Programming for Applications

Chapter 15: ggplot2

Yu-You Liou (NTU)

Shih Chien University

2026-08-03

The Grammar of Graphics

A Short Introduction

Hadley Wickham’s ggplot2 became one of R’s most popular packages — not just for pretty output, but because it speaks a language for describing plots: the grammar of graphics. You describe what you want to show, not how to draw it. The toy data (lattice’s, with renamed variables):

d <- data.frame(a=c(0:9), b=c(1:10), c=c(rep(c("Odd", "Even"), times=5)))
qplot(x=a, y=b, data=d)

We specified an x, a y, and a data set — never the words “scatter plot.” (qplot = quick plot; deprecated in recent ggplot2 but still working — the full ggplot() style follows shortly.)

Plot Objects

Like lattice, ggplot2 functions return an object; printing draws it. Inside functions and scripts: print it yourself.

first.ggplot2.example <- qplot(x=a, y=b, data=d)
summary(first.ggplot2.example)
data: a, b, c [10x3]
mapping:  x = ~a, y = ~b
faceting:  <empty> 
-----------------------------------
geom_point: na.rm = FALSE
stat_identity: na.rm = FALSE
position_identity

The summary lays the plot’s anatomy bare: the data, the mapping (x = a, y = b), the faceting, and the layer (geom_point + stat_identity + identity position).

Facets

Panels (lattice’s conditioning) are facets, specified by formula:

qplot(x=a, y=b, data=d, facets=~c)     # panels left-right

qplot(x=a, y=b, data=d, facets=c~.)    # panels top-bottom — easy to flip!

(Hadley’s preferred spelling of the first: qplot(x=a, y=b, data=d) + facet_wrap(~c).)

Colors Instead of Facets

Group-by-color is just another mapping — same data, one panel:

qplot(x=a, y=b, data=d, color=c)       # color instead of panels

One Variable: Histogram and Density

With only x supplied, qplot defaults to a histogram; ask for another geom by name:

set.seed(123456789)
e <- data.frame(f=rnorm(1000))
qplot(x=f, data=e)

qplot(x=f, data=e, geom="density")

The Grammar’s Components

Drawing a chart means doing many things at once; the grammar names each one, and ggplot2’s functions mirror the names:

  • Data — what is visualized;
  • Mappings — variables → chart components;
  • Geometric objects (geom) — points, bars, lines… (geom_point, geom_bar, …);
  • Aesthetic properties (aes) — how it looks: sizes, labels, ticks;
  • Scales — how variables map to aesthetics;
  • Coordinates — Cartesian (coord_cartesian), polar (coord_polar), map projections (coord_map);
  • Statistical transformations (stat) — summaries: stat_bin for histograms, stat_boxplot…;
  • Facets — partitioning into subplots;
  • Positional adjustments — fine placement (stack, dodge, jitter…).

summary of any plot object reports every component — your map when tuning a chart.

A More Complex Example: Medicare Data

The Data

Real, complicated data: US Medicare cost and outcome measures, packaged three ways — outcome.of.care.measures.national (national mortality/readmission rates for heart attack, heart failure, pneumonia), medicare.payments (average payment per hospital × 70 conditions; 140,722 records), medicare.payments.by.state:

temp <- tempfile(fileext = ".rda")
download.file("https://raw.githubusercontent.com/cran/nutshell/master/data/outcome.of.care.measures.national.rda",
              temp, mode = "wb")
load(temp)
outcome.of.care.measures.national
      Condition     Measure Rate
1  Heart Attack   Mortality 15.9
2 Heart Failure   Mortality 11.3
3     Pneumonia   Mortality 11.9
4  Heart Attack Readmission 19.8
5 Heart Failure Readmission 24.8
6     Pneumonia Readmission 18.4

Bar Charts: Facets…

One categorical x, so a univariate plot with bar heights from weight=Rate; facet by Measure, fill by Measure:

qplot(x=Condition, data=outcome.of.care.measures.national,
      geom="bar", weight=Rate, facets=Measure~., fill=Measure)

…or Dodging

Drop the facets, set position="dodge" to place bars side by side. The book wrote qplot(..., position="dodge"), but modern ggplot2 has removed qplot’s position argument — position adjustments now belong to the layer, so we write it the full-grammar way:

ggplot(outcome.of.care.measures.national,
       aes(x=Condition, weight=Rate, fill=Measure)) +
  geom_bar(position="dodge")

Both work — differently: faceting invites comparing conditions within each measure; dodging invites comparing measures within each condition. The display is an argument.

Payments vs. Volume

Question: do hospitals treating more cases charge Medicare more or less? Restrict to the three heart-failure diagnosis groups and plot payment vs. case count, colored by diagnosis:

temp <- tempfile(fileext = ".rda")
download.file("https://raw.githubusercontent.com/cran/nutshell/master/data/medicare.payments.rda",
              temp, mode = "wb")
load(temp)
heart.failure <- c("Heart failure and shock w/o CC/MCC",
                   "Heart failure and shock w MCC",
                   "Heart failure and shock w CC")
payment.plot <- qplot(x=Number.Of.Cases, y=Medicare.Average.Payment,
  data=subset(medicare.payments, Diagnosis.Related.Group %in% heart.failure),
  color=Diagnosis.Related.Group)
payment.plot

Unreadable — everything clumps left. (Small counts are censored by HIPAA.)

Fixing the Plot by Adding Layers

Recipe: log x-scale, semi-opaque points (alpha=I(1/10)), a smoothing line, trimmed y. The elegant route adds components with +:

payment.plot.alpha <- qplot(x=Number.Of.Cases, y=Medicare.Average.Payment,
  data=subset(medicare.payments, Diagnosis.Related.Group %in% heart.failure),
  color=Diagnosis.Related.Group, alpha=I(1/10), ylim=c(0, 20000))
payment.plot.scaled <- payment.plot.alpha + scale_x_log10() + geom_smooth()
payment.plot.scaled

summary of this object now shows two geom/stat/position triples — one per layer — with the alpha aesthetic threaded through.

