Madeline ran the model in brms as it’s much faster. She did this, assuming the same results as nimble, yet was surprised to see that although effect sizes seemed the same, the credible intervals were not.

You’ll be able to see below that it seems as if NIMBLE has a difficult time estimating the random effect of year (which does make sense as there are only four levels). We’re not exactly sure right now why the brms model does a better job at estimating the year effect, as they both have the same priors and use the same complier.

As we try and figure this out, I wanted to present both results so we can decide which one to use.

NIMBLE model results