Hi,
A question about multi-collinearity issue in some experimental designs (after reading "Orthogonalization of Regressors in fMRI Models" by Mumford et al. published in PLoS ONE 2015) has puzzled me for a while and I would appreciate for some experts' comments.
Consider a learning paradigms (with 3 types of stimulus and 3 types of feedback) like this:
stimulus (4 s) --> delay (3 s) --> feedback (4 s) --> jitter Inter-trial interval 4-7 s
Suppose the first-level GLM is:
Model 1: stimulus (3 regressors), delay (3 regressors), feedback (3 regressors)
If I'm interested in distinct responses during the stimulus, delay, and feedback period, Model 1 seems not allowed, because responses during the 3 periods would have a high degree of multicollinearity due to the slow response of the hemodynamic response (however, it seems that many papers used this kind of model)?
Consider another situation. If I'm interested in distinct responses during the stimulus and feedback period (not interested in delay period) and I constructed another model:
Model 2: stimulus (3 regressors), delay (totally 1 regressor), feedback (3 regressors)
Can I say that Model 2 is better because it reduced the multicollinearity problem between stimulus and feedback period? Also, if I want to further minimize collinearity between stimulus and feedback, is it better if I jitter the delay period?
Thank you. Manuel
|