Before diving into the mechanics of creating plots, consider why visualization matters. The human visual system excels at detecting patterns, spotting outliers, and perceiving relationships—abilities that summary statistics cannot replace.
Consider Anscombe’s Quartet—four datasets with nearly identical summary statistics (same mean, variance, and correlation) but completely different patterns:
# Reshape Anscombe's built-in dataset
anscombe_long <- anscombe |>
pivot_longer(everything(),
names_to = c(".value", "set"),
names_pattern = "(.)(.)")
ggplot(anscombe_long, aes(x = x, y = y)) +
geom_point(size = 2, color = "steelblue") +
geom_smooth(method = "lm", se = FALSE, color = "coral") +
facet_wrap(~set, ncol = 2) +
labs(title = "Same Mean, Variance, and Correlation—Different Stories") +
theme_minimal()
All four datasets have nearly identical statistical summaries, yet they represent fundamentally different phenomena: a linear relationship, a curved relationship, an outlier-driven relationship, and a vertical cluster with one outlier. Summary statistics alone would suggest these datasets are equivalent—only visualization reveals the truth.
Never trust summary statistics alone. Before running statistical tests, visualize your data to check assumptions, identify outliers, and understand the underlying patterns.
Different questions call for different visualizations. Matching your question to the right chart type is the first step toward effective communication:
| Question | Chart Type | Why |
|---|---|---|
| How are values distributed? | Histogram, density plot | Shows shape, center, spread |
| How do groups compare? | Box plot, bar chart | Side-by-side comparison |
| How do two variables relate? | Scatter plot | Shows correlation, patterns |
| How does a value change over time? | Line plot | Connects sequential observations |
| What is the composition? | Stacked bar chart | Shows parts of a whole |
Before creating any visualization, ask yourself: “What question am I trying to answer?” The chart type should emerge from the question, not the other way around.
Data visualization is both an art and a science. A well-designed graphic can reveal patterns, communicate findings, and guide analysis in ways that tables of numbers cannot. The ggplot2 package implements a coherent system for creating graphics based on Leland Wilkinson’s “Grammar of Graphics”—a framework that describes the fundamental components from which all statistical graphics can be built.
Just as grammar provides rules for constructing sentences from words, the grammar of graphics provides rules for constructing visualizations from components. Every graphic is composed of data, aesthetic mappings that connect variables to visual properties, and geometric objects that represent data points. Additional components like scales, statistical transformations, coordinate systems, and facets allow for sophisticated customizations.
The basic structure of a ggplot2 call begins with the ggplot() function, which creates a coordinate system. You add layers to this foundation using the + operator.
ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
geom_point()
This creates a scatterplot of highway fuel efficiency against engine displacement using the built-in mpg dataset. The aes() function establishes the aesthetic mapping—which variables map to which visual properties. Here, displ maps to the x-axis and hwy to the y-axis. The geom_point() function adds a layer of points.
Aesthetics are visual properties of the plot. Beyond position (x and y), common aesthetics include color, size, shape, and transparency (alpha). You can map variables to these aesthetics to encode additional information.
ggplot(mpg, aes(x = displ, y = hwy, color = class)) +
geom_point(size = 3, alpha = 0.7)
Now the color of each point indicates the vehicle class. The legend is created automatically. Note that aesthetics defined inside aes() are mapped to variables, while those defined outside (like size = 3) apply uniformly to all points.
Geometric objects, or geoms, determine what type of plot you create. Different geoms represent data in different ways.
Points are good for showing the relationship between two continuous variables:
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point()
Lines connect points in order, useful for time series or showing trends:
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point() +
geom_smooth()
The geom_smooth() function adds a smoothed conditional mean with confidence interval.
Bar charts show counts or summaries of categorical data:
ggplot(diamonds, aes(x = cut)) +
geom_bar()
Use fill to color bars by another variable:
ggplot(diamonds, aes(x = cut, fill = clarity)) +
geom_bar(position = "dodge")
Histograms show the distribution of a continuous variable:
ggplot(diamonds, aes(x = carat)) +
geom_histogram(binwidth = 0.1, fill = "steelblue", color = "white")
Boxplots summarize distributions and highlight outliers:
ggplot(mpg, aes(x = class, y = hwy)) +
geom_boxplot()
You can layer multiple geoms to create richer visualizations:
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(aes(color = class)) +
geom_smooth(se = FALSE, color = "black")
Faceting creates small multiples—separate panels for subsets of the data. This is powerful for comparing patterns across groups.
