Professional Documents
Culture Documents
Better Data Visualizations Scholars
Better Data Visualizations Scholars
VISUALIZATIONS
COLUMBIA UNIVERSITY PRESS ր ր NEW YORK
Columbia University Press
Publishers Since 1893
New York Chichester, West Sussex
cup.columbia.edu
Copyright © 2021 Columbia University Press
All rights reserved
Chapter 11, “Tables,” based on Jonathan A. Schwabish, “Ten Guidelines for Better Tables,”
Journal of Benefit-Cost Analysis 11, no. 2 (2020): 151–178. Reprinted with permission.
INTRODUCTION 1
5. TIME 133
Line Chart 133
Circular Line Chart 149
Slope Chart 150
Sparklines 152
Bump Chart 153
Cycle Chart 155
Area Chart 157
Stacked Area Chart 159
Streamgraph 162
Horizon Chart 164
Gantt Chart 166
Flow Charts and Timelines 170
Connected Scatterplot 175
Conclusion 177
6. DISTRIBUTION 179
Histogram 179
CONTENTSփ ix
7. GEOSPATIAL 217
Choropleth Map 220
Cartogram 233
Proportional Symbol and Dot Density Maps 243
Flow Map 245
Conclusion 248
8. RELATIONSHIP 249
Scatterplot 249
Parallel Coordinates Plot 263
Radar Charts 267
Chord Diagram 269
Arc Chart 272
Correlation Matrix 275
Network Diagrams 277
Tree Diagrams 284
Conclusion 287
9. PARTҺTOҺHOLE 289
Pie Charts 289
Treemap 297
Sunburst Diagram 299
Nightingale Chart 300
Voronoi Diagram 304
Conclusion 309
xփ CONTENTS
CONCLUSION 391
Acknowledgments 409
References 413
Index 431
BETTER DATA
VISUALIZATION
INTRODUCTION
R
aise your hand if your approach to creating a graph goes something like this: You ana-
lyze some data. Write up the results. Make a graph and drop it into the report, sur-
rounded by text. Label it something benign like “Figure 1. Average Earnings, 1990–2020.”
Save it as a PDF. Post it to the world.
It might have taken you months or even years to compile and analyze the data and write
the report. For many, it takes far less time to design the graphs that showcase that data. You
might open a program like Microsoft Excel, paste in the data, click through the drop-down
menu, select one you’ve used dozens or hundreds of times, accept the default formatting,
and paste it into the report.
But at any point in this sequence did you pause to consider what’s most important about
communicating the work? It’s the audience. People will read your report. People will listen to
you discuss your work. And yet many of us spend far too little time thinking about how we can
best present our findings. Instead we use whatever default approach is quickest and easiest.
Why is this? Maybe you don’t believe you have the technical skills or design know-how
to create complex, attractive graphs. Or you worry it’s not worth the effort, because your
managers or tenure committee or whoever else won’t see it as time well spent. Many people
simply think that their reader will just “get it,” as if everyone has seen the content a hundred
times before. But many readers, especially those who can make change or implement policy,
may have never seen this content before. In these cases—which are probably most of them—
thinking carefully about how data is presented is just as important as the data itself.
This book is about how to create better, more effective visualizations of your data. It aims
to expand your graphic literacy and put more graphs in your toolbox. The next time you open
2փ INTRODUCTION
Excel, Tableau, R, or whatever your software tool of choice, you won’t be bound by the graphs
in the dropdown menus or the tutorial manual. This book will guide you to choose the graph
that is the best fit for your data and will most effectively communicate your message.
People often tell me they can’t create some of these different, nonstandard graphs because
their colleague or manager or audience won’t understand them. We are not born knowing
instinctively how to read a bar chart or line chart or pie chart. As Scott Klein, deputy manag-
ing editor at ProPublica once wrote, “There is no such thing as an innately intuitive graphic.
None of us are born literate in reading visualizations.”
As data visualization creators, we must understand our audience and know when a differ-
ent graph can engage readers—and help them expand their own graphic literacy.
This book has three main parts. Part 1 covers general guidelines to creating effective visu-
alizations. We’ll learn the importance of our audience and how to consider what category
of graph will best meet their needs. No data visualization book will contain every lesson to
create effective graphs, but there are some best practices that can guide your work. As you go
Each of these six charts visualizes the same data: The share of people earning minimum
wage or less in each state.
INTRODUCTIONփ 3
forward creating more visuals and seeing their effect on your audience, you’ll develop your
own aesthetic and learn when to bend or break these guidelines.
Part 2 is the meat of the book. We will define and discuss more than eighty graphs, cat-
egorized into eight broad categories: Comparisons, Time, Distribution, Geospatial, Rela-
tionship, Part-to-Whole, Qualitative, and Tables. We will see how each graph works and the
advantages and disadvantages of each.
Graphs overlap between these categories—a bar chart, for example, can be used to show
changes over time or comparisons between groups. The categorizations here are based on
a graph’s primary purpose. But even that’s not an objective truth, and your perspective and
situation may differ. I do not discuss every single possible graph—there are many specialized
graphs in fields like architecture, biology, and engineering that are excluded here. Instead,
these chapters cover the most common and flexible graphs that can showcase the sorts of
data most people will need to display.
I tie these chapters together in part 3 with a chapter on building a data visualization style guide
and a chapter on how to pull the different lessons together in a series of graph redesigns. If you’ve
ever written a research paper, or even a book report, you are probably aware of the array of writing
style guides, from the Chicago Manual of Style to the Modern Language Association. These guides
break down writing into component parts and prescribe their proper use. A data visualization style
guide does the same for graphs—defines their parts and how to style and use them. In the final
chapter, we apply the lessons to redesign a series of graphs to improve how they communicate data.
This book will guide you as you explore your data and how it might be visualized. Now
more than ever, content must be visual if it is to travel far. Your clients and colleagues, and
your audiences of policymakers, decisionmakers, and interested readers are inundated with
a flow of information. Visuals cut through that.
Anyone can improve the way they visualize and communicate their data—and you don’t
need a graduate degree in marketing or design or advertising. Take it from me, I started my
career as an economist in the federal government.
I moved to DC in 2005 to join the Congressional Budget Office (CBO). My job was to
help work on the long-term microsimulation model that is used to examine the Social Secu-
rity system and forecast the long-term finances of the federal budget. The spring of 2005 was
an exciting time to work on Social Security—President George W. Bush had made Social
Security a central component of his second term. In his 2005 State of the Union address, he
said, “We must pass reforms that solve the financial problems of Social Security once and for
all.” Reform would stall later that year, but in the course of my first few months on the job,
my group at CBO estimated and analyzed the effects of dozens of policy proposals.
Five years later, I had expanded my work to include issues around policies that affected
disabled workers, immigration, and food stamps (now called the Supplemental Nutrition
Assistance Program or SNAP). In 2010, three of my colleagues were drafting a special report
on policy options for Social Security. In it, they would show the impact of thirty different
options for reform. One of the central figures in the report would show changes in taxes
received by the system, benefits paid out from the system, the balance between the two, and
other measures of fiscal solvency for these thirty options. It looked something like this:
Author’s rendering of early draft of exhibit from the Congressional Budget Office.
INTRODUCTIONփ 5
You don’t need to be a government economist to know that members of Congress are
unlikely to read something that looks like a spreadsheet. There are too many rows, too many
columns, too many numbers—too much information. It was right then that I first started
thinking about better ways to present this information.
This was the result. We replaced some numbers with small area charts, which give the
reader an immediate visual impression of each option—which ones increased the solvency
of the program and which ones did not.
Final version of that main exhibit in the Congressional Budget Office report on Social
Security. Notice that there is less data and more graphs.
Source: Congressional Budget Office.
The report worked. We received good feedback from colleagues at CBO and other agen-
cies, as well as readers on Capitol Hill and elsewhere, noting how easy it was to read and
digest the graphs. It was maybe the first time I (and perhaps the agency) thought carefully
6փ INTRODUCTION
and strategically about our data visuals. From there, I started reading books on data visual-
ization, design, color theory, and typography.
Working with our editorial department and designers, we began to improve the
graphs in our basic reports and started creating new report and graph types. We made
infographics—what was then a buzzword referring (sometimes derisively) to longer
graphics that combine data, text, images, and more into a single visual. In 2012, we cre-
ated this infographic to accompany and summarize The Long-Term Budget Outlook, a
109-page report.
One-page infographic about the 2012 Long-Term Budget Outlook from the Congressional
Budget Office.
Source: Congressional Budget Office.
INTRODUCTIONփ 7
That June, CBO’s director sat in front the U.S. House Budget Committee to relay the
results of our analysis. As the hearing played on a TV out in the hallway, I suddenly heard
yells of, “Jon! Jon! Come out! Your infographic is on TV!”
And, sure enough, Congressman Chris Van Hollen was holding up the infographic on
C-SPAN, covered with scribbles and notes. The visualization had captured and engaged the
attention of one of the busiest people in America, and someone who could do something
about the pressures facing the federal budget. That was the moment I knew that how we
presented our data could matter as much as the data itself.
In 2014, I moved to the Urban Institute, a nonprofit research institution in Washington,
DC, to spend half of my time conducting research and half of my time in the Communica-
tions department, helping colleagues present and visualize their data.
Since that time, I have conducted hundreds of workshops, delivered lectures around the
globe, and published two books on data communication. The world, it seemed, had seen
what I saw—better visual content and better presentations were the currency of research and
Maryland Congressman Chris Van Hollen holding up that Long-Term Budget Outlook info-
graphic in a House Budget Committee hearing.
Source: C-SPAN2.
8փ INTRODUCTION
policy adoption. The advance of computing power, social media platforms, and the expand-
ing media landscape made visual content more important, perhaps even necessary.
Today, I work with people in nonprofits, government agencies, private sector companies,
and everything in between to improve how they create their graphs and communicate their
content. I’ve worked with junior economists and analysts dealing with enormous data sets;
health care workers trying to communicate results to patients, families, and hospital admin-
istrators; human resource representatives working with databases of job-seekers; advertisers
and marketing executives selling products to clients; and many more.
I’ve seen hundreds of different kinds of data visualization challenges. The skills to meet
them, unfortunately, are not yet regularly taught in schools or professional development
programs. But these skills can be learned. We can learn how to read chart types we’ve never
seen before, even if they are complex. And we can learn how to communicate our work in
better and more effective ways.
Eventually, I discovered that one of the most important things I can show people is the
incredibly wide array of graphs available to them. And that is precisely the content of this
book, a survey of more than eighty types of data visualizations, from the familiar to the
nonstandard.
But before we get to the library of graph types, we’ll consider some of the science behind
how we process visual information and some best practices and approaches to visualizing data.
PART ONE
PRINCIPLES OF DATA VISUALIZATION
VISUAL PROCESSING AND
PERCEPTUAL RANKINGS
1
B
efore we start creating our charts and graphs, we need to cover some basic theory of
how the brain perceives visual stimuli. This will guide you as you decide what chart type
is most appropriate to visualize your data.
When we consider how to visualize our data, we must ask ourselves how accurately the
reader can perceive the data values. Are some graphs better equipped to guide the reader to
the specific difference between, say, 2 percent and 2.3 percent? If so, how should we think
about those differences as we create our visualizations?
There’s a thread of research in the data visualization field that explores this very ques-
tion. Based on original research over the past thirty years or so, the image on the next
page shows a spectrum of graphs—or more generally, types of data encodings like dots,
lines, and bars—arrayed by how easily readers can estimate their value. The encodings that
readers can most accurately estimate are arranged at the top, and those that enable more
general estimates are at the bottom.
The rankings are unsurprising. It is easier to compare the data in line charts, bar charts,
and area charts that have the same axis or baseline. Graphs on which the data are positioned
on unaligned axes—think of a pair of bars that are offset from one another on different
axes—are slightly harder for us to accurately discern the values.
Farther down the vertical axis are encodings based on angle, area, volume, and color.
You intuitively know this: it’s much easier to discern the exact data values and differences
between values when reading a bar chart than when reading a map where countries are
shaded with different colors.
ovbঞom-Ѵom]1ollomv1-Ѵ;v
Enable ovbঞom-Ѵom]b7;mঞ1-Ѵķmom-Ѵb]m;7v1-Ѵ;v
accurate
;vঞl-|;v
Length
Area
(oѴl;
"_-7bm]-m7v-|u-ঞom
May enable
general Color hue
;vঞl-|;v
Perceptual ranking diagram. What kind of data visualization you choose to create will
depend on your goals and your audience’s needs, experiences, and expertise. This image is
based on Alberto Cairo (2016) from research by Cleveland and McGill (1984), Heer,
Bostock, and Ogievetsky (2010), and others.
VISUAL PROCESSING AND PERCEPTUAL RANKINGSփ 15
Standard graphs, like bar and line charts, are so common because they are perceptually
more accurate, familiar to people, and easy to create. Nonstandard graphs—those that use
circles or curves, for instance—may not allow the reader to most accurately perceive the
exact data values.
But perceptual accuracy is not always the goal. And sometimes it’s not a goal at all.
Spurring readers to engage with a graph is sometimes just as important. Sometimes, it’s
more important. And nonstandard chart types may do just that. In some cases, nonstandard
graphs may help show underlying patterns and trends in better ways that standard graphs.
In other cases, the fact that these nonstandard graphs are different may make them more
engaging, which we may sometimes need to first attract attention to the visualization.
This graphic from information designer Federica Fragapane shows the fifty most vio-
lent cities in the world in 2017. The vertical axis measures the population of each city and
the horizontal axis captures the homicide rate per 100,000 people. The number of lines in
each icon represents the number of homicides, and additional colors, shapes, and markers
capture metrics like country of origin (the symbol in the middle of each), region (vertical
dashed line), and change since 2016 (blue for decreases, red for increases). It could be a bar
Graphic from Frederica Fragapane for La Lettura—Correier della Serra that shows the fifty
most violent cities in the world. See the next page for a closer look at the legend.
A zoom-in of the graphic from Frederica Fragapane. Notice all of the details and data ele-
ments included in each icon. It could be a bar chart or line chart, but would you then be
inclined to zoom in and read it closely?
VISUAL PROCESSING AND PERCEPTUAL RANKINGSփ 17
chart or a line chart or some other chart type. But if it were, would you be inclined to zoom
in, read it closely, and examine it?
Data visualization is a mix of science and art. Sometimes we want to be closer to the sci-
ence side of the spectrum—in other words, use visualizations that allow readers to more
accurately perceive the absolute values of data and make comparisons. Other times we may
want to be closer to the art side of the spectrum and create visuals that engage and excite the
reader, even if they do not permit the most accurate comparisons.
Sometimes you must make your visuals interesting and engaging, even at the cost of
absolute perceptual accuracy. Readers may not be as interested in the topic as we hope or
may not have enough expertise to immediately grasp the content. As content creators, how-
ever, our job is to encourage people to read and use the graph, even if we “violate” perceptual
rules that we know will hamper someone’s ability to make the most accurate conclusions.
Thinking about different audience types is not just about considering among decision mak-
ers, scholars, policymakers, and the general public—it also means thinking about different
levels of interest or engagement with the visual itself. As historian Cecelia Watson writes in
her book about the history and use of the semicolon, “What if we thought less about rules
and more about communication, and considered it our obligation to one another to try to
figure out what is really being communicated?”
We should not operate from the assumption that readers will pay attention to everything
in our visual, even if we use a common, familiar chart type. Let’s be honest: People see bar
charts and line charts and pie charts all the time, and those charts are often boring. Boring
graphs are forgettable. Different shapes and uncommon forms that move beyond the bor-
ders of our typical data visualization experience can draw readers in. Reading a graph is not
like the spontaneous comprehension of seeing a photograph. Instead, reading a graph has
more of the complex cognitive processes as reading a paragraph.
This isn’t to say we should not concern ourselves with visual perception or allowing our
readers to make the most accurate comparisons, but the goal of engagement can be worth a
lot in its own right. Elijah Meeks, a data visualization engineer, wrote that, “Charts, like any
other communication, need to be compelling to be convincing, and if your bar chart, as opti-
mal as it may be, has been reduced to background noise by the constant hum of bar charts
crossing a stakeholder’s screen, then it’s your responsibility to make it more compelling, even
if it’s not any more precise or accurate than a more simple form.”
Introducing a new or different graph type can also introduce a hurdle to your reader.
These can be big hurdles, like a completely new graph type or an exceptionally unusual
18 փ PRINCIPLES OF DATA VISUA LIZATION
representation of the data. Or they can be small hurdles, graphs that rank lower on the
perceptual-accuracy scale or graphs that people may have only seen a few times before.
To overcome these hurdles, you may need to explain how to read the graph. But that
might be worth it because sometimes different charts attract reader’s attention and pique
their curiosity.
When should you use a nonstandard graph? Likely not for many scholarly purposes,
because they do not enable the most accurate perceptions of the data. For scholarly writing,
accuracy is paramount. We want our reader to clearly and efficiently compare the values we’re
presenting. But in other cases—headline-style or standalone graphics, blog posts, shorter
briefs or reports, or graphs for social media—creating something different may draw people
in and hold their attention just long enough to convey your argument, data, or content.
VISUAL PROCESSING AND PERCEPTUAL RANKINGSփ 19
This visualization from artist and journalist Jaime Serra Palou is a lovely example of
this kind of nonstandard and creative data visualization. He plots his coffee consump-
tion every day over the course of a year by using the stains from his coffee cups. You can
immediately see those parts of the year when he needed an extra burst of caffeine. Yes, a
line chart might convey the same data, but would you pause to spend an extra moment
reading it?
Sometimes you can do both—a nonstandard, attention-grabbing graphic accompanied
by a more familiar graph next to it. What you present and how you present it depends
on your audience. The Serra piece might work as the lead graphic on a book or report
about coffee consumption, but more detailed charts inside might take the form of standard
charts and tables. Some academic research has shown that creating novel graphs, such as
Artist and journalist Jaime Serra Palou plotted his coffee consumption every day for a year
by using stains from his coffee cup.
20փ PRINCIPLES OF DATA VISUA LIZATION
those that enable the user to personalize the content (by inputting their own information) or
are simply more aesthetically appealing, encourages readers to actively process the content.
ANSCOMBE’S QUARTET
The value of visualizing data is best illustrated by Anscombe’s Quartet, published in 1973 by
statistician Francis Anscombe. The Quartet demonstrates the power of graphs and how they,
together with statistical calculations, can better communicate our data.
Examine the table below, which shows four pairs of data, an X and a Y.
We can make some basic observations about these data. We can see that the first three
series of X’s are all the same; the values of X’s in the last series are all 8 except for the one 19;
and the X’s are all whole numbers while the Y’s are not. We might even notice that the 12.7
value in the third column of Y is larger than the rest. In my experience, most people don’t
comment about the relationship between the different series, which, at the end of the day, is
what we want to understand. It turns out that each of the four pairs yield the same standard
information: the same average values of the X series and the Y series; the same variance for
each; the same correlation between X and Y; and the same estimated regression equation.
Data set 1 1 2 2 3 3 4 4
Variable x y x y x y x y
Known as Anscombe’s Quartet, this example demonstrates how difficult it is for us to pull
out basic patterns and summary statistics.
VISUAL PROCESSING AND PERCEPTUAL RANKINGSփ 21
When we see the same data presented in four graphs, however, we can immediately see
these relationships, for example, the positive correlation in all four pairs, the curvature in the
second pair that you couldn’t see in the table, and the outliers 12.7 and 19.0.
We are much more likely to remember these four small graphs than we are the origi-
nal table. In his bestselling book, Brain Rules, molecular biologist John Medina writes,
“The more visual the input becomes, the more likely it is to be recognized and recalled.”
The more we can make our data and content visual, the more we can expect our readers to
remember it and, hopefully, use it.
15 15
10 10
5 5
0 0
0 5 10 15 20 0 5 10 15 20
15 15
10 10
5 5
0 0
0 5 10 15 20 0 5 10 15 20
The data visualization representation of Anscombe’s Quartet. Notice how much easier
it is to see the positive relationship between the two variables, the curvature in the
pattern in the top-right graph, and the outliers in the bottom two graphs.
Source: Francis Anscombe (1973).
22 PRINCIPLES OF DATA VISUA LIZATION
How do we perceive information? And how, as chart creators, can we use these perceptual
rules to more effectively communicate our data? “Gestalt theory” is one such way we can
think about how our readers will look at our graphs. Gestalt theory was developed in the
early part of the twentieth century by German psychologists and refers to how we tend to
organize visual elements into groups. Further developments in the field were interrupted by
the rise of the Nazi regime in Germany and then by World War II, and after the war it was
criticized for not having rigorous methodological methods. But the ideas persist in many
disciplines, including information theory, vision science, and cognitive neuroscience.
These six organizational principles from Gestalt theory are especially useful for creating
graphs and visuals that tap into our reader’s visual processing network.
PROXIMITY
We perceive objects that are close to one another as belonging to a group. There are lots of
graphical elements that we can group together: labels with points, bars with each other, or,
like this graph, clusters of points in a scatterplot in which we can see two groups or clusters,
one in the top-right and the other closer to the bottom-left.
VISUAL PROCESSING AND PERCEPTUAL RANKINGS 23
SIMILARITY
Our brains group objects that share the same color, shape, or direction. Adding color to
the above scatterplot reinforces the two groups.
ENCLOSURE
Bounded objects are perceived as a group. Here, in addition to using color, we can enclose
the two groups with circles or other shapes.
24փ PRINCIPLES OF DATA VISUA LIZATION
CLOSURE
Our brains tend to ignore gaps and complete structures with open areas. In its basic form,
we don’t have a problem viewing a simple graph that has a horizontal axis and a vertical axis as
a single object because the two lines are enough for us to define the closed space. In a line chart
with missing data, for example, we tend to mentally close the gap in the most direct way pos-
sible, even if there might be something different going on in that missing area. For example, in
the line graph on the left, we mentally close the gap between the two segments with a straight
line even though the missing data might yield a pattern that moves up and then down.
CONTINUITY
Here, objects that are aligned together or continue one another are perceived as a group.
Hence, our eyes seek a smooth path when following a sequence of shapes. You don’t need
the horizontal axis line in this bar chart, for example, because the bars are aligned along a
consistent path between the labels and the bottoms of the bars.
A B C D E
VISUAL PROCESSING AND PERCEPTUAL RANKINGSփ 25
CONNECTION
According to this principle, we perceive connected objects as members of the same group.
Take this series of dots: At first, we perceive it as a single series, a mass of blue dots. Adding
color makes it clear there are two different series. Connecting the dots makes it clear how the
two initially track each other but then diverge.
PREATTENTIVE PROCESSING
The concept of “preattentive processing” is a subset of Gestalt theory, and it is the visual pro-
cess I consider most when creating my data visualizations. As we just saw, because our eyes
can detect a limited set of visual characteristics, we combine various features of an object
and unconsciously perceive them as a single image. In other words, preattentive attributes
draw our attention to a specific part of an image or, in our case, a graph.
For example, try to find the four largest numbers in this table.
Hard to do, right? Now try it with these versions that use color (on the left) and intensity
(on the right) to highlight those four numbers.
Table 1. Our sales grew to $600 million this year Table 1. Our sales grew to $600 million this year
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
Bob 26 35 72 84 Bob 26 35 72 84
Ellie 22 15 61 35 Ellie 22 15 61 35
Gerrie 19 20 71 55 Gerrie 19 20 71 55
Jack 22 95 13 64 Jack 22 95 13 64
Jon 83 62 46 48 Jon 83 62 46 48
Karen 30 65 98 82 Karen 30 65 98 82
Ken 38 28 45 71 Ken 38 28 45 71
Lauren 98 81 41 63 Lauren 98 81 41 63
Steve 16 50 23 41 Steve 16 50 23 41
Valerie 46 24 30 57 Valerie 46 24 30 57
Total $400 $475 $500 $600 Total $400 $475 $500 $600
Preattentive attributes here direct our attention to the large numbers immediately.
It’s easier to find the numbers in these two tables than the first because the numbers are
encoded using preattentive attributes: color and weight. Each distinction helps us effortlessly
identify the key number.
Examples of preattentive attributes that we can use in our visualizations to direct our
reader’s attention.
VISUAL PROCESSING AND PERCEPTUAL RANKINGSփ 27
Preattentive attributes are effects that seem to pop out from their surroundings. There are
many we can use to tap into our reader’s visual processing network to draw their attention:
shape, line width, color, position, length, and more.
Preattentive processing works in photographs too. Consider these images of fruits and
vegetables. In the photo on the left, the eye is drawn to the upper-right corner. The group of
tomatoes is slightly larger than the rest and positioned away from the group. In the photo-
graph on the right, however, the eye is not drawn to any specific position. This photograph
is more evenly balanced, so no one object stands apart from the rest.
Notice how your eye gravitates toward the four tomatoes in the top-right part
of the image on the left. The image on the right is balanced, so your eye doesn’t
immediately focus on any particular area. Photos by NordWood Themes (left) and Tim
Gouw (right) on Unsplash.
We can apply these attributes to data visualization. A line chart uses the position
of the points to indicate the data, while a bar chart uses length. You can use preatten-
tive attributes to draw your audience’s attention to aspects of your graphs, guiding
their focus.
For example, on the next page, I can enclose the ‘Forecast’ area of the line chart on the left
with the gray box—notice how it immediately draws your eye to the right side of the graph.
Similarly, I can use the color attribute to highlight a few points in the scatterplot on the right
(and keep the other dots gray).
28 PRINCIPLES OF DATA VISUA LIZATION
3.0
2.5
2.0
1.5
1.0
0.5
Actual Forecast
0.0
2010 2012 2014 2016 2018 2020 2022 2024
^ŽƵƌĐĞ͗ŽŶŐƌĞƐƐŝŽŶĂůƵĚŐĞƚKĸĐĞ
Applying simple preattentive attributes to these graphs directs your eye to the “Forecast”
area of the graph on the left and to the two highlighted countries in the graph on the right.
WRAPPING UP
With these basic rules of perception, we are now better equipped to recognize and interpret
the visual features we can use to encode and highlight our data. Before we start adding more
graphs to our data visualization toolbox, let’s lay out some basic guidelines of more effec-
tive data visualizations—things you should keep in mind no matter what kind of graph you
are creating.
FIVE GUIDELINES FOR BETTER
DATA VISUALIZATIONS
2
W
henever I create a data visualization, whether it’s static, interactive, or part of a
report or blog post or even a tweet, I follow five primary guidelines.
Showing the data and reducing the clutter means reducing extraneous gridlines, markers,
and shades that obscure the actual data. Active titles, better labels, and helpful annotations
will integrate your chart with the text around it. When charts are dense with many data
series, you can use color strategically to highlight series of interest or break one dense chart
into multiple smaller versions.
Taken together, these five guidelines remind me of the needs of my audience and how my
visuals can tell them a story.
Your reader can only grasp your point, argument, or story if they see the data. This doesn’t
mean that all the data must be shown, but it does mean that you should highlight the values
30 փ PRINCIPLES OF DATA VISUA LIZATION
that are important to your argument. As chart creators, our challenge is deciding how much
data to show and the best way to show it.
Consider this dot density map of the United States (see page 244 for more on this kind
of map). It uses data from the 2010 U.S. decennial census and places a dot for each of the
country’s 308 million residents in their census blocks (a census block roughly corresponds
to a city block). Notice how there is nothing in the image except for the data. There are no
state borders, roads, city markers, or labels for lakes and rivers. We still recognize it as the
United States because people tend to live along borders and coasts, which helps give shape
to the country.
This doesn’t mean we must show all of the data all the time. Sometimes charts show too
much data, making it hard to see which data points matter most. On the next page are
two line charts that both show the average number of years of schooling for fifty countries
around the world. In the graph on the left, each country is assigned its own color. This makes
The Gestalt principle of similarity helps us see the clusters of people around the country.
Source: Image Copyright, 2013, Weldon Cooper Center for Public Service, Rector and Visitors of the
University of Virginia (Dustin A. Cable, creator).
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 31
Average years of schooling has increased around the world Average years of schooling has increased around the world
(Number of years) (Number of years)
16 16
14 14 Germany
United States
12 12
10 10 Spain
Mexico
8 8 China
6 6
Nepal
4 4
2 2
0 0
1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017
Source: Our World in Data Source: Our World in Data
Highlighting just a few countries in the chart on the right makes it easier to read.
it busy and confusing, impossible to pick out a trend for any single country. In the graph on
the right, just six countries of interest are highlighted while the remaining are set in gray,
blending them into a neutral background. This gives the reader a clear view of the countries
we want to highlight. It’s not about showing the least amount of data, it’s about showing the
data that matter most.
The use of unnecessary visual elements distracts your reader from the central data and clut-
ters the page. There are lots of different types of chart clutter we might want to avoid. There
are basic elements like heavy tick marks and gridlines, which we should remove in almost
every case. Some graphs use data markers like squares, circles, and triangles to distinguish
between series, but when the markers overlap they jumble the patterns. Some use textured
or filled gradients when simple, solid shades of color work just as well. Some use unnec-
essary dimensions that distort the data. And others contain too much text and too many
labels, cluttering the space and crowding out the data.
Take this three-dimensional column chart of average schooling for the United States and
Germany for a few select years.
32 PRINCIPLES OF DATA VISUA LIZATION
Average years of schooling has grown faster in Germany than in the United States
(Number of years)
Germany United States
1000.00%
500.00%
0.00%
1997
2007
2017
You’ve seen these kinds of 3D charts before—they are distracting, hard to read,
and distort the data.
If you think that this looks so outlandish that no one would ever style a chart this way,
you’d be wrong. I’ve copied the exact style from another chart, even down to the gradient styl-
ing. The three-dimensional bars and shimmering stripes, mismatched data and axis labels,
the abundance of decimals that suggest a level of data precision that’s not actually there—
all these combine to create a graph that is difficult to read and, quite honestly, uncomfortable
to look at. Also notice how the three-dimensional view distorts the data. The first bar never
touches the gridline even though it should match it exactly. This distortion occurs because
the unnecessary third dimension requires adding perspective to the graph. Simplifying the
graph by discarding these extraneous, distracting elements and showing the data makes
your argument clear and comprehensible.
While much of our understanding of perception and how our eyes and brains work is
rooted in scientific research, our decisions of which graph to use, where we place labels and
annotation, which colors and fonts to use, and how we lay out our visualizations is mostly
subjective. There are cases where certain graphs are wrong, but many other cases call for
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 33
Average years of schooling has grown faster in Germany than in the United States
(Number of years)
Germany United States
14.1
13.7 13.4
12.7 12.9
10.0
A basic bar chart eliminates the clutter and the distortion caused by the
3D effect, so the graph is easier to read and understand.
Data Source: Our World in Data.
nothing more than your best judgment. As you create more visualizations and read more
graphs, you’ll develop your own eye and aesthetic—and your own balance of art and science.
Let’s start with the easiest type of annotation: Removing legends and directly labeling your
data. Many software tool defaults create a data legend and place it around the chart, discon-
nected from the data. This forces more work upon your reader to connect each line or bar to
its label. A better approach is to directly label your data series.
