Dear Vladimir,
There's an error in the last letter. The detailed models in family A and family C were not absolutely same. However, this made few effects on the final BMS results(RFX). I meant that family A was always much better than family C.
Considering the reliability, I did BMS (both RFX and FFX) again. This time Family A and family C, family B and family D had the same models in details.
Family A included three models: A1(modulation on backward connection from DLPFC to ACC ), A2 (modulation on forward connection from ACC to DLPFC ), A3 ( modulation on both backward and forward connection between DLPFC and ACC). Family C was the same with family A except for the disconnected source.
Return to my former question: if disconnected source doesn't affect the model, and if family A and family C have the same models except for the disconnected source. In this case, I think family A and family C will get similar performances. However, family A turned out to be much better than family C. It was this that confused me much. How do you think about this question?
May you a happy weekend!
Haoran.
At 2011-07-15 20:22:14，"Vladimir Litvak" <[log in to unmask]> wrote:
>Dear Haoran,
>
>From your schemes it's not clear what is the difference between A and
>C or between B and D. Also it's not clear how the models within each
>family differed so one thing you need to make sure is that you did
>your comparisons correctly. However, assuming the result is valid I
>would definitely interpret it with your interpretation (2). Namely,
>there is no evidence for the necessity of the hidden source to explain
>the data. Disconnected hidden source does not affect the model so your
>model A could as well be without the hidden source at all (but you
>were right to include the source without connecting it).
>
>Vladimir
>
>
>
>2011/7/15 飞鸟 <[log in to unmask]>:
>> Dear SPM's users,
>> I got some results using dynamic causal modeling (DCM for ERP,MEG).
>> However, I was surprised at the results. I built four basic families of DCM
>> (see attachment DCM_Model.jpg), expected the best model would appear in
>> family B or D. To my surprise, family A turned out to be the best (see
>> attachment bms_family). Then I inferred:
>> 1. The hidden source might latently present several other subcortical
>> sources which might closely related to my experiment. Moreover, family A had
>> the least limitations of the connections between hidden source and other
>> non-hidden sources. Finally, it became the best family of the four.
>> 2. Family A was had much less connections relative to Family B and D, so
>> its model complexity was lower. Finally, it became the best family of the
>> four.
>> Are the inferences above reasonable? Or what's the role of the hidden
>> source when it has no connections between other sources in DCM? Any of your
>> help will be greatly appreciated.
>> Thank you!
>>
>> --
>> Haoran LI (MS)
>> Brain Imaging Lab,
>> Research Center for Learning Science,
>> Southeast University
>> 2 Si Pai Lou , Nanjing, 210096, P.R.China
