Dear Matthias,

One should not compare parameter estimates between models that are not identical in structure.  Therefore, if your two groups differ wrt. the best model, then this is what you should report: that different mechanisms best explain the data in both groups.  Depending on your question, it may also make sense to statistically test parameter estimates within each group separately - but not between groups.

Best wishes,

----- Ursprüngliche Mail ----
Von: Matthias Schurz <[log in to unmask]>
An: [log in to unmask]
Gesendet: Mittwoch, den 30. Juli 2008, 09:22:52 Uhr
Betreff: [SPM] DCM model comparisons with 2 groups

Dear DCM experts,

I’ve got a question about DCM in the context of a group study:
Within 2 groups, I’ve have compared the same models, and I’ve found that
group 1 favors model A and group 2 favors model B. Normally, one compares
the parameter estimates between the 2 groups - for example, I could look at
the parameters of model A for group 1 (where this model is suitable) and for
group 2 (where it’s not) and then do t-tests on the parameters. Does this
make sense? Are the parameters for group 2 meaningful, although there is a
better model? Or would you rather only report the model-parameters for the
group where the model is suitable?

Many thanks in advance,

Gesendet von Yahoo! Mail.
Dem pfiffigeren Posteingang.