The t-test ignores the uncertainty about the parameters individually (their variance) and the dependencies between them (their covariance). Additionally, the BPA function (spm_dcm_average.m in older versions, spm_dcm_bpa.m in the most recent releases) has an option for disabling conditional dependencies - i.e. the covariance between parameters. Enabling this option will bring the average closer to the arithmetic mean. For more information, see https://en.wikibooks.org/wiki/SPM/Bayesian_Parameter_Averaging_(BPA) .
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Toshi, K
Sent: 14 February 2017 09:13
To: [log in to unmask]
Subject: [SPM] different DCM results between simple averaging and BPA averaging
Dear SPM experts.
I'm dealing with the spectral DCM analyses for resting-state fMRI data.
I think I finished the individual level analyses and got the values of connection strength (in Hz) for each individual from "DCM.Ep.A".
I took those values and t-tested to know whether the connection was different from zero.
For confirmation, I averaged those values across all participants but the value was totally different from the results of BPA.
Could you tell me why the value was different ??
In addition, was the t-test technically correct ?
Any comment would be appreciated.