Posthoc contrasts with emmeans, tidybayes, and brms

bayesian_posthoc.R

library(tidyverse)
library(emmeans)
library(brms)
library(tidybayes)

warp.brms <- brm(breaks ~ wool * tension, data = warpbreaks)

#get the adjusted means
warp_em <- emmeans (warp.brms,  ~ tension | wool)
warp_em

#get all possible contrasts
cont <- contrast(warm_em, "tukey")
cont

#get the posterior draws from the contrasts
cont_posterior <- gather_emmeans_draws(cont)

#plot
ggplot(cont_posterior,
       aes(y = contrast, x = .value)) +
  geom_halfeyeh() +
  facet_wrap(~wool) +
  geom_vline(xintercept = 0, color = "red", lty = 2)


ggplot(cont_posterior,
       aes(y = contrast, x = .value, fill = wool, group = wool)) +
  geom_halfeyeh(alpha = 0.5)+
  geom_vline(xintercept = 0, color = "red", lty = 2)

#need to figure out how to make this look all posterior-y
emmip(warp.brms,  ~ tension | wool, CIs = TRUE, cov.reduce = range)

Hey thank you so much for this code! This definitely helped me to get going myself. I am wondering though is there any reason why you used contrast(..., "tukey") instead of pairwise? I am getting the same results for both for my model but I want to be sure that I am not missing anything about using tukey for Bayesian mdoels.