Yes, that is what I also believe and the OP corrects the error when changing the likelihood from NGB to lognormal. However, the question is should the OP not have compared the difference in group1 vs group2 by taking the difference in of the posteriors of group2 and group1. Instead of looking at the Deterministic “diff_of_means” variable, that does not seem to be conditioned on the actual observations of the data?