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point() +
facet_wrap(~ class, nrow = 2)
Use facet_grid() for two-variable faceting:
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point() +
facet_grid(drv ~ cyl)
Add informative labels with the labs() function:
ggplot(mpg, aes(x = displ, y = hwy, color = class)) +
geom_point() +
geom_smooth(se = FALSE) +
labs(
title = "Fuel Efficiency Decreases with Engine Size",
subtitle = "Data from EPA fuel economy tests",
caption = "Source: fueleconomy.gov",
x = "Engine Displacement (liters)",
y = "Highway Fuel Efficiency (mpg)",
color = "Vehicle Class"
)
Themes control the non-data aspects of the plot—background, grid lines, fonts, etc. ggplot2 includes several built-in themes:
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point() +
theme_classic()
Other built-in themes include theme_minimal(), theme_bw(), theme_light(), and theme_dark(). The ggthemes package provides many additional themes.
Choosing an appropriate visualization depends on the types of variables you want to display and the message you want to convey.
For one categorical variable, use bar charts. For one continuous variable, use histograms or density plots. For two continuous variables, use scatterplots. For one continuous and one categorical, use boxplots or violin plots. For two categorical variables, use stacked or grouped bar charts or heat maps.
Edward Tufte articulated principles of graphical excellence that remain influential: “Graphical excellence is that which gives to the viewer the greatest number of ideas in the shortest time with the least ink in the smallest space.”
Key principles include:
Show the data. Above all else, make the data visible. Avoid chart junk that obscures what you are trying to communicate.
Encourage comparison. Design graphics to facilitate comparison of different groups or conditions.
Represent magnitudes honestly. The visual representation should be proportional to the numerical quantities being represented. Avoid truncated axes that exaggerate differences.
Minimize clutter. Remove unnecessary grid lines, borders, and decorations. Every element should serve a purpose.
Make displays easy to interpret. Use clear labels, appropriate colors, and logical organization.
By default, R orders categorical variables alphabetically, which is rarely the most informative arrangement. Use reorder() to order categories by a meaningful value:
# Create sample data
sample_data <- tibble(
treatment = c("Control", "Low Dose", "Medium Dose", "High Dose"),
response = c(12, 18, 25, 31)
)
# Default alphabetical order (not ideal)
ggplot(sample_data, aes(x = treatment, y = response)) +
geom_col(fill = "steelblue") +
labs(title = "Default Order (Alphabetical)",
x = "Treatment", y = "Response") +
theme_minimal()
# Order by response value (more meaningful)
sample_data |>
mutate(treatment = reorder(treatment, response)) |>
ggplot(aes(x = treatment, y = response)) +
geom_col(fill = "steelblue") +
labs(title = "Ordered by Value (More Meaningful)",
x = "Treatment", y = "Response") +
theme_minimal()
The reorder() function takes a categorical variable and a numeric variable, reordering the categories by the numeric values. For horizontal bar charts (which are often easier to read), add coord_flip():
sample_data |>
mutate(treatment = reorder(treatment, response)) |>
ggplot(aes(x = treatment, y = response)) +
geom_col(fill = "steelblue") +
coord_flip() +
labs(title = "Horizontal Bars (Good for Long Labels)",
x = NULL, y = "Response") +
theme_minimal()
Effective visualization depends on understanding how humans perceive visual information. We encode data using visual cues—properties like position, length, color, and shape. But not all visual cues are equally effective.
Research by Cleveland and McGill established that we perceive some visual encodings more accurately than others. From most to least accurate:
This hierarchy explains why bar charts work better than pie charts for comparing quantities—we judge lengths more accurately than angles or areas.
When possible, encode your most important data using position. Scatterplots, line graphs, and dot plots all use position effectively:
# Compare a pie chart vs. bar chart for the same data
library(patchwork)
category_data <- data.frame(
category = c("Engineering", "Medicine", "Natural Sciences", "Social Sciences"),
funding = c(35, 28, 22, 15)
)
# Bar chart - easy to compare
p_bar <- ggplot(category_data, aes(x = reorder(category, funding), y = funding)) +
geom_col(fill = "steelblue") +
coord_flip() +
labs(title = "Bar Chart: Easy Comparison", x = "", y = "Funding (%)") +
theme_minimal()
# Pie chart - harder to compare
p_pie <- ggplot(category_data, aes(x = "", y = funding, fill = category)) +
geom_col(width = 1) +
coord_polar("y") +
labs(title = "Pie Chart: Harder to Compare", fill = "Category") +
theme_void() +
theme(legend.position = "bottom")
p_bar + p_pie
The bar chart makes it immediately obvious that Engineering has the most funding. With the pie chart, you must work harder to compare the slice sizes.
