Awesome

ggplot tricks

The goal of this repository is to keep track of some neat ggplot2 tricks I’ve learned. This assumes you’ve familiarised yourself with the basics of ggplot2 and can construct some nice plots of your own. If not, please peruse the book at your leasure.

I’m not incredibly adapt in gloriously typesetting plots and expertly finetuning themes and colour palettes, so you’d have to forgive me. The mpg dataset is very versatile for plotting, so you’ll be seeing a lot of that as you read on. Extension packages are great, and I’ve dabbled myself, but I’ll try to limit myself to vanilla ggplot2 tricks here.

For now, this will be mostly a README-only bag of tricks, but I may decide later to put them into separate groups in other files.

Start-up
Splicing aesthetics
1. Colour-fill relations
2. Text contrast
Half-geoms
Midpoints in diverging scales
Labels as text
1. Labelling points
2. (Outdated)Facetted tags
Recycling plots

Let’s begin

By loading the library and setting a plotting theme. The first trick here is to use theme_set() to set a theme for all your plots throughout a document. If you find yourself setting a very verbose theme for every plot, here is the place where you set all your common settings. Then never write a novel of theme elements ever again¹!

library(ggplot2)
library(scales)

theme_set(
  # Pick a starting theme
  theme_gray() +
  # Add your favourite elements
  theme(
    axis.line        = element_line(),
    panel.background = element_rect(fill = "white"),
    panel.grid.major = element_line("grey95", linewidth = 0.25),
    legend.key       = element_rect(fill = NA) 
  )
)

Splicing aesthetics

The ?aes documentation doesn’t tell you this, but you can splice the mapping argument in ggplot2. What does that mean? Well it means that you can compose the mapping argument on the go with !!!. This is especially nifty if you need to recycle aesthetics every once in a while.

my_mapping <- aes(x = foo, y = bar)

aes(colour = qux, !!!my_mapping)
#> Aesthetic mapping: 
#> * `x`      -> `foo`
#> * `y`      -> `bar`
#> * `colour` -> `qux`

Relating colour and fill

My personal favourite use of this is to make the fill colour match the colour colour, but slightly lighter². We’ll use the delayed evaluation system for this, after_scale() in this case, which you’ll see more of in the section following this one. I’ll repeat this trick a couple of times throughout this document.

my_fill <- aes(fill = after_scale(alpha(colour, 0.3)))

ggplot(mpg, aes(displ, hwy)) +
  geom_point(aes(colour = factor(cyl), !!!my_fill), shape = 21)

Text contrast

You may find yourself in a situation wherein you’re asked to make a heatmap of a small number of variables. Typically, sequential scales run from light to dark or vice versa, which makes text in a single colour hard to read. We could devise a method to automatically write the text in white on a dark background, and black on a light background. The function below considers a lightness value for a colour, and returns either black or white depending on that lightness.

contrast <- function(colour) {
  out   <- rep("black", length(colour))
  light <- farver::get_channel(colour, "l", space = "hcl")
  out[light < 50] <- "white"
  out
}

Now, we can make an aesthetic to be spliced into a layer’s mapping argument on demand.

autocontrast <- aes(colour = after_scale(contrast(fill)))

Lastly, we can test out our automatic contrast contraption. You may notice that it adapts to the scale, so you wouldn’t need to do a bunch of conditional formatting for this.

cors <- cor(mtcars)

# Melt matrix
df <- data.frame(
  col = colnames(cors)[as.vector(col(cors))],
  row = rownames(cors)[as.vector(row(cors))],
  value = as.vector(cors)
)

# Basic plot
p <- ggplot(df, aes(row, col, fill = value)) +
  geom_raster() +
  geom_text(aes(label = round(value, 2), !!!autocontrast)) +
  coord_equal()

p + scale_fill_viridis_c(direction =  1)
p + scale_fill_viridis_c(direction = -1)

Half-geoms

There are some extensions that offer half-geom versions of things. Of the ones I know, gghalves and the see package offer some half-geoms.

Here is how to abuse the delayed evaluation system to make your own. This can come in handy if you’re not willing to take on an extra dependency for just this feature.

Half-boxplots

The easy case is the boxplot. You can either set xmin or xmax to after_scale(x) to keep the right and left parts of a boxplot respectively. This still works fine with position = "dodge".