Take the line chart of average schooling for fifty countries from earlier. Rather than the
default approach of putting a legend somewhere around the chart, as in the graph on the left,
in the version on the right, I directly label the lines at the right end of the graph.
Average years of schooling has increased around the world Average years of schooling has increased around the world
(Number of years) (Number of years)
16
Germany United States Spain Mexico China Nepal
16 14 Germany
United States
14
12
12
10 Spain
10 Mexico
8 China
8
6
6
Nepal
4
4
2 2
0 0
1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017
Help your reader more easily find the labels for the data values by placing the labels
directly on the chart.
In graphs that have fewer lines, we might also be able to place the labels directly on
the graph. In these cases, I try to align the labels instead of placing them in random
positions. Notice how in the graph on the left of the next page your eye needs to jump
around to find each label. And because we might start reading the graph with the title,
the proximity of the label for the United States could give that series greater emphasis.
In the version on the right, the labels are aligned along a single vertical line, making it
easier to read the entire visual.
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 35
Average years of schooling has increased around the world Average years of schooling has increased around the world
(Number of years)
(Number of years)
16 16
Germany
14 14
United States
United States
12 12
Germany
Spain
10 10
Spain Mexico
8 8
Mexico China
China
6 6
Nepal
4 4
Nepal
2 2
0 0
1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017
Align the labels and match the colors with the data as in the graph on the right rather than
placing them in random positions.
We can take a similar approach to labeling this bar chart of average schooling in Germany
and the United States. With just two countries, instead of a legend disconnected from the
data, what if we added the labels inside the bars or used color in the title of the graph itself
to link the title to the content of the graph?
By integrating the text and the data, we’re doing a better job of considering the reader’s
needs. Do they need to see every line equally, or will including all the lines clutter the graph?
Is it important to label every point in the scatterplot, or will highlighting just a few points
suffice? How can we integrate labels and chart elements to help the reader understand the
content quickly and easily?
13.7 14.1
13.4
12.7 12.9
10.0
These are just two examples of how to integrate labels into the graph.
Data Source: Our World in Data.
36 փ PRINCIPLES OF DATA VISUA LIZATION
Removing the legend isn’t always possible. A bar chart with several categories or a map
with different colors requires a legend, because directly labeling the chart will add too much
clutter to the visual. In these cases, at least keep the order of the legend consistent with the
order of the data. Notice the inconsistency between the order of the lines and the legend in
this line chart from the Social Security Advisory Board. Not only do we need to jump back
and forth between the lines and labels, there is an extra task of figuring out the order of
the two. A redesigned version removes much of those unnecessary data markers and extra
gridlines, and integrates the legend onto the chart by adding labels directly next to the lines.
We won’t be able to remove the legend on every single graph we create, but we should
strive to link the data and the labels as best we can, and that starts with labeling the data
series on our charts.
DI and SSI allowance rates have generally moved in tandem over the past 25 years
(Percent)
80%
70%
60%
50%
40% mbঞ-Ѵ;;Ѵ
30% !;1omvb7;u-ঞom;;Ѵ
;-ubm];;ѴŐ bvrovbঞomvő
20%
;-ubm];;ѴŐ ;1bvbomvő
10%
0%
86
88
90
92
94
96
98
00
02
04
06
08
10
19
19
19
19
19
19
19
20
20
20
20
20
20
Notice the inconsistency between the order of the lines and the legend in the chart on the
left. The redesigned version removes unnecessary clutter and directly labels the lines.
Source: Social Security Advisory Board, February, 2012.
Most titles are neutral descriptions of the data, as in “Figure 1. Labor Force Participation
Rate, Men and Women, 1950–2016.” But better titles capture the takeaway of the chart,
telling the reader what conclusions can be drawn from the data. I call these “active titles”
or “headline titles.” In my book on presentations, I follow the advice of author Carmine
Gallo and urge presenters to use “Twitter-like headlines” in their slides. These are concise,
active phrases that make it easy to understand what the slide—or chart—is aiming toward.
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 37
Too often, we attach a title to the chart that describes the data instead of the point or argu-
ment we want to make.
While “Labor Force Participation, Men and Women, 1950–2016” is certainly a correct
and accurate description of the data in this graph from the Pew Research Center, it does
not describe what the reader should learn about the labor participation rate among men and
women between 1950 and 2016. The more active title that Pew uses instead—“Labor force
participation rate has risen for women, fallen for men”—tells the reader exactly what they
should take away from the graph.
Do people even read titles? A 2015 study from researchers at Harvard University showed
that they do: “Titles and text attract people’s attention, are dwelled upon during encoding,
The active title in this chart from the Pew Research Center tells you exactly what you are
supposed to learn from it.
38փ PRINCIPLES OF DATA VISUA LIZATION
and . . . contribute to recognition and recall.” If it’s indeed the case that people read titles
(and text more generally), then we should treat a chart’s title as carefully as the chart itself.
But can so-called “active” titles make us seem biased or partisan? If we use active titles
only to faithfully represent the results and showcase the message of the graph, then no. I’ve
worked with many people who have debated my inclination for active titles by arguing that
such titles will make their work appear biased. In most of those cases, I can look at the text
around their chart and see a single argument for what’s being shown in the graph and how
to interpret the data. Their argument is right next to the graphic, but, like the legend we saw
earlier, it’s disconnected from it.
Active titles don’t make us biased, but descriptive titles do waste an opportunity to make
a clear, compelling case. Of course, short, active titles aren’t always possible—you may be
making more than one point or your sole goal is to simply describe the data. Generally
speaking, however, integrating your graphs as part of your argument creates a more cohesive
approach to making your argument and telling your story.
In the case above, Pew doesn’t leave it to the reader to search for a point in the graph,
but neither are they biasing the results by adding commentary in the title. They are simply
foregrounding the takeaway of the visual.
If you are having trouble coming up with a concise, active title, that may be a sign that
your chart doesn’t have a concise takeaway or—and maybe this is more common—you
haven’t thought through what you want the graph to communicate.
3. ADD EXPLAINERS
Once the chart is made and the title is settled, ask yourself, Would this chart benefit from
more text?
Sometimes data sets have peaks or valleys, outliers or variations that bear explanation.
Adding detail in graphs can push your argument, highlight points of interest, or (in cases of
nonstandard graphs) even explain how to read it.
Take this line chart of the popularity of the name “Neil” in the United States created by,
yes, Neil Richards, a data visualization consultant in the United Kingdom. Anyone could
make the simple line chart on the left—it’s only one data series—but with only a quick glance
the reader might immediately ask some obvious questions: Why did the decline stop in the
late-1960s? Why did the line spike upwards a few years later in the 1970s? And what halted
the decline in the early 2000s?
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 39
Short explainers in this graph on the right from Neil Richards explain some of the basic
features of the data.
Now look at this second version of the chart with short explainers. The late-1960s spike
might be attributed to Neil Armstrong landing on the moon, followed by the popularity of
musicians like Neil Young, Neil Sedaka, and Neil Diamond in the 1970s. The flattening of
the trend in the mid-2000s could be attributed to “modern Neils” like Neil DeGrasse Tyson.
These annotations are not complicated and don’t require complex programming or design
techniques—they are often just interesting points in the data thoughtfully added with short
sections of text.
Annotation allows readers—especially those who may have less experience with data
visualization —to grasp and understand the content quickly. The bubble chart from the Los
Angeles Times on the next page is a great illustration of how to do so. The change in the
violent crime rate is plotted along the horizontal axis and the change in the property crime
rate is plotted along the vertical axis for about thirty-five cities in California. The average LA
Times reader is probably not a bubble-plot expert, so the authors have added annotations to
help readers navigate the format of the graph and its content.
Notice the use of color and annotation to help the reader understand this graph. The
top-right quadrant is shaded red with the word “Worse” in large, red letters. The bottom-
left quadrant is shaded blue with the word “Better” in large, blue letters. Immediately, you
40 փ PRINCIPLES OF DATA VISUA LIZATION
understand that the cities in the top-right have worsened and the cities in the bottom-left
have improved. Short, bold headlines (“Reduced violent and property crime rate” in the
bottom-left quadrant) explain the substance of the changes. Then, a short sentence high-
lights a city or two and what has transpired over the past year. This graph does an expert job
of explaining how to read it and how to understand the content in it.
This graph from the Los Angeles Times is one of my favorite examples of how to use anno-
tation to explain how to read a graph and its content.
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 41
It’s obvious when a certain graph contains too much information—line charts that look like
spaghetti, maps with dozens of colors and icons, or bar after bar after bar littering a chart.
Sometimes we face the challenge of including lots of data in a single graph but we don’t need
to try to pack everything into a single graph.
Two examples of the small multiples approach. The graph on the left, from Zeit Online,
shows the average temperature in Germany over the past 140 years. The graph on the
right, from the Centers of Disease Control and Prevention, shows how facial hair can affect
the fit of respirators. The Gestalt principle of connection helps us track the changes from
one image to the next in both graphics.
42փ PRINCIPLES OF DATA VISUA LIZATION
One way to address the packed, single chart is to break it into smaller parts. Known as
grid charts or panel charts (also called facets, trellis charts, or, most commonly, small mul-
tiples) these are smaller charts that use the same scale, axes, and scope but spread the data
across multiple visuals. In other words, instead of putting all of the data on one graph, create
multiple, smaller versions with variations on the basic data.
The small multiples approach isn’t a new or revolutionary approach to communicating
data. In 1878, photographer Eadweard Muybridge (see page 44) was tasked with deter-
mining whether a horse becomes fully airborne when it gallops. Muybridge developed a
technique to take a sequence of fast-action photographs (what we now call stop-motion)
of a horse at gallop. His photos proved that horses do indeed leave the ground entirely.
The sequence of images, which also gives a sense of motion and animation, is an early
example of small multiples.
The small multiples approach has at least three advantages. First, once the reader
understands how to read one chart, they know how to read all the charts. Second,
you can display lots of information without confusing your reader. Third, small mul-
tiples let readers make comparisons across multiple variables. This example from
the Guardian shows voting results in 2016 for the Brexit resolution in the United
Kingdom across six different demographic variables. The horizontal axes stay
unchanged and it’s easy to see the direction of the relationships for each demographic
measure.
But there are pitfalls to the small multiples approach that will muddy the visual if not
avoided. First, the charts should be arranged in a logical order. Don’t make your reader
navigate around the page—use an intuitive arrangement based on something like geography
or alphabetical order.
Second, the graphs should share the same layout, size, font, and color. Remember, we
are breaking up one chart into many, so it should look like one chart replicated multiple
times. The vertical and horizontal axes may change, but you wouldn’t want to have one
chart where blue dots represent “no” and another where they represent “yes.” Third, small
multiples should be relatively easy to read. You are not necessarily asking your reader to
zoom in and uncover all the specific details in all of the graphs—the purpose is to give
them a view of the overall patterns. The graphs are intended to be small, so including
annotations and labels or repeating long axis labels and data markers features can over-
whelm the reader.
Small multiple scatterplots from the Guardian shows the relationship between voting
choice and six demographic variables. Notice how the Gestalt principle of similarity lets us
easily see the two clusters of circles within each scatterplot.
44փ PRINCIPLES OF DATA VISUA LIZATION
I end this section with a practical technique that I think can be an easy step to creating clear,
comprehensible visualizations: Start with gray. Whenever you make a graph, start with all-
gray data elements. By doing so, you force yourself to be purposeful and strategic in your use
of color, labels, and other elements.
Consider a simpler version of the average schooling chart from earlier, this time with
only ten countries as shown on the next page. With color and labels (top-left), I could
put this graph in my report or handout, and with a little work (and a more active title),
my reader could figure out which labels correspond to which lines. But if I make all the
lines gray (top-right), the reader can’t accomplish that same task because it’s impossible
to figure out which country is represented by which line.
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 45
14
United States
12
Germany
China
10
India
8 Indonesia
Brazil
6 Pakistan
Philippines
4
Russia
Japan
2
0
1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017
Now I can be purposeful about what I want to do with this graph. I could add color and
even vary the thickness of the lines to better highlight only, say, the two countries I want to
emphasize. (Leaving all the labels in the version on the bottom-left is less useful than label-
ing the lines directly as in the version on the right.) Starting with gray forces us to deliber-
ately choose what elements to put into the foreground.
DATA TYPES
The bedrock of any data visualizaধon is the data. Without data and a good understand-
ing of what our data is, how it was collected, and what it tells us, we are just painধng
pictures. This book is not the venue for a thorough review of data types and staধsধcal
methods, but a short primer can help us organize our data types just as we organize our
graph types.
There are two major groupings of data types: 7uanধtaধve and 7ualitaধve. Quanধ-
taধve data can be measured with numbers, for example, distance, dollars, speed, and
ধme. Qualitaধve data is non-numerical informaধon, usually descripধve text like “yes
or no,” “saধs)ed or dissaধs)ed,” or longer 7uotes or passages from interviews and
surveys.
We can further break down each major data type into subcategories. On the
7ualitaধve side, we have nominal and ordinal scales. Nominal scales are used to label
variables and don’t have an order or 7uanধtaধve value. In a data set of animal types,
the order of lion, ধger, and bear has no meaning (aside from the song, of course).
In ordinal scales, order does maħer, but the exact size in comparison between val-
ues is unknown. Consider a survey that asks people to select between 1. Strongly
Agree; 2. Agree; 3. Disagree; and 4. Strongly Disagree. These choices can be ordered,
but the diøerence between 1 and 2 is not necessarily the same as the diøerence
between 3 and 4.
On the 7uanধtaধve side, data can be either discrete or conࢼnuous. Discrete data are
whole numbers (integers) that cannot be subdivided. Despite naধonal averages, no one
has exactly 2.3 children. Conধnuous data are numbers that can be broken down into
smaller units, like weight, height, and temperature.
Conধnuous data can be further broken down into two major scales: interval and
raࢼo. The diøerence is what we can and cannot calculate. With interval scales, we
know both the order and the exact diøerences, but they do not have a true zero value.
This means we can add and subtract data measured in interval scales, but we can’t
mulধply or divide. A classic example is temperature in degrees Farenheit: The diøerence
between 10 and 20 degrees is the same as between 70 and 80 degrees, but we can’t
say that 20 degrees is twice as hot as 10 degrees, because 0 degrees is an actual value,
not absolute zero.
!aࢼo scales have all of the characterisধcs of all the other scales plus they have
an absolute zero, which means we can do all of our mathemaধcal calculaধons.
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 47
Weight is a good example of a raধo scale—a person who weighs 200 pounds is
twice as heavy as someone who weighs 100 pounds, and 0 pounds is the absence
of weight.
These guidelines lay out the basic approaches to effectively visualizing our data. While this
is not a book about data analysis—how and where to get data, how to analyze underly-
ing statistical properties, and develop statistical models—whenever we work with data it
is important to recognize that visual content can have a large influence on how people use
data and make decisions. As data communicators, it is therefore our responsibility to treat
our work and our data as carefully and objectively as possible. It is also our responsibility to
recognize where our data may suffer from underlying bias or error, or even implicit bias that
data creators may themselves not even be aware of.
48փ PRINCIPLES OF DATA VISUA LIZATION
There are many ways in which the data we use may be biased or not representative. In
their book, Data Feminism, Catherine D’Ignazio and Lauren Klein describe how standard
practices in data science reinforce existing power inequalities. They explore how data has
been used for both good and evil—to expose injustice and improve health and policy out-
comes, for example, but also to surveil and discriminate. By asking who is producing the
data and for whom it is being produced, we can be better stewards of our own data and our
own visualizations.
Many fields are squarely built on a model of the world in which men are the only—or
maybe just the most important—participants. In Invisible Women, author Caroline Criado
Perez reveals the hidden places where inequality in even basic data resides. There are
straightforward examples, like how the average smartphone is 5.5 inches long—too big for
most women’s hands and pants pockets. Or how the average temperature in many office
buildings is five degrees too cold for women because the formula to determine the ideal
temperature was developed in the 1960s based on the metabolic resting rate of a forty-
year-old, 150-pound man. There are more insipient examples as well, like how women in
Britain are 50 percent more likely to be misdiagnosed following a heart attack, or how car
crash test dummies are based on the male body, so even though men are more likely to
get into car accidents, women involved in collisions are almost 50 percent more likely to
be seriously injured.
In a similar vein, the era of big data, machine learning, and artificial intelligence use
more and more unseen algorithms and statistical techniques. We often know little about the
data that feed these algorithms and how the models themselves may perpetuate inequality.
Mathematician Cathy O’Neil explores this in her book Weapons of Math Destruction, from
teacher quality, creditworthiness, and recidivism risk, algorithms can develop and reinforce
discriminatory models of public policy.
When it comes to data visualization specifically, we must be mindful of the under-
lying biases and inequality in how we present our results. As just one example of how
data and visualizations have been used to discriminate, consider this map of Rich-
mond, Virginia, produced in 1937 by the Home Owners’ Loan Corporation (HOLC),
a federal agency tasked to appraise home values and neighborhoods across the United
States. As Richard Rothstein writes in his book, The Color of Law, “The HOLC created
color-coded maps of every metropolitan area in the nation, with the safest neighbor-
hoods colored green and the riskiest colored red. A neighborhood earned a red color
if African Americans lived in it, even if it was a solid middle-class neighborhood of
FIVE GUIDELINES FOR BETTER DATA VISUA LIZATIONSփ 49
This redlining map of Richmond, Virginia, demonstrates how data and data visualizaধon can be
wielded to further systemaধc discriminaধon. Informaধon Studies scholar Sa)ya Umoja Nobel
argues that modern internet search engines and other algorithms are enacধng new ways of dis-
criminaধon and racial pro)ling, creaধng a modern form of “technological redlining.”
Source: Naধonal Archives.
single-family homes.” Systematic discrimination is and can be generated by how we use and
misuse our data.
Finally, in addition to cultural differences that might arise from, say, using certain colors
in different cultures, we should also be mindful of the language, shapes, and images in our
visuals. Are we using language and images that are inclusive? When do we need to provide
historical and social context for problems people are facing? As with developments in acces-
sibility, diversity, and inclusion (see Chapter 12), these are all challenges with which the data
visualization field is always wrestling.
DATA VISUALIZATION LESSONS FROM
THE CO RO NAVIRUS PA NDEMIC
The manuscript for this book was delivered to the publisher in March 2020, just as
the coronavirus (COVID-19) pandemic was traveling around the world. As it spread, it
induced massive poliধcal, economic, and societal changes, and it brought new terms into
our lexicon, like “*aħening the curve.” In late February 2020 The Economist published a
version of the graphic by data journalist Rosamund Pearce, based on original work by
the Centers for Disease Control and Prevenধon. Graphics like this built awareness and
facilitated acধon, most notably on the relaধvely new concept of “social distancing”.
But for every graph that informed and educated, there were many others that mis-
represented data or spread misinformaধon. One pie chart, for example, inexplicably
summed contagion rates among eleven separate diseases to 100 percent and added
a separate note that the “numbers for COVID-19 remain a rough esধmate because
the long incubaধon period means we sধll have no idea how many people have been
infected.” This is not responsible data visualizaধon.
The unprecedented spread of the coronavirus gave us an opportunity to use
real-ধme data that could be used to beħer understand the virus and its spread.
But one reason why many graphs and charts around COVID-19 are problemaধc is
because too many of us assume we have ade7uate knowledge in a parধcular subject
area. Public health professionals, epidemiologists, and physicians have the training,
insight, and experience with the health care system and modeling disease transmis-
sion to provide useful data and informaধon. For the rest of us, without experধse in
these areas, our visualizaধon work—even as well-intenধoned as it might be—can
make things worse.
We ođen create—or are asked to create—visualizaধons in subject areas in which
we are not experts. Someধmes this is an opportunity to explore diøerent visualizaধon
forms and funcধons and try new tools. Other ধmes, though, we may be out of our
depth. We may not fully understand our data. Even if we have read the data dicধonary
or considered the data collecধon methods, we may not know enough about how the
data were modeled or simulated or the reliability of their collecধon methods.
Under ordinary circumstances, visualizaধon exercises might consider issues such
as unemployment rates or housing opধons or the distribuধon of wealth and not life-
threatening events like a viral pandemic. In these cases, we must be especially aware
of how our work might be misunderstood and how it may change the thinking or
behavior of our readers.
The converse of the above is also true. An epidemiologist may know a lot about
modeling disease spread, but he or she may not understand how best to visualize that
modeling, explain jargon, and annotate important data points. Here, it is incumbent
upon the scienধst to reach out to data visualizaধon experts and graphic designers to
ensure their visualizaধon work is accessible to readers.
There is a beħer way forward. Instead of thinking our limited knowledge is suă-
cient to weigh in on every topic and every dataset, we should strive to collaborate. In
the case of COVID-19, not knowing enough may lead to deadly outcomes. If we think
of ourselves as journalists and seek out domain speci)c experts, we can work to build
teams, groups, and organizaধons that can deliver beħer data, beħer visualizaধons,
and beħer decisions.
52 փ PRINCIPLES OF DATA VISUA LIZATION
NEXT STEPS
Now armed with basic guidelines and rules of perception, you are almost ready to start
adding more graphs to your data visualization toolbox. But there’s one more thing you
should consider before you start encoding your data with bars, lines, and dots: the purpose
of your graph.
In what format do you need to present your data to your reader or user? Do they need a
static graph where you present your argument or will an interactive visualization help them
explore the data and come to more and deeper conclusions? In the next chapter, we discuss
the different forms and functions for visualizations and then turn to the many ways we can
visualize our data.
FORM AND FUNCTION
LET YOUR AUDIENCE’S NEEDS DRIVE YOUR
DATA VISUALIZATION CHOICES
3
T
he primary focus of this book is on static data visualizations. That is, those visualiza-
tions that are not interactive and do not move. But interactive and animated visu-
alizations are becoming more common even in academic fields, and the tools to make
interactive visuals are, year by year, becoming easier to use, cheaper (even free), and more
powerful.
Aside from the standard design decisions about color, layout, and font that apply to static
visualizations, there is another set for interactive visualizations. Where will the button go?
What will it look like? How will the user be guided to different parts of the visualization?
Will the user need to scroll? If so, in which direction? How might the interactivity change
between a desktop computer and a tablet or phone?
Discussing all of these options for storytelling, interactivity, and animation is beyond the
scope of this book, but it is worth pausing to consider which graph—static or interactive—
best meets the needs of our audience. Some audiences want a long, in-depth report. Others
want a short brief or a blog post. And some just want to get their hands on the data to do
their own work.
The broad schematic of different data visualization types shown on the next page can help
you think through what kind of visualization is best for your audience. This space has two
perpendicular lines. The vertical line is a spectrum of the general FORM of the visualization
and runs from static to interactive.
54 փ PRINCIPLES OF DATA VISUA LIZATION
Static
FORM
FUNCTION
Explanatory Exploratory
Interactive
Static visualizations provide all the information at once and do not move. These are, for
example, your basic line, bar, and pie charts.
Interactive visualizations allow a transfer of information between the user and the inter-
face. The user clicks, swipes, or hovers, and something happens or additional information
is made visible.
Sitting somewhere between the two are animated visualizations. These—for example, ani-
mated GIFs (Graphics Interchange Format), movies, online slideshows—do not necessarily
permit the user to manipulate data points to create alternative results, but they might let the
user control the pace or transitions of the visuals.
FORM AND FUNCTIONփ 55
On the horizontal spectrum is the FUNCTION of the visualization. Here, visuals run
from explanatory to exploratory visualizations.
Explanatory visualizations bring the results to the forefront; they convey the author’s
hypothesis or argument to the reader.
Exploratory visualizations encourage the user to explore the data or subject matter to
uncover findings for themselves.
The intersection results in four quadrants.
Graphs that are not interactive or do not move are typically used to illustrate a point or re-
inforce an argument made in the text or presentation, like a line or bar chart. These are also
the visualizations we might make on our own or within our organization as we explore our
data and develop our findings. This book is primarily concerned with these visuals, and we’ll
explore a wide array in depth in Chapters 4 through 11.
These visualizations let readers interpret the data and find their own results in a static display
of information. The infographic on the next page from designer Krisztina Szűcs, for exam-
ple, encourages us to explore the data on our own. In each graph, the left vertical axis shows
a movie’s Rotten Tomatoes score (a critical score for movie quality). The right vertical axis
shows the profitability of each movie, the gap between its gross revenues and budget. She
doesn’t make a specific argument or point out specific details. Instead, she lets you explore
the data to draw your own conclusions.
Perhaps the easiest explanatory-interactive graph to consider is a static graph that has an
interactive hover or rollover layered on top of it. While supremely popular just a few years
ago, this approach seems to be falling out of favor as readers move toward smartphones and
the need to click and hover is less important. Interactivity can engage a reader to experience
a story or explain a process. To take just one example, consider the New York Times’ “Paths
This static infographic from designer Krisztina Szcs encourages us to explore
the data on our own.
FORM AND FUNCTIONփ 57
to the White House” graphic, which lets the user pick the paths by which either presidential
candidate could win the election.
This “Paths to the White House” graphic lets the user pick the paths by
which either presidential candidate could win the election.
Source: New York Times.
These visualizations graphically present a complete data set and ask users to find interesting
patterns or stories. As one example, Aaron Koblin’s Flight Patterns project shows all flights
in U.S. airspace in a twenty-four-hour period. Videos and snapshots (including his 2011 TED
talk) enables users to zoom in, pause, rewind, and otherwise explore the visualization. Other
times, the purpose of these graphics is to give people the data to play with or create visualiza-
tions on their own or to enable them to use the data in their own work.
58 փ PRINCIPLES OF DATA VISUA LIZATION
This map by Aaron Koblin shows all flight paths in US airspace over twenty-four hours.
Remember these axes are spectrums, so a visualization can land somewhere in the middle
of this space, or bring different aspects of these quadrants together. You’ve certainly seen
examples of this from the Guardian, New York Times, Washington Post, and elsewhere—a
combination of interactive and static visualizations paired with text, perhaps audio clips
from interviews and pictures, all sewn together to creative a visual experience that gives you
a deeper level of experience and understanding than the text alone might do. Through text,
video interviews, static, and animated visualizations, Bussed Out from the Guardian in 2018
tells the story of how U.S. city governments bus homeless people out of their cities.
Another way to think about how these quadrants blend together is with animated data visu-
alizations. Animations can take different forms and fulfill different purposes. In her book Visu-
alization Analysis and Design, Tamara Munzner distinguishes between three kinds of animation:
narrative storytelling (i.e., movies); transitions from one state to another; and video playback,
in which the user can control the sequence by playing, pausing, and rewinding. “Animation
is extremely powerful when used for transitions between two dataset configurations because
it helps the user maintain context,” she writes. Such transitions, therefore, help the reader (or
perhaps better put, the user) see a data point or points move from one position to another.
Through text, video interviews, static, and animated visualizations, “Bussed Out”
from the Guardian in 2018 tells the story of how US city governments bus homeless
people out of their cities.
60փ PRINCIPLES OF DATA VISUA LIZATION
A simple way to think about animated visualizations is the “animated GIF.” A GIF (Graph-
ics Interchange Format) is a number of images or frames in a single file. From a data visu-
alization standpoint, animated GIFs are powerful because they enable us to stitch together
different graphs into an animated whole. They are just like cartoons: A graph with one line,
followed by the same graph with another and another and another, builds up to a line graph
with four lines.
Animated visualizations are especially powerful on social media because they break the
stream of endless content and images. A set of Tweets from the New York Times during the
2016 Summer Olympics, for example, showed (sped up) swimming events so users could
A series of tweets with animated visualizations from the New York Times showed how US
swimmer Katie Ledecky beat her opponents by wide margins.
FORM AND FUNCTIONփ 61
quickly and easily “watch” an illustrated version of the event. Different than a static bar chart
or table of finishing times, these animations engaged viewers and held their attention, espe-
cially on a fast-paced platform like Twitter.
There is no “right” or “wrong” quadrant here—what you choose depends on what reac-
tion you want from your audience, where you publish your content, and what tools your
audience needs to understand the data.
In 1997, Ben Shneiderman, then a professor of computer science at the University of Maryland
at College Park, wrote what would become a mantra of online interactive data visualizations:
The theory was that you give users an overview of the visualization, then let them zoom
in or out and filter through it, and then allow them the opportunity to reveal the details as
they need (for example, through tooltips or downloads).
With the growing importance of mobile technology, however, Shneiderman’s mantra is
not as applicable as it once was. About a decade later, Archie Tse, the Deputy Graphic Direc-
tor at the New York Times, argued that the increase in mobile platforms means that people
just want to scroll. If the reader needs to click or do anything besides scroll, “something
spectacular has to happen.”
This argument makes perfect sense. When I think about scrolling through my newsfeed
on my phone during my morning commute on the train, I don’t want to sort and filter data
on my device. I want the author to tell me or show me the most important features of the
data. This is even more important when I’m reading technical or scientific work.
A 2018 project from Microsoft Research (see next page) tested how quickly and accurately
people could read different graphs on their mobile phones. They asked a hundred people to
read data from a weather app and a sleep app that used different graph types: a linear bar
chart layout and a radial layout, like a circular column chart. The researchers found that peo-
ple were slower to comprehend radial layouts but were no less accurate in their inferences.
Mobile use is only growing. According to the Pew Research Center, in October 2017
roughly six in ten adults (58 percent) got their news from their mobile device, compared
62 փ PRINCIPLES OF DATA VISUA LIZATION
Microsoft researchers Brehmer, Lee, Isenberg, and Choe (2018) tested how quickly and
accurately people could read different graphs on their mobile phones.
with 39 percent who got their news on a personal computer. As you consider where your
audience resides in this FORM-FUNCTION space, keep in mind that technology, behavior,
and news habits are always changing.
The chapters of the rest of this book will explore families of charts and graphs based on their
primary functions. Each chapter employs a different style guide borrowed from news orga-
nizations, nonprofits, and other organizations to showcase all the variety of subjective design
decisions available. In Chapter 12, we’ll learn more about specific styles and how to assemble
the components of your own data visualization style guide.
Remember to ask yourself about your audience. You may even take this a step further
and talk directly to your audience. Ask them what they want and what they need to better
FORM AND FUNCTIONփ 63
understand your data, content, and analysis. Ground your work in the lived experience of
communities and people you care about and want to reach. Some audiences want a thirty-
five-page PDF report. Others want a two-page brief. Others want an eight-hundred-word
blog post. Still others want a more immersive, narrative experience like you might find on a
major newspaper website. And some just want the data. An academic researcher, manager,
practitioner, policymaker, and a reporter may need very different things. Your visualization
should match the needs of your audience.
As you consider your audience’s needs, you may need to balance the accuracy of your
graph with how it engages your audience. One way to think of how to do this effectively is to
be empathetic to your audience’s needs. Take it from Alan Alda, in his book If I Understood
You, Would I Have This Look on My Face?: “Developing empathy and learning to recognize
what the other person is thinking are both essential to good communication.”
