5  Data Visualization

Data visualization is one of the most powerful tools in the data scientist’s toolbox. Visuals allow us to quickly summarize complex data, spot trends and outliers, and communicate results to both technical and non-technical audiences. A good visualization can illuminate patterns that might remain hidden in tables or numerical summaries, while a poor visualization can obscure the truth or even mislead. In this chapter, we explore both base R graphics and the ggplot2 package, emphasizing good practices and illustrating common pitfalls. We will also critique real world examples of misleading charts and learn how to improve them.

5.1 Base R Graphics: plot()

R has a built-in graphics system that allows us to create plots quickly. The plot() function is versatile: depending on the type of data it is given, it can produce scatterplots, line plots, or even factor-based displays. This makes plot() an excellent starting point for beginners.

5.1.1 Scatterplot

Scatterplots display the relationship between two continuous variables. In the example below, we investigate how car weight relates to fuel efficiency using the built-in mtcars dataset.

Code 
### simple scatter using built-in `mtcars`
plot(mtcars$wt, mtcars$mpg,
     main = "Fuel efficiency vs. weight",
     xlab = "Weight (1000 lbs)", ylab = "MPG",
     pch = 19, col = "steelblue")

5.1.2 Line plot

When data are ordered, such as time series or physical measurements, line plots are appropriate. The following plot shows how pressure changes with temperature.

Code 
## line plot via type='l'
plot(pressure$temperature, pressure$pressure,
     type = "l", lwd = 2,
     main = "Pressure vs. Temperature",
     xlab = "Temperature", ylab = "Pressure")

Tip

Tip. Use options like pch, col, cex, and type to control appearance in base graphics. These adjustments can make exploratory plots more readable and more informative.

Base R graphics are quick and convenient, but they can be inconsistent and limited when creating complex or publication-quality graphics. This motivates the use of a more systematic framework.

5.2 ggplot2 Basics

Although base R allows us to make plots quickly, its commands are not always consistent, and combining multiple layers can be challenging. This is where the Grammar of Graphics comes in (Wilkinson, 1999). The the strongest reason is that grammar teaches you to think systematically about how graphics are constructed, not just how to make a specific chart.

That means instead of thinking “I want a scatterplot” or “I want a bar chart,” you think in terms of layers of grammar:

  • Data: what dataset to use.
  • Aesthetics: how variables map to visual properties (x, y, color, size, etc.).
  • Geoms: what geometric objects to draw (points, lines, bars).
  • Stats: statistical transformations (counts, smoothing, regression fits).
  • Scales: how data values are translated into colors, axes, and sizes.
  • Facets: how to split data into subplots for comparison.
  • Themes: the non-data ink (fonts, grid lines, background).

The ggplot2package (Wickham, 2016) implements this grammar, making it possible to build complex visualizations piece by piece. Because of this structure, ggplot2 is:

  • Consistent: once you know the grammar, you can build almost any plot without learning a new function each time.
  • Extensible: the same framework supports extensions (e.g., gganimate, ggrepel, patchwork).
  • Reproducible: code expresses the intent clearly, which is especially useful for teaching and communication.

The syntax involves calling ggplot() with a dataset and aesthetic mappings (via aes()), then adding layers such as geom_point() or geom_line() with the + operator. Additional layers for smoothing, faceting, and themes give us rich control over the appearance of plots.

For more syntax, see ggplot Cheatsheet

5.2.1 Exploring the mpg dataset

We begin by loading the tidyverse, which includes ggplot2, and looking at the mpg dataset.

Code 
library(tidyverse)

mpg

Here, mpg is a data frame containing information on car models, including engine displacement, highway mileage, and class.

Code 
glimpse(mpg)
View(mpg)
?mpg

The functions glimpse() and View() allow us to quickly inspect the structure of the dataset, while ?mpg shows documentation.

