ggplot2: Reorder items in a legend

Question

I have a stacked bar graph and I'd like to reorder the way the legend appears. I'd like the legend to show the order in "Social", "Economic", and "Environmental", but keep the colors the same. I keep trying different solutions, but I'm a newb and am probably writing the wrong things!

I think I'm on the right track with this, but it's not working:

mortality_df$subsystem <- factor(mortality_df$subsystem, levels = c("Social", "Economic", "Environmental"))

The code without the reordering of the legend:

mortality_df <- results_df %>%
  pivot_longer(
    cols = starts_with("imp_"),
    names_to = "mortality",
    names_prefix = "imp_",
    values_to = "importance"
  ) %>%
  relocate(c(feature, subsystem, mortality, importance, label))

mortality_df

diabetes_plot <- mortality_df %>%
  filter(mortality == "diabetes") %>%
  filter(importance > 0) %>%
  ggplot(aes(x = reorder(label, importance), y = importance, fill = subsystem)) +
  geom_bar(stat = "identity") +
  coord_flip() +
  scale_y_continuous(name = "Importance Score") +
  scale_x_discrete(name = "") +
  theme_minimal() +
  theme(aspect.ratio = 1.45)

diabetes_plot

Which creates this:

Diabetes plot

Answer 1

You can change the order of the items in the legend in two principal ways:

1. Refactor the column in your dataset and specify the levels. This should be the way you specified in the question, so long as you place it in the code correctly.

2. Specify ordering via scale_fill_* functions, using the breaks= argument.

Here's how you can do this using a subset of the mpg built-in datset as an example. First, here's the standard plot:

library(ggplot2)
library(dplyr)

p <- mpg %>%
  dplyr::filter(class %in% c('compact', 'pickup', 'minivan', 'suv')) %>%
  ggplot(aes(x=factor(year), fill=class)) +
  geom_bar() + coord_flip()
p 

Change order via refactoring

The key here is to ensure you use factor(...) before your plot code. Results are going to be mixed if you're trying to pipe them together (i.e. %>%) or refactor right inside the plot code.

Note as well that our colors change compared to the original plot. This is due to ggplot assigning the color values to each legend key based on their position in levels(...). In other words, the first level in the factor gets the first color in the scale, second level gets the second color, etc...

d <- mpg %>% dplyr::filter(class %in% c('compact', 'pickup', 'minivan', 'suv'))

d$class <- factor(d$class, levels=c('compact', 'suv', 'pickup', 'minivan'))

p <-
  d %>% ggplot(aes(x=factor(year), fill=class)) +
  geom_bar() +
  coord_flip()

Changing order of keys using scale function

The simplest solution is to probably use one of the scale_*_* functions to set the order of the keys in the legend. This will only change the order of the keys in the final plot vs. the original. The placement of the layers, ordering, and coloring of the geoms on in the panel of the plot area will remain the same. I believe this is what you're looking to do.

You want to access the breaks= argument of the scale_fill_discrete() function - not the limits= argument.

p + scale_fill_discrete(breaks=c('compact', 'suv', 'pickup', 'minivan'))