PART TWO
CHART TYPES
COMPARING CATEGORIES 4
T
he graphs in this chapter are intended to help our readers compare values across
categories. Bars, lines, and dots can all let our readers compare within and between
groups. In some cases, we want our reader to see both levels and change, or some other
variable combination; in other cases, we want to focus their attention on one comparison
or another.
The challenge when comparing categorical data is deciding what we want the chart to
convey. Is there a primary argument or story? Is there something you can identify as the
most important comparison you want the reader to make? As chart creators, we need to
prioritize what we want our charts to do. By putting every bar or dot in the graph, we can
obscure the point we wish to convey.
This chapter starts with the bar chart. Like the line chart that will kick off the next chapter,
the bar chart is familiar to most readers, which makes it a convenient choice to guide readers
as they compare categories or view changes over time. It also sits at the top of the perceptual
ranking diagram. It’s not necessarily the case that we must always give our readers the exact
values, but when we do, the bar chart is an excellent choice.
Graphs in this chapter are styled roughly following the guidelines published by Eurostat,
the statistical office of the European Union. Eurostat’s seventy-six-page style guide covers
everything from color, typography, logos, tables, layout, and more elements of a comprehen-
sive style guide that we will discuss in Chapter 12.
68 փ CHART TYPES
BAR CHARTS
One of the most familiar data visualizations, the length or height of the rectangular bars in
bar and column charts depict the value of your data. The rectangles can be arranged along
the vertical axis so that the bars lie horizontally (often called a bar chart) or vertically on
the horizontal axis (often called a column chart). For the sake of brevity, and the fact that
whichever way you align them they are still bars, I call these bar charts throughout the book.
Bar charts sit at the top of the perceptual rankings list. With rectangles sitting on the same
straight axis, it’s easy to compare the values quickly and accurately. Bar charts are also easy
to make, even with pen and paper. This one shows the total population in ten countries from
around the world. It’s easy to find the least (Italy) and most (Brazil) populous countries in
the group, even when they are not labeled with the exact values.
The bar chart is a familiar chart that’s easy to read and make. It sits at the top
of the perceptual ranking matrix.
Data Source: The World Bank.
It’s even easier to see the highest and lowest values when the data are sorted according to
their data values. This strategy doesn’t always work, however. If, for example, I was showing
population levels for sixty countries, I might sort the values alphabetically, so that readers
COMPARING CATEGORIESփ 69
When possible, sort the data in your bar charts. This makes it easier for your reader
to find the highest and lowest values.
Data Source: The World Bank.
could more easily find the bar for a specific country. But if I was making an argument about
the population level in a specific country or set of countries, I might sort the data so that the
country or countries of interest are at one end of the graph. Alternatively, I could simply use
a different color to highlight whichever bar or bars I want to set apart from the rest.
There are a few strategies to creating bar charts, many of which will apply to other charts
in this chapter as well.
Starting the axis of bar charts at zero is a rule of thumb upon which many data visualization
experts and authors agree. Because we perceive the values in the bar chart from the length
of the bars, starting the axis at something other than zero may overemphasize the difference
between the bars and skew our perception.
Take the bar chart of population. Because none are lower than fifty million, we might
be tempted to start the axis at fifty million. After all, this would emphasize the difference
between the values.
70 փ CHART TYPES
Starting the vertical axis at 50 million overmphasizes the differences in values and skews
our perception of the data.
But notice what happens when we do that. The differences in values are emphasized—in
fact, they are overemphasized. Here, it looks as though Brazil is orders of magnitude larger
than Italy, when, in fact, it is only about three-and-a-half times greater. This isn’t a matter
of moving from accurate perception to general perception—it’s a matter of moving from
accurate to inaccurate.
If you want to take a more extreme view of this, imagine starting the graph at a hundred
million—and why not? If starting at fifty is OK, then we can pick any arbitrary number. Now
at a glance it looks like nobody lives in half of these countries!
There is emerging research in this area that suggests that perhaps starting bar charts at
something other than zero does not bias our perception of the data. In one recent study,
participants were better able to assess the sensitivity of the results (e.g., no effect, small effect,
medium effect, or big effect) and more accurate (e.g., the size of the effect) when the vertical
axis was set at a range more consistent with the variation of the data. Until more research
is conducted, however, my preference is to start the axis in bar charts at zero to avoid any
confusion or possibility of visual bias.
COMPARING CATEGORIESփ 71
If starting the y-axis at fifty is OK, then why not one hundred?
Data Source: The World Bank.
Another cardinal sin of data visualization is what is called “breaking the bar”—that is, using
a squiggly line or shape to show that you’ve cropped one or more of the bars. It’s tempting
to do this when you have an outlier (see Box on page 74), but it distorts the relative values
between the bars.
Let’s create a bar chart of population in the ten most populous countries of the world. In 2018,
China and India were the most populous countries on the planet with 1.39 billion and 1.34 bil-
lion people, respectively, followed by the United States with 327 million people. We can see how
dramatically larger China and India are relative to the rest of these countries in the top graph on
the next page. If we wanted to make the differences between the less populous countries larger,
we could break the bars, but this makes China and India look much less populous than they
are. Chopping the lengths of the bars is completely arbitrary—I can place those squiggly lines
wherever I like to zoom in on the other differences. But that’s not being honest with the data.
If you run into a case where you have outliers but want to show the detailed differ-
ences between the smaller values, try using more graphs. You might think of this as
72 փ CHART TYPES
a “zoom in” and “zoom out” approach—show all of your data so your reader can see
the magnitude of the largest values, and then zoom in for a detailed look that omits
the outliers. On the next page, I’ve highlighted the less populous countries to show the
China and India are the most populous countries in the world
(Millions of people)
China 1,393
India 1,353
Brazil 209
Russia 144
Japan 127
Mexico 126
Ethiopia 109
Germany 83
Turkey 82
China and India are the most populous countries in the world
(Millions of people)
China 1,393
India 1,353
Brazil 209
Russia 144
Japan 127
Mexico 126
Ethiopia 109
Germany 83
Turkey 82
Don’t break the bar in your bar charts. The break can be arbitrarily set anywhere
and distort our perception of the data.
COMPARING CATEGORIESփ 73
differences between them, which we can’t quite see in the main graph. Adding labels and
an active title is another good way to communicate the differences between smaller values
to the reader.
China and India are the most populous countries in the world
(Millions of people)
China 1,393
India 1,353
Brazil 209
Russia 144
Japan 127
Mexico 126
Ethiopia 109
Germany 83
Turkey 82
Brazil 209
Russia 144
Japan 127
Mexico 126
Ethiopia 109
Germany 83
Turkey 82
In cases where you have large values or outliers but want to show the detailed differences
between the smaller values, try using more graphs.
EXTREME VALUES OR OUTLIERS
An outlier is a data point that is far away from other observaধons in your data. It may be
due to random variability in the data, measurement error, or an actual anomaly. Outliers
are both an opportunity and a warning. They potenধally give you something very inter-
esধng to talk about, or they may signal that something is wrong in the data.
In 2014, Buzzfeed teamed up with the website Pornhub to look at pornogra-
phy viewing by state. Using geolocaধon data of people accessing their site, Porn-
hub calculated the number of page views per person in each state. People in
Kansas, they found, watched far, far more pornography than any other state
in the country: 194 page views per person. Nevada was second with 166 page
views.
The data went into the scaħerplot below, comparing blue-state and red-state porn
consumpধon. You can clearly see Kansas as an outlier in page views. Do people in Kan-
sas really watch that much more porn?
KS
175
HI
150
Each circle
represents
one state
125
VT
100
UT NM ME
75
AR
50
20 30 40 50 60 70 80
Percent voting for Obama, 2012
Source: Pornhub views from Buzzfeed; Voting percentages from The Guardian and NBC News.
Scatterplot originally created by Christopher Ingraham.
Turns out the answer is no. Apparently, Pornhub’s methodology assigned missing
geolocaধon data to the geographic center of the United States, which, as it turns out,
is Kansas.
Not all outliers are mistakes, however. As just one example, we can look at the rate
of physical violence from a )rearm across advanced countries. In 2017, more than 8 per
100,000 people were vicধms of )rearm violence in the United States, compared with
0.90 per 100,000 people in Canada and 0.49 per 100,000 in Belgium. In some cases,
outliers are truly outliers.
There are lots of ways to test for outliers in your data, some more complex than oth-
ers. One way is to simply look at your data. Exploring your data does not need to start
with complex math and staধsধcs—you should always visually inspect your data.
But that approach is hardly mathemaধcal. A standard method is to compare data
values to 1.5 ধmes the interquarধle range (IQR). The IQR is a simple summary of your
data and is the diøerence between the third and )rst quarধles (see the Box in Chapter 6
on percenধles).
Bar charts don’t need tick marks between the bars. White space is an effective separator and
deleting the tick marks reduces clutter.
One exception is if you have a “major” category label that spans multiple bars. In such
cases, larger tick marks can be helpful to group the labels (see the bottom chart on the next
page).
Gridlines help the reader see the specific values for each bar and are especially useful for
the bars farthest from the axis label. Because they serve as a visual guide, they can be ren-
dered in a lighter color so the reader’s eye stays on the data.
When it’s important for the reader to know the exact values, you can add data labels to
the chart. My preference is to forgo the gridlines and axis lines altogether in these cases.
In bar charts, tick marks are not necessary. The white space does
the job of separating the bars.
Omitting tick marks is part of removing as many non-data elements as possible.
Consider Italy in the next two graphs (highlighted in blue). Without the labels, the gridline
helps us see that there are more than fifty million people living in the country; with the label,
it is clear that it’s sixty million people and thus the gridlines are probably not necessary.
250
200
150
100
50
0
Brazil Russia Japan Mexico Ethiopia Vietnam Germany Turkey France Italy
These horizontal gridlines help the reader see that, for example, there are
more than fifty million people living in Italy.
Data Source: The World Bank.
Data labels make gridlines redundant and, by extension, the vertical axis.
COMPARING CATEGORIESփ 79
The graph may look too cluttered with labels if I had fifty or maybe even twenty countries, so
I might include a separate table or an appendix. Deleting the gridlines when including data
labels is primarily an aesthetic choice and as you continue to work with data and make your
own graphs, you will develop your own style for these graphic elements.
The default solution for long horizontal axis labels is to run the text vertically, as on the spine
of a book. But this approach forces your reader to turn their head to the side. One solution is to
rotate them 45 degrees, but the reader still has to turn their head. Another approach is to shrink
the font size so they are aligned horizontally—though this usually makes them too small.
The most elegant solution is to simply rotate the entire graph. This still uses the same pre-
attentive attribute—the length of the bars—but the axis labels are now aligned horizontally;
they are easy to read with no effect on data comprehension.
Brazil
Russia
Japan
Mexico
Ethiopia
Vietnam
Germany
Turkey
France
Italy
0 50 100 150 200 250
With long axis labels, consider rotating the chart to make the labels
horizontal and easier to read.
Data Source: The World Bank.
80 փ CHART TYPES
There are countless ways to modify the standard bar chart. One simple variation is to use
other shapes in lieu of bars. The lollipop chart, for example, replaces the bar with a line and
a dot at the end. This version lives a hair below the bar chart on our perceptual rankings,
because it’s not exactly clear which part of the circle encodes the value. But it removes a lot
of ink from the page and gives you more white space to add labels or other annotation.
This is just one example of an alternative shape. Triangles, squares, and arrows are other
options, as are bar-shaped images that reinforce your data. A chart showing data on urban
growth may use building-shaped bars, and a chart on climate change may use trees for bars.
Be careful with this approach, however, as readers may confuse the total area of the icons as
a value indicator rather than just the height.
Brazil
Russia
Japan
Mexico
Ethiopia
Vietnam
Germany
Turkey
France
Italy
0 50 100 150 200 250
The lollipop chart replaces bars with a shape (usually a dot) and a line.
Data Source: The World Bank.
800 m
700 m
600 m
500 m
400 m
300 m
200 m
100 m
0m
Great Pyramid Empire State Building Taipei 101 Willis Tower KVLY-TV Mast Burj Khalifa
(Giza) (New York) (Taipei) (Chicago) (Blanchard) (Dubai)
Eiffel Tower Petronas Towers World Trade Center CN Tower Warsaw Radio Mast
(Paris) (Kuala Lumpur) (New York) (Toronto) (Gabin)
Alternative shapes, like buildings or people, can be used in lieu of the basic bar shape.
Source: Based on Wikimedia user BurjKhalifaHeight Petronas Towers
The total population in Brazil exceeds that of other countries Change in Brazil’s population from 2008 to 2018
(Millions of people) (Millions of people)
9.5)
il (20
.5)
(209
2008
Braz
2018
(192
Italy
)
20 7.8
20
(60.4
) 09 (
4 4.5 (19 17
(1 3.9 20
)
ia )
Fra ss
nc
e( Ru
67
)
) 201
97.5 5 (2
1 (1 04.5
201 )
.9)
(82 Me
ny xic
r ma o(
12
Ge 6.2
)
)
9.3
20
)
(19
14
(95.5
Ethio
(20
2013 (201)
12
20
2.8
nam
pia (1
2)
Viet
09.2)
A radial bar chart wraps the standard bar chart around a circle. This chart type moves
down the perceptual ranking list because it is harder to compare the heights of the bars.
Data Source: The World Bank.
82 փ CHART TYPES
Another approach to the basic bar chart is to abandon the usual grid and instead place
the bars in a circle, called a radial layout. There are two common ways to do this: the radial
bar chart and the circular bar chart.
The radial bar chart, also called the polar bar chart, arranges the bars to radiate outward
from the center of a circle. This graph lies lowers on the perceptual ranking list because
it is harder to compare the heights of the bars arranged around a circle than when they
are arranged along a single flat axis. But this layout does allow you to fit more values in a
compact space, and makes the radial bar chart well-suited for showing more data, frequent
changes (such as monthly or daily), or changes over a long period of time.
W. E. B. Du Bois used a circular bar chart in his famous Exposition des Negres d’Ameriique
at the 1900 Paris Exposition. He included this radial bar chart in his set of infographics for
Source: W. E. B. Du Bois, Assessed Value of Household and Kitchen Furniture Owned by Georgia Negroes
(1900) via Library of Congress Prints and Photographs Division.
COMPARING CATEGORIESփ 83
The Georgia Negro: A Social Study, which shows the dollar value of household and kitchen
furniture held by African Americans in Georgia in six years (1875, 1880, 1885, 1890, 1895, and
1899). “The end result,” wrote Whitney Battle-Baptiste and Britt Rusert in their book about
Du Bois’s graphics, “is simultaneously easy to read and hypnotic.”
Perceptually speaking, the circular bar chart is problematic because it distorts our percep-
tion of the data—in this case, the lengths of the bars don’t correspond to their actual value.
Consider the case where the values of two bars are the same—the ends of the bars will line
up in the same position, but the lengths of the bars are not actually the same because they lie
along the circumference of two different circles. Author and data visualization expert Andy
Italy
France
Turkey
Germany
Vietnam
Ethiopia
Mexico
Japan
Russia
Brazil
Perceptually speaking, the circular bar graph is problematic because it distorts our percep-
tion of the data. In this case, the lengths of the bars don’t correspond to their actual value.
Data Source: The World Bank.
84 փ CHART TYPES
Kirk uses an Olympic footrace as a metaphor. Runners start at staggered positions on the
track, but they all end up running the same distance because the runner on the outside lane
has more distance to cover than the runner on the inside lane. Here, the visualization doesn’t
move down the perceptual ranking, but off of it altogether because it distorts the data and for
that reason, I recommend avoiding them altogether.
PAIRED BAR
A simple bar chart is perfect for making comparisons across categories, like comparing
populations across countries. If I want to show comparisons not just across but also within
countries, the paired bar chart is a good option. The paired bar chart will be familiar to most
readers and is easy to read, and the shared baseline makes it easy to make comparisons.
There are more women than men in each country except for Ethiopia
(Millions of people)
Men Women
Brazil
Russia
Japan
Mexico
Ethiopia
Vietnam
Germany
Turkey
France
Italy
0 20 40 60 80 100 120
A simple paired bar chart is familiar to most readers and easy to read.
Data Source: The World Bank.
COMPARING CATEGORIESփ 85
Instead of showing both data values, we could show the difference between them.
Data Source: The World Bank.
Say we want to show the number of men and women in each country in our sample. A
paired bar chart allows us to do so.
Note that the paired bar chart directs the reader’s attention not just to the levels, but also
to the difference. If it’s important that readers see both, this is a good option.
But if our goal is for the reader to focus only on the difference between the two values
within each category, this isn’t the most direct way to do so, because we are asking them to
compare the difference in lengths. Instead, we could just show the difference between the
two values in a single bar, like the one above.
In cases where you want the reader to see the level and the difference, you may need a
different chart entirely. I prefer the parallel coordinates plot (see page 263), the slope chart
(for data that vary over time; see page 150), and the dot plot (see page 97). Remember to
ask yourself, What is the goal of this graph? That question will guide you to the best way to
visualize your data.
86 փ CHART TYPES
200
150
100
50
0
Brazil Russia Japan Mexico Ethiopia
The paired bar chart can be used to show changes over time and can be used to examine
changes within and between countries.
Data Source: The World Bank.
Another use for the paired bar chart is to show changes over time. And although
I include the word pair in the title, these charts can have more than two values. The chart
below, for example, shows the population in five of our countries from 2014 to 2018. This
The line chart is a more familiar way to show changes over time.
Data Source: The World Bank.
COMPARING CATEGORIESփ 87
allows the reader to examine the population change within countries and the differences
across countries.
The patterns in your data can also drive your chart type selection. If the values
decline evenly across the different years for all categories, a paired bar chart may look
fine. But if the values move around over time, a line chart (as shown earlier) or cycle
chart (see Chapter 5) may make for better comparisons over time within and across
each group.
There are two instances when I prefer to use bar charts rather than line charts to show
changes over time. First, when there are few data points—for example, only five years—the
extra ink in the five bars gives the graph more visual weight. Second, when I have discrete
time intervals (and few observations), such as the first quarter of the year.
Clutter is the main issue to keep in mind when assessing whether a paired bar chart
is the right approach. With too many bars, and especially when there are more than
two bars for each category, it can be difficult for the reader to see the patterns and
determine whether the most important comparison is between or within the different
categories.
When it comes to whether a paired bar chart is too cluttered, trust your eyes and your
instincts. Put yourself in your readers’ shoes—try to imagine where their eyes will go when
they look at the graph for the first time. If there’s too much going on, you may need to break
up your data, use a different chart type, or try a small multiples approach.
STACKED BAR
Another variation on the bar chart is the stacked bar chart. While the paired bar chart shows
two or more data values for each category, this chart subdivides the data within each cat-
egory. The categories could sum to the same total, say, 100 percent, so that the total length
of the bar is the same for every group. Or the totals may differ across the groups, in which
case the total length of each bar may differ. Above, I’ve plotted the share of gross domestic
product (GDP) each of ten countries spends on support for health care, old age, and other
programs. The entire length of the bars shows how much each country spends on these pro-
grams as a share of GDP.
As with the bar charts we’ve looked at thus far, the stacked bar chart is familiar, easy to
read, and easy to create. The biggest challenge, however, is that it can be difficult to compare
88 փ CHART TYPES
The stacked bar charts shows how diøerent categories sum to a total. The interior series
in the chart, however, are harder to compare with one another because they do not sit
on the same baseline.
Data Source: Organisaধon for Economic Co-Operaধon and Development.
the different values of the segments within the chart. In the example above, it’s easy to com-
pare the values across the countries for the Health category, because the bar segments share
the same vertical baseline. But that’s harder to do with the two other series because they do
not share a baseline. Which country spends more on old-age programs, Italy or Greece?
You can quickly see that Italy spends more on health programs than Greece, because those
segments are left-aligned on the vertical axis, but it’s much harder to determine with the
segments for the other categories.
One way to address the changing baseline is to break the graph apart so that each series
sits on its own vertical baseline. This is a small multiples graph, arranged side by side. It’s
now easier to see that Greece spends more on Old Age programs than Italy. The tradeoff is
that it is harder (if not impossible) to see the total values. But that too can be overcome: You
can still break up the stacked graph and add a final segment that represents the total amount
(this is not an issue when all of the series sum to 100 percent because the summed segments
will all have the same length).
COMPARING CATEGORIESփ 89
Belgium
Norway
France
United Kingdom
Australia
Spain
Italy
Canada
Greece
Netherlands
Instead of stacking all the data series together, we can break them up (either with or
without the totals) to create a sort of small-multiples approach. Here, we move up to the
top of the perceptual ranking list because each series sits on its own baseline.
Data Source: Organisaধon for Economic Co-Operaধon and Development.
In both versions, the horizontal spacing for each segment should be the same width, oth-
erwise it might appear that a segment takes up a larger proportion of the space than it really
does. In cases where you add the total, the width does not need to be the same as the other
groups, but the increments along the axis should be the same. In other words, if the width of
each segment above in which the data range from 0 percent to 50 percent is one inch wide,
the total category that spans 0 percent to 100 percent should be two inches wide.
Belgium
Italy
Greece
Norway
1” 1” 1” 2”
When creating these sorts of small multiples bar charts, be sure that each
segment has the same width.
90 փ CHART TYPES
Even though different baselines in the standard stacked bar chart can make it more dif-
ficult to compare values, there are cases when the stacked bar chart is preferable. In this
stacked bar chart, I’ve included more spending categories and divided them into shares
of the total so the graph highlights the distribution. In this view, it becomes clear that
around three-quarters of total government spending in these countries goes to programs
for health care and old age programs. That observation is harder to see in the version on
the right, where each category is placed on its own vertical baseline. Even though it is
easier to compare differences in each category across countries, you don’t see large differ-
ences between them.
As always, identify what you want to show and where you’d like to focus your reader’s
attention. In these examples, the Health category is emphasized because the data are sorted
according to those values (shown as a percent of total spending on these specific programs)
and it is situated along the vertical baseline. In this layout, the other segments become sec-
ondary in comparison to health spending.
There is one other stacked bar chart that you may have come across that shows a single
set of data values and the gap between them and another value (often the total). The graph
on the next page uses this approach to show the share of women elected to the U.S. House
of Representatives from 1917 to 2018. The version on the left shows the raw percentages; the
vertical axis ranges from 0 to 25 percent. Here, you see a dramatic increase in the share of
In these examples, we can see how our ability to compare different values within and
across countries varies between these two views.
Data Source: Organisaধon for Economic Co-Operaধon and Development.
COMPARING CATEGORIESփ 91
20%
15%
10%
5%
0%
19 1-1 9
19 5-1 3
19 9-1 7
19 3-1 1
19 7-1 5
19 1-1 9
19 5-1 3
19 9-1 7
19 3-1 1
19 7-1 5
19 1-1 9
19 5-1 3
19 9-1 7
19 3-1 1
19 7-1 5
19 1-1 9
19 5-1 3
19 9-1 7
19 3-1 1
20 7-1 5
20 1-2 9
20 5-2 3
20 9-2 7
20 3-2 1
-2 5
9
2 91
2 92
2 92
3 93
3 93
4 93
4 94
4 94
5 95
5 95
6 95
6 96
6 96
7 97
7 97
8 97
8 98
8 98
9 99
9 99
0 99
0 00
0 00
1 01
17 01
01
19 7-1
1
19
A somewhat rare case where a stacked bar chart is used to focus attention on one series.
This technique may be particularly valuable where the relative proportion is as important
as the change.
women in Congress. The version on the right shows the same data, but stacks a gray series
on top of the data values to 50 percent. In this version, we can emphasize that the share of
women is still small even though that share is rising. It is in these cases—where the relative
proportion is as important as the change—that this technique may be particularly valuable.
((NEW-OLD)cOLD) ȅ 100.
Posiধve percent changes (that is, NEW>OLD) mean there is a percent (or percentage)
increase. Negaধve changes (NEW<OLD) mean there is a decline. You can calculate dif-
ferences over ধme or between groupsU all that really maħers is that you follow the for-
mula and know that you are comparing the change relaধve to the iniধal value of OLD.
92 փ CHART TYPES
NEW-OLD
But the poverty rate fell by 0.4 percentage points over the two years:
Obviously, those are two very diøerent numbers, but people confuse them all the
ধme. Clearly represenধng your data starts with clearly understanding your data, how
they were collected, and how to calculate basic descripধve staধsধcs.
A variation on the stacked bar chart is one in which the stacks diverge from a central
baseline in opposite directions. These are often found in surveys where the responses are
arrayed in ranges from, for instance, strongly disagree to strongly agree. These are often
called “Likert Scales,” named after the psychologist Rensis Likert, who invented the scales
in the early 1930s.
In this example, drawing on data from the International Social Survey Programme,
survey respondents were asked whether they believe it is the government’s responsibil-
ity to reduce income inequality. By grouping the “disagrees” and the “agrees” together
on either side of a central baseline, we can compare the total sentiment across the dif-
ferent countries.
The diverging bar chart can show differences in opposing sentiments or groups, such as
“agreecdisagree” or “truecfalse.”
One advantage of this chart is that the sentiments are clearly presented—the Disagrees
jut out to the left (in what we might typically think of as a negative direction) and the Agrees
out to the right. This works well if your audience is most interested in the total sentiment
of each side and not necessarily comparisons between each individual component. If the
individual comparisons are the primary focal point, then a paired bar chart could do the
job just as well.
Why do we perceive these values to the left as negative? Throughout western history, the
concept of left—and even left-handed people—has been plagued with negative connotations.
94 փ CHART TYPES
Consider the etymology of the word: left is derived from the Old English word lyft, which
means “weak.” In Latin the word sinister means the left or left-hand direction. The word
right comes from the Old English riht, whose original meaning was “straight” and thus not
bent or crooked. And this is why we have phrases such as “standing upright” or “do the right
thing” or “the right answer,” all of which connote goodness and correctness. You can also
see this in other languages: In Spanish, for example, the word derecha means “right” and the
closely-derived derecho means “straight.”
As with the stacked bar chart, the challenge with visualizing these kinds of data is that we
are comparing within and across the categories. Arranging the bars in opposite directions
makes it difficult to compare the totals of the two groups. In other words, it’s difficult to com-
pare the total share of people who disagree with the total share of people who agree. That task
is slightly easier in the paired bar chart, but then you lose the positive-negative connotation
of the diverging chart. Depending on the patterns in your data and the number of categories
and groups, you might find this chart looks cluttered and busy.
Taking the opposite sides of the diverging bar chart and placing them in a more standard
paired bar chart approach can also work and allows us to more accurately compare the totals.
COMPARING CATEGORIESփ 95
You must be especially careful using a diverging bar chart when you have a “neutral”
category. By definition, the neutral survey response is neither agree nor disagree, and should
therefore be grouped with neither category.
Placing the Neutral category of a diverging bar chart in the middle wrongly implies that the
neutral responses are split between the two sentiments.
Placing the neutral category in the middle of the chart along the vertical baseline creates
a misalignment between the two groups and implies the neutral responses are split between
the two sentiments. It also means that none of the segments sit on a vertical baseline. Placing
it to the side of the chart is a better strategy because the disagree, agree, and neutral catego-
ries now all sit on their own vertical axes, even though the neutral category is somewhat
emphasized as it sits to the side (see next page).
Another alternative—regardless of whether you have a neutral category—is the stacked
bar chart as shown on the next page. In this view, the different categories sum to 100 per-
cent, and one can more easily compare the totals between the countries. A good strategy is
to mark specific aggregate values to guide the reader. Here, for example, I have marked the
50-percent position to make it clear for which countries the total “agree” and “disagree” sen-
timents are at least half of the total.
A better approach is to place the Neutral category off to the side of the graph.
A stacked bar chart can be used to show these kinds of Likert scales.
COMPARING CATEGORIESփ 97
The small multiples bar chart is yet another way to visualize these kinds of data.
You could take this further and break the chart into its components, as we discussed
in the previous section. In general, as with many graphs, which variation you choose will
depend on your goals.
DOT PLOT
The dot plot (sometimes called a dumbbell chart, barbell chart, or gap chart) is one of my
favorite alternatives to a paired or stacked bar chart. Developed by William Cleveland, one
of the early pioneers in data visualization research, the dot plot uses a symbol—often but
not always a circle—corresponding to the data value, connected by a line or arrow. The data
values correspond to one axis and the groups to the other, which do not necessarily need to
be ordered in a specific way, though sorting can help.
The dot plot is an easy way to compare categories—especially many categories—when bars
might add too much ink and clutter to the page. For this example, let’s look at scholastic test
98 փ CHART TYPES
This simple bar chart shows math and reading scores across multiple countries. Bar charts
are often the default visual for these kind of data, but it looks heavy and dense.
scores around the world from the Programme for International Student Assessment (PISA),
an international set of achievement tests taken by fifteen-year-old students in reading, math-
ematics, and science. We can easily plot the mathematics and reading scores for a set of coun-
tries using a simple bar chart, but the twenty bars makes the chart heavy and dense.
By contrast, the dot plot shows the same data with a dot at each data value connected by
a line to show the range or difference. The circles use less ink than the bars, which lightens
the visual with more empty space. The country labels are placed close to the leftmost dot,
though they could also be set off to the left along the vertical axis. If necessary, data values
can be placed next to, above, or within each circle.
Dot plots are not restricted to two dots and a connecting line, nor are they restricted
to simply comparing different categories. You can use dot plots to show a change between
two years, for example. You could use different shapes or icons or arrows instead of lines to
denote direction. You can also use more than two objects. For example, we could add sci-
ence test scores to this plot, but we need to be sure to add sufficient labeling so our reader
COMPARING CATEGORIESփ 99
The basic dot plot places a dot for each data point and connects them with a line. Notice
how more white space lightens the visualization.
knows what each object on the graph represents. Axes and gridlines can be included or not,
depending on how important it is for the reader to determine the exact values.
A few points of caution about the dot pot. First, it’s not entirely obvious when the direc-
tion of the values change, as in the last chart. Did you notice that math scores were higher
than reading scores in four of the countries in the dot plot above? That difference is not
immediately evident unless the reader carefully examines the points and their coloring. In
this and other cases, we should consider how sufficient annotation, clear labeling, and high-
lighting colors can help clarify different directions. The data are sorted by math scores in the
dot plot on the next page, which helps organize the countries, but it is still not immediately
clear that in only the first four countries are math scores higher than reading scores.
One approach is to split the graph into two groups, one for countries in which math scores
are higher than reading scores and another for countries where the opposite occurred. In
these versions (page 100), the groups are split and then sorted with larger, bold headers to
distinguish them. We can also add data values—I will sometimes put them right inside the
100 փ CHART TYPES
As with the basic bar chart, sorting the data in a dot plot helps organize
the space for the reader.
circle—but be careful because the labels can clutter the chart. An alternative is to include ver-
tical gridlines, depending on how precisely we want to communicate the data to the reader.
When using a dot plot to show change over time, I prefer to change the linking lines to
arrows, which helps make the direction clear.