# A basic plot to reuse for examples
p <- ggplot(mpg, aes(class, displ, colour = class, !!!my_fill)) +
  guides(colour = "none", fill = "none") +
  labs(y = "Engine Displacement [L]", x = "Type of car")

p + geom_boxplot(aes(xmin = after_scale(x)))

Half-errorbars

The same thing that works for boxplots, also works for errorbars.

p + geom_errorbar(
  stat = "summary",
  fun.data = mean_se,
  aes(xmin = after_scale(x))
)

Half-violin

We can once again do the same thing for violin plots, but the layer complains about not knowing about the xmin aesthetic. It does use that aesthetic, but only after the data has been setup, so it is not intended to be a user accessible aesthetic. We can silence the warning by updating the xmin default to NULL, which means it won’t complain, but also doesn’t use it if absent.

update_geom_defaults("violin", list(xmin = NULL))

p + geom_violin(aes(xmin = after_scale(x)))

Combining

Not left as an exercise for the reader this time, but I just wanted to show how it would work if you were to combine two halves and want them a little bit offset from one another. We’ll abuse the errorbars to serve as staples for the boxplots.

# A small nudge offset
offset <- 0.025

# We can pre-specify the mappings if we plan on recycling some
right_nudge <- aes(
  xmin = after_scale(x), 
  x = stage(class, after_stat = x + offset)
)
left_nudge  <- aes(
  xmax = after_scale(x),
  x = stage(class, after_stat = x - offset)
)

# Combining
p +
  geom_violin(right_nudge) +
  geom_boxplot(left_nudge) +
  geom_errorbar(left_nudge, stat = "boxplot", width = 0.3)

Setting midpoints in divergent scales

Let’s say you have better colour intuition than I have, and three colours aren’t enough for your divergent colour palette needs. A painpoint is that it is tricky to get the midpoint right if your limits aren’t perfectly centered around it. Enter the rescaler argument in league with scales::rescale_mid().

my_palette <- c("dodgerblue", "deepskyblue", "white", "hotpink", "deeppink")

p <- ggplot(mpg, aes(displ, hwy, colour = cty - mean(cty))) +
  geom_point() +
  labs(
    x = "Engine displacement [L]",
    y = "Highway miles per gallon",
    colour = "Centered\nvalue"
  )

p + 
  scale_colour_gradientn(
    colours = my_palette, 
    rescaler = ~ rescale_mid(.x, mid = 0)
  )

An alternative is to simply center the limits on x. We can do that by providing a function to the scale’s limits.

p +
  scale_colour_gradientn(
    colours = my_palette, 
    limits = ~ c(-1, 1) * max(abs(.x))
  )

Labels as text

Labelling points

You can label points with geom_text(), but a potential problem is that the text and points overlap.

set.seed(0)
df <- USArrests[sample(nrow(USArrests), 5), ]
df$state <- rownames(df)

q <- ggplot(df, aes(Murder, Rape, label = state)) +
  geom_point()
q + geom_text()

There are several typical solutions to this problem, and they all come with drawbacks:

The {ggrepel} package is wonderful for solving this problem. But if you try to keep your dependencies to a minimum, you might not want to depend on it.
You can set the nudge_x and nudge_y parameters. The issue here is that these are defined in data units, so spacing is unpredictable, and there is no way to have these depend on the original locations.
You can set the hjust and vjust aesthetics. It allows you to depend on the original locations, but these have no natural offsets.

Here are options 2 and 3 in action:

q + geom_text(nudge_x = 1, nudge_y = 1)

q + geom_text(aes(
  hjust = Murder > mean(Murder),
  vjust = Rape > mean(Rape)
))

You might think: ‘I can just multiply the justifications to get a wider offset’, and you’d be right. However, because the justification depends on the size of text you might get unequal offsets. Notice in the plot below that ‘North Dakota’ is offset too munch in the y-direction and ‘Rhode Island’ in the x-direction.

q + geom_text(aes(
  label = gsub("North Dakota", "North\nDakota", state),
  hjust = ((Murder > mean(Murder)) - 0.5) * 1.5 + 0.5,
  vjust = ((Rape > mean(Rape)) - 0.5) * 3 + 0.5
))

The nice thing of geom_label() is that you can turn off the label box and keep the text. That way, you can continue to use other useful stuff, like the label.padding setting, to give an absolute (data-independent) offset from the text to the label.

q + geom_label(
  aes(
    label = gsub(" ", "\n", state),
    hjust = Murder > mean(Murder),
    vjust = Rape > mean(Rape)
  ),
  label.padding = unit(5, "pt"),
  label.size = NA, fill = NA
)

Facetted tags

This used to be a tip about putting facet tags in panels, which used to be complicated. With ggplot2 3.5.0, you no longer have to fiddle with setting infinite positions and tweaking the hjust or vjust parameters. You can now just use x = I(0.95), y = I(0.95) to place text in the upper-right corner. Open up the details to see the old tip.

Putting text annotations on facetted plots is a pain, because limits can vary on a per-panel basis, so it is very difficult to find the correct position. An extension that explores alleviating this pain is the tagger extension, but we can do a similar thing in vanilla ggplot2.