5.2.2 First ggplot calls

Before plotting, we might check available geoms:

Code 
?ggplot
?geom_point
?geom_line

5.2.3 Using pipes

Pipes make code easier to read by passing the result of one expression into the next. Instead of nesting functions, we can write a sequence of operations in the order we think about them. There are two main pipes in R: the base R pipe |> (available since R 4.1) and the magrittr pipe %>% (commonly used in the tidyverse).

Both pipes take the left-hand side and feed it into the first argument of the right-hand side.

Code 
# Base R pipe
mpg |> head()
Code 
# Magrittr pipe (needs library(magrittr) or tidyverse)
mpg %>% head()

We can now create a basic scatterplot of engine displacement vs highway mileage.

Code 
mpg |>
  ggplot() +
  geom_point(aes(displ, hwy))

This produces a scatterplot with displ on the x-axis and hwy on the y-axis. Swapping the variables simply flips the axes:

Code 
mpg |>
  ggplot() +
  geom_point(aes(hwy, displ))

5.2.4 Building layers

We can add additional layers. For example, combining points with a line layer:

Code 
mpg %>%
  ggplot() + 
  geom_point(aes(displ, hwy)) +
  geom_line(aes(displ, hwy), color = "tomato")

5.2.5 Adding aesthetics

Color can highlight categories such as car class:

Code 
mpg %>%
  ggplot() + 
  geom_point(aes(displ, hwy, color = class))

5.2.6 Adding smoothers

A smoothing curve helps reveal overall trends.

Code 
mpg %>%
  ggplot() +
  geom_point(aes(displ, hwy, color = class)) +
  geom_smooth(aes(displ, hwy))

Themes can alter the look of the plot:

Code 
mpg %>%
  ggplot() +
  geom_point(aes(displ, hwy, color = class)) +
  geom_smooth(aes(displ, hwy)) +
  theme_bw()

5.2.7 Customization

We can set fixed aesthetics outside aes():

Code 
mpg %>%
  ggplot(aes(displ, hwy)) + 
  geom_point(color = "steelblue", size = 3)

Transparency can improve clarity:

Code 
mpg %>%
  ggplot(aes(displ, hwy, color = class)) +
  geom_point(size = 2, alpha = 0.8)

5.2.8 Adding regression lines

We can fit smoothers with different methods:

Code 
mpg %>%
  ggplot(aes(displ, hwy)) +
  geom_point(aes(color = class)) +
  geom_smooth(se = FALSE)
Code 
mpg %>%
  ggplot(aes(displ, hwy)) +
  geom_point(aes(color = class)) +
  geom_smooth(method = "lm", se = FALSE)

5.2.9 Titles, labels, and themes

Code 
mpg %>%
  ggplot(aes(displ, hwy, color = class)) +
  geom_point(size = 2) +
  labs(
    title = "Fuel efficiency vs. engine displacement",
    x = "Engine displacement (liters)",
    y = "Highway MPG"
  ) +
  theme_minimal()

5.2.10 Faceting

We can split data into subplots by categories.

Code 
mpg %>%
  ggplot(aes(displ, hwy, color = class)) +
  geom_point(size = 2) +
  facet_wrap(~ class)
Code 
mpg %>%
  ggplot(aes(displ, hwy)) +
  geom_point() +
  facet_grid(drv ~ cyl)

5.2.11 Other plot types

Bar charts summarize categorical data:

Code 
mpg %>%
  ggplot(aes(class)) +
  geom_bar()

Histograms show distributions:

Code 
mpg %>%
  ggplot(aes(hwy)) +
  geom_histogram(bins = 20)

Density plots are another way to display distributions:

Code 
mpg %>%
  ggplot(aes(hwy)) +
  geom_density()

5.2.12 Other ggplot functions

  • coord_flip flips the x and y axis to improve the readability of plots
  • scales change the formatting of x and y axes
  • plotly makes plots interactive; you can hover over points/lines for more information
  • labs allows you to add/edit a title, subtitle, a caption, and change the x and y axis labels
  • gganimate allows you to animate plots into gifs

5.3 Misleading Charts

Visualizations can be abused to mislead. It is important to learn how to critically assess charts we see in the media. The following real-world examples show common problems and better alternatives.

