Although easy to generate a data frame with rvars from {emmeans} (similar to what's possible with marginaleffects::posterior_draws()), it would be nice to have this option built in - it will fit nicely with the other *_rvars() alternatives to the *_draws() functions.
library(dplyr)
library(brms)
library(emmeans)
library(tidybayes)
set.seed(5)
n <- 10
ABC = tibble(
condition = rep(c("A","B","C","D","E"), n),
response = rnorm(n * 5, c(0,1,2,1,-1), 0.5)
)
m <- brm(response ~ condition, data = ABC,
chains = 1, iter = 500)
eml <- emmeans(m, pairwise ~ condition)
gather_emmeans_draws(eml)
#> # A tibble: 3,750 × 7
#> # Groups: condition, contrast, .grid [15]
#> .grid condition .chain .iteration .draw .value contrast
#> <chr> <chr> <int> <int> <int> <dbl> <chr>
#> 1 emmeans A NA NA 1 -0.271 NA
#> 2 emmeans A NA NA 2 -0.00172 NA
#> 3 emmeans A NA NA 3 -0.167 NA
#> 4 emmeans A NA NA 4 0.0142 NA
#> 5 emmeans A NA NA 5 0.0934 NA
#> 6 emmeans A NA NA 6 -0.0276 NA
#> 7 emmeans A NA NA 7 -0.0110 NA
#> 8 emmeans A NA NA 8 0.249 NA
#> 9 emmeans A NA NA 9 0.162 NA
#> 10 emmeans A NA NA 10 0.165 NA
#> # ℹ 3,740 more rows
#> # ℹ Use `print(n = ...)` to see more rows
gather_emmeans_rvars <- function(object, ...) {
tidybayes::nest_rvars(tidybayes::gather_emmeans_draws(object))
}
gather_emmeans_rvars(em_condition)
#> # A tibble: 15 × 4
#> # Groups: condition, contrast, .grid [15]
#> condition contrast .grid .value
#> <chr> <chr> <chr> <rvar[1d]>
#> 1 A NA emmeans 0.1914 ± 0.18
#> 2 B NA emmeans 1.0134 ± 0.18
#> 3 C NA emmeans 1.8733 ± 0.18
#> 4 D NA emmeans 1.0220 ± 0.18
#> 5 E NA emmeans -0.9509 ± 0.17
#> 6 NA A - B contrasts -0.8220 ± 0.27
#> 7 NA A - C contrasts -1.6819 ± 0.26
#> 8 NA A - D contrasts -0.8306 ± 0.24
#> 9 NA A - E contrasts 1.1423 ± 0.25
#> 10 NA B - C contrasts -0.8600 ± 0.25
#> 11 NA B - D contrasts -0.0086 ± 0.26
#> 12 NA B - E contrasts 1.9643 ± 0.24
#> 13 NA C - D contrasts 0.8514 ± 0.25
#> 14 NA C - E contrasts 2.8242 ± 0.25
#> 15 NA D - E contrasts 1.9729 ± 0.26
Although easy to generate a data frame with
rvars from{emmeans}(similar to what's possible withmarginaleffects::posterior_draws()), it would be nice to have this option built in - it will fit nicely with the other*_rvars()alternatives to the*_draws()functions.