Dear Becky,
I'm not familiar with the R package FIAR.
I did a quick google search and got the R package FAIR:
http://rgm2.lab.nig.ac.jp/RGM2/func.php?rd_id=FAiR:00FAiR-package
If you are using this one, which seems to be based on Factor analysis, then I don't
think it will generate data from DCMs generative model (which uses differential equations).
But then maybe you are not using this one ?
Best, Will.
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
> On Behalf Of Becky van den Honert
> Sent: 28 April 2012 16:06
> To: [log in to unmask]
> Subject: [SPM] DCM fails to select model that generated the data
>
> Hi DCM experts,
>
> I recently conducted a couple of experiments to familiarize myself with
> DCM and got some really surprising results; I would appreciate any
> feedback to find potential errors in my implementation of DCM.
> First I applied DCM to an existing dataset in the lab that had been
> used to test granger, and got results that were inconsistent with a
> seemingly highly probable hypothesis. Conceding that we weren’t 100%
> sure of our effective connectivity hypothesis, I turned to simulations.
> I simulated data using the following parameters/procedure (using the R
> package FIAR). Everything (timing, scanning parameters etc.) was based
> on a template subject’s scan.
>
> - Single 5 min scan
> - TR = 0.240 sec (to be consistent with granger dataset parameters)
> - TE = .025 sec
> - A = [0 0 0 0; 0 0 0 0; .01 1 0 .01; 1 .01 .01 0]
> - B = zeros
> - C = [1 .5 0; .5 1 0; 0 0 0; 0 0 0]
> - SNR = 10, and then in another simulation, NO noise was added.
> - Each of the first two inputs had 10 unique onsets (~14 seconds apart,
> duration of .5 seconds).
>
> I input the timeseries generated from the above model into our SPM
> scripts by replacing the following fields in the template subject’s
> DCM.mat file:
>
> - DCM.Y.y(:, roi) = fake timeseries generated for that roi
> - DCM.xY(1,roi).y = fake timeseries
> - DCM.xY(1,roi).u = fake timeseries
>
> I included just three models in model space. The first was the one used
> to generate the data. The second was the “opposite” model in which the
> intrinsic connections pointed the other direction and the direct inputs
> entered in the opposite pair of regions. The third was the union of
> those two models.
>
> The BMS results (FFX with one subject, and FFX & RFX when I repeated
> this with two template subjects) always yield the third model (union
> model) as the very, very clear winner.
>
> How can this be? There are a number of things that we have done that I
> understand are atypical (fast TR, no modulations, use FIAR to generate
> data, short-cut timeseries replacement …), but nothing I can think of
> should have made the wrong model win so definitively.
>
> I would be happy to follow up with more details about what we have
> tried.
>
> Thank you for your time,
> Becky van den Honert
|