Dear Guoshi Li
> As a follow-up to our previous discussion (https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.jiscmail.ac.uk%2Fcgi-bin%2Fwebadmin%3FA2%3Dind1901%26L%3DSPM%26F%3D%26S%3D%26X%3DC088331A853E80BB55%26Y%3Dguoshi_li%2540med.unc.edu%26P%3D775888&data=02%7C01%7Cpeter.zeidman%40UCL.AC.UK%7Ca846ca26340c4f0603a708d6ccf124bc%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C636921733410579734&sdata=Wh5mM%2FDrwDd3hdBndk42TPWX2G6A30dYngFBV3cuibk%3D&reserved=0), I have a few additional questions:
> 1) My DCM model has 27 ROIs, which has a total of 729 connections (A matrix). After PEB analysis and BMR, 380 connections (for group difference) remain with a posterior probability of greater than 0.99. Does it mean all these 380 connections are different between the two groups? Does it suggest that the PEB analysis may be more suitable for model with a smaller number of ROIs?
Mixtures of parameters are disabled during the automatic search, only if doing so keeps the free energy the same or makes it higher. If you have lots of highly covarying parameters, then disabling one or a small number of parameters might only make a small change in the free energy. In other words, you may have lots of parameters working together to produce your observed effects, where their individual contributions cannot be confidently identified. Have you considered whether you can form some hypotheses - i.e. define a small number of predefined sets of connections which show group effects - rather than doing an automatic search? This would enable you to have a cleaner narrative to your paper, and will help to focus your ideas.
> 2) To find out the connections with the most salient group effect, should I focus on the connections with the largest size effect (beta value)? If the size effect (PEB model parameters) is not used for group comparison, why do they need to be estimated?
Yes, that sounds sensible.
> 3) The fMRI data was collected from two different centers. To include the center and other (age, gender) information into the PEB model, I have the following design matrix (assuming each group only have 3 subjects).
mean diff. gender age center
[ 1 1 1 age1 1
1 1 2 age2 1
1 1 2 age3 2
1 -1 1 age4 2
1 -1 1 age5 2
1 -1 2 age6 2 ]
> I will remove the mean for the last three columns. Is this correct?
Perfect :-)
Best
Peter
|