Descriptive Statistics — Part III: Building ggplot2 Charts

Before you begin, take a few minutes to check what you already know and how confident you feel. You will see the same questions again at the end of the module — this helps both you and us measure what you have learned. Click an option for every question and confidence rating, then click Next to continue.

In Part I you described variables in numbers — a mean here, a median and an IQR there. Numbers alone hide shape. A mean BMI of 28 says nothing about whether the values cluster tightly or trail off in a long tail. In this part you SEE the data: you learn how a ggplot is assembled from named parts, then you draw the standard figures a clinician reaches for first.

This part of the module builds a ladder from a blank plot to a labelled, faceted figure:

• The grammar of graphics — every ggplot is data, an aesthetic mapping, and a geom, stacked with the + operator; and why a constant belongs outside aes(), not inside it.
• Distribution plots — geom_histogram(), geom_density(), and geom_boxplot() to see the shape of one variable and spot skew and outliers.
• Relationships, categories, and small multiples — geom_point() for scatter, geom_col() versus geom_bar(), facet_wrap() for subgroups, and labs() to label every figure.

By the end of this part you will be able to build a scatter plot from scratch — choosing the data, an aes() mapping, and the right geom; draw a histogram, density curve, and boxplot to judge a variable's shape; use geom_col() versus geom_bar() correctly; split a figure into per-subgroup panels with facet_wrap(); and finish any figure with readable labels.

Try every snippet in the R Scratchpad on the right. Each block begins by building the same simulated 200-patient cohort with set.seed() and tibble(), so this part needs no data file — you can run any block on its own and watch the figure appear.

ggplot2 does not have one function per chart type. Instead it has a grammar of graphics: you describe a plot by naming its parts, and ggplot2 draws it. Three parts are enough to start.

• data — the tibble you are plotting, passed to ggplot().
• aes — the aesthetic mapping: which column goes on the x-axis, which on the y, which colours the points. You write it inside aes().
• geom — the geometric shape that draws the data: points, bars, boxes, a smooth density curve.

An aesthetic mapping is the link from a column in your data to something you can see on the plot — a position, a colour, a size. aes(x = age, y = bmi) says put age along the bottom and bmi up the side.

You assemble a plot by adding layers with +. Start with ggplot(), name the data and the mapping, then add a geom. The + always goes at the end of a line, never the start of the next one — R reads a line that looks finished as finished, so a leading + throws an error.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
ggplot(cohort, aes(x = age, y = bmi)) +
  geom_point()

Notice what each piece did. ggplot(cohort, ...) set the data. aes(x = age, y = bmi) mapped two columns to the two axes. geom_point() drew one point per patient. The figure appears because the whole expression is printed at the top level — if you save a plot into an object and never print it, nothing shows.

Putting a constant inside aes() is the classic beginner trap. aes() is for mappings from data. To colour every point the same fixed blue, set colour outside aes(), as a plain argument to the geom.

Right way: a column inside aes() maps to the legend (aes(colour = arm) gives one colour per treatment arm). Wrong way: aes(colour = "blue") does not paint things blue — it invents a one-level variable literally called "blue" and lets ggplot pick a colour for it.

Let’s build your first plot one layer at a time.

library(ggplot2)
d <- data.frame(age = c(45, 52, 61, 58), bmi = c(24, 27, 31, 29))
ggplot(d, aes(x = age, y = bmi)) +
  geom_point()

A scatter shows two variables at once. To see the shape of just one variable — is it symmetric, skewed, full of outliers? — you reach for a distribution plot. These are the first figures you make when you meet a new dataset.

A histogram slices a numeric variable into equal-width bins and draws a bar for how many values fall in each. geom_histogram() needs only an x mapping. The binwidth is the width of each slice — and it is your choice, not the data's.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
ggplot(cohort, aes(x = crp)) +
  geom_histogram(binwidth = 2)

Here is the trap that catches everyone. A histogram's shape depends on the binwidth you pick. Wide bins smooth real structure away; narrow bins turn random noise into spiky fake peaks. Always try two or three binwidths before you trust a shape. Run the block again with binwidth = 0.5, then binwidth = 10, and watch the story change.

A density plot is the smoothed cousin of the histogram: geom_density() draws one continuous curve instead of bars. It is gentler on the eye for comparing the shapes of two groups overlaid, but it still hides the raw counts, so quote it alongside a sample size, never alone.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
ggplot(cohort, aes(x = crp, fill = arm)) +
  geom_density(alpha = 0.4)

A boxplot summarises a distribution with five numbers: the box spans the 25th to 75th percentile (the IQR), the line inside is the median, the whiskers reach most of the range, and dots beyond them are flagged as outliers. It is the standard way to compare a continuous variable across groups.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
ggplot(cohort, aes(x = arm, y = crp)) +
  geom_boxplot()

But a boxplot is a summary, and a boxplot hides the shape inside each box. Two groups with identical boxes can hide a smooth hump in one and two separate clumps in the other — the box cannot tell them apart. When the sample is small or you suspect clustering, overlay the raw points with geom_jitter() so the reader sees what the box smoothed over.

Distribution plots show one variable. Now you connect variables and split a figure into subgroups — and you label everything, because an unlabelled figure is unreadable to anyone but you.

Both draw bars, and confusing them is the most common ggplot bar mistake. geom_bar counts rows for you: give it one categorical x and it tallies how many patients fall in each category. geom_col plots a value you already computed: give it x and y, and each bar's height is your y.

Right way to remember it: use geom_bar() when you want a count of raw rows; use geom_col() when you have already summarised the numbers yourself. Feeding a pre-computed mean to geom_bar() double-counts and gives nonsense heights.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
library(dplyr)
# geom_bar counts rows per arm for you:
ggplot(cohort, aes(x = arm)) +
  geom_bar()

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
library(dplyr)
# geom_col plots a value you computed first:
arm_means <- cohort |>
  group_by(arm) |>
  summarise(mean_hba1c = mean(hba1c))
ggplot(arm_means, aes(x = arm, y = mean_hba1c)) +
  geom_col()

A facet splits one plot into a grid of small panels, one per subgroup, all sharing the same axes. facet_wrap(~ sex) draws the same chart once for females and once for males, so you compare shapes side by side instead of overlaying them.

Always finish a figure with labs. labs() sets the axis titles, the legend title, and an overall plot title. A raw column name like crp on an axis is fine while you explore, but a figure you show anyone needs words a human can read.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(ggplot2)
ggplot(cohort, aes(x = age, y = crp, colour = arm)) +
  geom_point() +
  facet_wrap(~ sex) +
  labs(
    title = "CRP versus age, by sex and treatment arm",
    x = "Age (years)",
    y = "CRP (mg/L)",
    colour = "Arm"
  )

You have reached the end of the module. Try the same questions again — your answers here, paired with your pre-test answers, are how we measure what the module taught you. Answer every question and confidence rating, then click Submit and see results to view your score.

Here is how your post-test answers compare with your pre-test answers. The pre/post pairing is the most reliable way to see what this module actually taught you.