I have a question about some sort of odd DCM results. The winning model selected by rfx BMS included a B parameter modulation of connection X, let's say. In using the summary statistic approach for parameter interpretation, we found that connection X was not statistically significantly different from zero across the group, however, modulatory parameter B was statistically significant from zero.

I understand that B parameters are additive in nature but this result seems somewhat questionable to me. 

Might anyone have any suggestions for further interpreting this or would another parameter inference method be more appropriate?

Best,

Drew