A contentious issue in data visualization is whether the y-axis should always start at zero. The answer depends on the type of chart and what you’re trying to show.
For bar charts, the length of the bar represents the magnitude of the value. If the axis doesn’t start at zero, the visual representation misrepresents the data:
# Demonstration of misleading truncated axis
gdp_data <- data.frame(
country = c("A", "B", "C"),
gdp = c(45000, 47000, 49000)
)
p_trunc <- ggplot(gdp_data, aes(x = country, y = gdp)) +
geom_col(fill = "steelblue") +
coord_cartesian(ylim = c(44000, 50000)) +
labs(title = "Truncated Axis: Misleading",
subtitle = "Differences appear huge",
y = "GDP per capita") +
theme_minimal()
p_full <- ggplot(gdp_data, aes(x = country, y = gdp)) +
geom_col(fill = "steelblue") +
labs(title = "Full Axis: Honest",
subtitle = "Differences in proper context",
y = "GDP per capita") +
theme_minimal()
p_trunc + p_full
For position-based encodings like scatterplots and line charts, zero doesn’t need to be included if it would waste space and obscure meaningful variation:
# Temperature data - zero would be meaningless
set.seed(42)
temp_data <- data.frame(
day = 1:30,
temp = rnorm(30, mean = 72, sd = 5)
)
p_zero <- ggplot(temp_data, aes(x = day, y = temp)) +
geom_line(color = "firebrick") +
ylim(0, 100) +
labs(title = "Including Zero: Wastes Space", y = "Temperature (°F)") +
theme_minimal()
p_auto <- ggplot(temp_data, aes(x = day, y = temp)) +
geom_line(color = "firebrick") +
labs(title = "Natural Range: Shows Variation", y = "Temperature (°F)") +
theme_minimal()
p_zero + p_auto
The key question is: what would zero mean for this variable? For temperature in Fahrenheit, zero has no special significance for daily weather data. For proportions or counts, zero is meaningful and often should be included.
Bar charts: Always include zero—the bar length represents magnitude. Line charts and scatterplots: Include zero if it’s meaningful; otherwise, show the natural range of the data. When in doubt: Ask whether excluding zero could mislead readers about the magnitude of differences.
Sometimes the raw data doesn’t visualize well. Transformations can reveal patterns that are hidden in the original scale.
Many biological variables—gene expression, population sizes, concentrations—follow approximately log-normal distributions with long right tails. Log transformation can make patterns visible:
# Simulated gene expression data
set.seed(123)
expression_data <- data.frame(
gene_a = rlnorm(200, meanlog = 2, sdlog = 1.5),
gene_b = rlnorm(200, meanlog = 3, sdlog = 1.2)
)
p_raw <- ggplot(expression_data, aes(x = gene_a, y = gene_b)) +
geom_point(alpha = 0.5) +
labs(title = "Original Scale",
subtitle = "Pattern obscured by outliers",
x = "Gene A Expression", y = "Gene B Expression") +
theme_minimal()
p_log <- ggplot(expression_data, aes(x = gene_a, y = gene_b)) +
geom_point(alpha = 0.5) +
scale_x_log10() +
scale_y_log10() +
labs(title = "Log Scale",
subtitle = "Relationship visible",
x = "Gene A Expression (log)", y = "Gene B Expression (log)") +
theme_minimal()
p_raw + p_log
Consider log transformation when:
Be sure to label axes clearly when using transformed scales, and remember that zero cannot be log-transformed.
Color is a powerful but often misused encoding. Effective use of color requires understanding perception and accessibility.
Sequential: For ordered data from low to high. Use a single hue varying in lightness.
Diverging: For data with a meaningful midpoint. Two hues diverge from a neutral center.
Qualitative: For categorical data with no inherent order. Use distinct hues.
Approximately 8% of men and 0.5% of women have some form of color vision deficiency. Design for accessibility:
# Good practice: color + shape
ggplot(mpg, aes(x = displ, y = hwy, color = drv, shape = drv)) +
geom_point(size = 3) +
scale_color_viridis_d() +
labs(title = "Color + Shape: Accessible Design",
x = "Engine Displacement (L)", y = "Highway MPG",
color = "Drive Type", shape = "Drive Type") +
theme_minimal()
Rainbow color scales (like the default “jet” colormap in MATLAB) have serious problems:
Use viridis, plasma, or other perceptually uniform scales instead.
