Having inference flow "one way" in a model with conditional sampling

[StaffanBetner]

I have been tinkering with some external C++ to get a similar thing working for within-model bootstrapping, so that the (continuous) weights for the observations are generated in each draw. Not for any particular use case, but more along the lines of “can it be done”. Here is how I approached it:

Fully Bayesian Bootstrap

Some progress has been made! I will show an example using the bearingcage dataset from the fwb package, a bootstrap package for fractional weight bootstrapping. First we load some packages: library(tidyverse) library(fwb) library(cmdstanr) library(brms) library(posterior) library(ggdist) We start by with estimating the ordinary model, with default priors from ´brms`: > brm(formula = hours|cens(!failure)~1, + family = weibull, + data = bearingcage, + backend = "cmdstanr", + …