Hello. I’m trying to understand the output of arviz.plot_compare. Specifically, I’m trying to understand how the deltas should be interpreted.
Here, can I say that Model A is significantly better than Model B? Is there a way to compare two consecutively ranked models? I am not exactly sure if I understand what deltas mean. Any help will be greatly appreciated.
I’ve never used arviz, but the plot seems a bit confusing…
elpd_diff from the loo package is the difference in ELPD between the model with largest ELPD and the model being compared. This can easily be seen in the above table by doing the subtraction of ELPD of the models from the best model and comparing to
elpd_diff. I’m not sure why the plot would show this difference overlaid above the mean estimated ELPD for each model…? In any case, @avehtari 's Cross-validation FAQ may be helpful to you Cross-validation FAQ