressor, the difference between cond 1 (with its associated
modulating regressor) and the activation during condition two will depend on
the regressor.

You can take the (beta) coefficients for the two conditions and the parameter
for the first condition to the second level, in the usual way. (Of course,
usually people take contrasts, not the betas themselves.) You could do that
without thinking about the consideration you mention. Or you could take into
account the consideration. But that doesn't actually affect anything; the
remarks above also apply to the predicted (population) values at the second
level.

If you want to see what the effect you describe is, you can just write down
the model and plug in some values.


>I have had no luck finding this answer on the forums or documentation, and
>would appreciate any assistance. This is all being done in SPM5.
>
>Hope you can help,
>
>Brian
>