Inspired by the latest user meeting, I'm running some two level DSEM models. However, convergence based on the Gelman-Rubin PSR takes about 10 hours. A shorter number of iterations shows that the chains are not stationary for several parameters.
I don't recall whether it was discussed at the meeting how convergence can be hastened, and I can't find it in the slides. My first thought was that the key might be starting values and/or priors, but I'm not sure how to specify these. Any obvious time drains I should address?