Dear Prejaas,
Previous posteriors can be used as initialization point for the next
inversion. Usually the parameters are initialized to the prior means.
But this is not essential. You could also keep the same priors but use
different initial values. The idea of using the previous posteriors is
that in some cases you would like to be around the same area of the
parameter space for different inversions. When using previous
posteriors you are very likely to end up in the same local maximum of
F which probably explains higher consistency of delay posteriors when
using this option. Note,however, that all the differences you show are
under a millisecond.
Best,
Vladimir
On Thu, Jan 12, 2012 at 9:58 AM, Tewarie, P.K. <[log in to unmask]> wrote:
> Dear spm experts,
>
>
>
> I am using the EEG dataset from the website (mismatch negativity) to see how
> dcm works and what I can expect when applying it on my own dataset.
>
>
>
> It is repeatly asked if I want to use the previous posteriors and/or
> previous priors when inverting another dcm. I can understand that you want
> to start the next dcm with the same priors but the question about the
> posteriors is not clear to me. These are not used as priors for the next
> model?
>
>
>
> I tried to see how minimal changes in position of the sources could affect
> the estimated parameters. I looked for example to the delays DCM.Ep.D when
> doing this. I randomly moved a source with 0.5cm from the original position
> as mentioned in the reproducibility study and spm manual. I did the same for
> using the previous posteriors and priors and on the other hand only using
> the previous priors.
>
> See attached the results. Why do I get so much fluctuation in the estimation
> using only the previous priors? And what is the role of the posterior for
> calculating the next dcm. It is not clear to me from the papers of kiebel
> 2006, david 2006.
>
>
>
> Many thanks,
>
> Best regards,
>
>
> Prejaas Tewarie
>
>
>
>
>
>
|