How to set a fixed fill in geom_col equivalent to geom_bar in ggplot

Question

Consider the following code that should be making two identical bar charts, except for axis labels. The data.frame dat is representative of the kind of data I have, just colour responses. It would seem a good idea to use those colours in plotting.

colOrder = c('green', 'blue', 'red', 'orange', 'purple')
dat <- data.frame(q = rep(c('red', 'orange', 'green', 'blue', 'purple'), c(1:5)))

# using geom_bar
ggplot(dat, aes(x = factor(q, levels = colOrder) )) +
    geom_bar(fill = colOrder)

#using geom_col
dc <- as.data.frame(table(dat$q))
names(dc) <- c('q', 'n')
ggplot(dc, aes(x = factor(q, levels = colOrder), y = n )) +
    geom_col(fill = colOrder)

Why are these plots not identical? The geom_col seems to be filling using the colour order in dc$q while geom_bar correctly takes colOrder as requested. Am I missing something in geom_col documentation? bug?

Answer 1

The issue is that you pass the fill colors as an argument to geom_bar/col. Doing so the colors are assigned to the categories according to the order as they appear in the datasets used by geom_bar/col. While geom_col uses the data as is, geom_bar first computes the counts via stat="count" and reorders the data according to the variable mapped on x. As a result, when adding the vector of fill colors to these datasets, you get different colors for some categories.

You can see this clearly by using layer_data() which allows to retrieve the data used by a geom layer under the hood:

colOrder = c('green', 'blue', 'red', 'orange', 'purple')
dat <- data.frame(q = rep(c('red', 'orange', 'green', 'blue', 'purple'), 1:5))

library(ggplot2)

p1 <- ggplot(dat, aes(x = factor(q, levels = colOrder) )) +
  geom_bar(fill = colOrder)

dc <- as.data.frame(table(dat$q))
names(dc) <- c('q', 'n')
p2 <- ggplot(dc, aes(x = factor(q, levels = colOrder), y = n )) +
  geom_col(fill = colOrder)

layer_data(p1, 1)[c("x", "y", "group", "fill")]
#>   x y group   fill
#> 1 1 3     1  green
#> 2 2 4     2   blue
#> 3 3 1     3    red
#> 4 4 2     4 orange
#> 5 5 5     5 purple

layer_data(p2, 1)[c("x", "y", "group", "fill")]
#>   x y group   fill
#> 1 2 4     2  green
#> 2 1 3     1   blue
#> 3 4 2     4    red
#> 4 5 5     5 orange
#> 5 3 1     3 purple

As is evident for geom_bar aka the first plot the data is ordered according to colOrder after the statistical transformation, whereas for geom_col aka the second plot the data is not reordered.

If you want the same or consistent colors then map on the fill aesthetic and set your colors via scale_fill_manual (or if you have a column with color names as in your example by using scale_fill_identity()):

p3 <- ggplot(dat, aes(x = factor(q, levels = colOrder) )) +
  geom_bar(aes(fill = q)) +
  scale_fill_manual(values = colOrder)

p4 <- ggplot(dc, aes(x = factor(q, levels = colOrder), y = n )) +
  geom_col(aes(fill = q)) +
  scale_fill_manual(values = colOrder)

library(patchwork)

(p1 + p2) / (p3 + p4)