Computed `prop` is hard to use.

[teunbrand]

This issue was first brought to my attention by @yjunechoe a while back, and thought it was salient enough to merit an issue.
In stat_count() there is a computed variable prop that is basically the count divided by the total count in the group.
Because the default group is the interaction between all discrete variables, this makes vanilla-use of prop a bit pointless:

library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 4.3.1
p <- ggplot(diamonds, aes(cut, fill = color)) +
  geom_bar(aes(y = after_stat(prop)), position = "dodge")
p

It only starts making sense if you manually set groups:

p + aes(group = color)

However, you cannot compute the proportion within a position:

p + aes(group = cut)
#> Warning: The following aesthetics were dropped during statistical transformation: fill
#> ℹ This can happen when ggplot fails to infer the correct grouping structure in
#>   the data.
#> ℹ Did you forget to specify a `group` aesthetic or to convert a numerical
#>   variable into a factor?

^{Created on 2023-11-02 with reprex v2.0.2}

And there are variants of prop one might be interested in:

• Group-wise proportion (basically current prop)
• Position-wise proportion (difficult to compute as user)
• Panel-wise proportion (difficult to compute as user)
• Proportion of total (can easily be computed by aes(y = after_stat(count / sum(count))))

So the suggestion here is to add the last 3 variants as prop_position, prop_panel and prop_total (or prop_layer) as new computed variables to stat_count().

[larmarange]

Just for the history of the discussion, a similar issue was raised in the past: #3369 and #3376

stat_prop() is an alternative proposition, implemented in ggstats: https://larmarange.github.io/ggstats/articles/stat_prop.html and https://larmarange.github.io/ggstats/reference/stat_prop.html

Denominators are here defined according to a by aesthetic

[teunbrand]

Thanks for the history @larmarange! I hadn't realised this has been discussed several times before.
I'll guess I'll just leave some solutions to find for the next people to raise this issue:

library(ggplot2)
p <- ggplot(diamonds, aes(cut, fill = color)) +
  geom_bar(position = "dodge")

# Proportion per group
p + aes(y = after_stat(prop), group = color)

# Proportion of total
p + aes(y = after_stat(count / sum(count)))

# Proportion per x-position
p + aes(y = after_stat(count / ave(count, x, FUN = sum)))

# Proportion per panel
p + aes(y = after_stat(count / ave(count, PANEL, FUN = sum))) +
  facet_wrap(~ carat > 0.7)

# Proportion per panel and x position
p + aes(y = after_stat(count / ave(count, x, PANEL, FUN = sum))) +
  facet_wrap(~ carat > 0.7)

# Proportion per panel and fill
p + aes(y = after_stat(count / ave(count, fill, PANEL, FUN = sum))) +
  facet_wrap(~ carat > 0.7)

[teunbrand]

Or with some extra rlang spice:

library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 4.3.1

as.prop <- function(what, ...) {
  what / ave(what, ..., FUN = function(x) sum(abs(x)))
}

prop_by <- function(...) {
  dots <- rlang::enquos(...)
  if (length(dots) == 0) {
    expr <- rlang::expr(after_stat(count / sum(count)))
    return(expr)
  }
  expr <- lapply(dots, rlang::quo_get_expr)
  rlang::expr(after_stat(as.prop(count, !!!expr)))
}

p <- ggplot(diamonds, aes(cut, fill = color)) +
  geom_bar(position = "dodge")

# Proportion of total
p + aes(y = !!prop_by())

# Proportion by x and panel
p + aes(y = !!prop_by(x, PANEL)) +
  facet_wrap(~ carat > 0.7)

^{Created on 2023-11-02 with reprex v2.0.2}

[yjunechoe]

Speaking of abusing delayed aesthetic evaluation, you can also call prop_by() inside of after_stat() and kidnap the computed count variable from the after-stat context, then proceed to do stuff with it with just standard evaluation (I won't pollute the closed issue further I promise 😅):

library(ggplot2)

prop_by <- function(...) {
  count <- evalq(count, parent.frame())
  if (length(list(...)) == 0) {
    count / sum(count)
  } else {
    count / ave(count, ..., FUN = function(x) sum(abs(x)))
  }
}

p <- ggplot(diamonds, aes(cut, fill = color)) +
  geom_bar(position = "dodge")

# Proportion of total
p + aes(y = after_stat(prop_by()))

# Proportion by x and panel
p + aes(y = after_stat(prop_by(x, PANEL))) +
  facet_wrap(~ carat > 0.7)

^{Created on 2023-11-02 with reprex v2.0.2}

[teunbrand]

I was today years old when I learned about the evalq() function ^_^