Is there a way to easily calculate some estimate of the bias E(x - μ)
for some estimate in Arviz? Given that LOO is doing something similar to jackknife estimation already, it should be possible to calculate/return a jackknife-like estimate of the bias, I think.
(While debiasing a Bayesian estimate is usually a poor choice, I’m interested in the bias as a summary statistic for evaluating model fit – very high values for bias can indicate a bad model.)