Dear Robert,
On your first question - no. The BMA will, as you say, generate results for the winning family - but I don't think it would be valid to use it without much stronger XP for that family.
In general I would not treat these zeros as NaNs (assuming that the difference between zero and nan is the number of values to include when calculating degrees of freedom for some statistical model).
Best,
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Robert Schulz
Sent: 24 February 2015 12:19
To: [log in to unmask]
Subject: [SPM] BMA - statistics on mEPs - how to proceed with "zeros"
Dear DCM experts,
may I send again my recent question to the list. It's about how to treat "0" in the mEPs after BMA across winning models for a group of participants. Unfortunately so far, I have not received any answers.
I devided a model space of 6x12 models according to the C matrix in 6 families (x12 models according to B), ran BMS.
I have two questions:
1.) To restrict further BMA to models of one winning family, would an exceedance prob. XP of 0.45 for family 3 compared to 0.15 for the other families be sufficient to illustrate family superiority? I wonder, because if I enable BMA with "winning family", in fact, BMA will be executed over family 3. I thought that the XP should be over 0.9.
2.) If I conduct BMA, lets say over family 3: As the models (in occams window) considered by BMA differ between subjects, also the mEPs might differ. E.g. for selected subjects some .mEps value might remain "zero" because models with (!) an estimated modulatory value (according to "1" in B matrix) for this specific connection were not considered by occams and BMA. I wonder whether I can use these mEPs "zero" values in further group comparisons / statistics or I have to treat them as NaNs.
I hope that anybody can help again with this problem.
Thank you very much in advance.
Best wishes,
Robert
|