Dear McLaren,

I tried gPPI, but the result was very different from the one of the conventional PPI in SPM8. Because you demonstrated the similarity between two results. I am very curious why I've not a strange one.

Please find the attached for the PPI structure that I used.
The design includes 12 regressors of interest and 14 regressors (8 motion params + 6 compcor) for motion artifacts.
I tried to test the difference between the 3rd and the 1st regressors, and ran the conventional PPI for the same VOI and contrast.

The results are different, not just in different scales but only loosely correlated (r=0.36).

Should you need more details to answer my question, please let me know!

Best,
-SG

Dear Users and McLaren,

I'm glad that I can update my way-too-unspecific question. I finally figured out the reason for different regressors from traditional-PPI and coonditional-PPI was P.contrast option, which seems to be used to regress out effects of nuisance covariates from the physiological factor. I set zero for both cases, and got identical regressors (all of psi, phi, and PPI) an very close estimates of effects (r>0.9), which is quite neat! :D

My question is if P.contrast option is really for such regressor that I guessed, and how can it be useful when the nuisance covariates from the original GLM are included by default.
-SG

The regressing out of P.contrast is to remove non-neuronal sources from the physiological signal.

If you regressors are movement parameters, then you want to remove the effect of movement from the BOLD signal prior to deconvolution as the non-neuronal portion of BOLD signal doesn't have any relationship to the neural activity.

BOLD signal = conv(neural activity) + known non-neural noise sources + unknown noise sources

We measure the BOLD signal, but want to get neural activity, so we want to subtract "known non-neural noise sources", then perform the deconvolution process. Obviously, there is still some noise present, but its better than not removing any noise.

Hope this helps.

Best,
Donald

Dear Donald,

Thank you very much for response. Now I understand that the option can be useful for such cases. Is it also related to the colinearity? Because when using physiological regressor (phi) from the residuals after non-neural regression, possible correlation between the physiological term and the non-neural terms (or linear sum of them) should be reduced.

Best,
-SG

It's not related collinearity. It's related to what is being deconvolved to estimate the neural activity.