 |
 |
Dear all, From Klaas paper (2010 - ten simple rules) I understand that BMA is a "promissing method for comparing parameters across groups". After reading Penny (2010 - Comparing model families) I still am not able to understand: What is the best way to statistically compare the DCM modulatory parameter estimates obtained with rfxBMA for two groups apart? Is this at all justified to do? In our study, we expect differences in both modulatory and effective connectivity strengths between two groups. The BMS model comparison and BMA family comparison both show that there is difference in the best models for two groups (i.e TL group has clear preference for say model 1 (posterior density of 0.9), and group TCL has slightly better model 6 (posterior density 0.4, and for model 1 something like 0.31)). Different models have all effective connectivities the same, because we assume that the difference should be only in connectivity strengths. Comparing connectivity strengths with t-test for same models (e.g. model 1) shows no difference. But after performing BMA and plotting the posterior densities for average network parameters (see attached figure) there is obviously difference between goups TL and TCL for a given connection. But how to understand and evaluate statistically this difference? Thank you Branislava Curcic-Blake