I have run second level PEB/BMA analysis on DCMs estimated using the cross-spectral method run on resting-state fMRI data and am trying to examine second level effects using the spm_dcm_peb_review routine.

In particular, I am trying to compare the two thresholding methods: Probability that parameter > 0 and Free Energy (with vs without).

If i use the probability approach, the effects i see in my A matrix disappear as I increase the threshold - i.e., 0 0.5 0.75 0.95 0.99. Great.

But if I switch to the free energy approach, the threshold has no impact, all my effects appear to remain present all the way up to very strong evidence Pp > 0.99. While it would awesome to think all my effects are super strong, this seems very unlikely since the number of effects vary noticeably as a function of the probability threshold method. Can someone assist with troubleshooting this? 

This screenshot shows the effect of a covariate of interest on my A matrix, I have selected the effective connectivity parameter with the largest value to give a sense of the effect sizes I'm looking at.