Calculating model averaged estimates using loo

[avehtari]

@gunhanb

I am trying to use pseudo-BMA as described in “Using Stacking to Average Bayesian Predictive Distributions” by Yao et al 2018.

Please use stacking instead of Pseudo-BMA, as it’s much better.

# stacking method:
wts1 <- loo_model_weights(
  log_lik_list,
  method = "stacking",
  r_eff_list = r_eff_list,
  optim_control = list(reltol=1e-10)
)

The code example you listed is using stacking instead of Pseudo-BMA.

@gunhanb

How can I calculate the model averaged posterior estimates, in this example “model averaged” estimate of the sigma parameter?

Stacking (and pseudo-BMA) can be used to combine predictive distributions, but it’s sensible for averaging parameters only in some special cases. It’s likely that averaging sigmas doesn’t make sense. Instead you should sample from the combined predictive distribution and focus on the predictions.