Small multiples—the same chart repeated for different subsets of the data—are remarkably effective for comparison. Edward Tufte called them “the best design solution for a wide range of problems in data presentation.”
# Small multiples example
ggplot(gapminder::gapminder %>%
filter(continent != "Oceania"),
aes(x = gdpPercap, y = lifeExp)) +
geom_point(alpha = 0.3, size = 0.8) +
geom_smooth(method = "loess", se = FALSE, color = "firebrick") +
scale_x_log10(labels = scales::comma) +
facet_grid(continent ~ cut(year, breaks = c(1950, 1970, 1990, 2010),
labels = c("1952-1970", "1971-1990", "1991-2007"))) +
labs(title = "Life Expectancy vs. GDP Over Time by Continent",
x = "GDP per Capita (log scale)",
y = "Life Expectancy (years)") +
theme_minimal() +
theme(strip.text = element_text(size = 9))
Small multiples work because:
Beyond the principles discussed, watch out for these common errors:
Overplotting: Too many points obscure patterns. Use transparency, jittering, or density plots.
# Overplotting solution
set.seed(42)
overplot_data <- data.frame(
x = rnorm(5000),
y = rnorm(5000)
)
p_over <- ggplot(overplot_data, aes(x, y)) +
geom_point() +
labs(title = "Overplotting: Points Hidden") +
theme_minimal()
p_alpha <- ggplot(overplot_data, aes(x, y)) +
geom_point(alpha = 0.1) +
labs(title = "Transparency: Density Visible") +
theme_minimal()
p_over + p_alpha
Dual y-axes: These are almost always misleading. The relationship between the two scales is arbitrary and can be manipulated to show any desired pattern.
3D effects: Three-dimensional bar charts, pie charts, and similar decorations distort perception without adding information. Avoid them.
Excessive decoration: Gridlines, borders, backgrounds, and other “chart junk” should be minimized. Focus attention on the data.
Recognizing bad graphics helps you avoid making them.
Ticker-tape style displays make it hard to see patterns. Lines connecting unrelated points mislead. Pie charts make comparisons difficult because humans are poor at judging angles. Three-dimensional effects distort perception without adding information.
Charles Minard’s 1869 map of Napoleon’s Russian campaign is often cited as one of the best statistical graphics ever made. It displays six variables: the size of the army, its location (latitude and longitude), direction of movement, temperature, and date—all in a single coherent image.
The graphic tells a story. You can see the army shrink as it advances, the devastating losses during the retreat, and the correlation with plummeting temperatures. No legend is needed; the meaning is immediately apparent.
Save plots with ggsave():
# Create and save a plot
p <- ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point()
ggsave("my_plot.png", p, width = 8, height = 6, dpi = 300)
ggsave("my_plot.pdf", p, width = 8, height = 6)The function infers the format from the file extension. Specify dimensions and resolution for publication-quality output.
The best way to learn ggplot2 is to use it. Take a dataset you care about and try different visualizations. Experiment with aesthetics, geoms, and facets. Read error messages carefully—they often point directly to the problem.
Create your first ggplot visualizations:
library(ggplot2)
data(mpg)
# Basic scatterplot
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point()theme_minimal(), theme_classic(), theme_bw())Practice with different geoms:
geom_histogram()geom_boxplot()geom_bar()Explore different aesthetic mappings:
size = 3) outside of aes()aes() versus setting a fixed value outside?# Mapped aesthetic (variable determines color)
ggplot(mpg, aes(x = displ, y = hwy, color = class)) +
geom_point(size = 3)
# Fixed aesthetic (all points same color)
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(color = "steelblue", size = 3)Practice creating small multiples:
facet_wrap()facet_grid() with two variablesscales argument to allow different axis scales per facetggplot(mpg, aes(x = displ, y = hwy)) +
geom_point() +
facet_wrap(~ class, nrow = 2)Build complex visualizations by layering:
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(aes(color = class), alpha = 0.7) +
geom_smooth(method = "lm", color = "black", se = TRUE) +
labs(
title = "Engine Size vs. Fuel Efficiency",
x = "Engine Displacement (L)",
y = "Highway MPG",
color = "Vehicle Class"
) +
theme_minimal()Create a polished figure suitable for publication:
ggsave() with appropriate dimensions and resolution