Dear Donna,
>I'm running parametric modulations using SPM to look at how the level of
>one variable affects activation in one particular region. I have found
>that it modulates activation here. However, I ran another parametric
>modulation and found that a nuisance variable also modulates activation
>of this structure, though negatively.
>
>So the next thing I want to do is try and covary out the effect of this
>nuisance variable, so I can see what the parametric effect of my
>variable of interest is. While I know how to enter two different sets
>of parameters for one trial type when setting up the parametric
>modulation (e.g., variable of interest; nuisance variable) I am unsure
>of how to set up my contrasts so that the effect of age is factored out.
>(i.e., Im wanting to do what I would normally use a step-wise regression
>for with behavioural data).
>
>Normally, with only one covariate in the parametric modulation, I use
>the contrast [0 1 0 0 0 0 0 0] (i.e., a [0 1] to test for the positive
>parametric effect, and [0 0 0 0 0 0] to adjust data for the 6 motion
>parameters I entered as regressors).
>
>I'm wondering, in this case where I have entered nuisance covariate
>also, if I place a 0 for this (i.e., [0 1 0] to test for the parametric
>effect of the variable on interest only, and then [0 0 0 0 0 0] for
>motion)
>will the data be adjusted for the effect of nuisance variable.
Yes it will, effectively. Whenever you include another explanatory variable
in the design matrix, the parameter estimates change so that they
correspond to the estimators you would have got if you had covaried
the additional effect out of the data. This follows from the fact
that the esitmates reflect the contribution of their covariate that
is unique to that covariate (i.e. after it has been orthogonalised
with respect to the rest of the design matrix).
I hope this helps - Karl
|