Mapping alpha within geom_col changes the sorting of the column

[margaret-colville]

I found a problem with creating a column chart where I was trying to highlight some specific segments using transparency. Applying an alpha aes within the geom_col changes the order of the sorting for the fill - while doing so within the ggplot call leaves the ordering untouched.

I expected ordering to be unaffected by transparency in either case.

Here is the code to reproduce the bug:

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(ggplot2)

penguins %>%
  filter(!is.na(sex)) %>%
  group_by(species, island, sex) %>%
  summarise(count = n()) %>%
  ungroup() %>%
  mutate(flag = ifelse(species == 'Chinstrap' & sex == 'male', 'flag', 'no flag')) %>%
  ggplot(aes(x = species, y = count, fill = sex)) +
  geom_col(aes(alpha = flag)) +
  facet_wrap(~island)
#> `summarise()` has grouped output by 'species', 'island'. You can override using
#> the `.groups` argument.
#> Warning: Using alpha for a discrete variable is not advised.

penguins %>%
  filter(!is.na(sex)) %>%
  group_by(species, island, sex) %>%
  summarise(count = n()) %>%
  ungroup() %>%
  mutate(flag = ifelse(species == 'Chinstrap' & sex == 'male', 'flag', 'no flag')) %>%
  ggplot(aes(x = species, y = count, fill = sex, alpha = flag)) +
  geom_col() +
  facet_wrap(~island)
#> `summarise()` has grouped output by 'species', 'island'. You can override using
#> the `.groups` argument.
#> Warning: Using alpha for a discrete variable is not advised.

^{Created on 2025-05-13 with reprex v2.1.1}

[smouksassi]

this is not related to alpha it is related to using "flag" as an aesthetic and then it chagnes to ordering of the stacking see below also how to apply the alpha with scale_alpha_identity without affecting the stacking ordering.

library(ggplot2)
library(dplyr)
library(patchwork)

plotdata <-penguins %>%
filter(!is.na(sex)) %>%
group_by(species, island, sex) %>%
summarise(count = n()) %>%
ungroup() %>%
mutate(flag = ifelse(species == 'Chinstrap' & sex == 'male', 'flag', 'no flag')) %>%
mutate(alphavalue = ifelse(species == 'Chinstrap' & sex == 'male', 0.1,1))

ggplot(plotdata,
aes(x = species, y = count, fill = sex)) +
geom_col(aes(alpha = alphavalue)) +
scale_alpha_identity()+
facet_wrap(~island)

a<- ggplot(plotdata,
aes(x = species, y = count, fill = sex)) +
geom_col(aes(alpha = alphavalue)) +
scale_alpha_identity()+
facet_wrap(~island)
b <- ggplot(plotdata,
aes(x = species, y = count, fill = sex)) +
geom_col(aes(linetype=flag)) +
facet_wrap(~island)
a/b

[teunbrand]

Thank you for the report. This is due to how default groups are derived. By default, they're an interaction of all discrete variables, but order matters. In the 1st example, alpha is up front, whereas in the 2nd example, alpha is in the back. You can use aes(alpha = flag, x = species, y = count, fill = sex) in the second example to get the order the same as in the first example. If you want to intervene, you can set groups manually. I don't think changing ggplot2's behaviour with regards to this will generally affect people's plots positively, so I'm proposing to close this issue.