Is it ok to ravel() over all chains/draws to compare the posteriors of two means?

Yes ravelling is fine. The draws from the different chains should be statistically equivalent if the sampler converged

2 Likes