Dear Yuqian
Sorry for the delay in replying. Yes, you can test whether one connection or modulatory effect is bigger than another, and you only need one PEB model to do it. More specifically, your question is: what's the probability that the left-visual -> left-motor modulation is stronger than the left-visual to right-motor modulation?
For each modulation, you have a (marginal) normal distribution, specified by the posterior mean (PEB.Ep) and posterior variance (PEB.Cp(i,i)). You can then calculate the probability of a difference between these two normal distributions. This has previously been called a "Bayesian contrast".
To perform this analysis, please see the Matlab function spm_dcm_peb_con. If it's not in your copy of SPM yet, you can find it online at https://github.com/spm/spm/blob/main/spm_dcm_peb_con.m .
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Yuqian Yang
Sent: 10 May 2022 14:30
To: [log in to unmask]
Subject: [SPM] DCM-PEB: parameter comparison between different connections
⚠ Caution: External sender
Dear DCM-PEB experts,
I am currently using DCM-PEB to test if two conditions (A and B) modulate the effective connectivity between visual and motor areas in a different way. At the first level, bilateral visual and bilateral motor areas were involved, and the between-region connections were modulated by both conditions A and B. After the group-level analysis, I found that condition_A positively modulated the bottom-up connections from the left visual to the left motor, while condition_B positively modulated the bottom-up connections from the left visual to the right motor (you can find the figure of the results attached) across subjects.
Now I wonder if I could compare these two positive modulatory effects directly.
Because the different modulation effects on the sensorimotor connections due to the two conditions are the main interest of our study.
However, I realized these two parameters were calculated involving different target regions (one was calculated using the left motor area timeseries, and the other with the right motor area timeseries)...
If it is doable, I feel like, for each subject, two DCMs will need to be fitted, one for the network architectures only including the right motor area, and the other for that only including the left motor areas. That means, there will be GCM_RightMotor and GCM_LeftMotor. Then at the second level, a PEB model will be used to test the difference between the right and left motor related connections using 1 and -1 regressors.
Any suggestions would be greatly appreciated. Thank you very much in advance.
Best wishes,
Yuqian Yang
|