Another word of caution for dot plots that show changes over time. The dot plot is, by defi-
nition, a summary chart. It does not show all of the data in the intervening years. If the data
between the two dots generally move in the same direction, a dot plot is sufficient. But if the
data contain sharp variations year by year, a dot plot will obscure that pattern (as it also does
for bar charts). For example, if test scores decreased between 2015 and 2019 and then increased
sharply between 2019 and 2020, the dot plot would only show an overall increase, masking the
change in the intervening years. In some cases, you may not have a choice—if you are using
data from the decennial U.S. Census, by definition you will only have data for every ten years.
That’s something you can’t help, but if you are familiar enough with your content, you’ll know
whether showing only those points is enough to clearly and accurately make your point.
Labels and annotation can clarify differences in the relationships between the values. Grid-
lines are not always necessary.
Dot plots can show changes over time. In these cases, I will often make the linking line an
arrow to suggest the change over time.
102 փ CHART TYPES
Because dot plots are essentially a summary plot, be wary of using highly
variable data with dot plots.
Marimekko charts may look odd at first, but they are just an extension of the bar chart. This
type of chart is useful when you want to make comparisons between two variables: one com-
paring categories and one showing how they sum to a total. The name of the chart comes from
the Finnish design firm Marimekko, founded in 1951 by Armi Ratia and her husband, Viljo.
Early Marimekko style featured straight, oversized, geometric patterns and bright colors.
In the standard vertical bar (or column) chart, the data are measured along the height of
the vertical axis and the widths of the columns are identical. The Marimekko chart takes that
standard column chart and expands the width of each bar according to another data value. The
Marimekko chart is an easy way to add a second variable to your standard column or bar chart.
In this Marimekko chart, I show two variables for the ten most populous countries: the
share of people with less than $5.20 per day and the share of the total population among these
countries. The percent of people with less than $5.20 per day is plotted along the vertical axis
COMPARING CATEGORIESփ 103
Early Marimekko fabric styles like this one featured straight, oversized,
geometric patterns and bright colors.
as in a standard bar chart; the widths of each bar are then scaled according to the share of
each country’s population summing to 100 percent across these ten countries (an alterna-
tive is to show the raw counts rather than percentages). You can see that the most populous
countries in this sample—the widest bars—and their distribution of poverty. You can also
use color strategically here: if this graph were in an article about poverty in Brazil and China,
we might shade all the bars the same color except for those two, as in the graph on the right.
The Marimekko chart (sometimes called a Mekko chart) scales the widths of the bars in a
bar chart corresponding to another variable. Color can be used to highlight specific values.
104 փ CHART TYPES
The two variables could also be plotted separately in two bar charts, and while these
graphs are familiar and easy to read, they do not communicate the relationship between the
two variables as well.
Percent of people with less than $5.20/day Population (as percent of all countries)
India India
China China
Nigeria Nigeria
Bangladesh Bangladesh
Indonesia Indonesia
Pakistan Pakistan
Brazil Brazil
Mexico Mexico
Russia Russia
Putting two variables on a chart might get you thinking about the association or relation-
ship between the two variables. If that’s the case, you could visualize that relationship with
other chart types. I’ve plotted the same data in this scatterplot (see also page 249). You can
see how China and India are outliers, especially along the population axis, but the chart
doesn’t communicate the part-to-whole picture of population. The parallel coordinates plot
on the right (see also page 263) similarly shows how many more people live in China and
India, and how a greater share of the population in India lives on $5.20 per day. (One poten-
tial issue with the parallel coordinates plot is that the lines might suggest a change over time
to some readers when instead, in this case, it is being used to compare the two variables.)
A variation on the Marimekko is to have both the heights and the widths of the bars sum to
100 percent. This is sometimes called a mosaic chart, though many people don’t differentiate
between these two charts and use the terms interchangeably. In this definition of the mosaic
chart, you fill the entire graph space and can therefore provide a part-to-whole perspective
of the data along both dimensions. In this way, the mosaic chart is also closely related to the
treemap (see page 297), but does not necessarily show a hierarchical relationship.
COMPARING CATEGORIESփ 105
Other ways to plot two data series are a scatterplot (left) or parallel coordinates
plot (right). Both are discussed later in this book.
In this example, population is still plotted along the horizontal axis, and the vertical axis
now contains three categories for people with low levels of income: share of people with less
than $1.90 per day, $3.20 per day, and $5.20 per day.
The mosaic chart is a variation on the Marimekko where both the heights
and the widths of the bars sum to 100 percent.
106 փ CHART TYPES
Distribution of tax returns by size of tax change for the “Tax Cuts and Jobs Act”
Decrease Greater Decrease Less than Increase Increase Greater
than $500 $100-$500 $100 $100-$500 than $500
$500,000
$200,000
$100,000
$75,000
$50,000
$40,000
$30,000
$20,000
$10,000
0
0% 20% 40% 60% 80% 100%
Source: Joint Committee on Taxation
Notice the difference between the stacked bar chart on the left (where all the bars
are the same width) and the mosaic chart on the right. While the mosaic chart adds
another variable, it is harder to see the details in the top category.
A mosaic chart can also serve as an extension of a stacked bar chart. These graphs show gains
from the Tax Cuts and Jobs Act of 2017 for tax units at different points in the United States’
income distribution. The stacked bar chart on the left shows five categories of tax gains across
eleven income intervals, and each bar shares an equal width. If we scale the widths of the bars
to the number of tax units in each income interval—so that the total vertical height of the chart
sums to 100 percent—we can create the mosaic chart shown on the right. Notice that the mosaic
chart gives a better sense of the distribution of the number of taxpayers in different groups, but
because there are relatively few people in the top income group, it’s harder to see those values.
Unit charts show counts of a variable. Each symbol can represent an observation or a num-
ber of units. For example, if one symbol represents ten cars and there are ten car icons, the
reader mentally multiplies the two for the total of one hundred cars. You can use unit charts
to show percentages, dollars, or any other discrete amount. You can arrange them in differ-
ent directions or break them down into subcategories by using colors or outlines.
Another advantage of these charts is that they can lend themselves to a more human
connection. Bar charts, for example, are abstract and impersonal. They collapse all of the
people reflected in that data point into a single shape. These charts, on the other hand, offer
COMPARING CATEGORIESփ 107
an opportunity to connect with the subject by reminding readers of the number of people
represented, particularly if each dot represents one person.
Another simple way to show these kinds of discrete counts is to just show the numbers.
In The Big Book of Dashboards, authors Steve Wexler, Andy Cotgreave, and Jeff Shaffer
call this the BAN approach: “Big-Ass Numbers.” BANs might work best in a dashboard,
infographic, social media post, or slide deck, but personally, I use them more sparingly in
longer reports.
108 փ CHART TYPES
ISOTYPE CHARTS
Isotype charts are a subclass of unit charts that use images or icons instead of simple shapes.
The term Isotype—International System of Typographic Picture Education—was coined by
German philosopher and political economist Otto Neurath, his wife Marie Neurath, and
their collaborator Gerd Arntz in the 1920s. They used the Isotype system to visualize all
kinds of data, from workers in different industries, to population density and distribution,
to the number of machines used in specific factories. They believed that this kind of visual
system would help people communicate demographic, economic, and environmental issues
to a broader public regardless of people’s educational attainment.
Otto Neurath, Marie Neurath, and Gerd Arntz developed the Isotype chart in the 1920s.
COMPARING CATEGORIESփ 10 9
1990 47%
2015 14%
The graphic below is a classic example of their work. Each symbol represents a different
count of workers (home or factory) and pounds of production. Aligned along a single verti-
cal axis, it is easy to see how the values change over time.
We can take the same approach with the poverty data we’ve been using in this chapter.
Notice that there’s more than one way to use Isotype images in these two charts of extreme
poverty rates. The version on the left uses individual icons to show each group of ten percent-
age points (the lighter icons could be included or not). The version on the right essentially
orders the icons atop a bar chart. In either case, the icons connect the subject and content
with an immediately recognizable visual image.
47% 14%
in 1990 in 2015
Source: The World Bank
Icons can also be scaled according to their data values, but be careful because it’s hard to
know whether they are scaled according to the height, width, or area.
110 փ CHART TYPES
Instead of lining up the icons in rows or another arrangement, you can also scale them
according to their value. But be careful, as sometimes it’s hard to know whether the data are
scaled according to the height, width, or area. That may not matter to every audience—it’s clear
here that the 47 percent is much larger than the 14 percent—but in cases where accuracy is
paramount, this icon-scaling approach is inadvisable. Here, the vertical distance represents the
data values, but the area of the icon on the left is about 10 times the size as the icon on the right.
These icon-driven graphs may look nice and engage readers, especially for few data val-
ues, but they can be difficult for your reader to count and compare. In his 1914 book, Graphic
Methods for Presenting Facts, Willard Cope Bertin criticized this approach: “Charts of this
kind with men represented in different sizes are usually so drawn that the data are repre-
sented by the height of the man. Such charts are misleading because the area of the pictured
man increases more rapidly than his height.” More recently, data visualization author and
instructor Stephen Few wrote that unit charts force the reader “to either count, read the num-
bers, or do our best to compare the areas formed by each, which we can do poorly at best.”
But sometimes the downsides of imprecision and slow comprehension may be offset by
how memorable the chart is and how it engages readers, an issue that is borne out in recent
research. Viewers in one study had clear preferences for graphs that included stacked icons
rather than simple bars. They also found that images that sit in the background or are added
to a chart but do not depict data are distracting to the reader, so if you choose to use these
The total population in Brazil exceeds that of other countries The total population in Brazil exceeds that of other countries
(Millions of people) (Millions of people)
Brazil Brazil
Russia Russia
Japan Japan
Mexico Mexico
Ethiopia Ethiopia
Vietnam Vietnam
Germany Germany
Turkey Turkey
France France
Italy Italy
Both graphs show the population in ten countries. The graph on the left is
cluttered by an unnecessary backdrop of the world, while the one on the right uses
flag icons to add identifying detail.
COMPARING CATEGORIESփ 111
kind of small unit icons or images in your work, be sure to use them only to encode your
data and not for gratuitous decoration.
The graph on the left, which shows the population in ten countries, has a backdrop of a
world map, a superfluous and distracting decoration. The graph on the right uses flag icons
to add identifying detail and some visual engagement to a standard chart type.
Other research suggests that unit visualizations are intuitive and flexible, a way for readers
to slowly wade into a visualization. Some have found that “unit visualizations are mostly useful
when we want the readers to understand specific data item and encode its value (e.g., a unit,
a person, a currency, a region, etc.).” Too much data and too many units, however, can create
a visualization that is cluttered and obscures individual data points or the overall argument.
WA FFLE CHARTS
Waffle charts are another subclass of unit charts. They are especially good for visualizing
part-to-whole relationships. Waffle charts are arranged in a 10 × 10 grid in which each col-
ored cell represents one percentage point. You can use multiple waffle charts to show sepa-
rate percentages—so the graph both shows part-to-whole relationships and lets your reader
compare across the categories.
When creating unit or waffle charts, especially with icons, be mindful of your audience
and how symbols may not appropriately represent your content. If you are visualizing child
mortality rates in different countries, for example, an icon of a baby is not appropriate. Using
icons of men to represent counts of people may ignore women in your data set. Alternatively,
Waffle charts are a subclass of unit charts and, in this case, arrange
the squares in a 10 x 10 grid.
112 փ CHART TYPES
if you want to measure the different types of broccoli in your garden, simply line up the pic-
tures, as in this fun project from Zachary Stensell shown above.
HEATMAP
Heatmaps use colors and color saturations to represent data values. Simply put, a heatmap is
a table with color-coded cells. They are often used to visualize high-frequency data or when
seeing general patterns is more important than exact values.
These two heatmaps show the different components of total income for ten countries
using data from the Luxembourg Income Study. The version on the left uses the same color
scale for all six categories, where lighter colors encode smaller values and darker values
COMPARING CATEGORIESփ 113
Self-
Employment Capital Public Social Private
Earnings Income Income Benefits Transfers Unclassified
Australia (2014)
Brazil (2013)
Canada (2013)
Switzerland (2013)
Denmark (2013)
France (2010)
Italy (2014)
Netherlands (2013)
UK (2013)
US (2016)
Heatmaps use colors and color saturations to represent data values and can focus the
reader’s attention along the columns or across the rows.
encode larger values. In this view, you can see that people’s earnings account for the great-
est share of their total income, and, in most countries, public social benefits appear to be
the second-largest share. In the heatmap on the right, each category is assigned its own
color scale. Here, you can more clearly see that public social benefits (in the fourth column)
account for a smaller share of total income in Australia, Brazil, Switzerland, and the United
States. Which one is better depends on your goals. Do you want your reader to compare
across all of the values or within each category?
You can also use a heatmap to show changes over time. Imagine a spreadsheet that con-
tains infection rates from the measles disease for every state in America from 1928 to 2008.
If the spreadsheet had states ordered along the rows and years along the columns, your first
instinct might be to create the line chart on the next page.
11 4 փ CHART TYPES
You can see basic patterns in measles infections across the United States in this dense line
chart, but it’s hard to pick out any specific values.
There’s nothing inherently wrong with this next line chart—you can see the positive infection
rate from 1928 to about 1963 (marked with the black vertical line), the year when the measles
vaccine was introduced. Over the next five years or so, infections dropped quickly, and eventu-
ally reached around zero within about ten years. What you essentially get from this chart is that
there were infections—going up, going down, in a tangle of lines—until about 1963.
The Wall Street Journal looked at the same spreadsheet and, instead of creating a dense line
chart, they created a heatmap. I’ve created my own version here using a different color palette
and discrete categories of the infection rate. You can see the darker blue cells (mostly above 16
infections per 100,000 people) prior to the introduction of the vaccine, again marked with a
black line. After 1963, the colors quickly transition to lighter shades of blue, and ultimately to
the lowest rates of infections (zero infections and fewer than 1 infection per 100,000 people).
This chart may not be inherently better than the line chart, but it does let you more easily
examine each state or year far more easily than picking out a single line from the tangle in
the line chart. Also remember that sometimes being different is good in itself. How many
COMPARING CATEGORIESփ 115
This heatmap may not be inherently better than the line chart at showing measles infec-
tions rates, but it does let the reader more easily examine each state or year.
complex line charts have you seen and just immediately skipped over? The heatmap, with its
different look and color, can draw readers in. As artist and data visualization expert Giorgia
Lupi said, “Beauty is a very important entry point for readers to get interested about the
visualization and be willing to explore more. Beauty cannot replace functionality, but beauty
and functionality together achieve more.”
116 փ CHART TYPES
Another way to use the heatmap is to modify the layout, for example, applying it to a cal-
endar. In this example, vehicle fatalities in 2015 are plotted on heatmaps of months. Notice
how easy it is to see the higher fatality rate on Fridays and Saturdays along the right edge of
each column.
Vehicle fatalities in 2015
January February March
Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa
1/1/15 2/1/15 3/1/15 4/1/15 5/1/15 6/1/15 7/1/15 8/1/15 9/1/15 10/1/15 11/1/15 12/1/15 1/1/16
Source: National Highway Traffic Safety Administration
This line chart shows the same data as in the heatmap calendar, but it’s more difficult to
reach the same conclusion.
By comparison, consider plotting the same data in a line chart. Even with the additional
blue circles used to mark Saturdays, it’s difficult to reach the same conclusion about more
fatalities on the weekends.
Unlike the measles example, where both charts had advantages and disadvantages, in this case
the calendar heatmap is a better approach because it does a better job highlighting the important
pattern of deaths on the weekends—and it’s a more engaging graph placed in a familiar shape.
A final way to modify the heatmap is to change the rectangular layout altogether. On the
next page, each of the fifty states is plotted along each radii of a circle, grouped into five geo-
graphic regions (separated by the black lines). Each ring represents a different (binary) data
type, such as whether the state has right-to-work laws, an income tax, or a minimum wage.
This alternative to a set of six maps has some advantages and some disadvantages. On
the one hand, it is a compact representation of six different data series that the reader can
quickly and clearly see the categorical and part-to-whole data within and between states.
On the other hand, it’s not a familiar graph type, and that may turn off some readers. It’s also
118 փ CHART TYPES
Heatmaps can be arranged in different ways. The radial layout has some advantages and
disadvantages, but so does the six-pack of maps.
worth noting that the order of the rings can affect our perception of the data because the (red)
squares on the outer ring are by definition larger than the (yellow) squares on the inner ring.
The gauge chart (or gauge diagram) looks like the speedometer in your car’s dashboard.
Typically set up somewhere between a half-circle and a circle, it uses a pointer or needle to
indicate where your data fall within a particular range. Sections of the gauge are shaded to
illustrate sections such as poor, good, and excellent.
I see gauge diagrams most often in financial planning tools because they give an easy,
familiar way to visualize targets or progress towards a goal. They also frequently show up
in fundraising campaigns where the entire semicircle represents the goal, and the needle
and filled area represent money raised so far. This can be a good example of using a familiar
shape to support the metaphor of the visualization—everyone understands that “filling up”
the gauge means the fundraising effort has reached its goal.
Gauge diagrams do, however, introduce perceptual challenges because, again, people are
not very good at measuring and comparing angles. If you want to give your reader a general
COMPARING CATEGORIESփ 119
$12,763,095
Our Goal:
$16m
sense of the values, the gauge chart is a decent choice. But if enabling your reader to discern the
specific values and compare those values to the ranges is of utmost importance, then it is not.
Given the familiarity with the gauge and the obvious metaphor it represents, it’s perhaps
no surprise that they show up in a variety of settings. As just one example, I once received
This series of gauge charts shows four real estate trends in my Northern Virginia neighborhood.
Source: MountJoy Properধes, brokered by Keller Williams Realty.
120 փ CHART TYPES
a flyer from Mountjoy Properties (a residential real estate team that serves the DC Metro
Area) that consisted of a series of gauge diagrams showing current real estate trends in my
neighborhood of Northern Virginia. I could pretty quickly see the current state of the mar-
ket, but adding more data or more detailed data might be more difficult.
BULLET CHARTS
Because of these perceptual challenges with the gauge chart, author Stephen Few invented
the bullet chart, which is a linear, more compact way to show similar kinds of data. The basic
bullet chart contains three different data elements:
1. First, there is the actual or observed value, shown here as the black horizontal bar.
In this illustration, the bar represents an average customer satisfaction score of 4.0.
2. Second, there is a target value, shown here as the black vertical line. Here, we were
aiming for a satisfaction score of 3.5.
3. Finally, there is the background range, which shows grades or bands of success, such
as poor, good, and excellent. These sit behind the other two series so the reader can
compare the actual and target values. Here, poor scores are 1.5 and below, good scores
are from 1.5 to 3.0, and excellent scores are anything above 3.0.
The different components of the bullet charts can vary. There might be a scale of five ranges
instead of three, or there may not be a target value. The scales can also show the underlying
Credit Score
Securities
Bonds
Combining different bullet charts is a compact way to let the reader make a series
of different comparisons.
distribution of the data—for example, showing quartiles or ranges of percentiles (see Box
on page 183 for more discussion of percentiles). Because the bullet chart is so compact, it’s
easy to create multiple versions and stack them together. The bullet charts above show three
metrics you might find in a financial report, but they are more compact than gauge charts
and, because the rectangles are aligned, it is easier to compare across the different categories.
In the basic bubble graph, circles represent values. Like a bar chart, the purpose of these types
of charts is to compare values between categories. But unlike the bar chart, humans are not
very good at accurately comparing the sizes of circles (remember the perceptual ranking dia-
gram from page 14). Still, circles may be more visually interesting, they can reinforce a visual
or metaphor, and they are a good choice when discerning exact quantities is not paramount.
Another drawback of proportionally sized circles is that you cannot visualize negative
values. While bars can go in both directions—typically right or upward for positive and left
or downward for negative—that’s much harder to visualize with circles.
In any case, we are not very good at making accurate estimates from the circles even when
they are sized by area. Instead, what I think we try to do is make the comparison based on the
diameter of the circle—like in a bar chart—which gives us incorrect conclusions. Take a look at
the two graphs on the next page and try to guess the percent of people living on less than $1.90
per day in Pakistan. Do you think that task is easier in the bubble chart or in the bar chart?
This is not to say that you should never use circles. Remember, again, it’s all about your
audience. A bubble comparison chart, inserted in a short article off to the side with the
122 փ CHART TYPES
Pakistan
Vietnam 2.0%
We are better at discerning differences from bars than by areas of circles. By the way, the
percent of people with less than $1.90 a day in Pakistan is 4.0 percent.
numbers placed prominently in the middle of each circle may be more engaging than a stan-
dard bar chart. Too many circles, however, may make it difficult for your audience to discern
any quantities or relationships. In this next example, yes, you can see that India, the Demo-
cratic Republic of the Congo, and Nigeria have the largest number of people in poverty, but
it’s difficult to quickly assess how different they are or the numbers of the next set of countries.
Nigeria
84.8
Indonesia
15.1
Kenya
17.8
A bubble comparison chart can be engaging and interesting, but it can also
be hard to discern the values.
CALCULATING THE AREA OF A CIRCLE
Remember to size the circles by area, because using the radius or diameter generates
circles that overemphasize diøerences (the radius or diameter scales in a linear way,
but the area scales quadraধcally). The )rst black circle is sized relaধve to the gray circle
using the diameter while the second black circle uses the area. As you can clearly see,
using the diameter skews our percepধon and makes the diøerence between the two
values look much larger.
A simple example will demonstrate the importance of using the area rather than the
radiuscdiameter for sizing these circles. In case you don’t remember your middle-school
math, the diameter is any straight line that passes through the middle of the circle. The
radius (r) is half the diameter. And the area (A) is equal to the constant pi (Q) ধmes the
radius-squared, or A Qr2.
So, say the data value for the gray circle is 1 and for the black circle is 2. If we start
with the gray circle and set the radius equal to 1, we can )nd the area is equal to: AO
Qr2 Q12 Q.
To )nd the size of the black circle the correct way, we say that the area of that circle
is twice the size of the gray circle, corresponding to the diøerence in their data values.
So, if we double the area of the black circle so that it is now 2Q, we can then rearrange
the formula and )nd the radius of the black circle to draw it: rB AB / à 2à / à 2.
Let’s do this the wrong way and use the radius instead of the area. In this case, the
radius of the gray circle is sধll 1, so let’s make the radius of the black circle 2, again cor-
responding to the diøerence in their values. This makes the area of the black circle now
AB Qr2 Q(2)2 4Q. In other words, the area of the black circle sized in this way is four
ধmes the size of the gray circle instead of twice the size, as the data suggest.
It’s easier to pick out the countries with the most and least poor populations in this bar
chart, but it takes up the entire page.
COMPARING CATEGORIESփ 125
The nested bubble chart can sometimes mask circles in the back—but it can
also make for easier comparisons.
If we were to use a more perceptually accurate representation of these same data in, say, a
bar chart, the visual becomes much larger as on the previous page. Each bar is labeled here,
so a reader could find Madagascar, but is that important? Again, is the goal to show all of the
countries or just a subset? As always, consider your goals and whether your reader needs a
perceptually accurate view to understand your argument.
The bubble charts shown above are known as bubble comparison charts. Layering cir-
cles on top of each other, as in the graph above, are often called nested bubble charts. The
nested bubble chart can sometimes mask circles in the back, but it can also make for easier
comparisons.
You can use bubbles to demonstrate correlations (see the bubble chart in Chapter 8)
or add bubbles to a map to encode another variable (see the point map in Chapter 7). In
general, while there are perceptual issues with using circles and bubbles in data visualiza-
tion, they can also be more engaging and enjoyable than yet one more bar or line chart.
As Amanda Cox, Data Editor at the New York Times said, “There’s a strand of the data viz
world that argues that everything could be a bar chart. That’s possibly true but also possibly
a world without joy.”
126 փ CHART TYPES
SANKEY DIAGRAM
Sankey diagrams—named for their creator Matthew Henry Phineas Riall Sankey in 1898—
are especially useful for showing how categories compare to one another and flow into other
states or categories. Arrows or lines display the transition from one state to another, and the
width of the lines denote the magnitude of each transition. Changes can occur over time or
as comparisons between categories. For example, a Sankey diagram could be used show how
different companies in an industry have merged, broken apart, or failed in different years.
This Sankey diagram shows how fifty-two students tried to spell the word camouflage.
The first blue segment shows that all fifty-two students started with the letter “C”, fifty
then went to “Cam”, followed by thirty-seven with “Camof ”, and so on. Ten students
spelled the word correctly, shown in the orange segment near the top of the graph.
Germany
Sub-Saharan Africa
This Sankey diagram shows flows in federal aid from the United States, United Kingdom,
and Germany to regions around the world.
More applicable to the sort of content we’ve looked at thus far in this chapter, the graph
above shows flows in federal aid from the United States, United Kingdom, and Germany to
regions around the world. You can see that countries in Sub-Saharan Africa and unspecified
countries receive the bulk of the aid and that the United Kingdom and United States contrib-
ute more to Sub-Saharan African countries than does Germany.
Presenting these data as paired or stacked bar charts give us a different perspective. In the
paired bar chart on the left, the obvious first comparison is across the funder countries for
each region of the world. In the stacked bar chart, by comparison you’re more likely to com-
pare funding across the recipient regions—that, for example, the greatest share of spending
on Sub-Saharan African countries is from the United States. The Sankey draws our atten-
tion in a different way, reading horizontally across the page in a way that mixes these two
comparisons, privileging neither. Neither view is “right” or “wrong,” but may serve different
audiences differently, highlight different patterns, and answer different questions.
Sankey diagrams can be layered together with other chart types. In this example, the aid
flows from the United States are presented on a world map. This provides a geographic view
of the data, but also simplifies things—imagine how cluttered the visual would look if it
included flows from the United States, United Kingdom, and Germany.
12 8 փ CHART TYPES
Financial support flows from Germany, the United States, Financial support flows from Germany, the United States,
and the United Kingdom to different areas of the world and the United Kingdom to different areas of the world
(Percent of total support) (Percent of total support)
Germany UK US Sub-Saharan Africa Southern and Central Asia Other Asia and Oceania
45 Middle East and North Africa Latin America and Caribbean Europe
40 Unspecified
35
Germany
30
25
20
US
15
10
5 UK
0
Sub-Saharan Southern and Other Asia Middle East and Latin America Europe Unspecified
Africa Central Asia and Oceania North Africa and Caribbean 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Source: Organisation for Economic Co-Operation and Development Source: Organisation for Economic Co-Operation and Development
Presenting the financial flow data as a paired or stacked bar chart give us a different per-
spective than in the Sankey diagram.
This flow map provides a geographic view of the financial flow data, but it also simplifies
things by only showing flows from the United States.
The biggest problem with Sankey diagrams—and many charts for that matter—is plot-
ting too many series, as in this version that includes more countries. With too many groups
or too many crossings, the chart becomes difficult to navigate. If you find yourself with too
COMPARING CATEGORIESփ 129
UK Southern and
Central Asia
Other Asia
and Oceania
US Middle East
and North Africa
France Europe
Spain Unspecified
The biggest problem with Sankey diagrams—and many charts for that matter—is plotting
too many series makes it difficult to identify any patterns or trends.
many lines or crossings, try simplifying your data, using multiple Sankey diagrams, or using
a different chart type.
WATERFALL CHART
A waterfall chart shows a basic mathematical equation: adding or subtracting values from
some initial value to produce a final amount. It is essentially a bar chart, but each subsequent
bar starts where the previous one left off, showing how they accumulate across the graph.
Typically, negative values are given a different color than positive values, and so are the totals
at the beginning and end. Including lines that connect the bars can guide the reader through
the visualization. Because the lines are guides and not actual data, they should be lighter and
thinner than the other elements.
This next chart shows contributions to total gross income and total net income for Aus-
tralia in 2016. The data are the same as those used in the heatmap example from earlier, but
in that approach, I could fit data for ten countries in the same view. Imagine trying to do the
130 փ CHART TYPES
A waterfall chart shows a basic mathematical equation: adding or subtracting values from
some initial value to produce a final amount.
same with a waterfall chart—we would need ten different graphs—something that might be
useful under certain scenarios, but is certainly less compact than the heatmap.
Waterfall charts can also show changes over time. You might show, for example, con-
tributions to total GDP from one year to the next, and how different values contribute to
the change over the course of the year. Any data series that are added or subtracted to one
another can be presented in this way, though, again, it is a nonstandard chart type and may
require your reader more time to navigate it.
CONCLUSION
From single bars to groups of bars to stacks of bars, the bar chart is one of the most famil-
iar data visualizations for showing categorical comparisons. It also ranks at the top of our
COMPARING CATEGORIESփ 131
perceptual ranking scale from earlier. But the bar chart also poses certain challenges: too
many bars can make the visual seem overwhelming and cluttered, and stacking the series
on top of one another makes it more difficult to compare series that are not aligned on the
same axis.
The basic bar shape can be organized in many ways. They can sit next to each other or
diverge from a central baseline. They can be stacked on top of one another on a horizontal or
vertical dimension, or both as in a mosaic chart. They can also be arranged to show simple
mathematical equations, as in a waterfall chart. We are generally good at discerning the data
values from the lengths of the bars, so many of these chart types will make it easy for your
reader to perceive the exact value.
There are other ways to let your reader make comparisons. I’m especially fond of dot plots
because they remove a lot of the heavy ink from a standard bar chart and free up space to
add annotation and labels. Using icons, squares, or other shapes can engage our audience in
ways that standard charts may not, but may be less data dense.
While bar charts sit at the top of the perceptual ranking list, let’s be honest: They can be
very boring. We see bar charts every day. As chart creators, sometimes our challenge is to
find ways to engage our audience, and deploying less common chart types from our data
visualization toolbox can do just that. It’s up to us to determine where we want to focus our
reader’s attention, on the level or the difference, the single or multiple comparison, or the
relative or total values.
TIME
5
T he graphs in this chapter show changes over time. Your readers will most likely be
familiar with visuals like line, area, and stacked area charts. But others, like connected
scatterplots and cycle charts, may need more labeling and annotation for the reader to navi-
gate them successfully.
Many of the visuals in this chapter are variations on the line or area chart. Some let us
include more data on the page than usual, while others allow us to combine changes over
time with some other view of the data. With horizon charts and streamgraphs, for example,
we can include more data in a single visual, but they are probably not best suited for detailed
comparisons. Other graphs, like flow charts and timelines, may use qualitative data or nar-
rative text and visual clues to guide the reader.
Graphs in this chapter are styled following the guidelines published by the Urban Insti-
tute, a nonprofit research institution based in Washington, DC. Urban’s style guide outlines
their color palette, fonts, and guidance for different chart types.
LINE CHART
The line chart and the bar chart may be the most common charts in the world. The line
chart is easy to read, clear in its representation, and easily drawn with pen and paper. Data
values are connected by lines to show values over a continuous period, tracking trends and
patterns.
13 4 փ CHART TYPES
Total health care spending in the United States grew from 12.5 percent
to 16.8 percent between 2000 and 2015
(Percent of GDP)
20
15
10
0
2000 2005 2010 2015
Source: The World Bank
This line chart shows the percent of gross domestic product (GDP) spent on health care
in the United States over the sixteen-year period from 2000 to 2015.
As with the bar chart, the line chart sits near the top of the perceptual ranking scale. With
lines relative to the same horizontal axis, it is easy to compare the values to each other and
between different series.