Luckily, there is a mechanic in ggplot2’s position axes that let’s -Inf and Inf be interpreted as the scale’s minimum and maximum limit respectively³. You can exploit this by choosing x = Inf, y = Inf to put the labels in a corner. You can also use -Inf instead of Inf to place at the bottom instead of top, or left instead of right.

We need to match the hjust/vjust arguments to the side of the plot. For x/y = Inf, they would need to be hjust/vjust = 1, and for x/y = -Inf they need to be hjust/vjust = 0.

p + facet_wrap(~ class, scales = "free") +
  geom_text(
    # We only need 1 row per facet, so we deduplicate the facetting variable
    data = ~ subset(.x, !duplicated(class)),
    aes(x = Inf, y = Inf, label = LETTERS[seq_along(class)]),
    hjust = 1, vjust = 1,
    colour = "black"
  )

Unfortunately, this places the text straight at the border of the panel, which may offend our sense of beauty. We can get slightly fancier by using geom_label(), which lets us more precisely control the spacing between the text and the panel borders by setting the label.padding argument.

Moreover, we can use label.size = NA, fill = NA to hide the textbox part of the geom. For illustration purposes, we now place the tag at the top-left instead of top-right.

p + facet_wrap(~ class, scales = "free") +
  geom_label(
    data = ~ subset(.x, !duplicated(class)),
    aes(x = -Inf, y = Inf, label = LETTERS[seq_along(class)]),
    hjust = 0, vjust = 1, label.size = NA, fill = NA,
    label.padding = unit(5, "pt"),
    colour = "black"
  )

Recycling plots

Let’s say we’re tasked with making a bunch of similar plots, with different datasets and columns. For example, we might want to make a series of barplots⁴ with some specific pre-sets: we’d like the bars to touch the x-axis and not draw vertical gridlines.

Functions

One well-known way to make a bunch of similar plots is to wrap the plot construction into a function. That way, you can use encode all the presets you want in your function.

I case you might not know, there are various methods to program with the aes() function, and using {{ }} (curly-curly) is one of the more flexible ways ⁵.

barplot_fun <- function(data, x) {
  ggplot(data, aes(x = {{ x }})) +
    geom_bar(width = 0.618) +
    scale_y_continuous(expand = c(0, 0, 0.05, 0)) +
    theme(panel.grid.major.x = element_blank())
}

barplot_fun(mpg, class)

One drawback of this approach is that you lock-in any aesthetics in the function arguments. To go around this, an even simpler way is to simply pass ... directly to aes().

barplot_fun <- function(data, ...) {
  ggplot(data, aes(...)) +
    geom_bar(width = 0.618) +
    scale_y_continuous(expand = c(0, 0, 0.1, 0)) +
    theme(panel.grid.major.x = element_blank())
}

barplot_fun(mpg, class, colour = factor(cyl), !!!my_fill)

Skeletons

Another method of doing a very similar thing, is to use plot ‘skeletons’. The idea behind a skeleton is that you can build a plot, with or without any data argument, and add in the specifics later. Then, when you actually want to make a plot, you can use the %+% to fill in or replace the dataset, and + aes(...) to set the relevant aesthetics.

barplot_skelly <- ggplot() +
  geom_bar(width = 0.618) +
  scale_y_continuous(expand = c(0, 0, 0.1, 0)) +
  theme(panel.grid.major.x = element_blank())

my_plot <- barplot_skelly %+% mpg + 
  aes(class, colour = factor(cyl), !!!my_fill) 
my_plot

One neat thing about these skeletons is that even when you’ve already filled in the data and mapping arguments, you can just replace them again and again.

my_plot %+% mtcars + 
  aes(factor(carb), colour = factor(cyl), !!!my_fill)

Ribcage⁶

The idea here is to not skeletonise the entire plot, but just a frequently re-used set of parts. For example, we might want to label our barplot, and pack together all the things that make up a labelled barplot. The trick to this is to not add these components together with +, but simply put them in a list(). You can then + your list together with the main plot call.

labelled_bars <- list(
  geom_bar(my_fill, width = 0.618),
  geom_text(
    stat = "count",
    aes(y     = after_stat(count), 
        label = after_stat(count), 
        fill  = NULL, colour = NULL),
    vjust = -1, show.legend = FALSE
  ),
  scale_y_continuous(expand = c(0, 0, 0.1, 0)),
  theme(panel.grid.major.x = element_blank())
)

ggplot(mpg, aes(class, colour = factor(cyl))) +
  labelled_bars +
  ggtitle("The `mpg` dataset")

ggplot(mtcars, aes(factor(carb), colour = factor(cyl))) +
  labelled_bars +
  ggtitle("The `mtcars` dataset")

