Hi Helmut,
"> Also I think I should report decision-related regions identified in (A). But confirmation is needed
If you don't care about average activations during DEC you don't have to look at it. However, it might make a difference for interpretation whether the activation level is modulated by the PM in a certain region, with its average activation level being sig. negative or positive. Besides, even if it's not the main topic of your study the average activation pattern can serve as a confirmation that the design worked, that you find regions that have been reported before."
" If you want to find out something about the unique variance associated with ATT and DIF you will have to go with something like Analysis 2 (but e.g. step-wise model building might be an alternative depending on exact purpose)."
---> We want to know which brain regions are activated during DEC, and the brain activity specifically modulated by ATT, by DIF, or by both. To this end, my plan is to go for Analysis 2:
(A). For (i), I construct a first-level GLM without any parametric regressor (PM) to identify decision-related regions.
(B). For (ii), I construct a separate first-level GLM, in which the PMs contain first, the DIF and second, ATT. A contrast focusing on ATT is then set.
(C). For (iii), I construct a third first-level GLM, in which the PMs contain first, the ATT and second, DIF. A contrast focusing on DIF is then set.
Then,
(1). In (B), the parametric modulation contrast "0 1" identifies regions specifically modulated by ATT.
(2). In (B), the parametric modulation contrast "1 0" identifies regions modulated by DIF.
(3). In (C), the parametric modulation contrast "0 1" identifies regions specifically modulated by DIF.
(4). In (C), the parametric modulation contrast "1 0" identifies regions modulated by ATT.
So the regions commonly modulated by ATT and DIF would be the conjunction of (2) and (4).
Are my procedures correct? (Actually not very clear about "step-wise model building"...)
"> I won't be able to identify brain regions that are correlated to both ATT and DIF?
The shared variance is always attributed to the first PM."
As described above.
> a study published in NeuroImage (Mùˆller-Pinzler et al., 2015)
I had a look at the study to get an impression of what you might want to implement, and I stumbled across the following:
"The parametric modulators of hemodynamic responses during feedback contained first, the exact percent values of PERFORMANCE, second, the deviation from 50% to control for any within-subject variance in neural activation due to PERFORMANCE, and third, the trial-by-trial pupil dilation value"
"Not sure, but the second PM sounds like a rescaled version of the first PM to me. If serial orthogonalization is disabled the two PM regressors would be collinear then (they would only differ in their amplitude, if at all), and they would correlate with the condition regressor (as all three predict a BOLD response after trial onset). If serial orthogonalization is enabled the second PM regressor would be a constant term, as it has no unique variance. If mean-centered performance values were entered, with the serial orthogonalization turned off, then the two PMs would not correlate (highly) with the condition regressor, but the two PMs should still be collinear then. So basically I'm not sure what their approach with the two performance PMs is about. In any case, it can be quite useful to add a figure with the predictors as soon as designs become more complex."
--> My understanding is, in that paragraph, the authurs wanted to identify brain activity modulated by trial-by-trial pupil dilation value (representing arousal) after regressing out the potential confounding effect of the first and second PM.
In my SPM8 GUI, if I enter two PMs which are not independent from each other (e.g., both PMs are correlated with each other), it won't allow me to execute the parametric modulation analysis. Btw, how can I know my serial orthogonalization is enabled ot disabled?
Mike
|