Dear SPM experts,
I currently try my hands on Bayesian models in SPM. The tutorials on the
matter seem rather sparse and I cannot find out what my problem is.
I firstly tried to use Classically estimated 1st Level and 2nd level
data (as I learnd you first have to estimate the 2nd level model with
the Classic/Frequentistic method) as it is mentioned in Neumann &
Lohmann (2003, https://doi.org/10.1016/S1053-8119(03)00443-9) in the 2nd
level Bayesian Model and got the error notion:
"Index exceeds the number of array elements (0)." after 3 ReML
estimations, as looking at the source code the iQ matrix in the
"Bayesian model reduction (successive removal of parameters)" step gets
reduced to a 0-dimensional matrix and thus the following steps seem to
fail.
I thought maybe it is better to use already Bayesian 1st level data. But
after painstakingly estiamting all the models again using the
not-smoothed just normalised data (as that was mentioned in a previous
mailinglist post), I today arrived at the exact same error message when
trying the 2nd level Bayesian estimation. Also just looking at the
results on a single subject level and also the contrast on the 2nd level
using Classically estiamted model reveals results that can just be
interpreted as noise; just very scattered single voxels all across the
brain.
Does that mean there the model gets reduced to 0 because there is just
noise in the data an no parameter explains anything, what I highly doubt
since the Classic analysis did show results?
Or did I just do something terribly wrong here? Does anybody per chance
knows a good source for Bayesian modelling of task data in SPM?
Thank you very much in advance and best wishes,
Falko
--
Falko Mecklenbrauck, M.Sc.
PhD Student
University of Münster
Institute of Psychology
Biological Psychology
Fliednerstr. 21
Room 310b
Tel.: +49 251 83 34101
|