Hi Mihaly,
Please see: http://www.ncbi.nlm.nih.gov/pubmed/20300649
see page 7 of the paper: Missing parameters are assumed to be zero, with
a postrior density modeled as a delta function at zero. The height of
the delta function corresponds to the total posterior probability mass
of models which assume that the connection is zero...
Cheers,
Mohamed
On 06/04/2011 16:40, Bányai Mihály wrote:
> Dear Mohamed,
>
> I went over the paper you suggested and some other material, but I
> nowhere found an exact description about what happens during model
> averaging when the models have different number of parameters. Do we
> consider lacking ones zero?
> If you could suggest any references on this, it would be extremely
> helpful to me.
>
> Thank you,
> Mihály
>
> 2011/3/31 Mohamed Seghier<[log in to unmask]>:
>> Dear Mihaly,
>>
>> You can try BMA (Bayesian Model Averaging; see Penny et al. PLoS Comp Biol
>> 2010). Basically, because the posterior densities of your connectivity
>> parameters are conditional on the selected models, BMA weights the
>> contribution of each model to the mean by its evidence.
>> BMA is available in the BMS menu...
>>
>> I hope this helps,
>>
>> Mohamed
>>
>>
>> On 31/03/2011 10:30, Bányai Mihály wrote:
>>> Dear SPMers,
>>>
>>> my earlier question might have slipped the attention of the one's
>>> knowing the answer, if you can, please point me to the direction I can
>>> find an answer to it:
>>>
>>> is there a proper way to average DCMs with different connection
>>> patterns (same areas and data, of course)? DCM8 doesn't do that and I
>>> don't see such a solution in the DCM10 changelog either. Is there a
>>> mathematically consistent way to do so at all?
>>>
>>> Thanks for the help,
>>> Mihály Bányai
>>>
>>>
>>
>
|