02 · Advanced ggplot2

Level 2 covered the ggplot2 grammar — aes(), geoms, and basic themes. This module covers the tools you reach for once a single scatter plot isn't enough: faceting to split one chart into a grid of small multiples, custom themes you can reuse across a project, and the factor-ordering gotcha that quietly reorders bars and legends on almost every real dataset at some point.

library(ggplot2)
library(dplyr)

Faceting

sales <- tibble(
  month = rep(1:6, times = 2),
  region = rep(c("East", "West"), each = 6),
  revenue = c(100, 120, 115, 140, 160, 155, 90, 95, 105, 110, 130, 128)
)

ggplot(sales, aes(x = month, y = revenue, color = region)) +
  geom_line() + geom_point() +
  facet_wrap(~ region) +
  labs(title = "Revenue by region", x = "Month", y = "Revenue ($k)") +
  theme_minimal()

facet_wrap() splits one plot into a panel per level of a categorical variable, each sharing the same x/y scales by default — useful for comparing shape and trend across groups without cluttering one panel with overlapping lines. facet_grid(rows ~ cols) does the same thing along two variables at once, arranging panels in an actual grid rather than a wrapped sequence. Pass scales = "free_y" to facet_wrap() when the groups have very different magnitudes and forcing a shared y-axis would flatten the smaller ones into near-invisibility.

Reusable custom themes

Repeating the same theme() tweaks on every plot in a project is a sign you want a named theme object instead:

my_theme <- theme_minimal(base_size = 12) +
  theme(
    plot.title = element_text(face = "bold"),
    legend.position = "bottom"
  )

ggplot(sales, aes(month, revenue, color = region)) +
  geom_line() +
  my_theme

Building on theme_minimal() (or any built-in theme) rather than theme() alone means you inherit sensible defaults for everything you didn't explicitly override — start from a base theme and only override what your project's style guide actually specifies.

Custom scales

ggplot(sales, aes(month, revenue, color = region)) +
  geom_line() +
  scale_y_continuous(labels = scales::dollar_format(scale = 1, suffix = "k"))

scale_* layers control how a variable maps to a visual property — axis tick labels and breaks (scale_y_continuous), color palettes (scale_color_manual(), scale_fill_brewer()), or which values get plotted at all. The scales package's dollar_format() and friends turn raw axis numbers into formatted labels without touching the underlying data — the values plotted are still plain numbers, only the display of the axis text changes.

The factor ordering trap

ggplot2 draws categorical axes and legends in factor level order, not the order values happen to appear in your data or any "natural" reading order. If you never set factor levels explicitly, R falls back to alphabetical order — which is rarely the order you want for anything with an inherent ranking:

grades <- c("Low", "High", "Medium")
levels(factor(grades))
# [1] "High"   "Low"    "Medium"     -- alphabetical, not a ranking

f <- factor(grades, levels = c("Low", "Medium", "High"))
levels(f)
# [1] "Low"    "Medium" "High"       -- the order you actually meant
as.integer(f)
# [1] 1 3 2

Trap: a bar chart or legend built from an un-leveled character column of "Low"/"Medium"/"High" will silently render High, Low, Medium in that alphabetical order — no warning, no error, just a chart that reads oddly to anyone who expects a ranking. Always set levels = c(...) explicitly on any categorical column with a meaningful order before plotting it, whether via factor() beforehand or scale_x_discrete(limits = c(...)) at plot time.

Bar charts with position control

ggplot(sales, aes(month, revenue)) +
  geom_col(aes(fill = region), position = "dodge")

position = "dodge" places same-x bars for different groups side by side; the default "stack" piles them on top of each other instead — pick based on whether you want the reader comparing group totals or group magnitudes at each x value.

Cheat sheet

Task	Function
Split into a panel per category	facet_wrap(~ var)
Split into a 2D grid of panels	facet_grid(rows ~ cols)
Let each facet panel scale independently	facet_wrap(..., scales = "free_y")
Build a reusable theme	my_theme <- theme_minimal() + theme(...)
Format axis labels without changing data	scale_y_continuous(labels = scales::dollar_format())
Fix categorical plotting order	factor(x, levels = c(...))
Side-by-side bars per group	geom_col(position = "dodge")
Stacked bars per group	geom_col(position = "stack") (default)

How It Actually Works

Custom ggplot2 themes work by modifying a nested list of element_*() specification objects (theme() returns a theme object that's really a named list of drawing instructions — line widths, colors, margins) that gets merged with the currently active default theme via inheritance rules: each element can inherit unset properties from a parent element (e.g. axis.text inherits from text unless overridden), resolved at render time by walking this inheritance graph before any pixel is drawn. This is why setting text = element_text(family = "serif") cascades to axis labels, titles, and legends simultaneously — they're all descendants in that same inheritance tree.

Faceting (facet_wrap()/facet_grid()) works by first computing, from your faceting variable(s), the full set of unique panel combinations and their layout grid position, then re-running the entire per-layer stat/scale/geom rendering pipeline (from Module 4) independently within each panel, but sharing scale ranges across panels by default so panels stay visually comparable — that shared-scale computation is why faceting a plot with scales = "free" renders noticeably differently: each panel gets its own independently-computed axis range instead of one global one.

Exercise

1. Take the sales tibble above and facet it by region with scales = "free_y" — compare the result to the shared-scale default and note when each is the better choice.
2. Build a my_theme object with a title, subtitle, and caption styled consistently, and reuse it across two different plots of sales.
3. Create a factor column with levels "Small", "Medium", "Large" from data where those strings appear in a different order, plot it with geom_bar() unordered vs. explicitly ordered, and compare the two charts.