As simple as the line chart can be to create and read, there are a number of considerations
to take into account, some of which are aesthetic, and some of which are substantive.
There is no hard rule to dictate the number of series you can include in a single line graph.
The key is not to worry about the sheer amount of data on the graph, but instead about the
purpose of the graph and how you can focus your reader’s attention to it. For example, in a
line graph with many series, you can highlight or emphasize a subset of your data.
Say we were interested in showing the share of government spending on health care for
the United States and Germany, but we also wanted to show them in relation to the other
TIMEփ 135
There is no hard rule to dictate the number of series you can include in a single line graph.
thirty-four countries that make up the Organisation of Economic Co-operation and Devel-
opment (OECD). To do that, instead of giving each line the same color saturation and thick-
ness, we might only color and thicken the lines for the United States and Germany and leave
the lines for the other thirty-two countries gray and thin. The “Start with Gray” strategy
from Chapter 2 comes in handy here.
Recall the section on preattentive processing: color and line width are two of the preat-
tentive attributes (page 25), thus our attention is drawn to the thicker, colored lines. The
advantage of the gray strategy is that the reader can appreciate the general pattern for the
entire sample and yet focus on the two lines of interest.
We might also take the line graph and break it into multiple graphs, the small multi-
ples approach. We might include just the line of interest in each small graph, or include
all of the lines and use the gray strategy. The set of small multiple line graphs on the
next page uses the former approach and shows spending on health care for nine of the
thirty-four OECD countries. Instead of forcing all nine lines on a single graph, each
country has its own panel. While we might lose some perspective of the relative values
136 փ CHART TYPES
The small-multiples approach breaks up a dense line chart into separate components.
of spending in each country, this layout provides more space for each country and thus
the opportunity to provide more detail, labels, or other annotation.
One of the few rules of thumb of visualizing data is that bar chart axes must start at zero (see
Chapter 4). Because we perceive the values in the bar from the length of the bars, starting the
axis at something other than zero overemphasizes the differences in values.
This does not hold true for line charts. The axis of a line chart does not necessarily need
to start at zero. As with many aspects of visualizating data, there are complications and dif-
ferent perspectives. If we say the axis does not need to start at zero, what is an appropriate
range? Where should we start and end the axis?
TIMEփ 137
To illustrate, let’s look more closely at changes in health care spending in the United
States. Each of the four charts below uses a different range in the vertical axis. As you can
plainly see, those ranges affect our perception of the level and the change in spending. In
the top-left graph, where the axis ranges from 0 to 20, we see a slight increase in spending.
As you move clockwise through the graphs and the axis range gets smaller and smaller, the
change in spending looks increasingly dramatic.
There is no right answer to the choice of the vertical axis; the answer depends on the data
and your goal. If you need to demonstrate that the economy will falter if spending reaches
17 percent of GDP, then the bottom-right chart may be best. If you are telling a more general
story, then one of the graphs in the top row might be preferable, because it still clearly shows
the increase in spending over time. If you need to show a detailed examination of spending
in each year, you might want to consider the graphs on the right.
There is no right answer to the choice of the vertical axis. The answer depends
on the data and your goal.
13 8 փ CHART TYPES
Personally, I try to avoid the approach in the bottom-left graph where the axis does not
start at zero and either the top or bottom value equals the minimum or maximum data
value. In this case, the graph feels too “tight” to me and I think it can suggest the data can go
no lower or higher than what’s pictured, which is rarely the case.
Another way to think about this is how the data, context, content, and units all work
together. A change in spending from 12 percent of GDP to 17 percent of GDP is a large
change in the context of health care reform. But it’s not as important that my kids can beat
me seventeen times in a board game now instead of twelve times when they were a bit
younger (for me—though it’s pretty important for them!).
It’s also worth noting that our perception of where zero lies in this space is affected
by how we draw the vertical axis. Without looking carefully at a line chart, you are
probably inclined to think the bottom of the vertical axis is zero, and in some cases—
especially where the data series are both positive and negative—this can be especially
important. Take these charts showing the year-to-year percentage-point change in health
care spending instead of the level of spending as a portion of GDP. In the case on the
left, it’s not immediately clear that there are any spending declines over this period,
because the zero baseline is not clearly delineated. By just darkening that axis line a bit
more than the rest as in the graph on the right, it is more evident that there are three
years when health care spending as a share of GDP declined year-over-year.
Year-to-year change in U.S. health care spending: Zero axis not marked
(Percent of GDP)
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
2001 2003 2005 2007 2009 2011 2013 2015
Source: The World Bank
We are inclined to think the bottom of the vertical axis is zero, so using a different color or
thickness for the axis makes it clear where zero lies.
TIMEփ 139
With line charts (and for other time-series charts, for that matter), we tend to misestimate
the differences between two curves.
Take this graph from William Playfair, a Scottish engineer and political scientist who is
often credited as the founder of graphic methods of statistics. In his chart from 1785, Playfair
plots exports and imports (in millions of British pounds) between England and the East Indies
from 1700 to 1780. The top line denotes imports, and the bottom line denotes exports. The ver-
tical distance or gap between the two lines shows England’s (positive) trade balance with the
East Indies. Starting in 1700 on the left, you can see the balance grow over the first thirty years
or so. Then, starting around 1730, the gap starts to shrink, reaching its narrowest point around
1755. The trade balance then appears to grow for a time and expands rapidly after around 1770.
In his chart from 1785, William Playfair, a Scottish engineer and political scientist who is
often credited as the founder of graphic methods of statistics plots, shows exports and
imports between England and the East Indies.
1 40 փ CHART TYPES
Did you notice the hump in the trade balance after 1760? In the three-year period between
1762 and 1764, imports rose quickly while exports grew more slowly, creating a larger trade
balance. Between 1764 and 1766 exports to the East Indies shoots up and brings the trade
balance right back down. But the spike between 1762 and 1764 is hard to see in Playfair’s
original chart. Those changes are much easier to see in this line chart, which plots the gap
between imports and exports.
This is the line-width illusion at work: we tend to assess the distance between curves at
the closest point rather than the vertical distance. A variety of scholars have demonstrated
this effect and have suggested alternative graph types, but the easiest solution may be to plot
the difference whenever it’s the metric of interest.
Gap between exports and imports between the UK and the East Indies
(Millions of pounds)
1.0
0.8
0.6
0.4
0.2
0.0
1700 1710 1720 1730 1740 1750 1760 1770 1780
Source: Based on Cleveland and McGill, 1984; data from Michael Friendly
The line-width illusion at work here—the bump in the gap between exports and imports is
easier to see here than in Playfair’s original.
Data markers are just what they sound like: symbols along the line to mark specific points
in the series. There isn’t a right answer to the question of when to deploy them. Personally,
I include data markers when I have only few lines or data points, or for specific points I want
to label or annotate. The data markers give the graph more visual weight.
TIMEփ 141
I like to add data markers to my line charts when I don’t have a lot of data or
when I want to highlight or label specific values.
These charts, for example, show health care spending as a share of GDP for Germany and
Spain. There are so few data points and the changes in the series are so subtle that the addi-
tion of the circular data markers give the lines more visual heft.
I prefer to make my data markers circles rather than triangles, squares, or other shapes.
This is partly an aesthetic preference, but there’s also a logic to it. Circles are perfectly sym-
metrical, and so it never matters where the line intersects the circle. With other shapes, like
triangles, the line might intersect the thinner top part or the thicker bottom part.
Other shapes may be necessary if you or your organization are required to comply with
certain rules or laws that enable screen readers to differentiate between objects on a screen
for people with vision disabilities. In the United States, federal government agencies are
required to follow Federal Section 508 regulations that make websites accessible to people
with disabilities (see Chapter 12 for more on data visualization accessibility). Even with dif-
ferent colors, most screen readers cannot differentiate between the different series if the
shapes are all the same. In these cases, different data markers are a good choice.
Circles are a symmetrical shape, which is why I prefer to use them as data markers.
142փ CHART TYPES
At some level, there is always missing data. People change jobs every day, not just when unem-
ployment numbers are published. Lots of things happen in the ten years between the publication
of each U.S. Census. Most data are a snapshot in time, but we often treat them as continuous.
Missing data are truly missing when a regular series is interrupted because the data were
not collected. In these cases, we should make it clear that the data are incomplete. In line
charts, we can change the format of the line (for instance, with dashes) or not connect the
points at all to signal that those data points are missing. We can also place a note on the chart
or below the chart to explain that those data values are unavailable.
Here are two ways to signal missing data: a dashed line or annotation for the gaps.
What we should never do is ignore the missing values altogether and make it appear as
though we have a continuous, uninterrupted series.
9
0
2004 2007 2008 2009 2011 2012 2013 2014 2015 2016
This chart ignores the missing data points and is misleading. It gives the false impression
of a continuous, uninterrupted series.
TIMEփ 1 43
You might be inclined to add another vertical axis to your line chart when comparing
changes in two or more series of different units. Resist that urge. Consider this dual-axis line
chart that shows the share of income devoted to paying for housing on the left axis and the
quarterly unemployment rate on the right axis in the United States from 2000 to 2018. It’s
not immediately obvious that the unemployment rate is the blue line and associated with the
right axis and housing debt is the yellow line plotted along the left axis. The purpose of this
graph is to show that the economic climate for consumers in 2017 and 2018 is quite good—
low unemployment rates and low housing debt.
But there are three problems with plotting the data like this.
First, they are often hard to read. Did you intuitively know which lines corresponded to
which axis? I didn’t. Even if the labels and axes were colored to match the lines (which many
dual-axis charts don’t include), it’s hard to discern patterns in the data. They’re extra work
for the reader, especially when the labeling is not obvious.
Second, the gridlines may not match up. Notice how the horizontal gridlines in this graph
are associated with the left axis, which leaves the numbers on the right axis floating in space.
At the crossing point in 2009, it’s hard to see that the value of the unemployment rate (the
blue line) is just shy of 9 percent.
Third, and most importantly, the point where the lines cross becomes a focal point,
even though it may have no real meaning. In this graph, the eye is drawn to the middle of
The economic climate for consumers in 2017 and 2018 was quite good
Housing Debt Unemployment Rate
13.5% 11.5%
13.0% 10.5%
9.5%
12.5%
8.5%
12.0%
7.5%
11.5%
6.5%
11.0%
5.5%
10.5% 4.5%
10.0% 3.5%
9.5% 2.5%
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Source: Household debt service ratio, Federal Reserve Board of Governors; Unemployment rate,
Bureau of Labor Statistics. Unemployment rate averaged to quarters.
The dual-axis chart introduces a series of perceptual issues, maybe the most important of which
is that the eye is drawn to where the lines intersect, even though that may be meaningless.
144փ CHART TYPES
the chart where the two lines intersect, because that’s where the most interesting thing is
happening. But there’s nothing special about 2009, it’s just a coincidence that they crossed
at that time. The intended takeaway of the chart is how much the economic climate has
improved since the 2007–2009 recession, but that’s not what draws the eye.
The vertical axis in a line chart does not need to start at zero, so this chart—with the left axis
starting at 9.5 percent and the right axis starting at 2.5 percent—is a perfectly reasonable way
to plot the two series. By that logic, we could arbitrarily change the dimensions of each axis to
make the lines cross wherever we like. And this is the problem with dual-axis line charts: the
chart creator can deliberately mislead readers about the relationship between the series.
Because there is no hard and fast rule about how to set the dimensions of the vertical axis,
we can arbitrarily change the dimensions to make the lines cross wherever we like.
TIMEփ 1 45
Each of these four graphs are reasonable ways to set the vertical axes, and by manipulating
those ranges, I can make the series look like they (a) are closely matched for a few years around
2010 and 2012; (b) cross in the middle and the end; (c) intersect around 2003 and then again a
few years later; and (d) are closely related in the first half of the period but then diverge.
By arbitrarily choosing the axes range, we can make different data series look as correlated
as we like. On his website, Spurious Correlations, Tyler Vigen shows all kinds of dual-axis
charts in which arbitrary vertical axis scales creates erroneous—and humorous—correlations.
Similar difficulties, though to a slightly different extent, exist in dual-axis charts that com-
bine different graph types. On the next page you can see the same graph of the unemploy-
ment rate and housing debt, now with the unemployment rate plotted as an area chart. The
right axis starts at zero so the gridlines on both sides match, but it’s still not immediately
obvious which variable goes with which axis. And though it is more obvious that there are
two separate trends being visualized, the same perceptual pitfalls still exist, leading readers
to see correlations that might not really be there.
There are a few solutions to the dual-axis chart challenge.
First, try setting the charts side by side. Remember, not everything needs to be packed in
a single graph. We can break things up and use a small multiples approach. Although ideally
side-by-side graphs should have the same vertical axis to facilitate easier comparisons, we’ve
already determined that approach is impossible here, so splitting them up and using differ-
ent axis ranges can work.
1 46 փ CHART TYPES
The issue with dual-axis charts is not resolved by combining area charts and line charts.
Housing debt in the United States has declined since 2008 The unemployment rate has declined since about 2010
14% 12%
13% 10%
12% 8%
11% 6%
10% 4%
9% 2%
8% 0%
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Source: Federal Reserve Board of Governors Source: Bureau of Labor Statistics. Unemployment rate averaged to quarters.
One alternative to the dual axis line chart is to use two, side-by-side charts.
If it’s important to annotate a specific point on the horizontal axis, you could also verti-
cally arrange the two and draw a line across both. This will change the rotation of the final
graphic, but is an easier way to label a specific value or year.
Second, we might calculate an index or the percent change from some value or year (see page
148). This way the reader can see the change over time for both series and compare them along
the same metric. In the data we’ve been looking at here, we calculate the difference between
each year and 2000, the first year of the period (thus the percentage-point change). The obvious
trade-off here is that we lose the level presentation of the data and instead present the change.
TIMEփ 147
13%
12%
11%
10%
9%
8%
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
10%
8%
6%
4%
2%
0%
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Source: Household debt service ratio, Federal Reserve Board of Governors; Unemployment rate,
Bureau of Labor Statistics. Unemployment rate averaged to quarters
One alternative to the dual axis line chart is to use two charts aligned vertically, which
makes it a little easier to mark a specific data point on both charts.
Third, try a different chart type. If showing the changes in the associations between the
two series is important, try a connected scatterplot. The connected scatterplot—which has
its own section at the end of this chapter—is like a scatterplot with a horizontal and vertical
axis, but each point represents a different unit of time, such as a quarter or a year. As you can
see on the next page, it’s easier to see how the relationship has changed over time between
these two metrics. You can also see how I have added more labels and annotation (along
with different colors) to help the reader navigate the visual.
The economic climate for consumers in 2017 and 2018 was quite good
(Percent pointchange since 2000)
7
-1
-2
-3
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Source: Household debt service ratio, Federal Reserve Board of Governors; Unemployment rate,
Bureau of Labor Statistics. Unemployment rate averaged to quarters.
Another alternative to the dual-axis chart is to normalize the data or calculate the percent
change from some value.
$_;&ĺ"ĺ;1omol-rr;-uvvrrouঞ;o=|_;1omvl;ub|_
ѴoŊm;lrѴol;m|u-|;-m7_ovbm]7;0|
(Household debt service ratio)
14%
ƑƏƏƐ!;1;vvbom u;-|!;1;vvbom (4Q2007-2Q2009)
(1Q2001-4Q2001) 4Q2007 &m;lrѴol;m|u-|;ubv;vvb[Ѵ-m7_ovbm]7;0|v|-0bѴb;v
13% ovbm]7;0|-m7|_;
m;lrѴol;m|u-|;
0o|_bm1u;-v;
12% 4Q2009
=;;-uv-[;u|_;ƑƏƏƐ
1Q2000 !;1;vvbom;m7vķ|_;
11% m;lrѴol;m|u-|;v|-u|v|o
=-ѴѴ-m7_ovbm]7;0|ubv;v
10%
Ƒ ƑƏƐѶ
$_;m;lrѴol;m|u-|;bv-0o| ƑƏƏƖŊƑƏƐѶ
9% _;u;b|-vbmƑƏƏƏķ0||_;v_-u; ;u|_;Ѵ-v|7;1-7;ķ|_;m;lrѴol;m|u-|;_-v
o=bm1ol;7;o|;7|o_ovbm]7;0|bv =-ѴѴ;mv|;-7bѴ_bѴ;7;0|v_-u;v_-;Ѵ;;Ѵ;7o@
Ɛĺƕr;u1;m|-];robm|vѴo;u
8%
3% 4% 5% 6% 7% 8% 9% 10%
&m;lrѴol;m|!-|;
Source:ov;_oѴ77;0|v;ub1;u-ঞoķ
;7;u-Ѵ!;v;u;o-u7o=o;umouvĸ&m;lrѴol;m|u-|;ķ
u;-o=-0ou"|-ঞvঞ1vĺ&m;lrѴol;m|u-|;-;u-];7|ot-u|;uvĺ
Yet another alternative to the dual-axis chart is to use the connected scatterplot in which
one data series corresponds to the horizontal axis and another to the vertical axis.
TIMEփ 149
A possible exception to the “no dual axis chart” rule is if you are showing a translation of
a single measure, for example Fahrenheit and Celsius temperatures. In these cases, we are
not trying to track two different variables but showing how one maps directly onto another.
In those cases, the usual pitfalls don’t apply.
The radial bar chart and circular bar chart in Chapter 4 showed us how to take a bar chart
and wrap it in a circle. The same can be done with lines showing changes over time. As
before, using a circle may be less perceptually accurate for the reader, but it can be used, for
example to improve a visual metaphor.
These two graphs show the percent of hospital emergency room visits for the flu in the United
States for each week of the year from 2014 to 2017. Starting at the beginning of the flu season in
October, the line chart on the left gives us the standard view: an increase in the flu during the
winter months, which fades as we enter summer. The radial chart on the right shows the same
data but with a different perspective—the “lean” toward three o’clock on the chart when more
infections occur during the fall and winter, and fewer infections during the summer months on
the left side of the circle. The radial chart is more compact than the standard line chart, but it is
also harder to make precise comparisons because the lines do not sit on a single horizontal axis.
3HUFHQWRI(5YLVLWVIRUWKHćXå
2014 2015 2016 2017
0
Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep
Two ways of showing the same time series data—as a standard line chart or by wrapping
the lines around a circle.
150 փ CHART TYPES
SLOPE CHART
In some cases, it may not be necessary to show all of the data in your time series. In these
cases, a slope chart—which is really just a simplified line chart—is a useful alternative.
The paired bar chart is a standard way to visualize two data points for multiple obser-
vations (also see page 84). As an example, consider these charts of changes in the unem-
ployment rate for six states in the United States between 2000 and 2018. With this kind of
visualization, we ask our reader to process the level and change in the unemployment rate
between and across the six states. There is a lot of ink in the graph, and it asks the reader to do
lots of mental math. We could, of course, just plot the change between the two time periods,
but we often want to show both the level and the change.
The slope chart addresses this challenge by plotting each data point on a separate vertical axis
and connecting the two with a line. In this example, the left vertical axis represents the first month
of data (January 2000) and the right vertical axis represents the last month of data (January 2018).
We can easily see the relative values of each data point. Here, for example, we can see—perhaps
even more easily than in the paired bar chart—that Montana had the highest unemployment rate
in the first month and Connecticut had the lowest. The line that connects the two data points
visualizes the change over time. We can more easily see that the unemployment rate in Montana,
Hawaii, and Idaho fell between 2000 and 2018 while it rose in the other three states.
There are many ways to style the slope chart. We can use two colors to denote increases
and decreases. We can include or exclude labels for levels and changes. We can even adjust the
The paired bar chart asks the reader to make several comparisons on their own simultane-
ously, while the slope chart visualizes these comparisons for the reader.
TIMEփ 15 1
thickness of the line to correspond to a third variable. We could also rely on the Start with Gray
strategy and add more data to the basic slope chart. Here, I’ve included every state in the nation
but highlighted and thickened the six states of interest and the national average.
In graphs like these, consider whether a taller chart will make it easier for the reader to see
all of the detailed colors, labels, and annotations. As with the dot plot (see page 97), be careful
about using the slope chart when a summary of the time series may mask changes in the inter-
vening years. Of course, this is the same consideration when using the paired bar chart.
There are many ways to style a slope chart. This Start with Gray strategy can be especially
useful here to show many observations while highlighting only a few.
152 փ CHART TYPES
SPARKLINES
There is a specific style of small multiples for line charts called sparklines. Invented by author
and statistician Edward Tufte, sparklines are “small intense, simple, word-sized graphics
with typographic resolution.” They are typically used in data-rich tables and may appear at
the end of a row or column. The purpose of sparklines is not necessarily to help the reader
find specific values but instead to track general patterns and trends.
Let’s use sparklines with the health spending data. The numbers in the two table col-
umns show spending in 2000 and 2015, while the sparklines show the values for the
entire sixteen-year period. This way, readers can see some specific values as well as
the patterns over the entire period. Here, for example, you can quickly see that health
spending rose for all of these countries except for Turkey, which I’ve also highlighted
so it stands out.
Sparklines are a type of small multiples line chart typically used within data-rich tables.
TIMEփ 153
BUMP CHART
A variation on the line chart is the bump chart, which is used for plotting changes in ranks
over time, for example, political polling or positions in a golf tournament from hole to
hole. When we want to show relative ranks rather than absolute values, the bump chart is a
good choice.
A bump chart is, of course, a compromise. It does not show the raw values, which are
often preferred, but it can be especially useful if your data have outliers. By plotting the
ranks, we abstract from the large differences in magnitude.
These two bump charts show changes in health care spending across the ten countries
that have the highest spending on health care as a share of their GDP in 2015. Those coun-
tries appear in the far right position of the horizontal axis, above the 2015 label. The differ-
ence between the two charts is that that the one on the left shows the patterns and ranks for
only these ten countries for every year. You can see some gaps in certain years where other
countries would appear in the rankings, but this chart only tracks those countries that end
up in the top ten in 2015. The chart on the right, by comparison, includes every country in
every year, which requires more labeling so the reader can understand why new countries
(with different colors) suddenly appear in the chart. We could emphasize certain countries
by changing line colors or the colors of the inside of the data marker circles, or even the
thickness of the lines.
Use color, data markers, or line thickness to highlight specific data series.
Compare these bump charts to the line chart above that shows all of the OECD coun-
tries in gray and highlights the United States and Germany with color. In this case,
the United States stands far above the rest of the countries that here appear clumped
together in a swirl of lines. As is often the case, there is a tradeoff between the bump
chart and the line chart: In the standard line chart, you can see the relative differ-
ences between the series, but they’re stacked together and hard to disentangle. In
the bump chart, by comparison, we can’t see the relative differences, but we can see
relative ranks.
A modification on the bump chart is a ribbon effect. Here, in addition to rank, the widths
of the ribbons are scaled according to the actual data values. Like the streamgraph, which
we’ll see later, this approach has a more organic, flowing look. This chart from the Berliner
Morgenpost shows the rank, amount, and change in different sentiments around political
problems in Germany.
TIMEփ 155
The ribbon effect is a modification on the standard bump chart. The title in this chart from
the Berliner Morgenpost translates to “These are the 15 most important political problems
in Germany”. Arbeitslosigkeit translates to Unemployment; Einwanderung to Immigra-
tion; Bildung to education; and so on. It shows the changes in each ranks (and amounts) of
these different sentiments.
CYCLE CHART
Cycle graphs typically compare small units of time, such as weeks or months, across a mul-
tiyear time frame. They are most commonly used to display strong seasonal trends. Here,
we see the number of births in the United States in each month from 2007 to 2017. A yellow
line marks each month’s average value (a general but not necessary characteristic of the cycle
chart). We can see the downward trend for births over the decade for every month and the
156 փ CHART TYPES
higher birth rate during the summer months, July, August, and September. I’ve added a dot
at the end of each line to mark the most recent year.
By comparison, this same data displayed as a standard line graph is less clear. We can see a
spike in each year, but without more labeling, it’s not clear in which month that spike occurs.
Even though the cycle chart has more information on it—the average values shown in yel-
low and the point at the end of each line—it still feels less busy than the standard line chart.
A cycle graph can also split up a dense bar or line chart to give each series more space—
something like a small multiples chart. Take this column chart of the unemployment rate
for four groups in the United States. This kind of graph—with multiple years for different
Cycle graphs compare small units of time, such as weeks or months, across a
multiyear time frame.
TIMEփ 15 7
groups pushed together—can be difficult for the reader to make comparisons within or
across years. The cycle graph on the right separates each racial group into its own panel,
sorted by the value in the most recent year. You could argue that the graph on the right is a
small multiples line graph, but the organization and design make it more like a cycle graph.
AREA CHART
Area charts are line graphs with the area below the line filled in, giving the series more visual
weight. The area chart on the left and the line chart on the right both show the number of
people who died from prescription opioid overdoses in the United States between 2000 and
2016. You might think of the area chart as a bar chart where the bars have infinitely thin
widths and thus, as we saw in the previous chapter, the vertical axis should always start at zero.
Placing two or more series in an area chart can be difficult because one series can hide (or
“occlude”) the other, an effect we will see more in the coming chapters. On the next page, the
area chart on the left, for example, shows overdose deaths from cocaine and heroin, but the
data series for heroin is hidden behind the series for cocaine. Even if the order of the data
series are changed so that heroin overdose deaths are in front, now the heroin-deaths series
blocks the cocaine-deaths series. Compounding that difficulty is that some readers might
Area charts are line graphs with the area below the line filled in, giving the series
more visual weight.
158 փ CHART TYPES
Placing two or more series in an area chart can spell trouble, because one series can hide
(or “occlude”) the other.
mistake the two as summing to a total rather than as separate data series. Here, it’s important
to use the title and annotation to make it clear there are two distinct series.
One strategy to address this overlap is to add a transparency to the color of one (or both)
series. But be careful: by adding a transparency to only one series, we deemphasize its impor-
tance. Another alternative is to use a line chart, as in the graph on the right.
One strategy to address the overlap between series on an area chart is to add a transpar-
ency to the color of one (or both) series. Another alternative is to use a line chart, as in the
graph on the right.
TIMEփ 159
Stacked area charts build on the typical area chart by showing multiple data series simulta-
neously. Instead of sitting independently of one another as in the previous chart, the data in
a stacked area chart sum to a total or a percentage.
The stacked area chart on the left shows the total number of drug overdose deaths between
1999 and 2016. The version on the right shows the same data, but presented as percentages
that sum to 100 percent.
The reader will take away different conclusions from these two representations. In the
graph on the left, the eye is drawn to the large increase in overall deaths over the period. In
the version on the right, it is drawn to the changes in the distribution of deaths—a decline
in overdoses from cocaine, but an increase in heroin, Benzodiazepines (drugs that are often
used to treat anxiety, insomnia, and seizure disorders), and other drugs.
There are three disadvantages with the stacked area chart on the left. First, as earlier, we
again see the line-width illusion—we tend to view steep changes as bigger than they actu-
ally are. Second, only the bottom series sits on a horizontal axis, so it is hard to accurately
compare the changes over time for the other series. (Remember, this is the second row in
the perceptual ranking table showed earlier.) Third, the ordering of the data series can affect
Stacked area charts build on the typical area chart by showing multiple data series simulta-
neously and sum to a total, often 100 percent.
160 փ CHART TYPES
Recall that the perceptual ranking list suggests we can better compare values when they sit
on the same axis. That’s why it’s easiest to compare values for the bottom series.
our perception of the shares of the total and move the reader’s attention around from one
series to another.
To demonstrate, consider the two stacked area charts above. The version on the left is
the same as before while the version on the right changes the order. Notice how in the new
version, it is easier to compare changes in overdose deaths caused by Benzodiazepines (the
yellow series), because the series sits along the same horizontal axis.
There isn’t necessarily a right way to stack the series in a stacked area chart, but how you
decide to arrange them will influence how your reader perceives the data.
This isn’t to say there is a “right” way to stack the data in an area chart, or that the most important
data series should sit along the horizontal axis. If, say, you were telling a story about the declining
TIMEփ 161
share of overdose cocaine deaths over this period, you could keep it in the same position as above,
but use the “start with gray” strategy and use color in just the cocaine series. Even with the line-
width illusion, you can still see the share of deaths has declined. If it’s important for your reader to
see the exact change in the share, then putting that series along the horizontal axis is a better strat-
egy. By placing the series in the middle of the chart, you can’t compare the values to the horizontal
baseline and thus less accurately perceive the values. (Also notice how I directly labeled the seg-
ments instead of using a legend so the reader can quickly and easily identify the different series.)
A small multiples approach (here, six different graphs) more clearly shows the exact pat-
terns in each category, but does not as clearly show the relative shares of each. You make a
couple of tradeoffs here. On the one hand, the stacked chart is more compact than the small
multiples—you pack all of the information into a single visualization in which you can see
the changes in shares. On the other hand, the small multiples gives you a more perceptually
accurate view—because each series sits on its own horizontal baseline—but it is harder to
compare across the different categories.
20,000
15,000
10,000
5,000
20,000
15,000
10,000
5,000
0
2000 ’02 ’04 ’06 ’08 ’10 ’12 ’14 2016 2000 ’02 ’04 ’06 ’08 ’10 ’12 ’14 2016 2000 ’02 ’04 ’06 ’08 ’10 ’12 ’14 2016
The small multiples approach more clearly shows the exact patterns in each series, but
does not as clearly show the relative values of each.
162 փ CHART TYPES
90% All
others
80%
Respiratory System
70%
60% External
Causes Circulatory
50% System
40%
30% Nervous
System
20%
Cancer Mental
10% disorders
0%
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Stacked area charts can also show changes in the distribution of a data series.
Finally, the stacked area chart can also show changes in the distribution of some data
series. This stacked area chart, for example, shows all of the different ways people from age
zero to one hundred died in the United States in 2017. Instead of years or months along the
horizontal axis, this graph shows the number of deaths for each single year of age, a different
measure of time. Categorized into fifteen groups, most people who die around age 25 do so
of “external causes,” (the green series) such as falling or drowning, while most people who
die around age sixty die of some form of cancer (blue). As before, we could modify the colors
or the arrangement of the data to focus our reader’s attention on specific patterns or trends.
STREAMGRAPH
Like the stacked area chart, a streamgraph also stacks the data series, but the central hori-
zontal axis does not necessarily signal a zero value. Instead, data can be positive on both
sides of the axis. Together, the streamgraph illustrates fluctuations in data over time in a
TIMEփ 163
flowing, organic shape. They are therefore best used for time series data when the series
themselves have high volatility.
Streamgraphs are well suited for showing patterns that have peaks and troughs. Both
the stacked area chart on the left and the streamgraph on the right show the total number
of deaths rather than shares of deaths, as in the earlier stacked area chart. The streamgraph
gives us a slightly different view of the data and may point us more toward overall increases
rather than changes in specific series. The idea behind the streamgraph is to minimize the
distortion in each layer’s baseline that accumulates more rapidly with a stacked area chart.
