Setting Up

Since ggplot2 is part of the tidyverse, we can load it in like so:

library(tidyverse)

ggplot2 Syntax

Here’s an example ggplot2 command, and associated output:

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length)) +
  geom_point()

• iris is the dataset. This is an example dataset built into R, with data from the Fisher (1936) paper, The use of multiple measurements in taxonomic problems. The data consists of measurements on the width and length of different iris flowers’ sepals and petals, as well as their species. Each row refers to a different flower.

• %>% feeds our data into the ggplot() function.

• ggplot() creates the ggplot object, and tells the plot which variables are plotted on which axes using the aes() function. On the x axis, we’ve put flowers’ sepal length, and on the y axis, their petal length.

• geom_point() is a geom object which says what to actually do with all this information. Here, we tell ggplot2 to plot a point for each row (flower) in the dataset.

Geom Objects

Ggplot2’s “geom” objects say what features to actually add to the plot. As well as the geom_point() object above, there are many other possible geom objects. A full list of possible geoms is available on the ggplot2 function reference page, though most useful geoms can be seen on the ggplot2 cheatsheet.

Another nice geom we could use here is geom_smooth(). By default, this geom draws a smooth line estimated using the LOESS method (basically a curvy line of best fit). Other methods can be specified with the methods argument (see the ggplot documentation), but for now we’ll keep this default. Ggplot2 will also draw a grey shaded area showing the 95% confidence intervals (essentially, the certainty with which we can be sure the fit is correct across the x axis).

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length)) +
  geom_smooth()

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Here’s where it gets really cool! Geoms are additive, which means we can put multiple geom objects in one plot. Let’s superimpose our data points on top of the line we just fit.

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length)) +
  geom_smooth() +
  geom_point()

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

And finally, to visualise this relationship with a straight line (fit with a linear model), we can just add method = "lm" to the geom_smooth() function.

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length)) +
  geom_smooth(method = "lm") +
  geom_point()

Aesthetics

Aesthetics are specified with the aes() function, and tell ggplot2 what variables to plot, and which visual features should represent them. Many possible aesthetics are possible, but some of the main ones are:

• x

• y

• colour (or color as both spellings are accepted by ggplot2)

• fill

• shape

As an example, let’s colour the points from our original plot by which species the flowers belong to, with the colour argument.

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length, colour = Species)) +
  geom_point()

If we add other geom objects, you’ll notice that these features are also now coloured by Species.

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length, colour = Species)) +
  geom_point() +
  geom_smooth()

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Finally, we can again add method = "lm" to the geom_smooth() object to visualise these relationships with straight lines.

iris %>%
  ggplot(aes(x = Sepal.Length, y = Petal.Length, colour = Species)) +
  geom_point() +
  geom_smooth(method = "lm")

Visualising Distributions

Now that we’ve got the basics down, it would be good to introduce some key understandings about distributions. Whenever you have a numerical variable, you will have a range of possible values, often centred around an average.

iris %>%
  ggplot(aes(x = Sepal.Width)) +
  geom_histogram(binwidth = 0.1)

iris %>%
  ggplot(aes(x = Sepal.Length)) +
  geom_density()

iris %>%
  ggplot(aes(x = Sepal.Length, fill = Species)) +
  geom_density()

iris %>%
  ggplot(aes(x = Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.5)  # 50% transparency

Ban the Bar Graph

When visualising differences between groups in numerical values, people have historically used bar graphs. There is currently a shift taking place, with people moving away from bar graphs, and towards more informative visualisations.

Now that we understand distributions, can you see what would be the problem with a bar graph for comparing different iris species’ sepal length?

The bar graph shows us the average for each species, but doesn’t tell us much about the distribution of sepal length. It includes error bars for standard deviation, but even this can be misleading, as it looks as though there is almost no overlapsetosa and virginica, and the SD is symmetrical, whereas in fact, our data may be skewed, making values towards above the mean more likely than values below it (or vice versa).

Alternatives

So if we’re not going to use a bar graph, what should we use instead? Do not despair! Alternatives are out there, which can tell people much more about your data than a bar graph ever could.

One nice alternative is the violinbox plot.

iris %>%
  ggplot(aes(x = Species, y = Sepal.Length, fill = Species)) +
  geom_violin(alpha = 0.5) +
  geom_boxplot(width = 0.2)

1. Import Data

Once you’re in the correct working directory, the first step is to import the data from the file, haggis.csv (Hint: read_csv()).

haggis <- read_csv("haggis.csv")

2. Visualise Height by Breakfast Group

Create a violinbox plot, showing how height differs between haggis and porridge eaters.

haggis %>%
  ggplot(aes(x = group, y = height, fill = group)) +
  geom_violin(alpha = 0.5) +
  geom_boxplot(width = 0.1)

3. Check the Distributions

Notice anything strange? We seem to have a much broader distribution of heights for porridge eaters, while the heights of haggis eaters seem to be much more consistent. What could be causing this? Well, as it happens, we also collected data on the music preferences of our subjects, because we’d read some previous literature suggesting relationships between these factors and height. Maybe these could help us answer our question! (Note that all our subjects we exclusively fans of either pop or classical music.)

