Hi there,
Please could you comment on the appropriateness of the following design for a PPI design matrix?
My understanding is that the authors have a single psychological factor, 'context', with 2 levels, high and low. In their main analysis they used a parametric modulator across these 2 levels, and obtained significant activation in their ROI (that GLM included other regressors of no interest).
Their PPI question was about the connectivity between the ROI and other brain regions,and the hypothesis was that this connectivity would differ across the 2 levels of the psychological factor.
Their PPI design matrix includes:
1) event onset (across both levels of the psychological factor)
2) psychological regressor indicating when 'high' context occurred
3) psychological regressor indicating when 'low' context occurred
4) physiological regressor extracted from the ROI obtained above. My understanding is that this was extracted across all conditions of the experiment using an identity matrix to create an 'effects of interest' contrast. Alternatively this could have been extracted from the contrast corresponding to the parametric modulator.
5) the PPI regressor for the 'high' context (I presume they multiplied regressors 2 and 4)
6) the PPI regressor for the 'high' context (I presume they multiplied regressors 3 and 4)
The reasons I'm confused are
(1) Is it ok to extract the physiological variable using the 'effect of interest' contrast? My concern is that this contrast is not orthogonal to the psychological effect. However, from reading previous threads I understand when the physiological and a psychological regressors are correlated, this simply renders the PPI analysis more conservative, rather than confounding it or being an instance of 'double dipping'.
(2) How can the PPI still be considered an interaction if it only relates to one level of the psychological factor? Shouldn't the psychological factor always include at least two levels?
I'm used to much simpler PPI designs, where the physiological and psychological factors are orothogonal and there's only a single one of each, so would appreciate your input about this more complex design.
Cheers
Deborah
|