Researchers are aware of how unusual the streamgraph looks, and how it may be more
difficult for readers to understand. In a review of a streamgraph published by the New York
Times in 2008, researchers noted that they “suspect that some of the aesthetically pleasing—
or at least engaging—qualities may be in conflict with the need for legibility. The fact that the
New York Times graph does not look like a standard statistical graphic may well be part of its
appeal.” Thus, while this kind of graph—or any different-looking graph—may at first con-
fuse or confound readers, they may ultimately find the shapes, colors, and other attributes
more interesting and engaging. It all depends on your audience.
On the next page, you can see a more recent example of a streamgraph. This visualization
was published by the Hindustan Times in 2016 and shows the number and type of the high-
est civilian awards the Indian government confers. Additional streamgraphs in the original
news story showed breakdowns by state, nationality, gender, and discipline.
80,000
Rx
Psychostimulants
60,000
Other
40,000
Heroin
20,000 Cocaine
Benzodiazeprine
Antidepressants
0
2000 2005 2010 2015
Streamgraphs are a variation on the area chart and are well suited for showing patterns
that have peaks and troughs.
164 փ CHART TYPES
Source: Hindustan Times. This image was cropped and condensed from the original for purposes of this book.
This streamgraph from the Hindustan Times shows patterns in the number and type of the
highest civilian awards conferred by the Indian government.
HORIZO N CHART
A horizon chart is an area chart that is sliced into equal horizontal intervals and collapsed
down into single bands, which makes the graph more compact and similar to a heatmap (page
112). The horizon chart is split into bands with positive numbers collapsed down and nega-
tive values flipped above the horizontal axis. Multiple horizon charts—which is how they are
typically arranged—can condense a dense dataset into a single visualization. Horizon charts
are especially useful when you are visualizing time series data that are so close in value so that
the data marks in, for example, a line chart, would lie atop each other. Aligning the charts in
this way allows us to include our data in a more compact space than in a series of area charts.
Color is the most important attribute in a horizon chart. Darker colors represent larger
values and lighter colors smaller values. Like sparklines and to some extent heatmaps, the
purpose of the horizon chart is not necessarily to enable readers to pick out specific values,
but instead to easily spot general trends and identify extreme values.
TIMEփ 165
This horizon chart uses the same data we’ve been using on changes in the percent of GDP
spent on public health care. An area chart is built for each country, split and collapsed, and
then arranged all together in rows. Notice how much data are packed into the single visu-
alization (ten countries and fifteen years), and—recalling the importance of preattentive
processing—see how your eye is drawn to the brighter and darker colors.
I’ll use the series for Sweden to show how the horizon chart is built. The change in public
health spending for Sweden (the third country from the bottom in this horizon chart) is shown
on the next page as an area chart and sliced into equal increments (every 0.5 percentage points).
Larger values have darker shades, and negative and positive values have different colors. The neg-
atives are flipped above the horizontal axis and all are collapsed down to the first interval or band.
Color is the key here. The same visualization as a series of line charts does not have the
same punch. The eye scans the entire visualization for important trends, but no particular
region of the visual draws our attention. That could be modified by adding color to the
The horizon chart is an area chart divided into equal intervals and collapsed into a band.
166 փ CHART TYPES
These charts show how the area chart for Sweden is divided and collapsed
to create the horizon chart.
A line chart is sufficient, but the use of color in the horizon chart attracts
and directs the reader’s attention.
different lines to highlight extreme values, but the horizon chart does a much better job of
directing the eye by highlighting values through colors.
GANTT CHART
Another way to show changes over time is to use horizontal lines or bars to show the dura-
tion of different values or actions. Gantt charts are often used as schedule-tracking devices,
for example, to track different phases of a project or budget. Invented by Henry Laurence
TIMEփ 167
5 AM 6 AM 7 AM 8 AM 9 AM 10 AM 11 AM 12 PM 1 PM 2 PM 3 PM 4 PM 5 PM
Ellie
Jack
Lauren
Karen
Steve
Marc
Jon
Leah
Gantt, an engineer working around the turn of the twentieth century, the charts were first
used by production foremen and supervisors to track production schedules.
This Gantt chart shows staffing shifts at a coffee shop over the course of a day, denoting
breaks with the white gaps, lunch breaks with the gray breaks, and other time away from the
store with stripes.
Gantt charts can be extended by modifying the width of the bars to denote another vari-
able. For example, this Gantt chart modifies the hypothetical chart above to scale the widths
according to the pay of each employee.
Joseph Priestley, an eighteenth-century philosopher, chemist, and educator, published
A Chart of Biography in 1765, showing the lifespans of approximately two thousand states-
men, poets, artists, and other notables who lived between 1200 bc and the mid-1700s. Often
called a timeline, Priestley’s chart looks more like a Gantt chart because of the use of hori-
zontal bars/lines and the concrete beginnings (births) and ends (deaths).
5 AM 6 AM 7 AM 8 AM 9 AM 10 AM 11 AM 12 PM 1 PM 2 PM 3 PM 4 PM 5 PM
Ellie
Jack
Lauren
Karen
Steve
Marc
Jon
Leah
An extension to the Gantt chart is to adjust the widths of the bars to correspond to
another data series.
Joseph Priestley’s A Chart of Biography (1765) shows the lifespans of about two
thousand statesmen, poets, arধsts, and others. It covers an enormous ধme period,
from 1200 BC to 1800 AD. A horizontal line for each person shows his or her lifes-
pan. Dots indicate wherever such there is uncertainty around those dates.
Source: Library Company of Philadelphia.
1 70 փ CHART TYPES
Flow charts and timelines are two examples of an array of visuals that can show changes
over time or different kinds of processes, sequences, or hierarchies. This class of charts and
diagrams can be explicitly tied to data or can be less quantitative and more illustrative, a way
to demonstrate different structures or processes. In PowerPoint, for example, you can look
through the “SmartArt” menu for a wide selection of layouts.
A timeline shows when certain events take place. It can be basic and flag events with a line,
icon, or marker, or it can be more involved and include annotation, images, or even graphs.
Though horizontal timelines are common, timelines can also be vertical or even a variety
of different shapes. This timeline, based on work I conducted at the Congressional Budget
Office, shows major milestones and data for Supplemental Nutrition Assistance Program
(SNAP, formerly known as food stamps). The text in the gray boxes on the right gives details
on specific legislation or program changes, and the information on the left presents changes
in spending, number of program participants, and their share of the total population.
Flow charts are slightly different. They are not necessarily tied to time in the sense of
days, months, and years, but instead they map a process, often step by step. Flow charts
make it easier for readers to understand the paths of a process rather than reading through
long passages of text or navigating a convoluted table. The flow chart on the next page
shows the process by which people can apply and receive benefits through the U.S. Social
Security Disability Insurance program (DI). Applicants start the program at what is called
the “Disability Determination Services” stage. If their application is approved, they are
“Allowed” onto the program; if not, they can either appeal the decision or exit the pro-
cess altogether. The program is designed in such a way so applicants may appeal a denied
request at each stage.
The shapes in a flow chart may carry different meanings, so we can use them strategically
to denote different attributes of the system. For example, in a flow chart with rectangles,
other shapes can denote choke points or decision points, and rounded rectangles might sig-
nal the beginning or end of a process. Adding different colors can help readers understand
and differentiate the parts of the graph from each other.
If, for example, we wanted to highlight the different parts of the DI application system, we
could use different colors and shapes, as in the version in the middle. Labels in a flow chart
can sit alongside the lines and inside the boxes, but they should be large enough and have
enough color contrast to be easily read. We could take this even further and scale aspects of
the flow chart according to some data values; in the version on the right, for example, the
branches are all scaled according to the respective shares at each stage, similar to a Sankey
diagram (page 126).
I place the flow chart in the time chapter because these processes often occur over time,
one by one. But that’s not always the case. An organizational chart or org chart, for example,
172փ CHART TYPES
Shapes color, and other elements can help readers understand the paths
of a process in a flow chart or timeline. This chart is based on work from
the Congressional Budget Office.
is a type of flow chart that shows the hierarchy or management structure of an organiza-
tion, and how work flows from the top down. We’ll see more examples of an org chart in
Chapter 8.
Just as we saw with the line chart with many lines, it’s not about the amount of content
you place in the graph, but what meets the needs of your reader. The flow chart on page 174,
TIMEփ 173
for example, was prepared in 2010 by the Republican staff of the Joint Economic Committee
in response to President Obama’s Affordable Care Act proposal. The implicit purpose of this
flow chart is to show how complex the proposal (and health care in general) is in the United
States. In that sense, the chart does its job!
In these types of graphs, the amount of notes, text, icons, and other visual elements
should, as always, meet the needs of the reader. Are specific details necessary at each point?
Would an image anchor the moment in the reader’s mind? Consider what information your
reader needs most and provide it as engagingly as you can.
CONNECTED SCATTERPLOT
Imagine showing two line charts side-by-side. You may be asking your reader to examine the
relationship between the two. Do they move together? Do they diverge or converge? How
are they related?
One way to bring two time series together without using a dual-axis chart is the con-
nected scatterplot. A connected scatterplot shows two time series simultaneously—one each
along horizontal and vertical axes—and are connected by a line to show relationships of the
points over time.
As an example, the line chart on the left shows life expectancy in South Africa from 1996
to 2016. Over that twenty-year period, life expectancy followed a U-shape pattern, first fall-
ing from about sixty-three years old to fifty-three years old, and then increasing over the
next decade, reaching about sixty-one years old in 2016. On the right, per capita GDP is
plotted over that same period—that series was flat in the first few years, then increased until
about 2008, when it dipped slightly before rising at a slightly slower rate.
With these two charts, we can make some basic visual comparisons—even as life expec-
tancy fell, economic growth continued. When life expectancy started growing again, eco-
nomic growth flattened out.
62
60
58
56
54
52
50
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Source: The World Bank
A common challenge is how to clearly show the association between two time series.
176փ CHART TYPES
Now, notice what happens when the two lines are combined and plotted in a single graph.
Here, life expectancy is shown along the horizontal axis and economic growth is plotted
along the vertical axis.
Instead of jumping back and forth between the two charts, we can see that during the
first half of the period life expectancy fell (moved to the left along the horizontal axis) and
the economy grew (moved up along the vertical axis). Starting around 2006, life expectancy
started to increase again (now moving to the right along the horizontal axis) while the econ-
omy grew, but at slower rate (the slope of the line along this latter period is flatter than before).
Because this graph is different than standard graph types, your reader will need more
time to understand how to read it. Annotation can help: Consider adding more axis labels,
arrows and a label for the year on the first and last point. But when your reader understands
this graph, it becomes part of their graphic toolbox, just like it has now become part of yours.
The connected scatterplot is one way to show how two time series are related to one
another. One series corresponds to the horizontal axis and the other to the vertical axis.
TIMEփ 177
Though less familiar to some readers, the connected scatterplots can be used to show
more data series across two metrics.
You can also use connected scatterplots to show more groups. The connected scatter-
plot on the left shows the levels of economic growth and life expectancy for ten different
countries. Here, the higher per capita GDP in the United States stands out, but patterns in
the other countries are harder to see. The graph on the right shows percent growth in both
variables since 1996. In this view, the United States is hardly visible, while the large gains in
China and Ethiopia, for example, are clearly visible.
Which graph is better? As always, that depends on your audience, your argument, and
what pattern, trend, or finding you want to bring to your reader’s attention.
CONCLUSION
The graphs we covered in this chapter show changes over time. There are simple and familiar
graphs, like line, area, and stacked area charts. But there are also more complex, less familiar,
but equally useful chart types.
The line chart may be the most basic and familiar chart type to show changes over time.
There is no limit to the number of lines you can plot, but if there are many lines, consider
using color and line thickness to draw your reader’s attention to the most important ones.
Consider using data markers to add nuance to subtle or small data sets or ways to mark
important points. As with many graphs we will explore, but perhaps even more so with line
charts, use visual cues to signal missing data.
178փ CHART TYPES
Alternative chart types are useful when you have too many data series to track in a
single graph. Try sparklines, a small multiples approach, cycle charts, or horizon charts
when you have a lot of data to visualize. For some of these approaches, enabling the
reader to discern exact values is less important than showing them the overall trend or
pattern.
Other graph types, like flow charts and timelines, have infinite varieties and styles. Hori-
zontal layouts may work for some people, content, and platforms, while vertical layouts may
be better online where it matches the natural scrolling motion. Compact layouts are best for
mobile platforms.
Whichever graph you use to plot your data, consider how much detail your reader needs
and how you can guide them to the point you wish to convey. Many of these chart types are
well-known and understood, so our challenge is to make them engaging and interesting
without sacrificing accuracy.
DISTRIBUTION
6
HISTOGRAM
The histogram is the most basic graph type for visualizing a distribution. It is a specific kind of bar
chart that presents the tabulated frequency of data over distinct intervals, called bins, that sum to the
180 CHART TYPES
50,000 50,000
40,000 40,000
30,000 30,000
20,000 20,000
10,000 10,000
0 0
0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,000 0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,000
Histograms divide the entire sample into intervals (also called “bins”). The height of the bin
shows the number of observations within it.
total distribution. The entire sample is divided into these bins, and the height of each bar shows the
number of observations within each interval. Histograms can show where values are concentrated
within a distribution, where extreme values are, and whether there are any gaps or unusual values.
We can layer histograms together to show how different distributions compare. The two
histograms above depict the distribution of earnings for men and women working in the
United States in 2016. We can make some general comparisons between them, but that task
is made easier in the next two graphs, where the distributions are placed on top of each other.
MEN’S AND WOMEN’S EARNINGS DISTRIBUTIONS IN 2016 MEN’S AND WOMEN’S EARNINGS DISTRIBUTIONS IN 2016
Men Women Men Women
60,000 60,000
50,000 50,000
40,000 40,000
30,000 30,000
20,000 20,000
10,000 10,000
0 0
0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,000 0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,000
Histograms can be layered together by using saturated colors (left) or different encodings
like bars and lines (right).
DISTRIBUTIONփ 181
The graph on the left uses two column charts and transparent colors so both are visible.
The graph on the right combines a column chart and line chart, which has the advantage of
not using transparent colors, but the balance of how the two groups are presented is now
unequal. You might also notice that the line intersects the middle of each bin as opposed
to spanning the entire bin as the columns do—it’s a minor difference but one that you may
want to keep in mind.
A key consideration in creating a histogram is how wide to set the bins. Bins that are
too wide may hide patterns in the distribution, while bins that are too narrow may obscure
the overall shape of the distribution. While there is no “correct” number of bins, there are a
number of statistical tests (using, for example, square roots, logarithmics, or cube roots) that
80 9
70 8
7
60
6
50
5
40
30
3
20
2
10
1
0 0
$0-$4ok $40k-$80k $80k-$120k $120k-$160k $160k-$200k $4k $20k $36k $52k $68k $84k $100k $116k $132k $148k $164k $180k $196k
6 5
5
4
1
1
0 0
$2k $18k $34k $50k $66k $82k $98k $114k $130k $146k $162k $178k $194k $1k $21k $41k $61 $81k $101k $121k $141k $161k $181k
A data distribution will look different depending on the number bins, as it does here as the
number of bins increases from 5 to 30 to 50 to 120.
182 փ CHART TYPES
can help determine the optimal bin width. In this example, notice how the distribution looks
different as the number of bins increases from 5 to 30 to 50 to 120.
Histograms can help us understand whether our data lean to one side or another. A distri-
bution in which more data are pushed to the left is known as right-skewed. A histogram with
more observations to the right is called left-skewed. Distributions that have two peaks are
called bimodal and distributions with multiple peaks are called multimodal. Symmetric dis-
tributions are those with a roughly equal number of observations on either side of a central
value and uniform distributions in which the observations are roughly equally distributed.
When we understand the distribution of our data and its possible skew, we’re better prepared
to conduct more accurate statistical tests. Two completely different distributions may have the
same mean and median, but if we don’t understand the spread and structure of our data, our
results may not paint a complete picture. This is where visualizing our data can be invaluable.
A modification to the basic histogram is the Pareto chart, named after the Italian engineer
and economist Vilfredo Pareto. The Pareto chart (next page) consists of bars that represent
Histograms can help us understand the shape of the distribution of our data. Here we see
six such forms of distribution.
DISTRIBUTIONփ 183
18
16 80%
14
12 60%
10
8 40%
4 20%
0 0%
Education Manufacturing Retail Public Arts & Information Active
Trade Admin. Entertainment Military
Professional FIRE Agriculture Transportation Wholesale Other
& Scientific & Trade Services
Warehousing
Source: U.S. Census Bureau
Note: FIRE = Finance, Insurance, and Real Estate
The Pareto chart shows values for each group (usually in bars) and the cumulaধve total as a line.
individual data points and a line that represents the cumulative total. The Pareto chart may
be the exception to the rule against dual-axis charts because its purpose is to show the com-
plementary distributions overlaid on each other. Of course, the two metrics are really not
two different measures—it’s the same metric, one as a marginal distribution (separate values
of each group) and one as a cumulative distribution (where the values sum to a total).
This Pareto chart shows total earnings in thirteen different major industries in the United
States—the bars show the total earnings in each industry and the line shows how the shares
add up to total earnings in the economy.
UNDERSTANDING PERCENTILES
Imagine one hundred people standing on an auditorium stage. You stand in the audi-
ence as they line up from your leđ to right, arranged by their earnings. The person
with the lowest earnings stands at the leđ side of the stage, and the person with
1st person 50th person 51st person 100th person
(lowest earnings) (highest earnings)
the highest earnings stands at the right. Together, they represent the enধre earnings
distribuধon.
The )rst person in the line has the lowest earnings. At their posiধon, 99 percent
of people—all those to their leđ—have higher earnings. They are said to be in the 1st
percenধle. Similarly, there is a worker on the other side of the stage in the 100th posi-
ধon. To their right stands 99 other people—99 percent of all people on the stage with
lower earnings. They are in the 99th percenধle of the distribuধon, in the top 1 percent.
Finally, there is a point in the middle of the stage that splits everyone into two equal
groups, 50 percent of the distribuধon on either side. That point (or more precisely, the
earnings at that point) represents the 50th percenধle or the median of the distribuধon.
The mathemaধcs of increasing the number of people on stage from 100 to 200 to
1,000 to 150 million does not change—the middle of the ordered distribuধon is sধll the
median and the person standing at the 10 percent posiধon is at the 10th percenধle.
Because percenধles are independent of the populaধon, you can compare them across
any group such as country or industry.
While percenধles idenধfy a speci)c locaধon in the distribuধon, there are other met-
rics that will give you an overall measure of the distribuধon. The mean or average is
equal to the sum of all values divided by the number of observaধons. Because we add
up all the data, large values can generate a distorted picture of the true distribuধon.
In the example above, the mean would change dramaধcally if we replaced one of the
people on stage with someone who earned $100 million, but—take note—the median
would not change because that person would sধll stand on the far right edge of the
stage and the rest of the people in line would stay in the same posiধon.
The variance is another metric of a distribuধon and measures how far each obser-
vaধon in a data set is spread out from the mean value. A large variance indicates the
values in the data are far from the mean and from each other; a small variance, by con-
trast, indicates the opposite. A full decomposiধon of the variance and related formulas
are beyond the scope of this book, but if you are working with data and creaধng data
visualizaধons, it is certainly worth a bit of ধme and study to understand how they can
be used to beħer understand the shape and scope of your data.
DISTRIBUTIONփ 185
PYRAMID CHART
Most often used to show changes in population-based metrics such as birth rates, mortal-
ity rates, age, or overall population levels, pyramid charts put two groups on either side of
a center vertical axis. The pyramid chart is a subcategory of the diverging bar chart (see
page 92), but the name is reserved for comparing distributions, most often ages. As with the
diverging bar chart, the layout may cause some confusion, because your reader may assume
the leftward bars represent negative values, and the rightward bars represent positive values.
The advantage of the pyramid chart is that we can assess the overall shape of the distribu-
tion because both groups sit on the same vertical baseline. While many pyramid charts use
different colors for the two groups, that’s not a necessary characteristic.
Pyramid charts are a type of diverging bar chart, typically used to show population-based
metrics like birth rates, mortality rates, age, or overall population levels.
186 փ CHART TYPES
These pyramid charts show the distribution of ages in the United States and Japan in 2016.
In both graphs, women are represented on the left branch of the vertical axis and men on the
right. Each row represents a different age group: 0–4 years old, 5–9 years old, and so forth.
The shape of the chart means we can immediately see that Japan has a greater share of older
people, while there is a larger share of younger people in the United States.
Because the bars are not next to each other, it is difficult to compare the total shares of
men and women. But again, whether that’s a problem depends on the goal of your visual. If
you want your reader to compare the shares of the two genders, then a different chart type—
such as a paired bar chart or dot plot—would be a better choice. But if you want your reader
to see the overall shape of the distribution, the pyramid chart is perfect.
A natural alternative to the pyramid chart is the dot plot or the lollipop chart. Dots for each
gender are positioned along the horizontal axis corresponding to the data value and con-
nected by a line. Or, as shown on the right, we could simply use a lollipop chart (also see
page 80), replacing the bars with lines and dots. With either approach, we can still use dif-
ferent colors or just use a single color for the entire graph.
Alternatives to the basic pyramid chart are the dot plot or lollipop chart.
DISTRIBUTIONփ 187
Combining distributions in one chart can be accomplished by adjust colors and combining
different encodings.
One challenge with the pair of pyramid charts is to accurately compare the age distribu-
tions across the two countries. We can infer the main relationship, that Japan is, on average,
older than the United States, but it’s harder to make a more detailed comparison. Placing
the charts atop each other—a technique we saw earlier with the histogram—makes that task
easier. But be mindful that in using this approach the data for the two countries are shown
with different encodings (bars and lines), which can risk emphasizing one set over the other.
There are lots of types of uncertainty in data and statistics. Before we learn the differ-
ent ways to visualize uncertainty, it is worth pausing to understand what we mean by
the term. Even if you’re not a statistician or mathematician, it’s important to understand
how such uncertainty and measurement error can affect our results and, ultimately our
188 փ CHART TYPES
shows that reporting error (especially in government program participation) in some of the larg-
est, most trusted government household surveys has been increasing over the past several years.
We must also recognize that for this survey, the Census Bureau only asks a share of
Americans (so our calculations from these data also suffer from uncertainty from random-
ness). Consider all the reasons why that “sample” may not be truly representative. Maybe
some people don’t want to answer the survey. Maybe they moved and didn’t get the form, or
changed their phone number and didn’t get the call.
Whenever we work with data, we should consider how these kinds of uncertainty may
lead to some “error” around our final estimates. Not mistakes, but uncertainty. This error is
fundamental to being careful with data and ultimately visualizing and explaining our results
carefully. Line charts and bar charts suggest certainty with their sharp boundaries and crisp
edges, but that certainty is rarely actually the case.
In his book How Charts Lie, Alberto Cairo remarks that, “Uncertainty confuses many
people because they have the unreasonable expectation that science and statistics will
unearth precise truths, when all they can yield is imperfect estimates that can always be
subject to changes and updates.” We should not expect flawless data, and we should be ready
to explain those imperfections to our readers as best we can.
We now move to the specific challenge of conveying uncertainty around data estimates or
results from statistical models. This is a common problem: In a survey of ninety data visual-
ization authors and developers, information visualization researcher Jessica Hullman found
that graph creators did not include uncertainty in their work for four main reasons. First,
they did not want to confuse or overwhelm viewers. Second, they did not have access to
information about the uncertainty in their data. Third, they did not know how to calculate
uncertainty. And fourth, they did not want to make the data appear questionable. Hullman
argues that visualizing uncertainty is important because “a central problem is that authors
often omit or downplay information, such that data are interpreted as being more credible
than they are.” More effectively conveying such uncertainty—especially when making statis-
tical claims—builds trust and credibility.
This section will familiarize you with visual signals of uncertainty. There are many chart
types that can be used to show uncertainty around a central estimate, and this section intro-
duces a few of them: error bar charts, confidence interval charts, gradient charts, and fan charts.
190 փ CHART TYPES
ERRO R BARS
Perhaps the simplest and most common way to visualize uncertainty is to use error bars: small
markers that denote the error margin or confidence interval. Error bars are not really a visu-
alization on their own, but are an addition to other charts, often bar or line charts. The ends
of the error bars can correspond to any value you choose: percentiles, the standard error, the
95-percent confidence interval, or even a fixed number. And because error bars can convey
these multiple statistical measures, recent research has shown that this can invite confusion on
the part of the reader, making incorrect conclusions about the data. We must therefore clearly
label the intervals, either in a chart note or, preferably, on the chart itself.
This bar chart shows average earnings in each of thirteen industries in 2016. Error bars
denote the 25th and 75th percentiles.
The simplest and most common way to visualize uncertainty or distributions is to use error
bars, small markers that denote the margin of error or confidence interval.
DISTRIBUTIONփ 191
Applying error bars to bar charts raises a potentially interesting complication: some
research suggests that we tend to judge the points that fall within the bar as more likely than
those outside the bar (“within-the-bar” bias). In the previous chart, that would mean a reader
is more likely to assume that the salary of a worker in the Finance, Insurance, and Real Estate
(FIRE) sector is less than $80,000 and not more than $80,000. Other research has found that
we can better judge uncertainty and the distribution with other types of graphs, such as the
violin plot, stripe plot, or gradient plot.
While it may be a familiar approach for many readers and requires less data than some
of these other charts, existing research shows that we are not particularly good at assessing
uncertainty through these kinds of visual approaches.
CONFIDENCE INTERVAL
A confidence interval chart typically uses lines or shaded areas to depict ranges or amounts
of uncertainty, often over time. The basic confidence interval chart is literally a line chart
with three lines: one for the central estimate, one for the upper confidence interval value,
and another for the lower confidence interval value (these upper and lower lines can be
confidence intervals, standard errors, or a fixed number). The lines may be solid, dashed, or
colored, but if the central estimates are the primary numbers of interest, that line should be
thicker or darker to highlight it against the confidence interval values.
MEDIAN INCOME IN THE UNITED STATES FROM 1967 TO 2017 MEDIAN INCOME IN THE UNITED STATES FROM 1967 TO 2017
70,000 70,000
65,000 65,000
60,000 60,000
55,000 55,000
50,000 50,000
45,000 45,000
40,000 40,000
1967 1972 1977 1982 1987 1992 1997 2002 2007 2012 2017 1967 1972 1977 1982 1987 1992 1997 2002 2007 2012 2017
Lines or shaded areas around a central line can visualize ranges of uncertainty.
192 փ CHART TYPES
The two charts on the previous page show median earnings in the United States between
1967 and 2017. The standard error around those estimates is depicted by two separate lines in
the left graph and by a shaded area in the right.
GRADIENT CHART
A gradient chart (sometimes called a stripe plot) shows distributions or differences in uncer-
tainty. There are many ways to use the gradient plot, but the basic technique is to plot the
primary number of importance and add a color gradient on one or both sides to visually
demonstrate the measure of uncertainty around that single point. The plot is named not
necessarily after the shape of the graph but after the use of the color gradient.
The gradient plot can show changes over time or, as in the graph here, the distribution
around individual observations. This gradient plot shows the exact same data as the error
75,000
50,000
25,000
0
Information FIRE Transportation Wholesale Education Other Arts &
& Warehousing Trade Services Entertainment
Public Professional Manufacturing Agriculture Active Retail
Admin. & Scientific Military Trade
Gradient charts use a color gradient on one or both sides of the primary number of interest
to show distributions or uncertainty.
DISTRIBUTIONփ 193
bar chart above, but instead of bars with error lines jutting out in both directions, average
earnings are encoded with the dark horizontal line and the 25th to 75th percentiles of the
distribution are shown with the gradient. The color gradient might illustrate, for example,
multiples of the standard error, which can be a signal to the reader that the outcomes are less
certain the further they are from the central estimate line.
Stripe charts can also be an effective way to show changes over time. A strong example of this
is the series of stripe charts created by Dr. Ed Hawkins, a climate scientist at the University of
Reading, that showed temperature changes from 1850 to 2018. Each bar (stripe) showed a differ-
ent temperature level, ranging from cooler blues to hotter reds. Together, readers could quickly
and easily see the marked increase in temperatures around the world as a whole and in their
specific region of the world by using an online tool. These stripe charts were published by a
multitude of websites, television stations, and even became a cover of the Economist magazine.
In a 2019 interview on the Data Stories podcast, Hawkins said that he “was looking for a
way to communicate to audiences that aren’t used to seeing graphs, or axes, or labels—things
that we see day-to-day, but are complicated to them. It may look too mathematical to them,
so it turns them off straight away.” In a later interview on that same podcast, Jennifer Chris-
tiansen, the senior graphics editor at Scientific American, described them as, “every region’s
version of the climate stripe pattern progresses from cool blue to a warm red. No labels
This stripe chart from ShowYourStripes.com shows global temperatures from 1850 to
2018. The simple colored stripes are easy to see and understand.
194 փ CHART TYPES
are needed; no caption is needed. It’s a visceral and accessible nod to our warming planet
with color representing annual temperature. And it prints legibly on everything from social
media profiles, to pins, neckties, magazine covers, mugs, and concert screens.”
FAN CHARTS
If the color or saturation of the shaded area between the confidence interval lines changes
based on the value, it is often called a fan chart. Fan charts are like gradient plots for line
charts, and they are great for visualizing changes in uncertainty over time. In the fan chart,
values closest to the central estimate are the darkest and they lighten as the values move
outward. The use of color distinguishes the move from higher levels of statistical confidence
to lower levels. The advantage here is that it signals to the reader how the estimates become
less certain the further they are from the central estimate.
65,000
60,000
55,000
50,000
45,000
40,000
1967 1972 1977 1982 1987 1992 1997 2002 2007 2012 2017
Like the gradient chart, the fan chart shows distributions (in this case, standard errors)
around a central estimate.
DISTRIBUTIONփ 195
This fan chart shows the change in median household income over the last fifty years.
The color bands show the standard error divided into eight segments, though they could
also show bands of percentiles or other measures. Similar to the gradient chart, the change
in color saturation can show multiples of the standard error and signal less certainty as we
move farther from the central estimate, which is here represented by the black line.
Һ
One last strategy to suggest uncertainty is not a visualization technique per se but a design
technique. The hand-drawn, “sketchy,” “gooey,” or “painty” techniques can be used to add an
uneven edge or fuzziness to graph objects that will create a sense of uncertainty or impreci-
sion. Research suggests that sketchy graphs generate more engagement in the graph and that
we can “tie sketchiness to uncertainty or significance values.” These two examples, the first
from journalist Mona Chalabi at the Guardian and the second from Jo Wood at the Univer-
sity of London, both demonstrate these techniques in action.
ҺҺ
When you visualize the distribution of your data, you can show the entire distribution or just
specific points within it. The box-and-whisker plot (or boxplot), originally called a schematic
plot by its inventor John W. Tukey, uses a box and line markers to show specific percentile
values within a distribution. You can also add markers to show outliers or other interesting
data points or values. It is a compact summary of the data distribution, though it displays
less detail than a histogram or violin chart.
