Dear Fabienne,
So you've had two suggestions, depending on how you'd like to do the inference:
1. Include all models in the BMA (select 'all families'), calculated separately for each group, then compare the posterior distributions for each connection between groups (from the BMS.DCM.rfx.bma.Ep if I remember correctly)
2. As Martin suggested, put all the subjects into one group, perform a family-level inference on this group with BMA set to the winning family, then use the subject-specific estimates (BMS.DCM.rfx.bma.Eps, if I remember correctly) to compare the subjects
I don't think one approach is right or wrong - option 2 depends on whether you hypothesise finding a winning family when you collapse the subjects into a single group. Of course a third option would be to put all the subjects into one group but use 'all families' rather than 'winning family', so that all models are included.
Best,
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Martin Dietz
Sent: 14 July 2015 14:10
To: [log in to unmask]
Subject: Re: [SPM] BPA for parameter comparison between groups
Dear Fabienne,
It is only valid to perform inference on the parameters of the same model or Bayesian model average of the same family of models between subjects.
This means that you could do model selection on all subjects regardless of group and then draw inference on the difference between groups.
I hope this helps
Martin
> On 14 Jul 2015, at 13:53, Fabienne Jung <[log in to unmask]> wrote:
>
> Does anybody have an answer to this?
>
> How can I compare family parameters if two groups of subjects use different winning families?
>
> Any comments or advice?
>
> Thank you in advance!
> Fabienne
|