Hi Chris,
the reference for that is Will's Paper in Computational Biology (see
link below). The very first interesting statistic are the family
posteriors (stored in BMS.DCM.[fr]fx.family.post).
For the interpretation of family posteriors it is worthwhile to keep in
mind that (usually) the priors are chosen to be flat across families. A
family of models with a much less number of members as compared to other
families therefore has single-model priors that are larger than those of
models belonging to bigger families.
The last point may become particularly important in cases where you'd
like to go beyond family level inference, e.g. by looking at the
posteriors for the single models (BMS.DCM.[fr]fx.model.post).
For RFX-analyses the excedance probability may be another interesting
statistic to report (BMS.DCM.rfx.family.xp).
However, if you havn't read the mentioned paper yet, you should do so
first (maybe more than once...). I guess many of other upcoming
questions about family level inference are likely to be addressed there.
Good luck,
Thilo
http://www.ploscompbiol.org/article/fetchObjectAttachment.action;jsessionid=5C7C786FC324CECDB0EC513B1C50D803?uri=info%3Adoi%2F10.1371%2Fjournal.pcbi.1000709&representation=PDF
On Sun, 2012-01-15 at 22:22 +0000, Christopher Lee Friesen wrote:
> Excellent, thanks Thilo, that worked well! One more query for those better with MATLAB/SPM than I: upon running the analysis, what are the important stats to report and how can I produce them with MATLAB?--providing 'proof' both for the selection of the winning family as well as the selection of the winning model(s) within the winning family. Thanks again, Chris.
|