Christian H. Kasess1,2, Andreas Weissenbacher1, Lukas Pezawas2, Ewald Moser1,3, Christian Windischberger1,3
1MR Center of Excellence, Medical University Vienna, Vienna, Austria; 2Division of Biological Psychiatry, Department of Psychiatry and Psychotherapy, Medical University Vienna, Vienna, Austria; 3Center for Biomedical Engineering and Physics, Medical University Vienna, Vienna, Austria
A number of different approaches have been proposed to infer group statistics for dynamic causal models: (a) classic random effects (RFX) second level analysis based on the mean parameter estimates ignoring intra subject variance, (b) Bayesian averaging, combining multivariate single subject posterior parameter distributions according to Bayes theorem and (c) temporal averaging. Here, we compared these methods based on simulated data. We found that RFX and temporal averaging analysis are more robust against heterogeneity in the study population than Bayesian averaging which, for a heterogeneous population, can yield results that only partly resemble the input statistics.