After some more troubleshooting and finding this helpful answer on a related post from @bwiernik here, I am suspicious that the right answer is just that re_formula = NA
instead of Null
will do what I want? So something like this?
x <- df %>%
# dplyr::filter(!is.na(site), !is.na(week)) %>%
modelr::data_grid(year, site, week) %>%
tidybayes::add_epred_draws(bayes_fit, re_formula = NA)
ggplot(x, aes(x = .epred, y = year, fill = year)) +
tidybayes::stat_halfeye(.width = 0.95) +
scale_fill_manual(values = c(rep("lightpink", 16))) +
labs(
x = "Count", y = "Year",
subtitle = "Posterior predictions"
) +
theme(legend.position = "bottom") +
theme_bw()
which gives this:
Which feels right??