Reordering Factors for Dot Plots

Costs rise with volume — geography, perhaps? State-level data, with State levels reordered by payment (sorting the levels, not the rows) before a dot plot:

temp <- tempfile(fileext = ".rda")
download.file("https://raw.githubusercontent.com/cran/nutshell/master/data/medicare.payments.by.state.rda",
              temp, mode = "wb")
load(temp)
medicare.payments.by.state.hf <- subset(medicare.payments.by.state,
  Diagnosis.Related.Group %in% heart.failure)
medicare.payments.by.state.hf$State <- with(medicare.payments.by.state.hf,
  reorder(State, Medicare.Average.Payment.Maximum, mean))
qplot(x=Medicare.Average.Payment.Maximum, y=State,
      data=medicare.payments.by.state.hf,
      color=Diagnosis.Related.Group)

Top of the list: Northern Mariana Islands, Alaska, Virgin Islands — remote and expensive — then NY, MD, CA: high cost of living and big hospitals. Cheapest: Puerto Rico, American Samoa. Cost and volume are both correlated with geography.

A Choropleth Map

Are adjacent states similar? Merge payments with state polygons (maps package) and fill by payment — geom="polygon" plus a repositioned legend:

library(maps)
states <- map_data("state")
state.name.map <- data.frame(abb=state.abb, region=tolower(state.name),
                             stringsAsFactors=FALSE)
states <- merge(states, state.name.map, by="region")
toplot <- merge(states, medicare.payments.by.state,
                by.x="abb", by.y="State")
toplot <- toplot[order(toplot$order), ]
qplot(long, lat,
      data=subset(toplot,
        Diagnosis.Related.Group=="Heart failure and shock w/o CC/MCC"),
      group=group, fill=Medicare.Average.Payment.Maximum,
      geom="polygon") +
  theme(legend.position="bottom", legend.direction="vertical")

(The book wrote opts(...) — long since replaced by theme().)

Quick Plot and the Full Grammar

qplot: Arguments

qplot(x, y, ..., data, facets, margins, geom, stat,
      position, xlim, ylim, log, main, xlab, ylab, asp)
Argument Description Default
x, y, data values and (optional) data frame
facets facet formula: one-sided → facet_wrap, two-sided → facet_grid NULL
geom geoms as character vector "auto": point (x,y) / histogram (x)
stat / position statistics / position adjustments (book era — removed in ggplot2 2.0) list(NULL)
xlim, ylim, log limits; log = ""/"x"/"y"/"xy"
main, xlab, ylab, asp titles, labels, aspect deparsed expressions
... other aesthetics passed to layers

ggplot + Layers: the Verbose, Flexible Way

Beyond qplot: create the object with ggplot(data, mapping=aes(...)), then add layers and options with +:

plt <- ggplot(data=d, mapping=aes(x=a, y=b)) + geom_point()
summary(plt)
data: a, b, c [10x3]
mapping:  x = ~a, y = ~b
faceting:  <empty> 
-----------------------------------
geom_point: na.rm = FALSE
stat_identity: na.rm = FALSE
position_identity
  • aes(x, y, ...) declares the mappings (relatives: aes_string, aes_auto in the book’s era).
  • A bare ggplot(...) has no layers — printing it once errored (“No layers in plot”); modern ggplot2 draws an empty panel.
  • Book-era layer("point") added a layer by short-name; modern layer() needs explicit geom/stat/position — in practice, use geom_*.

The geom Vocabulary

geom_abline line by slope+intercept geom_errorbar(h) error bars geom_polygon polygons
geom_area area plot geom_freqpoly frequency polygon geom_quantile quantile-regression lines
geom_bar bars geom_hex hexagonal binning geom_rect rectangles
geom_bin2d 2-D bin heatmap geom_histogram histogram geom_ribbon ribbons (à la Minard)
geom_blank nothing geom_hline / geom_vline h/v lines geom_rug rug
geom_boxplot box plots geom_jitter jittered points geom_segment segments
geom_contour contours geom_line / geom_path lines/paths geom_smooth smoothed mean
geom_crossbar boxless boxes geom_linerange vertical intervals geom_step steps
geom_density(2d) densities geom_point / geom_pointrange geom_text text
geom_tile tiles

stats, scales, coords, facets, positions

  • stat_: bin, bin2d, binhex, boxplot, contour, density, density2d, function, identity, qq, quantile, smooth, spoke, sum, summary, unique, plus line helpers abline/hline/vline (modern ggplot2 folds these into geoms).
  • scale_: alpha, brewer (ColorBrewer), continuous, date/datetime, discrete, gradient/gradient2/gradientn, grey, hue, identity, linetype, manual, shape, size (today spelled e.g. scale_color_brewer, scale_x_continuous).
  • coord_: cartesian, equal, flip, map, polar, trans.
  • facet_: facet_grid (grid layout) and facet_wrap (wrapped 1-D list).
  • position_: dodge, fill, identity, jitter, stack.

Learning More

  • Hadley Wickham’s book ggplot2: Elegant Graphics for Data Analysis — the definitive treatment.
  • The official documentation site (now https://ggplot2.tidyverse.org) and the R Graphics Cookbook.

Tip

Course note on API drift. Since the book: opts() → theme(); qplot is soft-deprecated in favor of ggplot() + geom_*(); scales are written scale_<aesthetic>_<type>. The grammar itself — data, mapping, geom, stat, scale, coord, facet, position — is unchanged, which is exactly the point of having a grammar.