Primer 5 · Your first plot

The grammar of graphics, in three lines

If the cells below never become ready

This page runs R inside your browser. It needs about 37 MB on first load (this is the heaviest primer, because ggplot2 is a large package) plus the R engine, the machinery that runs and marks your code, and the cohort file itself. Give it a minute or two. If you are on a slow or shared connection, do this one at home rather than at the venue.

If nothing happens after that, your network is probably blocking it. Work through the R primer in RStudio instead. Everything taught here is taught there too.

There is no Stop button. If you write a loop that never ends, reload the page.

Day 1 opens with visualisation at 11:15. Work through this page first and that will not be the first ggplot() call you have ever written. You will be recognising a plot you have already drawn, rather than meeting one cold.

Every ggplot2 plot is built from three parts, and you have to name all three:

the data: which table,
the aesthetics: which column goes on which part of the picture,
the geom: what shape to draw.

Miss one and you get a blank panel rather than an error, which is confusing the first time. So we will add them one at a time and watch the picture appear. For the full explanation of any of this, see the R primer.

1. Data: which table

ggplot() on its own gives you a grey rectangle. That is not a failure. It is R saying I have your data and no instructions about what to do with it.

That is the same 240-student cohort, read with the same read_csv() line, from the same data/meded_students.csv path you will type on Day 1.

2. Aesthetics: which column goes where

aes() maps a column onto a feature of the picture: the x axis, the y axis, the colour, the fill. Map attendance_pct to x and an axis appears, correctly scaled to the range of the data:

Still no bars, because we have not yet said what to draw. But look at the axis: ggplot2 has already read the column, found it runs from about 57 to 100, and chosen the breaks for you.

3. Geom: what shape to draw

geom_histogram() cuts a numeric column into bins and draws a bar per bin. It is the standard first look at any single number.

Attendance clusters around 80%, with a tail of students below 65%, the ones a course convenor would want to know about. One line of code and you have seen something you could not see in the spreadsheet.

+, not |>: the one exception you will meet

Everywhere else in the workshop you join steps with the native pipe |>. In ggplot2 you join layers with +.

This is history, not logic: ggplot2 was written in 2007, years before R had a pipe, and it already had its own way of stacking one layer onto another. It was never changed, because millions of lines of published code use +.

The rule that keeps you out of trouble:

|> moves a table from one verb to the next: filter(), mutate(), summarise().
+ adds a layer or setting to a plot you have already started with ggplot().

You can use both in the same expression, and often will: pipe the data into ggplot(), then add layers with +.

students |>
  filter(batch == 2024) |>
  ggplot(aes(x = attendance_pct)) +
  geom_histogram(binwidth = 5)

Getting these the wrong way round produces one of two errors. ggplot(...) |> geom_histogram() complains about a missing argument; students + ggplot(...) says non-numeric argument to binary operator. Everybody does it once. Now you know what the message means.

Bin width changes what you see

binwidth is not a detail. It decides how much of the shape of the data survives. Change the 2 below to 5, then to 10, and rerun each time:

At 2 the histogram is spiky and you can see individual clumps of students; at 10 it is smooth and you may have flattened away something real. There is no correct answer, which is exactly why you should look at more than one before you decide what the data says.

Comparing two groups

The workshop cohort has two teaching arms, so the obvious question is whether they differ. A boxplot puts one box per group side by side: the thick line is the median, the box holds the middle half of the students, and the dots beyond the whiskers are outliers.

fill = teaching_arm maps the group onto the colour of the boxes; scale_fill_manual() then says which colour each group gets. These are the conclave’s own two colours.

Two boxes, and they sit at almost the same height: median burnout is 49.7 in the flipped arm against 49.1 in the traditional one. That is a real and useful finding: no visible difference is a result, and seeing it in ten seconds is better than testing for it in ten minutes.

x = NULL inside labs() removes an axis title that would only repeat the group names underneath it. fill = NULL removes the legend title for the same reason. Deleting labels that say nothing is most of what makes a figure look professional.

The warning under that plot is not an error

ggplot2 printed something like Removed 7 rows containing non-finite outside the scale range. Seven students in this cohort have no burnout_score: the value is NA, R’s marker for missing, and you cannot draw a box around a value that does not exist, so ggplot2 dropped them and told you so.

A warning means I did it, here is something you should know. An error means I could not do it, there is no result. Beginners often stop at the first warning; read it instead, and check the number it gives you is the number of missing values you expected.

Exercise 1: a histogram that explains itself

Time: 10 minutes

Draw a histogram of pretest, the score every student sat before the course began. Then give it a title and readable axis labels with labs(), so that somebody who has never seen the dataset can understand the figure on its own.

ggplot(students, aes(x = pretest)) +
  geom_histogram(binwidth = 5) +
  labs(
    title = "Pre-test scores across the 2022–2024 cohort",
    x = "Pre-test score (%)",
    y = "Number of students"
  )

Exercise 2: a scatter plot in three mappings

Time: 10 minutes

Do students who study more gain more? Put study_hours on the x axis and gain on the y axis, colour the points by rural_background, and draw them with geom_point().

This one maps three columns at once, which is the moment aes() starts to earn its keep. Expect a warning about eleven removed rows: gain is missing for the eleven students who never sat the post-test.

ggplot(students, aes(x = study_hours, y = gain, colour = rural_background)) +
  geom_point()

Check yourself

Four questions on the Moodle course

Take Check yourself · Primer 5 on the Moodle course. Four questions, unlimited attempts, no marks. Every answer comes back with an explanation, so a wrong one costs you nothing and teaches you something.

→ Datasets: download the files, ready for Day 1.

You have now written R, built a table and drawn a plot without installing anything. Day 1 does all of this again, on your own machine, with the project structure that makes it reproducible.