Thomas,
Yes, this is perfectly valid.
In Bayesian model comparison you make an inference about
the probability of the model, m, given the data Y ie. p(m|Y).
The idea is to have different models compete to explain
the same data. The constraint is therefore that Y be the same
for all models. In model comparison for DCM for fMRI, the data
are the fMRI time series. So you must have your models explain
the fMRI time series in the same regions (you can't therefore
have different numbers of regions, or different regions).
But because the inputs are explanatory variables, not the data Y, your
approach is valid.
Best wishes,
Will.
Thomas Ethofer wrote:
> Dear Will Penny,
>
> in your paper you describe comparisons of models with different
> intrinsic connectivity pattern and different modulary inputs. Is it also
> valid to use Bayes factors to guide choices about the input region? I
> tried this with 3 regions (one in the secondary auditory cortex, and two
> in the lateral prefrontal cortex) by not specifying any constraints on
> the intrinsic connectivity pattern (fully connected model) and comparing
> models in which direct inputs were specified for the secondary auditory
> cortex with models in which direct inputs were specified for one of the
> frontal regions. As expected, in nearly all cases fully connected models
> assuming direct inputs in the secondary auditory regions were superior
> to models with direct inputs in one of the frontal regions.
> However, is this a valid approach?
>
> Best greetings
>
> Tom Ethofer
>
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|