Multi-ci-levels don't match rows of "grid" like inputs

[qdread]

I noticed that after updating from v0.14.0 to v0.15.0, the behavior when printing an object of class describe_posterior created by calling describe_posterior() on an object of class emmGrid is incorrect, if I specify more than one CI width. There was clearly a breaking change having to do with the reshaping that is done to print the output. An incorrect median is printed for some of the rows, and some of the credible intervals are not printed. Warnings from reshapeWide are printed. I have kept R version and all other package versions constant, and only changed the bayestestR version, to produce the following output.

Thanks for the excellent package. Please let me know if I can help fix the bug!

Code to run

library(emmeans)
library(bayestestR)
library(brms)

mod <- brm(sales1 ~ factor(store), data = oranges, seed = 1)
emm <- emmeans(mod, ~ store)

describe_posterior(emm, centrality = 'median', ci = c(0.90, 0.95))

Output using bayestestR 0.14.0

This is the expected behavior.

Summary of Posterior Distribution

Parameter | Median |         90% CI |        95% CI |     pd |          ROPE | % in ROPE
----------------------------------------------------------------------------------------
1         |  12.75 | [ 8.35, 17.26] | [7.41, 18.08] |   100% | [-0.10, 0.10] |        0%
2         |   8.30 | [ 3.78, 12.83] | [2.84, 13.89] | 99.83% | [-0.10, 0.10] |        0%
3         |   6.59 | [ 2.06, 11.15] | [1.13, 11.93] | 98.95% | [-0.10, 0.10] |        0%
4         |   8.30 | [ 3.85, 12.57] | [2.86, 13.45] | 99.78% | [-0.10, 0.10] |        0%
5         |   9.76 | [ 5.30, 14.39] | [4.52, 15.34] |   100% | [-0.10, 0.10] |        0%
6         |  15.25 | [10.90, 19.57] | [9.96, 20.41] |   100% | [-0.10, 0.10] |        0%

Output using bayestestR 0.15.0

This is not the correct output. As you can see we get duplicated medians, incorrect CIs, and some blank CIs. The output is followed by warnings from reshapeWide.

Summary of Posterior Distribution

store | Median |        90% CI |        95% CI |     pd |          ROPE | % in ROPE
-----------------------------------------------------------------------------------
1     |  12.75 | [8.35, 17.26] |               |   100% | [-0.10, 0.10] |        0%
2     |  12.75 |               | [7.41, 18.08] |   100% | [-0.10, 0.10] |        0%
3     |   8.30 | [3.78, 12.83] |               | 99.83% | [-0.10, 0.10] |        0%
4     |   8.30 |               | [2.84, 13.89] | 99.83% | [-0.10, 0.10] |        0%
5     |   6.59 | [2.06, 11.15] |               | 98.95% | [-0.10, 0.10] |        0%
6     |   6.59 |               | [1.13, 11.93] | 98.95% | [-0.10, 0.10] |        0%
Warning messages:
1: In reshapeWide(data, idvar = idvar, timevar = timevar, varying = varying,  :
  some constant variables (Median,pd) are really varying
2: In reshapeWide(data, idvar = idvar, timevar = timevar, varying = varying,  :
  multiple rows match for CI=0.9: first taken
3: In reshapeWide(data, idvar = idvar, timevar = timevar, varying = varying,  :
  multiple rows match for CI=0.95: first taken

[strengejacke]

@mattansb ?

[mattansb]

Good catch!

Should be fixed now

library(bayestestR)
library(brms)


data("oranges")
oranges$store <- factor(oranges$store)
mod <- brm(sales1 ~ store, data = oranges, seed = 1)




# emmeans
emm <- emmeans::emmeans(mod, ~ store)

describe_posterior(emm, centrality = 'median', ci = c(0.90, 0.95))
#> Summary of Posterior Distribution
#> 
#> store | Median |         90% CI |        95% CI |     pd |          ROPE | % in ROPE
#> ------------------------------------------------------------------------------------
#> 1     |  12.75 | [ 8.35, 17.26] | [7.41, 18.08] |   100% | [-0.10, 0.10] |        0%
#> 2     |   8.30 | [ 3.78, 12.83] | [2.84, 13.89] | 99.83% | [-0.10, 0.10] |        0%
#> 3     |   6.59 | [ 2.06, 11.15] | [1.13, 11.93] | 98.95% | [-0.10, 0.10] |        0%
#> 4     |   8.30 | [ 3.85, 12.57] | [2.86, 13.45] | 99.78% | [-0.10, 0.10] |        0%
#> 5     |   9.76 | [ 5.30, 14.39] | [4.52, 15.34] |   100% | [-0.10, 0.10] |        0%
#> 6     |  15.25 | [10.90, 19.57] | [9.96, 20.41] |   100% | [-0.10, 0.10] |        0%



pr <- marginaleffects::avg_predictions(mod, variables = "store")

describe_posterior(pr, centrality = 'median', ci = c(0.90, 0.95))
#> Summary of Posterior Distribution
#> 
#> store | Median |         90% CI |        95% CI |     pd |          ROPE | % in ROPE
#> ------------------------------------------------------------------------------------
#> 1     |  12.75 | [ 8.35, 17.26] | [7.41, 18.08] |   100% | [-0.10, 0.10] |        0%
#> 2     |   8.30 | [ 3.78, 12.83] | [2.84, 13.89] | 99.83% | [-0.10, 0.10] |        0%
#> 3     |   6.59 | [ 2.06, 11.15] | [1.13, 11.93] | 98.95% | [-0.10, 0.10] |        0%
#> 4     |   8.30 | [ 3.85, 12.57] | [2.86, 13.45] | 99.78% | [-0.10, 0.10] |        0%
#> 5     |   9.76 | [ 5.30, 14.39] | [4.52, 15.34] |   100% | [-0.10, 0.10] |        0%
#> 6     |  15.25 | [10.90, 19.57] | [9.96, 20.41] |   100% | [-0.10, 0.10] |        0%




# data frame with rvar
grid <- data.frame(store = 1:6)
grid$pred <- posterior_epred(mod, newdata = grid) |> posterior::rvar()

describe_posterior(grid, rvar_col = "pred", centrality = 'median', ci = c(0.90, 0.95))
#> Summary of Posterior Distribution
#> 
#> store | Median |         90% CI |        95% CI |     pd |          ROPE | % in ROPE
#> ------------------------------------------------------------------------------------
#> 1.00  |  12.75 | [ 8.35, 17.26] | [7.41, 18.08] |   100% | [-0.10, 0.10] |        0%
#> 2.00  |   8.30 | [ 3.78, 12.83] | [2.84, 13.89] | 99.83% | [-0.10, 0.10] |        0%
#> 3.00  |   6.59 | [ 2.06, 11.15] | [1.13, 11.93] | 98.95% | [-0.10, 0.10] |        0%
#> 4.00  |   8.30 | [ 3.85, 12.57] | [2.86, 13.45] | 99.78% | [-0.10, 0.10] |        0%
#> 5.00  |   9.76 | [ 5.30, 14.39] | [4.52, 15.34] |   100% | [-0.10, 0.10] |        0%
#> 6.00  |  15.25 | [10.90, 19.57] | [9.96, 20.41] |   100% | [-0.10, 0.10] |        0%