1

I got stuck when trying to add percentage labels to a faceted bar plot with bars filled by another variable, such as the example below:

mtcars %>% 
  ggplot(aes(x = factor(gear) %>% droplevels(), fill = factor(am))) +
  facet_grid(
    cols = vars(cyl), scales = "free_x", space = "free_x", margins = TRUE
  ) +
  geom_bar(position = "fill") +
  geom_text(
    aes(label = ..count.., y = ..count..), stat = "count",
    position = position_fill(vjust = .5)
  )

Created on 2021-02-26 by the reprex package (v0.3.0)

In the example, the labels are counts instead of percentages of am by gear for each cyl. I therefore tried to replace the label = argument in the aes() of geom_text() as

label = scales::percent(..count.. / tapply(..count.., list(..PANEL.., ..x..), sum)[..PANEL.., ..x..], accuracy = 1)

but it didn't work.

This seems to be asked a lot, but after reviewing many similar questions including the following:

I still didn't manage to correctly reference the tapply() sums for creating the percentage labels as illustrated in my code above, and I think the overall panel makes it more complicated if I have to pre-calculate the percentages before plotting (I may need to duplicate the whole dataset and mutate cyl into a new variable facet, and then use facet_wrap() on the new variable instead of facet_grid()), as illustrated in my attempt below:

mtcars %>% 
  bind_rows(mutate(mtcars, facet = "(all)")) %>% 
  mutate(
    facet = if_else(is.na(facet), as.character(cyl), facet) %>% 
      factor(levels = c("4", "6", "8", "(all)"))
  ) %>% 
  group_by(facet, gear, am) %>% 
  summarise(freq = n()) %>% 
  summarise(am = am, freq = freq, pct = freq / sum(freq), .groups = "drop_last") %>% 
  ggplot(aes(x = factor(gear) %>% droplevels(), y = pct, fill = factor(am))) +
  facet_grid(cols = vars(facet), scales = "free_x", space = "free_x") +
  geom_col(position = "stack") +
  geom_text(
    aes(label = scales::percent(pct, accuracy = 1L)),
    position = position_stack(vjust = .5)
  )
#> `summarise()` regrouping output by 'facet', 'gear' (override with `.groups` argument)

Created on 2021-03-02 by the reprex package (v0.3.0)

However, it looks more verbose than the first solution, although my duplication of the data for including the "(all)" panel may not be the best way.

Any help fixing my first solution (with a little explanation) and improving the second solution will be greatly appreciated!

elarry
  • 521
  • 2
  • 7
  • 20

1 Answers1

3

I managed to do it, but it's not pretty.

I still think the best way is to pre-process the data before plotting.

mtcars %>% 
   ggplot(aes(x = factor(gear) %>% droplevels(), fill = factor(am))) +
   facet_grid(
     cols = vars(cyl), scales = "free_x", space = "free_x", margins = TRUE
   ) +
   geom_bar(position = "fill") +
   geom_text(
     aes(label = unlist(tapply(..count.., list(..x.., ..PANEL..), 
                               function(a) paste(round(100*a/sum(a), 2), '%'))),

     y = ..count.. ), stat = "count",
     position = position_fill(vjust = .5)
)

The general idea is that you have to do the tapply on the counts based on ..x.. and ..PANEL.. (in that order), which generates vectors of counts for each bar. You then generate the labels per bar from that vector by getting the percentage, rounding or whatever you need. Finally, you have to unlist the tapply results so that ggplot takes it like a given vector of labels.

This outputs the following plot :

enter image description here

RoB
  • 1,833
  • 11
  • 23
  • Thanks so much for your solution - it worked perfectly for me. Your emphasis on the correct order of the grouping variables is very useful, and how didn't I think of including all the calculation in a custom function before! I've edited my question to add my attempt at plotting using pre-calculated summaries, but I feel the part in my code that duplicates the data to include the "(all)" panel may not be the best way. Would you mind helping me to improve it? Many thanks again! – elarry Mar 02 '21 at 11:34
  • This is amazing. I spent so much time trying to figure something similar out, and this code just works directly in my case. Incredible. – Lukasz Jan 10 '22 at 00:26