Scaling weights necessary for Bayesian multi-level regression?

Hi everyone,

for frequentist multi-level model, it is usually recommended to scale sampling weights (e.g. with parameters::rescale_weights()), as the variance parameter estimators obtained by the pseudo-maximum-likelihood method can be biased with small sample sizes (among other reasons, https://assets.researchsquare.com/files/rs-53309/v1_covered.pdf?c=1623082270).
I was wondering whether such a recommendation exists for Bayesian multi-level models as well or whether the raw (non-scaled) weights can be used.

Many thanks in advance.
Sandra

I glanced over the citation. My hunch is that the rescaling is proposed to improve the estimation procedure. Bayesian methods solve this at the expense of specifying a “good” prior distribution.

Many thanks for taking the time to take a look at the citation. And thanks for the advice!