Dear Audreyana
You are quite right that you can test for an age effect by fitting two PEB models, with and without the age effect, and comparing the free energy. You would do this by loading each PEB structure in the Matlab workspace and running:
models = {PEB1 PEB2};
post = spm_dcm_bmc(models);
Where post is the posterior probability for each model. However, you don't need to fit two PEB models. You can fit one 'full' model containing the age effect, and derive the evidence (free energy) for the second model using Bayesian Model Reduction. This is done on a per-connection basis using spm_dcm_peb_bmc (you'll get a probability for each connection with vs without the age effect). Alternatively, to compare all age effects switched on vs all age effects switched off, you can use spm_dcm_bmc_peb. (Sorry that the function names are quite similar!) This will try all mixtures of switching on and off all covariates.
Let me know if you need more help with that.
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Audreyana Jagger
Sent: 16 January 2019 02:41
To: [log in to unmask]
Subject: [SPM] PEB Model Selection
Hello SPM team,
I am testing age-related effects in a spectral dynamic causal modeling project.
I want to use BMS to select between two PEB models (one PEB with age effects included as a covariate and the other PEB with no age effects). With this test, I believe I am testing if there is more evidence to support an age effect of effective connectivity over no age effect on effective connectivity. Is there a simple way to use the current spm_run_dcm_bms to test this hypothesis? If not, is there another way to test for an age-related effect at the group level?
Thank you in advance,
Audreyana
|