The basic box-and-whisker plot consists of a rectangle (the box), two lines (the whiskers)
that emanate from the top and bottom of the box, and dots for outliers or other specific data
points. Most standard box-and-whisker plots have five major components:
Each of these components can vary depending on which parts of the distribution we wish
to show. Some creators replace outliers with fixed quantiles such as the minimum and maxi-
mum values or the 1st and 99th percentiles. Some use the semi-interquartile range (Q3-Q1)/2,
which can generate asymmetric whiskers. And some add other descriptive statistics like the
mean or standard error. We can also vary the color, line thickness, and how and which parts
of the chart are labeled.
As a practical example, the box-and-whisker plots on the next page show the distribution
of earnings in our thirteen industries. The vertical line in the middle of each box represents
the median and the edges of the box represent the 25th and 75th percentiles. The ends of the
lines (the whiskers) show the 10th and 90th percentiles.
The graph on the left sorts the industries alphabetically while the one on the right sorts
them by the median value. I generally find it preferable to present data sorted by their val-
ues as opposed to alphabetical or some other arbitrary sorting. There may be cases when
alphabetical sorting is best—for example, if we were presenting earnings across all fifty U.S.
states, readers would find it easier to find individual states if they’re sorted alphabetically. If,
however, the goal of that graph is to discuss the high/low earnings of a particular state, we
would sort the data by their values to make those comparisons easier.
As with all of these visualizations, showing percentiles and statistical or data uncertainty
will often depend on the experience, interest, and expertise of our audience. In scientific or
research applications, for example, communicating uncertainty is especially important to
demonstrate whether a finding is statistically meaningful. But in other cases, where our data
only have a single value for each observation—say, a single estimate of per capita GDP in the
United States—we may not be able to visualize parts of the distribution.
In the case of the box-and-whisker plot, by plotting these specific percentile points, we
are explicitly deciding not to visualize the entire distribution. This may not be entirely prob-
lematic, especially if other percentile points aren’t particularly important, or if the data fol-
low a fairly standard distribution. But we must always fully explore the data we’re certain
we’re not hiding important patterns from our reader—or ourselves!
140,000 140,000
120,000 120,000
100,000 100,000
80,000 80,000
60,000 60,000
40,000 40,000
20,000 20,000
0 0
Active Arts & FIRE Manufacturing Professional Retail Wholesale Information FIRE Transportation Wholesale Education Other Arts &
Military Entertainment & Scientific Trade Trade & Warehousing Trade Services Entertainment
Agriculture Education Information Other Public Transportation Public Professional Manufacturing Agriculture Active Retail
Services Admin. & Admin. & Scientific Military Trade
Source: U.S. Census Bureau Warehousing Source: U.S. Census Bureau
Note: FIRE = Finance, Insurance, and Real Estate Note: FIRE = Finance, Insurance, and Real Estate
These charts show the distribution of earnings in thirteen industries either sorted alpha-
betically (left) or by median value (right). The edges of the box show the 25th and 75th
percentiles and the whiskers show the 10th and 90th percentiles.
DISTRIBUTIONփ 199
Candlestick or stock charts look like box-and-whisker plots, but they visualize different
content. Whereas box-and-whisker plots visualize uncertainty or a distribution, candlestick
charts visualize changes in the prices of stocks, bonds, securities, and commodities over
time. Bars and lines show opening and closing prices and highs and lows in a day, plotted
along a horizontal axis that measures time.
There are two elements of the candlestick chart. The central box—sometimes called the
“real body”—shows the gap between the opening and closing prices. The lines that extend
upward and downward from the real body—sometimes call the “wick”—show the low and
high value for the day. Like the box-and-whisker chart, the candlestick chart includes spe-
cific points and does not show all of the activity during the day, such as price volatility.
The candlestick chart is like a box-and-whisker chart but is typically reserved to refer to
prices of stocks, bonds, securities, and commodities over time.
200 փ CHART TYPES
Specific characteristics of the candlestick chart can vary in some obvious ways: color can
be changed to differentiate between a drop in price during the day (i.e., the closing price
is greater than or less than the opening price) and icons or other symbols can identify the
high and low prices. Although I’ve placed this chart in this chapter, because of its relation
to box-and-whisker plots, it could easily have also appeared in the Time chapter or even the
Comparing Categories chapter.
The candlestick chart on the previous page shows overall daily trading patterns for shares of
Alphabet, Inc.—the parent company of the Google search engine—from January to February
2018. The bar underneath shows trading volume. In both graphs, blue bars signal an increase
in price over the day and yellow bars signal a decrease. Notice how the two graphs are stacked
together as opposed to using a dual axis chart, which might be confusing or just plain cluttered.
VIOLIN CHART
Instead of showing selected percentiles from the distribution, the goal of the next set of
charts is to show the entire distribution. Unlike the box-and-whisker plot, in which we
choose specific points in the distribution, or the histogram in which values are grouped
together into intervals, the violin chart shows the shape of the whole distribution.
These violin charts use the same data as above, the average earnings in thirteen industries
in 2016. The thicker areas mean that there are more values in those sections while the thinner
parts imply lower frequency of observations. I’ve added a dot in the middle to mark average
earnings within each industry. Notice again the differences in the view when the chart is sorted
alphabetically (on the left) versus by the mean income (on the right).
KERNEL DENSITY
One consideration in creating this chart type is that it requires estimating what is called the
kernel density of each distribution. Kernel densities are a way to estimate the distribution of
a variable—akin to a histogram—but can be smoothed or made to look more continuous
using different algorithms. For the violin plot, those density estimates are plotted to mirror
each other around an invisible central line.
Think of it this way: A histogram plots a summary view of a distribution along a single
axis. The violin plot mirrors a smoothed version of the histogram on either side of that single
DISTRIBUTIONփ 201
Instead of showing select points (percentiles) in a data distribution, the violin chart shows
the estimated shape of the entire distribution using kernel densities.
axis. How that smoothing is accomplished will depend on what sort of kernel density esti-
mator you choose, which can vary based on the data, underlying function, and more.
Violin charts, then, are richer than the box-and-whisker plot, but can also be more diffi-
cult to create and for our audience to understand. In modern versions of Excel, for example,
the box-and-whisker plot is a default graph, while a violin plot requires manually calculating
the probability densities and then finding a graphing solution.
RIDGELINE PLOT
The ridgeline plot is a series of histograms or density plots shown for different groups aligned
along the same horizontal axis and presented with a slight overlap along the vertical axis. In
a basic sense, the ridgeline plot is like a small multiples histogram or a horizon chart where
the histograms are aligned in particular way.
The ridgeline plot on the next page shows the earnings distribution across the thirteen
different industries. The horizontal axis is shared across all thirteen industries and the dis-
tributions sometimes overlap along the vertical dimension. Depending on the color scheme
and density of the data, there may be more or less overlap between the series, but as we have
seen in other graph types (for example, sparklines and horizon charts), sometimes showing
the overall pattern is more important than the reader being able to pick out all of the specific
202 CHART TYPES
Retail Trade
Other Services
Active Military
Education
Agriculture
Manufacturing
Wholesale Trade
Information
Professional, Scientific
FIRE
The ridgeline plot is a series of histograms shown for different groups aligned along the
same horizontal axis and presented with a slight overlap along the vertical axis.
values. These overlaps can be a problem, but one that is mitigated by how quickly and eas-
ily readers see how the different distributions line up with one another along the same axis.
The most famous ridgeline plot is one that you didn’t even know was a ridgeline plot: The
album cover for Unknown Pleasures, the 1979 debut album of English post-punk band Joy Division,
which had white lines on a black background with no band name, album title, or other identifiers.
In 2015, Jen Christiansen, the Senior Graphics Editor at Scientific American, tracked down
the original image to the 1970 doctoral dissertation of Harold D. Craft, Jr., a radio astrono-
mer at Cornell University. The original chart graphed the distribution of consecutive radio
pulses emanating from a pulsar, a type of neutron star. The album cover designer, Peter
Saville, called it “a wonderfully enigmatic symbol for a record cover.”
Data that lend themselves well to a ridgeline plot are those in which the distributions dif-
fer from one category (row) to another so that the reader can see the shift up and down the
page. With data in hand, variations on color, font, and layout can help engage and interest
your reader. The ridgeline plot on the next page was published by the Guardian in 2018 and
DISTRIBUTIONփ 203
Source: Photograph by Jen Christiansen (featuring figure 5.37 from “Radio Observations of the Pulse
Profiles and Dispersion Measures of Twelve Pulsars,” by Harold D. Craft Jr., September 1970).
shows the distribution of the gap in pay between men and women in more than ten thou-
sand companies and public bodies in the United Kingdom. Using different colors on either
side of the vertical zero-percent line (perfect equality) and sorting the data from industries
that have the highest pay gaps (e.g., Construction) to smallest pay gaps (e.g., Accommoda-
tion and Food Services) also helps direct the eye.
This ridgeline plot from the Guardian shows earnings distributions for men and women in
different industries.
204 CHART TYPES
The graphs presented in this chapter so far summarize data distributions with lines, points,
bars, and color. To different degrees, this is what the histogram, violin, and ridgeline plots all
do. Another way to visualize the distribution of your data is to just show the data.
STRIP PLOT
The basic way to show your data is with what is known as a strip plot. In this graph type, the
data points are plotted along a single horizontal or vertical axis.
In this example, average earnings for each state are shown for each of the thirteen indus-
tries (the U.S. average is denoted with the vertical black line). We have already seen similar
data presented in the box-and-whisker plot and violin chart, but here you can see individual
FIRE
Information
Professional, Scientific
Wholesale Trade
Public Administration
Manufacturing
Agriculture
Education
Active Military
Other Services
Retail Trade
In a strip plot, data points are plotted along a single horizontal or vertical axis. This strip
plot encodes the data with circles, but small lines are also often used.
DISTRIBUTIONփ 205
points. Notice, however, that I’m showing average earnings for the fifty states, not for indi-
vidual workers. Plotting earnings for everyone in my data (some 1.3 million people) would
look like a single, dark line. There are too many values.
It’s true that some of the data is obscured, but, especially by virtue of the overlapping
transparent colors that make the patterns darker, it becomes clearer where the bulk of earn-
ings lie in the distribution. There’s no rule for how many data points are too many, but as you
plot your data, you can always tell when you’ve passed that threshold.
This static image from an interactive strip plot from NPR is a good example of how this
visualization can be richer than a standard bar chart or histogram. Here, they plot a point
for every school district in every state. Darker orange dots (and below the black horizontal
line) are districts in which spending per student is below the national average. Darker green
dots are districts in which spending per student is above the national average. An interesting
and useful design choice to make the dots transparent (with a solid border) lets us see where
districts are close enough to overlap. Also notice that only Alabama, Florida, and Alaska are
This strip plot from NPR shows the distribution of spending in different school districts
around the United States.
206 CHART TYPES
labeled along the horizontal axis. Those three states are explicitly annotated in the chart.
Other state labels only appear along the horizontal axis when, in the interactive version, the
user hovers over a stack of circles.
BEESWARM PLOT
If we want to plot individual data points rather than distributions, one way to make the data
more visible is to use a technique called “jittering.” This is when we alter individual values
slightly so that the data points don’t lie on top of one another.
Consider the strip plot on the left, which leaves all of the data along the same horizontal axis.
We can see clusters, but not all of the individual values. In the version on the right, the data are
jittered along the horizontal and vertical axes to help make each point visible. There are differ-
ent algorithms and approaches to jittering the data, and the most important consideration is
to manipulate the values just enough to make them visible but not so much that it changes the
overall view of the distribution. As with choosing a kernel density estimator for violin charts,
the choice of jittering technique (for example, do we jitter both x and y variables and if so, do
we jitter each variable independently?) will depend on the data and its underlying distribution.
100 105 110 115 120 125 130 100 105 110 115 120 125 130
Jittering doesn’t work in every case, of course. We are limited by how many points we can
plot. With too many data points, jittering would require so much movement that it would
modify the underlying distribution. Plotting everyone’s earnings in each industry, for exam-
ple, creates a mass of dots that requires so much jittering that it modifies the presentation
of the data by moving the points too far away from their true position. But showing average
earnings in each of the fifty states across the thirteen industries is not as overwhelming, and
you can see where the bulk of the distribution lies in each. Of course, if you’re interested in
just showing that there are a lot of points in your dataset and you can maintain the overall
shape of the data, plotting many points may help make your exact argument.
To create what is called a beeswarm plot—because the clustering of the data points resembles
a swarm of bees—we jitter the values so that they don’t overlap and each point is visible. As
with other charts in this chapter (and coming up in Chapter 8), there are different calculations
DISTRIBUTIONփ 207
we can use to arrange the dots—for example, arranging the points in increasing order or plac-
ing them in a square or hexagonal grid. Here, similar to the ridgeline plot, each industry shares
the same horizontal axis so that we can easily compare across the different sectors.
Notice how I have added some simple annotation and labeling to some of the outlier
values. These clearly stand out in the graph, and a curious reader will wonder what’s going
A beeswarm plot “jitters” all of the points in the data set so they don’t overlap and each
point is visible.
208 փ CHART TYPES
on with those points. Are they errors? If not, what is that state, and why are earnings so high
relative to the rest of the country? I haven’t labeled every outlier point, but enough to assure
the reader I’ve thought about those values.
Beeswarm plots can also show changes over time. This beeswarm plot from Axios—really
eight beeswarm plots aligned together—shows spending at properties owned by The Trump
Organization before and after the 2016 election. The combination of color (spending origin),
size (amount of spending), and the density of points (the time dimension) makes this an
effective visualization to show the patterns around Election Day.
This beeswarm plot from Axios shows spending at properties owned by The Trump Organi-
zation before and after the 2016 election.
DISTRIBUTIONփ 20 9
The wheat plot, developed and named by Stephen Few, is a richer version of what is called a
dot histogram or a Wilkinson Dot Plot (which is named after Leland Wilkinson author of the
seminal data visualization book The Grammar of Graphics, though Wilkinson himself actually
referred to these charts as histodot plots, a name that clearly did not stick). A Wilkinson Dot
Plot is like a regular histogram except that instead of showing a single bar encoding all of the
observations, data points are stacked on top of each other within their relative bins—some-
thing like combining a histogram and unit chart. With this approach, the points do not show
their actual value (measured along the horizontal axis) because they are stacked in a single
column. In other words, each dot represents an observation in each bin, not its actual value.
INCOME DISTRIBUTION
100 105 110 115 120 125 130 135 140 145 150 155 160 165 170
A dot histogram plots individual points within the bins of a histogram. Of course, it can
only handle so many individual points.
2 10 փ CHART TYPES
The Wilkinson Dot Plot can be modified to create a wheat plot in which we show the actual
data values, stacking them within their separate bins. The actual data values are plotted along
the horizontal axis—still grouped into their bins—and stacked vertically to show the total
number of observations. Stephen Few writes that, “The curved alignment of the dots is mean-
ingful, for it graphically displays the distribution of values within each interval, based on their
positions. Although this looks odd at first glance, it takes only a minute to understand and
learn how to read.” As with some of the previous distribution graphs, one of the limiting fea-
tures of the wheat plot is that too many data values may lie on top of each other.
The wheat plot on the next page shows the distribution of earnings for a single industry
for about two hundred workers. The histogram on the right is shown as a comparison—you
can still see the relative share of observations in each part of the distribution, but not the
actual data. There is an obvious tradeoff. On the one hand, the wheat plot shows more detail
for the reader to explore and the graph can look more interesting and engaging. On the other
hand, the histogram is likely more easily and immediately understandable to readers.
INCOME DISTRIBUTION
100 105 110 115 120 125 130 135 140 145 150 155 160 165 170
Note: Sample distribution of 200 workers.
The wheat plot, designed by Stephen Few, adjusts the dot histogram by showing the exact
values, still within each bin.
DISTRIBUTIONփ 211
The tradeoff between wheat plots and simple histograms: the wheat plot has more detail
but may be harder for people to understand.
We can see the difference between say, a wheat plot and a ridgeline plot in this visualiza-
tion from the Guardian. Leading off the same article that has the ridgeline plot on page 201,
this graph includes a dot for every company in their data set. It doesn’t quite have the “lean”
that a true wheat plot might have (because there are so many points), but you get a good
sense of the overall distribution and the greater number of firms towards the right side of the
graph. Plotting each individual company also lets the chart creators add labels to highlight
This wheat plot from the Guardian includes a dot for every company in their data set.
2 12 CHART TYPES
RAINCLOUD PLOT
Sometimes it’s useful to show both the distribution density of your data and the actual data
points. The raincloud plot, perhaps first named by neuroscientist Micah Allan, shows the
distribution (think violin chart) with the actual data plotted below. In this arrangement, it
looks like a cloud raining data.
Raincloud plots show us a summary of the data and all the individual data points, so we
can spot outliers and patterns. Again, the tradeoff is that this might require more work on
the part of the reader to understand how to read the graph.
This raincloud plot shows the distribution of earnings across the fifty states with data values
plotted below.
While the raincloud plot may seem like an esoteric chart—and, to be honest, right now it is—
there are certainly scenarios and data for which this chart would be a useful choice.
Education
Professional, Scientific
FIRE
Information
Transportation and
Warehousing
Retail Trade
Wholesale Trade
Manufacturing
Agriculture
The raincloud plot shows the summary histogram with the actual data plotted below.
DISTRIBUTIONփ 2 13
An example of a raincloud plot from Len Kiefer that shows the distribution
of weekly mortgage rates for different years. The visualization gives you
both an overall summary view of the data and shows the
specific data points.
This raincloud plot from Len Kiefer, the Deputy Chief Economist at Freddie Mac, shows
the distribution of mortgage rates from 2010 to 2019, with weekly observations shown in the
dots just below. This view gives us both an overall perspective of the data and the detailed
values in the points below.
214փ CHART TYPES
ҺҺ
The stem-and-leaf plot is a table that shows the place values of each data value. Values are
typically shown listed down the “stem” column with the first digit or digits. The rest of the
table is reserved for the “leaf ” that shows the last digit (or digits).
Stem-and-leaf plots show the place values of each data value. They are sometimes
used in transportation schedules, like this train schedule from the Tokorozawa
station in Saitama, Japan.
DISTRIBUTIONփ 2 15
As an example, take a simple dataset with just seven values: 4, 9, 12, 13, 18, 24, and 27. The
data are arranged in downward-ascending order with the first digit on the left side and the
second (tens) digit on the right. Obviously, for more detailed and complex data, the stem-
and-leaf plot may not be a useful approach.
Stem-and-leaf plots are most useful as a reference tool, like a public transportation schedule,
or to highlight basic distributions and outliers in a more limited set of data. The Japanese train
schedule for the Tokorozawa station in Tokyo on the facing page shows the timing of train
arrivals over the course of a day. The hour digit is shown in the far-left column, minutes are
shown to the right. The first train begins running at 5:00 am, the next train leaves at 5:08 am,
then 5:18 am, and so on. Because the stem-and-leaf plot is a table, it loses some of the advantages
of a traditional data visualization, but the leaves illustrate some basic view of the distribution.
CONCLUSION
The collection of graphs in this chapter demonstrates how we can show the distribution of
our data or uncertainty around specific values. Some of these charts show summary mea-
sures or specific values. We can aggregate the distribution into bins to visualize the distri-
bution in a histogram. Or we might use specific percentiles to generate a box-and-whisker
chart, for example, or show stock price variation in a candlestick chart.
With better data visualization tools and faster computers, we can show more data than
ever before. Beeswarm charts, wheat plots, and raincloud plots include specific data points.
While these visualization types are useful for presenting the full data set to our readers, they
have their limits: We can only show so many data points before they begin to overlap and
obscure one another.
The graphs in this chapter can introduce challenges to readers who are not familiar with
statistical concepts and measures of dispersion. As always, the most important thing you can
do when creating your graphs is to remember your audience. If you are a PhD economist
presenting your work at your university lunchtime seminar, you don’t need to explain the
median or variance or even the 95-percent confidence interval. If you were to present the
same results to a general audience, however, you would include definitions and annotation.
This isn’t to say you should avoid presenting these numbers or that you need to dumb things
down, but rather that you may need to take time explaining concepts within the visual. The
planning, testing, and conceptualizing of your visualization will pay off in the long run as
you more effectively communicate your work with your audience.
GEOSPATIAL
7
T
here is an obvious advantage to plotting geographic data on a map—people can find
themselves in the data. They can literally see themselves in the data, a connection with
the subject matter that other visualizations cannot muster. Plotting geographic data may
mean adding color to geographic areas likes states or countries, or adding circles, squares,
lines, or other shapes on top of a geographic map.
Data-driven maps are not new. In 1922, the “Maps and Sales Visualization” on the next
page shows the reader thirty-six different ways to place data on a map. The author writes:
The use of maps has to do with the visual representation of space. Therefore, in all map work
we start with an outline . . . The fact that the Earth is a globe makes visualization of maps a dif-
ficult process; arbitrary methods have to be used to get the surface of the ball represented in a
flat picture.
This chapter begins with some of the basic challenges of visualizing geographic data
and then presents some alternatives to the basic map, the modern version of this 1922
visualization.
The maps in this chapter use a color palette often used by the Washington Post. Maps of
the U.S. political system in this chapter use the basic red-blue color palette used by many
newsrooms.
2 18 փ CHART TYPES
E.P. Hermann’s (1922) “Maps and Sales Visualization” shows an early example of the multi-
tude of ways to place data on a map.
Aaron Koblin’s Flight Patterns is one of those maps that can entrance the reader. Koblin plots
all flight paths in the skies above the United States in twenty-four hours. The static version of
the map (there is an interactive version in which you can zoom into any area of the country)
shows the entire country, major airports, and the activity of the skies above. It’s not a visual-
ization that ranks the biggest airports or tells you how to avoid delays, but it quickly shows
the traffic patterns in American skies.
GEOSPAT IALփ 2 19
Aaron Koblin’s Flight Patterns project, which consists of static and interactive maps, shows
all flight paths in the skies above the United States over twenty-four hours.
There are some distinct challenges when creating maps. The biggest is that the size of a
geographic area may not correspond to the importance of the data value. Russia is more than
6.6 million square miles, almost twice the size of Canada, and so it takes up a lot of space
on a map. At 270,000 square miles, Texas is roughly the size of California and Colorado put
together, but it’s actually less than half the size of Alaska (665,000 square miles), which you
might not know because most maps of the United States tend to distort it and arbitrarily
position it out to sea, south of California. The point is that the data values for Russia, Texas,
and Alaska may not correspond to their importance in the data, and the map can distort our
perception of the important values being visualized.
I find that many people should approach their desire to create a map with more skepti-
cism and critical thought. Is a map truly the best way to present geographic data? Or are you
just showing where people live? Does it show the relationships we want to explore or are we
simply relying on the fact that we have geographic identifiers? This chapter will explore the
perceptual issues of maps and why they are not always the best medium through which to
demonstrate our points. This isn’t to say we should never make a map—in many cases we
need to make a map to better understand our data—but, especially with maps, we should
always take a step back and consider whether it is the right visualization choice.
The message is simple: there are many ways to present geographic data, and there are lots
of objects, shapes, and colors you can add to maps. Which kind of map you use to visualize
your data will depend on two questions: How important are the geographic patterns? And
how important is it for your reader to see a familiar map?
220 փ CHART TYPES
CHOROPLETH MAP
Perhaps the most familiar data map has perhaps the most unfamiliar name: a choropleth
map. Choropleth maps use colors, shades, or patterns on geographic units to show propor-
tionate quantities and magnitudes. You likely already know how to read this choropleth
map of per capita GDP around the world—it’s a simple and recognizable shape that lets you
quickly and easily find countries (and, by extension, yourself) in the visualization.
This color palette is also easy to understand—smaller numbers correspond to lighter col-
ors and larger numbers to darker colors (and what is sometimes called a “color ramp”). More
often than I care to count, map creators will use an incorrect color palette. For example,
for this map, someone might use a diverging color palette in which colors progress out-
ward from a central midpoint. Unless we are comparing per capita GDP above and below
some midpoint number, such as the average GDP, the diverging color palette is a bad choice.
Instead, we should follow what has become a simple standard of lighter colors to darker
colors. We discuss color palettes in more detail in Chapter 12.
The choropleth map is perhaps the most familiar data map. Colors correspond to data
values and are assigned to the different geographic areas on the map.
GEOSPAT IALփ 22 1
It may be bit difficult to find smaller countries or countries that you don’t already know,
but the overall shape is familiar, well-known, and easy to understand. Did you know Luxem-
bourg had the highest per capita GDP in 2017 at more than $104,000 per person (for refer-
ence, per capita GDP in the United States was $59,500)? It’s a country of only one thousand
square miles (for reference, France is almost 250,000 square miles) and difficult to find on
this map, but it has the highest per capita income.
Maps like these introduce a geographic distortion—the size of the geographic area may
not correspond to the importance of the data value. Even with this distortion, however,
maps are an easy and familiar way to present geographic data to our readers.
There are a variety of alternative map types that correct this geographic distortion. Carto-
grams, for example, resize the geographic units according to their data values (see page 233)
and a tile grid map uses a series of equal-sized squares (see page 238), not to mention the
other chart types we can use to plot geographic data, such as a heat map (see page 112). The
tradeoff with any of these alternative approaches is that the map is no longer as familiar to
the reader as the standard map. But, as Kenneth Field, author of the data visualization and
cartography book Cartography, once noted, “None of these maps are right and none of these
maps are wrong. They are all just a different representation of the truth.”
One challenge with mapping data is the map projection the creator chooses. The world is a
globe, but maps are flat. A mapmaker must choose a map projection to transform the world’s
sphere into a two-dimensional plane. All maps distort the surface of the planet to some
degree and there is considerable debate about which projection does the best job depicting
the earth in two dimensions.
The one you are likely most familiar with is the Mercator projection. This is the map used
in early versions of Google Maps and is the default in many data visualization tools like
Tableau and PowerBI. Developed in 1569 by Flemish geographer and cartographer Gerardus
Mercator, it became the standard map projection for nautical purposes. A sailor could draw
a straight line between two points and measure the angle between that line (called a rhumb
line) and the meridian (the vertical line that runs between the north and south poles) to find
their bearing. While it may be useful for sailing the seas, the Mercator projection distorts the
size of objects as the latitude increases from the equator to the north and south poles. Thus,
countries closer to the poles—like Greenland and Antarctica—appear much larger than they
actually are. In the Mercator map shown on the next page (on the left), Greenland looks to
222 փ CHART TYPES
Map projections can influence our perception of the map. The Mercator projection on the
left, for example, looks considerably different than the Robinson projection on the right.
Source: Wikimedia user Strebe.
be about the same size as South America, when it is in fact about one-eighth the actual size.
In the Robinson projection on the right, country areas are closer to their true sizes.
There are three major categories of map projections:
CONIC
Conic maps are as though a cone was placed over the Earth and unwrapped.
Conic projections are best suited for mapping long east-west geographies, such
as the United States and Russia, because the distortion is constant along common
parallels. The Albers Equal Area Conic and the Lambert Conformal Conic are two
of the more well-known projections.
CYLINDRICAL
Cylindrical maps work like conic maps but use cylinders instead of cones. Like the
Mercator projection, cylindrical maps inflate geographic areas farther from the
center.
GEOSPAT IALփ 223
PLANAR ӇAZIMUTHALӈ
With this approach, the planet is projected on a flat surface. All points are at the
same proportional distance from the center point, such as the north pole, but the
distortion gets larger as you move further from that center point.
To some degree, all maps distort the surface of the planet. Alberto Cairo’s diagram from
The Truthful Art shows a selection of different types of map projections.
22 4 փ CHART TYPES
There isn’t necessarily a right or wrong map projection, though I’m sure some cartog-
raphers would disagree! But each has tradeoffs, and serious mapmakers dig deep into the
different properties of these projections and weigh the different options. Many people in
the data visualization field shy away from the Mercator projection because of its obvious
drawbacks, though it can work well for small areas. For the United States, you can tell a map
is using the Mercator projection because the top left border of the country is a straight line.
The Albers projection, by comparison, preserves the east-west perspective, which you can
see by the slight curvature in the northern border.
Notice how the northern border of the United States looks straight in the Mercator map on
the left but is more curved in the Albers projection on the right.
As we start to add data to a choropleth map, our first consideration is the choice of intervals
(or “bins”) that will shade the geographic units. Placing data into discrete categories is, at its
core, an aggregation problem. By combining several states or countries into a single bin, we
don’t know how different those units are from one another.
The map of the United States on the facing page, for example, shows median household
income in 2018 in each state. How the states are placed into groups (the “bins”), the map
shading, and ultimately our reader’s perception of the data depends on our choices. Mas-
sachusetts ($86,345) and Maryland ($86,223) had the highest median household incomes
in 2018 and fall into the highest bin with the darkest color; New Mexico ($48,283) and
GEOSPAT IALփ 225
Mississippi ($42,781) are at the other end of the distribution and are shown with the light-
est shades.
There are four primary binning methods for creating maps.
NO BINS
This is essentially a continuous color palette (or “ramp”) in which each data value receives its
own unique color tone. On the one hand, this is easy because we don’t need to think too much
when we create the map—the colors ramp up from the lightest color for the lowest value to
the darkest color for the highest value. On the other hand, the resulting color gradient may
generate spatial patterns masked by subtle changes in color. In this example, it’s hard to dis-
tinguish the differences between Iowa ($68,718), Nebraska ($67,515), and Wyoming ($62,539).
80,000
70,000
60,000
50,000
The continuous color palette (or “ramp”) seamlessly goes from lighter colors (smaller val-
ues) to darker colors (larger values).
226 փ CHART TYPES
In maps with a discrete number of bins, the default approach is to typically divide the data
range into an equal number of groups. For example, in a map with four bins and a data range
from 1 to 100, we end up with four equal groups (1–25, 26–50, 51–75, and 76–100).
This approach more clearly distinguishes geographic units (such as states) than the
continuous (no bins) option, but it can mask the magnitudes of those changes by putting
states in the same or different bins. In cases where the distributions are highly skewed, this
approach may unevenly distribute the geographic units across the bins. In this map, the bins
are split into equal units of $10,000, which results in five states in the bottom category, ten in
the next, seventeen in the middle, fourteen in the next, and five in the top category.
>$80,000
$70k–$80k
$60k–$70k
$50k–$60k
<$50k
Dividing the data into equal intervals—such as $10,000 gaps—is one way to add color to a
data map.
GEOSPAT IALփ 227
We could also cut the data into different bins. For example, instead of having a bin at equal
intervals, we could arrange the bins to hold the same number of observations, such as quar-
tiles (four groups), quintiles (five groups), or deciles (ten groups). Or we could use other
measures to collapse the data into groups, such as the variance or standard deviation.
The data distribution approach clearly shows differences between the geographic units,
but the created cutoffs may not be numerically meaningful. In this map that divides the
country into five equal groups (quintiles), Connecticut is just barely in the top group with a
median household income of $72,812, while Minnesota is placed in the next lower bin, even
though it has a very close income estimate of $71,817.