<details style="margin-bottom:10px;"> <summary> Session info </summary>

#> ─ Session info ───────────────────────────────────────────────────────────────
#>  setting  value
#>  version  R version 4.3.2 (2023-10-31 ucrt)
#>  os       Windows 11 x64 (build 22631)
#>  system   x86_64, mingw32
#>  ui       RTerm
#>  language (EN)
#>  collate  English_Netherlands.utf8
#>  ctype    English_Netherlands.utf8
#>  tz       Europe/Amsterdam
#>  date     2024-02-27
#>  pandoc   3.1.1
#> 
#> ─ Packages ───────────────────────────────────────────────────────────────────
#>  package     * version    date (UTC) lib source
#>  cli           3.6.2      2023-12-11 []  CRAN (R 4.3.2)
#>  colorspace    2.1-0      2023-01-23 []  CRAN (R 4.3.2)
#>  digest        0.6.34     2024-01-11 []  CRAN (R 4.3.2)
#>  dplyr         1.1.4      2023-11-17 []  CRAN (R 4.3.2)
#>  evaluate      0.23       2023-11-01 []  CRAN (R 4.3.2)
#>  fansi         1.0.6      2023-12-08 []  CRAN (R 4.3.2)
#>  farver        2.1.1      2022-07-06 []  CRAN (R 4.3.2)
#>  fastmap       1.1.1      2023-02-24 []  CRAN (R 4.3.2)
#>  generics      0.1.3      2022-07-05 []  CRAN (R 4.3.2)
#>  ggplot2     * 3.5.0.9000 2024-02-27 []  local
#>  glue          1.7.0      2024-01-09 []  CRAN (R 4.3.2)
#>  gtable        0.3.4      2023-08-21 []  CRAN (R 4.3.2)
#>  highr         0.10       2022-12-22 []  CRAN (R 4.3.2)
#>  htmltools     0.5.7      2023-11-03 []  CRAN (R 4.3.2)
#>  knitr         1.45       2023-10-30 []  CRAN (R 4.3.2)
#>  labeling      0.4.3      2023-08-29 []  CRAN (R 4.3.1)
#>  lifecycle     1.0.4      2023-11-07 []  CRAN (R 4.3.2)
#>  magrittr      2.0.3      2022-03-30 []  CRAN (R 4.3.2)
#>  munsell       0.5.0      2018-06-12 []  CRAN (R 4.3.2)
#>  pillar        1.9.0      2023-03-22 []  CRAN (R 4.3.2)
#>  pkgconfig     2.0.3      2019-09-22 []  CRAN (R 4.3.2)
#>  R6            2.5.1      2021-08-19 []  CRAN (R 4.3.2)
#>  ragg          1.2.7      2023-12-11 []  CRAN (R 4.3.2)
#>  rlang         1.1.3      2024-01-10 []  CRAN (R 4.3.2)
#>  rmarkdown     2.25       2023-09-18 []  CRAN (R 4.3.2)
#>  rstudioapi    0.15.0     2023-07-07 []  CRAN (R 4.3.2)
#>  scales      * 1.3.0      2023-11-28 []  CRAN (R 4.3.2)
#>  sessioninfo   1.2.2      2021-12-06 []  CRAN (R 4.3.2)
#>  systemfonts   1.0.5      2023-10-09 []  CRAN (R 4.3.2)
#>  textshaping   0.3.7      2023-10-09 []  CRAN (R 4.3.2)
#>  tibble        3.2.1      2023-03-20 []  CRAN (R 4.3.2)
#>  tidyselect    1.2.0      2022-10-10 []  CRAN (R 4.3.2)
#>  utf8          1.2.4      2023-10-22 []  CRAN (R 4.3.2)
#>  vctrs         0.6.5      2023-12-01 []  CRAN (R 4.3.2)
#>  viridisLite   0.4.2      2023-05-02 []  CRAN (R 4.3.2)
#>  withr         3.0.0      2024-01-16 []  CRAN (R 4.3.2)
#>  xfun          0.41       2023-11-01 []  CRAN (R 4.3.2)
#>  yaml          2.3.8      2023-12-11 []  CRAN (R 4.3.2)
#> 
#> 
#> ──────────────────────────────────────────────────────────────────────────────

</details>

Well, you need to do it once at the start of your document. But then never again! Except in your next document. Just write a plot_defaults.R script and source() that from your document. Copy-paste that script for every project. Then, truly, never again :heart:. ↩
This is a lie. In reality, I use aes(colour = after_scale(colorspace::darken(fill, 0.3))) instead of lightening the fill. I didn’t want this README to have a dependency on {colorspace} though. ↩
Unless you self-sabotage your plots by setting oob = scales::oob_censor_any in the scale for example. ↩
In your soul of souls, do you really want to make a bunch of barplots though? ↩
The alternative is to use the .data pronoun, which can be .data$var if you want to lock in that column in advance, or .data[[var]] when var is passed as a character. ↩
This bit was originally called ‘partial skeleton’, but as a ribcage is a part of a skeleton, this title sounded more evocative. ↩