 |
 |
Dear Anderson,
thanks for your reply!
Actually, I have a within-subjects blocked design, where subjects were tested in two sessions/conditions
-
experimental condition: blocks of stimulation with a certain perceptual aspect present only in this condition and a subjective rating
for each block (the potential parametric modulation).
-
control condition: blocks of stimulation without the perceptual aspect under investigation present (so no parametric modulation and a
‘dummy’ rating).
I want to figure out BOLD responses specific to that perceptual aspect only present in the experimental condition.
In my situation subjects rated quite constantly across stimulation blocks according to the perceptual aspect under investigation in the experimental
conditions.
Can I still use the block-wise rating covariate to disentangle BOLD-responses specific to that perceptual aspect only present in the experimental
condition or
simply compare experimental and control condition without the rating covariate which gives temporally more fine-grained information according
to the phenomenon of interest?
I really appreciate your valuable advice!
Kindly regards,
Christopher
Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
Im Auftrag von Anderson M. Winkler
Gesendet: Dienstag, 1. September 2015 11:37
An: [log in to unmask]
Betreff: Re: [FSL] collinearity between intercept and parametric modulator (covariate) at 1st level
Hi Christopher,
If other subjects all have a similar design, that is, with two EVs coding the response, one modulated and the other not, I'd say leave both there, even for the subjects in which the variability is very small and there
is a high colinearity. That's how I would do.
All the best,
Anderson
On 31 August 2015 at 13:39, Milde, Christopher <[log in to unmask]> wrote:
Dear FSL Experts,
Do you have any suggestions how to properly construct the design matrix for a 1st-level analysis when being confronted with high degree of collinearity
between the parametric regressor and the unmodulated regressor (‘intercept’).
In specific, the covariate (parametric modulator) shows low trial variability given rise to a high degree of collinearity between the intercept term and the parametric
modulation.
1.)
In this case is it appropriate simply to use amplitude modulation (the slope) without the unmodulated regressor assuming an intercept of zero?
2.)
A follow up question: Am I right that the parametric regressor is automatically demeaned at the first-level (I’m aware that one has to demean or centre manually at higher level analyses)
I’m quite thankful for any advice and literature suggestions!
Kindly regards,
Christopher