@ldeschamps, that is a good thought but Omega_cond[1,2] really is distributed around 0.27. See included histogram.
@LeoGrin, could you elaborate a little bit more about the difference between a mixture of normal distributions with equal proportions vs a sum of normal distributions? It sounds like I have made a rookie mistake in specifying my model’s likelihood. What kind of stan code are you thinking of for a sum of normals likelihood?
Edit: @LeoGrin is right. My observation is the sum of two gaussian random variables. The mixture model assumes the observation is one variable or the other with a certain mixing probability – but never the sum. Obviously, then, my model is going to give me unexpected results. Thank you for helping me reason through this!
I am still having some trouble understanding how to model a sum of Gaussians, for which I have started a separate thread.