Create two density plots to see:

1. How the distribution of height differs between pop fans and classical fans.

haggis %>%
  ggplot(aes(x = height, fill = music_taste)) +
  geom_density(alpha = 0.5)

2. How the distribution of age differs between pop fans and classical fans.

haggis %>%
  ggplot(aes(x = age, fill = music_taste)) +
  geom_density(alpha = 0.5)

4. Does Age predict Height?

So, it looks like height differs slightly between our pop lovers and classical aficionados, but even more clearly, pop fans seem to be much younger than classical fans. We might wonder, therefore, whether age predicts height.

Draw a scatter plot to see how age predicts height. Add a line showing the linear relationship between age and height.

haggis %>%
  ggplot(aes(x = age, y = height)) +
  geom_point() +
  geom_smooth(method = "lm")

5. Does Age interact with Music Taste?

Okay, there’s some degree of relationship. Now, if age explains the weird distribution of heights we saw for porridge eaters, we might see a difference in the age-height relationship between haggis and porridge eaters.

Recreate the previous graph, but colour the points and line by participants’ breakfast habits.

haggis %>%
  ggplot(aes(x = age, y = height, colour = group)) +
  geom_point() +
  geom_smooth(method = "lm")

6. Does Music Taste interact with Breakfast

Finally, it might be that the relationship between breakfast habits and height is mediated by music taste in some way.

Create a violinbox plot as you did in Practice Question 2, but split by music taste as well as breakfast group.

This is meant to be a tricky question, but here are some hints:

• You can specify different variables as the x and fill aesthetics

• By default, the boxplots won’t overlay the violin plots as you might expect, but you can fix this with some googling (as I did). To work this out I did a quick google and found these links which you might find useful:
• A Google Search (stackoverflow results on google are always the best thing in the world)

• ggplot2 reference (have a look through the examples they give)

haggis %>%
  ggplot(aes(x = group, y = height, fill = music_taste)) +
  geom_violin(alpha = 0.5) +
  geom_boxplot(width = 0.2, position = position_dodge(width = 0.9))

Labels

The main types of labels you can add to a ggplot2 plot are for the:

• title
• subtitle
• x axis
• y axis
• caption

Labels can be added to a plot with the labs() function like so:

haggis %>%
  ggplot(aes(x = group, y = height, fill = music_taste)) +
  geom_violin(alpha = 0.5) +
  geom_boxplot(width = 0.2, position = position_dodge(width = 0.9)) +
  labs(
    title = "Eat Yer Porridge",
    subtitle = "The Effect of Breakfast Habits on Height",
    x = "Breakfast Group",
    y = "Height (cm)",
    caption = "Based on 2019 Data from the Silver Spoon Porridge Company Survey"
  )

Themes

ggplot2 allows the use of various themes, which can be used to quickly change the appearance of your plots. Themes can be added to a plot by simply adding a theme_... function to the code. For example, to apply the dark theme to your plot, you could use theme_dark():

haggis %>%
  ggplot(aes(x = group, y = height, fill = music_taste)) +
  geom_violin(alpha = 0.5) +
  geom_boxplot(width = 0.2, position = position_dodge(width = 0.9)) +
  theme_dark()

A full list of ggplot2 themes is available here.

Saving to file

When we’ve finished crafting our beautful ggplot2 figure, we can save the plot to file with the ggsave() function. This lets us save our figure as a .png image file, with the desired height in mm, cm, or inches.

The following code saves the last plot rendered to a file called my_pretty_plot.png.

ggsave("my_pretty_plot.png", width = 9, height = 7, units = "in")

Logos

Here’s a really cool fun fact! The logo for Hack Your Data Beautiful (as proudly displayed on the home page) was in fact made in ggplot2, with the geom_density() function! Here’s some reproducible code that we used to generate it:

# how many data points for each of the density plots
n <- 1e6
# where on the x axis each density plot will be
means <- c(40, 60, 80, 100, 120, 140)
# how wide each density plot will be
sds <- c(10, 7.5, 5, 5, 7.5, 10)
# colours from the psyTeachR colour scheme
psyteachr_colours_vec <- c("#983E82", "#E2A458", "#F5DC70", "#59935B", "#467AAC", "#61589C")

tibble(
  grp = rep(LETTERS[1:6], each = n),
  val = rep(means, each = n) +
        rnorm(n*6, 0, rep(sds, each = n))
) %>%
  ggplot(aes(val, fill = grp)) +
  geom_density(alpha = 0.5, show.legend = F, trim = FALSE) +
  scale_fill_manual(values = psyteachr_colours_vec) +
  xlim(-20,200) +
  theme_void()

Is there anything ggplot2 can’t do!?