04 · ggplot2 Basics

Level 1's base plotting (plot(), barplot(), hist()) is quick and dependency-free, but each function has its own quirky set of arguments, and combining multiple layers (points plus a trend line plus grouping by color) gets awkward fast. ggplot2 implements a different idea — the "grammar of graphics" — where every plot is built by adding layers to a base declaration of data and mappings. Once the grammar clicks, it scales to far more complex plots with barely more code than a simple one.

library(ggplot2)

sales <- data.frame(
    quarter = rep(c("Q1", "Q2", "Q3", "Q4"), each = 2),
    region = rep(c("East", "West"), times = 4),
    revenue = c(120, 90, 150, 130, 170, 140, 200, 160)
)

The grammar: ggplot() + aes() + geom_*()

Every ggplot2 plot follows the same skeleton:

ggplot(data = sales, mapping = aes(x = quarter, y = revenue)) +
    geom_col()

• ggplot(data, aes(...)) declares which data and which columns map to which visual properties (x position, y position, color, size, ...) — this alone draws nothing.
• geom_col() (or geom_point(), geom_line(), ...) is a layer that says how to draw it — bars, points, lines. + adds layers.

This separation is the entire idea: swap geom_col() for geom_line() and you get a completely different chart from the exact same data declaration.

Scatter plots — geom_point()

students <- data.frame(
    hours_studied = c(1, 2, 3, 4, 5, 6, 7, 8),
    score = c(52, 58, 63, 70, 75, 82, 88, 91)
)

ggplot(students, aes(x = hours_studied, y = score)) +
    geom_point(color = "steelblue", size = 3) +
    labs(title = "Study Hours vs. Score", x = "Hours Studied", y = "Exam Score")

labs() sets the title and axis labels — the ggplot2 equivalent of base R's main/xlab/ylab arguments, but added as its own layer rather than passed to the plotting call itself.

Bar charts — geom_col() vs. geom_bar()

This is a very common point of confusion:

# geom_col(): the data ALREADY has one row per bar with the height you want
ggplot(sales, aes(x = quarter, y = revenue)) +
    geom_col(fill = "seagreen")

# geom_bar(): counts rows per category automatically -- no y needed
ggplot(sales, aes(x = quarter)) +
    geom_bar()   # height = number of rows per quarter, not revenue

geom_col() plots a value you already computed; geom_bar() computes a count for you. Using geom_bar() when you meant geom_col() produces a chart of row counts instead of your actual values — a genuinely common mistake, since both draw bars and the error is silent (no error message, just a wrong-looking chart).

Mapping a column to color — grouping for free

ggplot(sales, aes(x = quarter, y = revenue, fill = region)) +
    geom_col(position = "dodge") +
    labs(title = "Revenue by Quarter and Region", x = "Quarter", y = "Revenue")

Mapping region to fill inside aes() automatically colors and legends each region — no manual loop over groups, no manual legend construction. position = "dodge" places same-quarter bars side by side instead of stacked (the default for geom_col() with a fill grouping).

Line plots over time

temps <- data.frame(
    day = 1:7,
    temp = c(15, 17, 16, 19, 22, 21, 18)
)

ggplot(temps, aes(x = day, y = temp)) +
    geom_line(color = "darkred", linewidth = 1) +
    geom_point(color = "darkred") +
    labs(title = "Temperature Over a Week", x = "Day", y = "°C")

Layers stack: this plot has both a line and points on top of it, from two separate geom_*() calls sharing the same aes() mapping declared in ggplot().

Histograms and boxplots

set.seed(42)
exam_scores <- data.frame(score = rnorm(200, mean = 75, sd = 10))

ggplot(exam_scores, aes(x = score)) +
    geom_histogram(bins = 15, fill = "cornflowerblue", color = "white")

groups <- data.frame(
    group = rep(c("A", "B"), each = 5),
    score = c(88, 92, 79, 65, 85, 70, 95, 60, 55, 80)
)

ggplot(groups, aes(x = group, y = score, fill = group)) +
    geom_boxplot()

Facets — one small plot per category, automatically

ggplot(sales, aes(x = quarter, y = revenue)) +
    geom_col(fill = "steelblue") +
    facet_wrap(~ region)

facet_wrap(~ column) splits one plot into a grid of small multiples, one panel per unique value of column, all with matching axes by default — the ggplot2 way to compare groups side by side without a manual loop or a cluttered single plot.

Themes — controlling overall appearance

ggplot(sales, aes(x = quarter, y = revenue, fill = region)) +
    geom_col(position = "dodge") +
    theme_minimal() +     # a clean built-in theme -- try theme_bw(), theme_classic() too
    labs(title = "Revenue by Quarter and Region")

Saving a plot — ggsave()

p <- ggplot(sales, aes(x = quarter, y = revenue)) +
    geom_col(fill = "seagreen") +
    labs(title = "Quarterly Revenue")

ggsave("revenue.png", plot = p, width = 8, height = 6, dpi = 150)

Unlike base R's png()/dev.off() pair, ggsave() takes a plot object directly and infers the file format from the extension — no explicit device open/close needed. Saving p as a variable first (instead of printing the plot immediately) is also useful for reusing or modifying a plot before deciding to save it.

ggplot2 vs. base R cheat sheet

Task	Base R	ggplot2
Declare data + mapping	(implicit in each call)	ggplot(data, aes(x=, y=))
Scatter	plot(x, y)	+ geom_point()
Line	plot(x, y, type="l")	+ geom_line()
Bar (pre-computed values)	barplot(x)	+ geom_col()
Bar (row counts)	table(x) then barplot()	+ geom_bar()
Histogram	hist(x)	+ geom_histogram()
Boxplot	boxplot(x, y)	+ geom_boxplot()
Group by color	manual loop/legend	aes(fill = group)
Small multiples	manual par(mfrow=) + loop	+ facet_wrap(~ group)
Save to file	png() / dev.off()	ggsave("file.png", plot)

How It Actually Works

ggplot2 doesn't draw anything the moment you write ggplot(df, aes(...)) — it builds up a ggplot object that's just a list: a data frame, a mapping, and a growing list of layers, scales, facets, and a theme. Actual rendering happens only when the object is printed (auto-triggered at the console, or explicitly), at which point ggplot2 runs its layered grammar pipeline: for each layer it (1) applies any statistical transformation (stat_*, e.g. binning for histograms), (2) maps the transformed data through each aesthetic's scale (a function from data values to visual properties like x-position or color, built by inspecting the full range of the data first), and (3) hands the result to a geom_*'s draw_panel() method, which emits actual grid graphical objects (grobs).

This is why scales are computed globally across all layers before any drawing happens — ggplot2 has to see every layer's data to know the overall x/y range and color domain before it can decide where "halfway" on an axis or "middle of the color gradient" actually falls, which is also why adding a layer with wildly different data can suddenly rescale a plot you thought was already finished.

Exercise

Using the sales data frame from the top of this page: build a single ggplot2 chart that shows revenue per quarter, grouped/colored by region, faceted is not required here — instead use position = "dodge" so both regions' bars sit side by side within each quarter. Add a title and axis labels with labs(), apply theme_minimal(), and save the result to quarterly_revenue.png with ggsave(). Then make a second version that uses facet_wrap(~ region) instead of dodged bars, and write one sentence comparing which version communicates the East/West comparison better.