Problemas de convergência em brms

Modeling
1

Hi,
I am currently working on a hierarchical quantile regression model with three levels (student, class, and school). At this stage, I am testing a null model with random effects at the class and school levels. However, I am facing convergence issues, as the Rhat values are consistently above 1.05, with many exceeding 2 or even 3.

Below is the R code I used:

m03 ← brm(
formula = Proficiencia ~ 1 + (1 | ID_escola/ID_classe/ID_aluno),
family = asym_laplace(),
data = dados,
chains = 4,
cores = 4,
iter = 2000,
seed = 123
)

Additionally, I attempted to include control = list(adapt_delta = 0.99, max_treedepth = 15) to improve model adaptation. However, the execution time became excessively long, and I still have not achieved the desired results.
I would greatly appreciate any suggestions that might help resolve this issue.

Thanks,
Marcos Machado

2

Hey and welcome.

A couple things that will help folks answer this.
Can you share the data or simulated data if the data is not shareable?
Can you share the full R code?
What version of R and brms are you using? What OS are you on?

Specifically a me thing following:
I wouldn’t set seed at all. I don’t know if that’s current best practices.
For all my new models (like new to me), I always start with the simplest but interesting model and see if that works.

So something like:
Proficiencia ~ 1 + (1 | ID_escola)

and what is the reasoning behind family = asym_laplace()?
and did you set your priors? or use get_priors() first? I really recommend setting your own priors.