5.3.1 Example 1: Truncated Bar Chart

Code 
approval <- data.frame(year = c(2000, 2005), percent = c(77, 65))
barplot(approval$percent, names.arg = approval$year,
        ylim = c(60, 80), col = "tomato",
        main = "Approval Ratings (misleading)")

Code 
barplot(approval$percent, names.arg = approval$year,
        ylim = c(0, 100), col = "steelblue",
        main = "Approval Ratings (truthful)")

5.3.2 Example 2: Straw Poll Graphic

Code 
poll <- data.frame(candidate = c("A", "B", "C", "D"),
                   support = c(22, 18, 15, 10))
barplot(poll$support, names.arg = poll$candidate,
        col = c("red", "blue", "green", "purple"),
        main = "Poll Results (misleading)")

Code 
library(ggplot2)
poll |> ggplot(aes(x = reorder(candidate, support), y = support)) +
  geom_col(fill = "steelblue") +
  coord_flip() +
  labs(title = "Poll Results (truthful)",
       x = "Candidate", y = "Support (%)")

5.3.3 Example 3: Donut vs Bar Chart

Code 
shares <- data.frame(group = c("X", "Y", "Z"), value = c(30, 50, 20))
shares |> ggplot(aes(x = 2, y = value, fill = group)) +
  geom_col(width = 1, color = "white") +
  coord_polar(theta = "y") +
  xlim(0.5, 2.5) +
  theme_void() +
  labs(title = "Shares (donut, misleading)")

Code 
shares |> ggplot(aes(x = group, y = value, fill = group)) +
  geom_col() +
  labs(title = "Shares (bar chart)",
       x = NULL, y = "Value")

Pie charts (and donut charts, which are essentially pies with a hole in the middle) are widely criticized because humans are not good at accurately comparing angles or areas. Judgments based on angles are much less precise than those based on position or length. This makes pie charts poor at conveying quantitative comparisons, especially when slices are similar in size. Donut charts exacerbate the problem by removing the center, which eliminates a natural visual baseline (the full radius), making angle judgments even harder. For these reasons, most visualization experts recommend bar charts instead, where lengths aligned to a common baseline support more accurate comparisons.

5.4 Good Practice

5.4.1 Bad Plots

  • Truncated axes exaggerate differences.
  • 3D effects distort perception.
  • Pie/donut charts hinder comparisons.
  • Overplotting or excessive colors obscure patterns.
  • Inconsistent scales or ordering confuse the audience.

5.4.2 Good Plots

  • Start bar charts at zero to preserve proportion.
  • Use simple, clear chart types.
  • Provide informative labels and titles.
  • Keep scales and colors consistent.
  • Avoid unnecessary clutter.

5.5 In-Class Example

Consider the data of Chetty et al. (2014).

  1. Visualize the relationship between social capital and absolute mobility. Do you see a correlation? Is it what you expected from the Chetty et al. (2014) study executive summary?

  2. Add an aesthetic to your graph to represent whether the CZ is urban or not.

  3. Separate urban and non-urban CZ’s into two separate plots.

  4. Add a smooth fit to each of your plots above. Experiment with adding the option method="lm" in your geom_smooth. What does this option do?

  5. Which variables in the chetty data frame are appropriate x variables for a bar graph?

  6. Make two separate bar graphs for two different x variables.

  7. Make two more bar graphs that display proportions rather than counts of your selected variables.

  8. Make a bar graph that lets you compare the number of urban and rural CZ’s in each of the four regions.

5.6 Summary

We have explored both base R plotting and the ggplot2 grammar of graphics. Base R offers quick and simple plotting functions, but lacks consistency for more advanced tasks. ggplot2 provides a flexible and layered system, allowing us to build complex visualizations step by step. By studying both good and bad visualizations, we learn not only how to make effective charts but also how to critically evaluate visuals we encounter in practice.