Brm_multiple not converging though separate brm models do

I am attempting to fit multiple imputed datasets with brm_multiple, however, when I provide the three datasets as the list for brm_multiple, they do not converge. When I fit them separately using brm they converge. Can you help me understand why this might be the case? For reference, I am also running the imputation within brm, using mi() but these models take over a week with my 62,000 observations and 5 step imputations, thus why I’m pursuing imputed prior to fitting.

Here is my general code for the brm_multiple (which doesn’t converge and has very small effective sample sizes) vs the brm model (which converges with many effective sample sizes).

output6_a <- brm(formula6, prior = set_prior("normal(0, 1)", class = "b"), data = imp_a, chains = 4, iter = 4000, warmup = 2000, seed = 02152019, cores = 4)
output6_b <- brm(formula6, prior = set_prior("normal(0, 1)", class = "b"), data = imp_b, chains = 4, iter = 4000, warmup = 2000, seed = 02152019, cores = 4)
output6_c <- brm(formula6, prior = set_prior("normal(0, 1)", class = "b"), data = imp_c, chains = 4, iter = 4000, warmup = 2000, seed = 02152019, cores = 4)

imps <- list(imp_a,imp_b,imp_c)
output <- brm_multiple(formula6, data = imps, prior = set_prior("normal(0, 1)"), chains = 4, iter = 4000, warmup = 2000, seed = 02152019, cores = 4)

Thank you for brms and all the hard work. It’s an amazing product. Any insights are greatly appreciated.

Please also provide the following information in addition to your question:

  • brms Version: 3.8.0

See ?brm_multiple for the explanation.

Thanks Paul. Sorry I missed this. Just to clarify, the statements:

The combined model may issue false positive convergence warnings, as the MCMC chains corresponding to different datasets may not necessarily overlap, even if each of the original models did converge. To find out whether each of the original models converged, investigatefit$rhats, wherefitdenotes the output ofbrm_multiple.

Is what you’re referring to? My question is if brm_multiple is providing false **negative** convergence warnings. If the fit$rhats are all between 0.99 and 1.01, does that indicate that the combined model converged or am I misunderstanding (which is totally possible)? Thanks for the clarification.

Your understanding seems correct.