Q5
Q4
Q3
Q2
Q1
Another way to divide the geographic units is to divide the data into an equal number of
observations, such as quartiles (four groups) or quintiles (five groups).
22 8 փ CHART TYPES
ARBITRARY BINS
In this approach, the map creator chooses the bin cutoffs based on round numbers, natu-
ral breaks, or some other arbitrary criterion. This method lets us avoid some of the odd
breaks that might occur, such as the Connecticut-Minnesota example above, but it can also
be misleading. A method in which the selected bins are based on larger groups or round
numbers—even without looking at the data—might look like this:
>$80K
$60K–$80K
<$60K
Depending on the goals of your data map, you can also divide the data into an arbitrary
number of bins.
ALTERNATIVE OPTIONS
This isn’t to say any of these maps are right or wrong, but they highlight the importance of
binning decisions in choropleth maps. To make the best binning decision, we can draw on
Mark Monmonier’s 2018 book, How to Lie with Maps. Instead of using equally sized bins
arbitrarily or letting the software tool decide which breaks to use, try considering—and
showing—the actual distribution. If I add a column chart to the arbitrary bin method map,
GEOSPAT IALփ 229
the bin breaks and the distinct differences between the values is apparent. Adding another
graph takes up more space, but it gives readers a clear picture of the data.
Including the number of observations in each bin somewhere in the visualization can
also reveal the distribution to your reader. In this version, the legend is converted to a small
bar chart to show the number of observations in each bin.
One way to help readers better understand the data in a map is to pair it with another visu-
alization type, like a bar chart. This visualization has a small bar chart embedded within the
legend to make clear how many states are in each group. This is not a necessary compo-
nent, but one that can help the reader understand the distribution of the data on the map.
Another consideration when we create a map is how we label the bins. Let’s use the visualiza-
tion we just created of the map and bar chart of median household income.
230 փ CHART TYPES
In that map, the definition of the bins is arbitrary. The top bin is defined as “>$86,000”
but because the maximum value in the next category $81,346 (New Hampshire), the top bin
label could just as easily be “>$85,000,” “>$82,000,” or even “>$81,346.”
The legend for this map can be defined several ways.
1. Instead of round numbers, we could use the actual income amounts, which still
leaves us with arbitrary bins. In this case, it isn’t clear whether $86,000 is in the
fourth or fifth group. There are a few ways to do this, for instance with separate boxes
or a single image with labels just below. (One possible solution is to explicitly note
which bins are inclusive or exclusive of the upper and lower bounds.)
This has the advantage of clearly showing the data values—for example, we can see the
gap between $49,973 and $50,573 in the first two bins. The disadvantage is that the leg-
end is overly detailed and complex. As the reader, you might wonder why all this preci-
sion is necessary and, depending on the content, you might wonder what’s going on in
the gap between the maximum of one bin and the minimum of the next.
$42,781– $49, 974– $50,573– $62,630– $63,938– $74,177– $77,066– $81,347– $86,223–
$49,973 $50,572 $62,629 $63,937 $74,176 $77,065 $81,346 $86,222 $86,345
GEOSPAT IALփ 231
Although accurate and comprehensive, this legend feels busy and obscures the five origi-
nal bins by adding four non-data bins that connect them.
There is no one-size-fits-all solution to this challenge, but here we have laid out the issues
and tradeoffs of creating and labeling bins. Always consider the necessary level of precision
(i.e., the number of decimal places), the overall number of bins, and the smoothness of the
data (that is, if the data show no visible jumps, round-numbered bins may be fine).
SHOULD IT BE A MAP?
Before we start exploring other types of maps, let’s first ask ourselves whether a map accom-
plishes our goals and best communicates our argument. Many maps are made simply because
the creator has geographic data, not because the map is the best medium for that content.
Take this simple example. A 2016 Washington Post story examined the relationship
between rates of suicide and gun ownership in the United States. The story explains:
One 2006 study found that from the 1980s to the 2000s, every 10 percent decline in gun owner-
ship in a census region accompanied a 2.5 percent drop in suicide rates. There are numerous
other studies that show similar results.
This pattern becomes clear when looking state by state. The states that have higher rates of gun
ownership, where people have more access to guns, also have higher rates of suicide. Suicides are
twice as common in states with high gun ownership than those with low gun ownership, even
after controlling for rates of mental illness and other factors, according to a 2007 study.
Below these two paragraphs were two maps, which I have recreated on the next page as
choropleth maps. Do these maps help you see the positive relationship between the gun
ownership rate and the suicide rate? Can you pick out states that have the highest suicide
rates and the highest gun ownership rates? I can’t. Instead, I found myself jumping back and
forth between the two maps trying to identify individual states and regions.
Instead, what if we placed the same data in a bubble plot? We could put the suicide rate
on the vertical axis and the gun ownership rate on the horizontal axis, and scale the size of
the circles by population. Adding color to differentiate between areas of the country makes it
clearer that states in the upper-right area of the graph tend to be western and southern states,
where gun ownership is higher and, it turns out, gun-control laws are weaker.
GUN OWNERSHIP RATE
Note: The original maps in the Washington Post story were actually tile grid maps—see page 238. Data were
extracted from the Washington Post story and Miller et al. If you or someone you care for is in distress,
suicide prevention and crisis resources are available at the National Suicide Prevention Lifeline in the United
States at 1-800-273–8255. Many other countries have similar hotlines.
Bubble plots may be less familiar to readers than simple choropleth maps, but with a clear
title and a little annotation, this chart can better demonstrate the relationship between gun
ownership and suicide. As you prepare to create your next map, ask yourself, “Is a map the
best visualization to communicate my argument?”
CARTOGRAM
One way to adjust for the geographic distortion of a typical choropleth map is with a carto-
gram, which reshapes geographic areas based on their values. There is an obvious tradeoff
here: On one hand, these adjustments more accurately visualize the data because cartograms
correlate the data and the geographic size. On the other hand, these graphs are not like the
standard maps that we know and recognize. They are therefore not as intuitive for your
reader. Your decision about whether to use a standard map or a cartogram will, as always,
depend on your goals and your audience.
In his book Cartography, Kenneth Field summarizes the purpose of cartograms:
The intent of most thematic maps is to provide the reader with a map from which compari-
sons can be made, and so geography is almost always inappropriate. This fact alone creates
problems for perception and cognition. Accounting for these problems might be addressed
in many ways such as manipulating the data itself. Alternatively, instead of changing the data
and maintaining the geography, you can retain the data values but modify the geography to
create a cartogram.
There are four primary types of cartograms: contiguous, noncontiguous, graphical, and grid-
ded. One of the best ways to demonstrate the value of a cartogram is to examine the U.S.
electoral college. In the U.S. election system, each state is assigned a number of electoral
votes corresponding to its population, not its geographic size. Thus, states like Idaho, Mon-
tana, and Wyoming, which are very large in terms of square miles (325,412 square miles in
total) but home to relatively few people, only have ten electoral votes between them. Mas-
sachusetts, by contrast, has eleven electoral votes and is 7,838 squares miles, less than 2.5 per-
cent of the size of those three states. In this choropleth map of the 2016 presidential election,
Idaho, Montana, and Wyoming take up a disproportionate share of space on the map relative
to their electoral votes.
23 4 փ CHART TYPES
WA
MT
ND ME
OR MN VT
NH
ID WI
SD NY MA
WY
MI CT
RI
IA PA
NV NE NJ
OH
IL IN MD
UT DE
CO DC
WV
KS VA
MO
CA
KY
NC
TN
OK
AZ AR
NM SC
MS AL GA
TX LA
FL
AK
HI
A standard choropleth map, this one shows the results of the 2016 U.S. Presidential
election. Notice how much space large, but less populous states, like Idaho (ID), Montana
(MT), and Wyoming (WY) in the northwest are compared with smaller but more populous
states, like Massachusetts (MA) and Rhode Island (RI) in the northeast.
CONTIGUOUS CARTOGRAM
The contiguous cartogram adjusts the size of each geographic unit according to the data. In
the map at the top of the next page, for example, each state is sized to its number of electoral
votes (or population, if you’d rather think of it that way). The version on the bottom-left
of the next page uses squares to scale each state while retaining the original approximate
geographic location and borders. The third map scales the counties in each state according
to the vote share, which generates a more purple shade to the country, reflecting the split
between the two political parties. Each of these approaches distorts the overall shape of the
country as the data warps the geography, so they will look somewhat foreign to readers. The
tradeoff becomes clear here—we can more accurately scale the states according to their data
values, but the geography no longer looks as familiar.
To address the fact that large but less populous states take up a disproportionate amount
of space on the map, a cartogram scales the size of geographic units according to another
data value. This map scales the states to their number of electoral votes.
ME
AK
NH
VT
MA
WI
MI CT RI
NY
MN
WA
IA PA
IL
IN OH NJ
MT ND WV
OR ID NE
WY SD DE
MD
NV DC
KS MO KY
CO
UT VA
OK AR TN
NC
AZ NM MS
AL GA SC
TX
LA
CA
FL
HI
Other contiguous cartograms (using squares on the left or scaling counties on the right)
are alternative ways to try to overcome the geographic distortions that occur in the stan-
dard choropleth map. But these maps are almost surely to be less familiar to readers than
the standard map.
236 փ CHART TYPES
NONCONTIGUOUS CARTOGRAM
Now that you’ve seen a contiguous cartogram, you can probably guess what a noncontiguous
cartogram looks like. In this approach, the size of the geographic units are based on the data
value, such as population, but the units are broken apart and not kept adjacent to one another. In
this way, we maintain the shape of the individual units but distort the overall view. One advantage
of the noncontiguous cartogram is that we can build in more space for labels and annotation.
The map on the left scales each U.S. state according to its number of electoral votes and
includes color to denote which candidate won those votes. The exact shape of each geo-
graphic unit also isn’t necessary—the map on the right uses collections of squares for each
state, here scaled to the number of electoral votes. In this version, we can again see how
Idaho, Montana, and Wyoming look a lot smaller, while New York becomes much larger.
The noncontiguous cartogram was invented in the mid-1970s by Judy Olson, a geogra-
pher then at Boston University. “Probably one of the most interesting aspects of the non-
contiguous cartogram,” she wrote in her 1976 paper, “is that the empty area between units
is meaningful. If the highest-density unit is used as the anchor [in other words, if the most
dense geographic unit is used to scale all the other units], then the empty areas reflect the
degree of discrepancy between the density in the most-crowded unit and the density in
other units. The effect can be quite dramatic.”
N.H. Me.
Alaska Vt.
Mass.
Wash. N.Y.
Mont. N.D. Conn. R.I.
Minn.
Mich.
Wyo. S.D. Pa.
Ore. Idaho Wisc. N.J.
Colo. Neb.
Utah
Iowa Ohio
Ill. Ind. Md. Del.
Nev. Kan.
Okla.
Ga. N.C.
Ky.
Ark.
Texas Tenn.
Calif.
S.C.
Fla.
Hawaii Miss.
La. Ala.
Noncontiguous cartograms like these break up the geographic areas. This helps for adding
labels and annotation, but is a more unfamiliar chart type.
GEOSPAT IALփ 237
GRAPHICAL CARTOGRAM
Graphical cartograms do not maintain the original shape of the geographic units and instead
use other shapes sized to the data values. Perhaps the most well-known graphical cartogram
is the Dorling map—named for geographer Danny Dorling at the University of Leeds—
which uses circles sized by area to the data.
A variant on the Dorling map is the DeMers Cartogram (or tilegram), which uses squares
instead of circles. One advantage with the DeMers approach is that it minimizes the space
between the geographic units. A disadvantage is that the entire geography becomes less rec-
ognizable. These two graphical cartograms again show the number of electoral votes in each
state, and the colors again show which candidate won the state’s electoral votes.
The Dorling (left) and Demers (right) cartograms move further away from the standard
geographic maps and use shapes instead.
GRIDDED CARTOGRAMS
The fourth and final cartogram is the gridded cartogram, in which different shapes are scaled
to the data and arranged so they maintain the general shape of the major geography. People
most often use squares or hexagons to create these kinds of maps.
Take these hexagon grid maps, for example. The advantage of the hexagon over other
shapes is that it offers us more flexibility to arrange the tiles closer to the true geography
of the country. The map on the left shows one hexagon per state, shaded to encode median
238փ CHART TYPES
MEDIAN HOUSEHOLD INCOME IN THE UNITED STATES, 2018 MEDIAN HOUSEHOLD INCOME IN THE UNITED STATES, 2018
AK ME
VT NH
WA MT ND MN WI MI NY MA RI 80,000
ID WY SD IA IL IN CT 70,000
OH PA NJ
60,000
OR NV CO NE MO KY WV VA MD DE
50,000
CA UT NM KS AR TN NC SC
AZ OK LA MS AL GA
HI TX FL
The hexagon grid map is named exactly for what it is: A gridded set of hexagons—either
one for each geographic unit or scaled to its data value.
household income. The version on the right uses multiple hexagons per state with the color
and number of hexagons corresponding to the data value.
Another popular gridded cartogram uses a single square for each geographic unit. This
is often called a tile grid map. Here, median household income is divided into four groups
(or quartiles).
The tile grid map uses a single square for each geographic unit.
GEOSPAT IALփ 239
As with all visualizations, there are tradeoffs. The advantage of the tile grid map is that
each state is the same size, which abstracts from the geographic distortion. The disadvantage
is that the geographic units are now not necessarily in the right place. In the tile grid map on
the previous page, South Carolina is located east of North Carolina, California doesn’t touch
Arizona, and Wisconsin is north of Minnesota, all of which are not their real geographic
relative locations. The arrangement of the tiles can be changed of course, but any decision
is going to be arbitrary because we have moved away from true geography. But this map
can also be easier to construct (it can be made in Excel with resized spreadsheet cells) than
choropleths or cartograms.
Another advantage of the tile grid map is that it enables you to add more data in a
consistent shape. In this tile grid map of the United States, small lines (or sparklines) are
included in the square of each state showing the change in median household income
between 2008 and 2018.
Another advantage of tile grid maps (and graphical cartograms, for that matter) is that we
can add other shapes to the different geographic areas. Both of the tile grid maps on the next
page use emojis to categorize the same household income estimates shown so far. Again,
there’s a clear tradeoff here: the emojis are a little more fun and a little more visual, but it has
One advantage of the tile grid map is because each state is the same size, we can add
small lines, bars, or other graph types to each square.
2 40 CHART TYPES
GROWTH IN MEDIAN HOUSEHOLD INCOME, 2008-2018 GROWTH IN MEDIAN HOUSEHOLD INCOME, 2008-2018
AK ME AK ME
WI VT NH WI VT NH
WA ID MT ND MN IL MI NY MA WA ID MT ND MN IL MI NY MA
OR NV WY SD IA IN OH PA NJ CT RI OR NV WY SD IA IN OH PA NJ CT RI
CA UT CO NE MO KY WV VA MD DE CA UT CO NE MO KY WV VA MD DE
AZ NM KS AR TN NC SC AZ NM KS AR TN NC SC
OK LA MS AL GA OK LA MS AL GA
HI TX FL HI TX FL
Another advantage of tile grid maps is that we can use other shapes, such as emojis.
a very different tone and makes it visually more difficult to immediately pick out values for
groups of states or larger geographic patterns as in the original map. Adding some boundar-
ies or shading (as in version on the right) merges the two approaches.
There is one other class of cartograms worth mentioning. Maps do not always have to encode
data, and they don’t necessarily need to encode them accurately. A non-area based carto-
gram (or distance cartogram) distorts the physical geography by displaying relative time
and distance. This version of the Washington, DC metro (subway) map, for example, shows
relatively constant distances between stops, when in fact the distances vary considerably.
Check out the western part of the Orange line that runs concurrently with the Silver line.
The distance between my metro stop at East Falls Church and Ballston-MU, and between
the Ballston-MU and Virginia Square-GMU stops are the same on the map, but the first trip
is 2.7 miles long and the second trip is only half a mile long. Stretching the stops out to their
actual distances is unnecessary here because the purpose of the map is to provide a compact
view of the subway lines so riders can quickly and efficiently plan their trip.
GEOSPAT IALփ 241
transfer
out of station transfer Rockville
Amtrak MARC
off-peak service only
regional rail connection montgomery
Twinbrook county
key Glenmont prince george’s
White Flint
county
Wheaton
Greenbelt
Forest Glen MARC
Grosvenor-Strathmore
Silver Spring College Park-U of Maryland
MARC MARC
Takoma
Medical Center
Prince
George’s
Plaza
Bethesda Fort Totten
West New
Georgia Av Hyattsville
Van Carrollton
Friendship Heights Ness- Brookland-CUA Amtrak
UDC
Tenleytown-AU NW Columbia NE Landover MARC
Heights
Cleveland Pk U St Shaw- Rhode Island Av
Wiehle-
Woodley Pk Howard U Cheverly
Reston E Dupont Cir McPherson
Spring Hill Farragut N Square NoMa-Gallaudet U
Foggy Bottom-GWU Mt Vernon
Square Deanwood
Greensboro Gallery
Ballston- Farragut Place Union Station
Tysons Corner
Rosslyn West Amtrak MARC VRE Morgan
MU Clarendon
McLean Metro Center Judiciary Sq Minnesota Av Blvd
Federal Triangle Archives
Vienna Dunn W Falls E Falls Virginia Court Smithsonian Capitol S
Stadium-Armory
Loring Church Church Sq-GMU House SW Benning
Road
Capitol Addison
Heights Road
Largo Town
Center
Arlington Cemetery L’Enfant Plaza Federal Eastern Potomac Av
VRE Center SW Market SE
arlington Waterfront Navy Yard
Anacostia
Pentagon Ballpark
Congress Heights
Pentagon City Naylor Rd
Southern Av
Crystal City Suitland
d.c.
VRE
virginia National Airport
Branch Av
alexandria Braddock Rd
Franconia-Springfield
VRE
This version of the Washington, DC, metro map from designer Jacob Berman shows rela-
tively constant distances between stops, even though that’s not geographically the case.
A familiar map like this also gives us the opportunity to add data. As an example, the next
map shows the number of passengers entering and exiting each station in the morning rush
hour. Each station is turned into a pie chart in which the blue segment represents the share
of people entering each station and the orange represents the share of people exiting. Even if
We can add other graphs to maps—pie charts here represent the share of people entering
and exiting each metro stop in the morning. Even if you’re not familiar with the DC metro,
you can see how people tend to commute from the outer parts of the area (more blue
areas in the pie charts) to downtown (more orange areas in the pie charts).
GEOSPAT IALփ 2 43
you’re not familiar with this subway system, you can see the movement from the outer areas
to the center city in the morning.
But let’s make sure we remember our audience. These maps would be of interest to people
who regularly use the DC metro system, but they would not be as valuable if I was an urban
planner making an argument in Atlanta or Dallas or Berlin, because those audiences might
not be familiar with the shape and arrangement. As always, we must consider the needs,
expertise, and expectations of our audience.
Color and size are not the only encoding techniques we can use to visualize data on a map.
Different shapes and objects—lines, arrows, points, circles, icons, even small compact bar
graphs and pie charts—can all be placed on a map. These are known as proportional symbol
maps, because the symbols are sized proportionate to the data. Be careful not to clutter the
map or the reader will have difficulty identifying the most important information.
Different shapes and objects—lines, arrows, points, circles, and more—can all be placed on
a map, sized according to the data value.
244փ CHART TYPES
The two maps on the previous page show per capita GDP for European countries encoded
with circles and squares, rather than using color to shade countries. Notice the importance
of using a transparent color to make overlapping shapes visible, a technique we’ve seen in
previous visualizations. Where there are dense clusters of areas, such as around Belgium and
Netherlands, it can be difficult to find the shapes for individual countries. Aside from our
inherent difficulty of discerning exact quantities from shapes like circles, the dense clusters
can be a barrier to seeing specific geographic units.
A dot density map or dot distribution map takes the proportional symbol map in a slightly
different direction by using dots or other symbols to show the presence of a data value. Sym-
bols can either represent a single data value (one-to-one) or a many values (one-to-many).
These kinds of maps can be data intensive, but they can also illuminate spatial patterns and
clusters that would otherwise be difficult to visualize in a choropleth map or cartogram.
Dot density maps are valuable because they quickly and easily show geographic densities
through the clustering of the symbols. The primary challenge, however, is that because these
Dot density maps include a dot or other symbol for a single (or many) data values. The
“similarity” Gestalt principle helps us see the clusters of people around the country.
Source: Image Copyright, 2013, Weldon Cooper Center for Public Service, Rector and
Visitors of the University of Virginia (Dusধn A. Cable, creator).
GEOSPAT IALփ 2 45
maps require exact geographic locations like addresses or longitude-latitude pairs, which are
not usually available (or publishable), the symbols must be placed in a random or arbitrary
position within a specific geographic area.
Consider the dot density map of the United States on the previous page. It uses data
from the 2010 U.S. decennial census and places a dot for each of the country’s 308 million
residents in their Census blocks. Colors denote different racial and ethnic groups: blue for
White people; green for Black people; red for Asian people; orange for Hispanic or Latino
people; and brown for Native American people and people of multiple or other races. As
with the map shown in Chapter 2, there is nothing in this map except for the data—no state
borders, city markers, or other labels. We can still recognize it as the shape of the United
States because people cluster in cities, and on borders and coasts.
FLOW MAP
Flow maps show movement between places. Arrows and lines denote the direction of the
flow, and the width of the line can correspond to the data value. Flow maps can also encode
qualitative data, but in those cases the width of the directional symbols may not be scaled to
a data value. We saw one such example of a flow map on page 128 to show import and export
trade flows between the United States and other areas of the world.
There are different types of flow maps. Radial flow maps (also called origin-destination
maps), show flows from a single source to many destinations. A distributive flow map is
similar, except that the flow from the single source can fork into many different lines. I like
to think of these maps as the ones in the back pages of the airplane magazine, like this one
in the back of the Delta Airlines magazine (see next page). This is a distributive flow map
because it shows all of the various connections.
As a slight aside here, you might take a look at this map and think, “Wow, that’s cluttered!
How am I supposed to track my flight here?” But that’s not the purpose of the map—instead,
the intention of the map (or at least how I infer it) is to demonstrate Delta’s domestic “unri-
valed coverage”—all the many flights Delta offers around the United States. Showing the
tangled web of all the flights communicates that point.
Perhaps the most famous flow map (at least in the data visualization field) is Charles-
Joseph Minard’s 1869 map of Napoleon’s Russian Campaign of 1812 to 1813 and the con-
nected, but lesser-known map of Hannibal’s 218 BC march through the Alps to Rome.
Minard was a French civil engineer who conducted in-depth studies over many decades to
2 46 փ CHART TYPES
Flow maps show movement between places. Even though this flow map from Delta Air-
lines looks cluttered, it meets the airline’s goal: demonstrating “unrivaled coverage.”
Charles Joseph Minard’s Hannibal and Napoleon maps show the march of two different
armies. Photo courtesy of Ecole nationale des ponts et chaussées.
In her book on Minard’s graphs, The Minard System, author Sandra Rendgen writes the following:
Minard created his visualization more than fifty years after the [Napoleon] campaign. It is a
brilliant conceptual transfer: in applying the flow method to a military campaign, Minard shifts
248փ CHART TYPES
his entire focus to a single variable: the number of people in the flow. This variable sees only one
type of variation—a sharp and steady decline. It seems to have been this potent and poignant
message that made these two maps (and particularly the Napoleon one) so successful in telling
a story about the cataclysm of war.
CONCLUSION
In this chapter we surveyed the promises and perils of visualizing geographic data. Some-
times the varying sizes of geographic units may distort the data. Other times sizing the
geography to the data may make a familiar geography look foreign.
When working with geographic data, your instinct may be to create a map. But take a
moment to consider: Is a map the best way to present your data? Does your reader need
to see the exact differences between data values? If so, the aggregation problem inherent in
many maps may make that difficult. Are there clear geographic patterns to be seen in the
data? If not, then the map may not actually help the reader see your point.
If a data map is the right approach, carefully consider the map projection you use and
whether the standard choropleth map is the best choice. Maybe some kind of cartogram—
even with all its flaws—would be a better fit for your context and reader.
You may also determine that the best approach is to combine visualization types. Depend-
ing on your final publication type, you might use multiple visualizations, say, a map with a
bar chart or table. This approach can help give your readers a familiar visualization type in
which they can identify themselves and their location, but also help them gain a better, more
detailed view of the actual data.
RELATIONSHIP
8
T he charts in this chapter show relationships and correlations between two or more vari-
ables. Perhaps the most familiar chart type in this class is the scatterplot, a chart in
which the data are encoded to a single horizontal and vertical axis. Other shapes and objects
can also be used to visualize the relationship between two or more variables—a parallel co-
ordinates plot uses lines, while a chord diagram uses arcs within a circle. These charts can
show the reader correlations and even causal relationships.
For this chapter, I use the color palette and font from a famous bubble chart created by the
Swedish academic Hans Rosling and his colleagues at Gapminder, a foundation dedicated to
visualizing statistics. Rosling’s Gapminder project didn’t lay out a specific data visualization
style guide, but the visualizations in this chapter use the basic colors and font (Bariol), with
additional styles based on other visualizations from the Gapminder website.
SCATTERPLOT
The scatterplot is perhaps the most common visualization to illustrate correlations (or lack
thereof) between two variables—one variable is plotted along a horizontal axis, and the
other along a vertical axis. The specific observations are plotted in the created space. Unlike
a bar chart, the scatterplot axes do not necessarily need to start at zero, especially if zero is
not a possible value for the data series.
Source: Based on a free chart from www.gapminder.org
RELATIONSHIPփ 25 1
Source: XKCD
One of the most famous graphs that shows the relationship between two variables is
the set of scatterplots created by Rosling and his colleagues at Gapminder. A physician by
training who spent roughly two decades studying public health in rural areas across Africa,
Rosling is perhaps best known for engaging presentations and data visualizations, and for
promoting the use of data to explore issues around international development. In Rosling’s
2014 TED Talk, he showed an animated scatterplot of the relationship between the fertility
rate (number of births per woman) and life expectancy at birth from 1962 to 2003 for coun-
tries around the world.
As I’ve noted previously, there are many nonstandard graph types—scatterplots
included—with which your reader may be unfamiliar. This doesn’t mean you can’t use these
visualizations, but it does mean you should be mindful that your reader may be unfamiliar
with them, and consider how to prepare them to understand your graphs.
Some readers may be familiar with scatterplots (or other nonstandard graph types for
that matter) even if they are not familiar with reading data in scatterplots. New York Mag-
azine’s weekly Approval Matrix on the next page, for example, is their “deliberately over-
simplified guide to who falls where on our taste hierarchies.” Bits of text, images, and icons
are plotted in a space defined by the “Highbrow-Lowbrow” vertical axis and “Despicable-
Brilliant” horizontal axis. Though it’s a light-hearted way to list popular news tidbits, it is,
at its core, a scatterplot.
Moving to scatterplots that are a bit more, shall we say, data-driven, these two scatterplots
on page 253 show net immigration (defined as the number of people migrating into a region
divided by the total number of migrants moving in and out of a region) plotted along the
horizontal axis and per capita gross domestic product (GDP) along the vertical axis. The
version on the left uses a single color with a slight transparency (or “opacity”) so the reader
New York magazine’s “Approval Matrix” is a scatterplot. Not truly data driven, but a scat-
terplot nonetheless.
RELATIONSHIPփ 253
Both scatterplots show the association between net immigration and per capita GDP, using
either a single transparent color (left) or different colors for regions of the world (right).
can see overlapping values. The same transparent effect is used in the scatterplot on the right,
but this time colors capture the different regions of the world.
A scatterplot can help the reader see whether two variables are associated with one
another. If the two variables move in the same direction—to the right along the horizontal
axis and up along the vertical axis—they are said to be positively correlated. In other words,
when both variables get bigger or smaller simultaneously, they are positively correlated. If
they move in opposite directions, they are said to be negatively correlated. And if there is no
apparent relationship, then they are not correlated (see Box on the next page). In the two
scatterplots above, you get a visual sense that the two metrics are positively correlated—that
net immigration is higher for regions with higher per capita GDP, in particular Western and
Northern Europe, Australia/New Zealand, and North America.
One way to make the correlation even clearer is to add what statisticians call a line of best
fit to the scatterplot. These are also called “regression lines” or “trendlines,” and they show
the general direction of the relationship. The statistical calculations to create lines of best fit
are beyond the scope of this book, but the point is that you can make it even clearer to the
reader in what direction (and to what magnitude) the two variables are correlated by calcu-
lating and including this line.
While the scatterplot is becoming a more common chart type, readers may still have
difficulties reading and understanding it. A 2016 Pew Research Center survey showed that
about 60 percent of people could correctly identify what they were seeing in a scatterplot.
A line of best fit visualizes the correlation between the two variables.
CO RRELATIONS
You have probably heard the old adage that “Correlaধon does not imply causaধon.”
We hear this so ođen because people regularly assign a causal relaধonship between
variables that is actually coincidental. People eat more ice cream when it’s hot out-
side, but that doesn’t mean that more ice cream consumpধon causes the temper-
ature to rise. As you look at your data and visualize the relaধonship between the
observaধons within, be careful to understand when something may be correlated and
when it might be causal. The less we know, the more we observe correlaধons and not
causaধon.
Correlaࢼon is a measure of the strength of the linear associaধon between two quan-
ধtaধve variables. The most common measure of correlaধon is the Pearson correlaࢼon
coeLcient, which measures the linear associaধon between variables and is typically
denoted with the Greek leħer rho (ű).
The sign and value of a linear correlaধon coeăcient describes the direcধon and mag-
nitude of the relaধonship between the variables. The value of the correlaধon coeăcient
ranges between Ȅ1 and 1. Values of Ȅ1 signal a perfectly negaধve correlaধon, 1 signals
a perfectly posiধve correlaধon, and 0 represents no linear correlaধon. Posiধve correla-
ধon coeăcients denote a posiধve correlaধon, which means that if one variable gets big-
ger, the other variable also gets bigger; negaধve coeăcients mean the variables move in
opposite direcধons, one variable gets bigger as the other gets smaller.
This discussion is related to these linear associaধons (or relaধonships) and it is
also possible for two variables to have a nonlinear relaধonship. A linear associaধon is
a staধsধcal term that describes a straight-line relaধonship between one variable and
another. A simple example is how we might calculate distance as rate ধmes ধme. In this
case, if we were driving sixty miles per hour for two hours, we would travel 120 miles.
The driving speed does not change over ধme, so the relaধonship is linear.
A nonlinear associaধon, by comparison, refers to paħerns in data that curve or break
from the straight linear trend. Consider, as an example, the pro)t a company makes
from a new product. When it is )rst released, there is liħle compeধধon and sales grow.
As sales conধnue to rise, public awareness increases, and pro)ts start to roll in. Com-
peধng companies then start making their own version of the product and prices for the
original fall to keep up, so pro)ts decline. The company then develops a new version,
and the whole cycle starts again.
As you can see in the images below, the two data values lie on a single diagonal
line when they are perfectly posiধvely or negaধvely correlated. When the two values
move in tandem, either up or down, they are said to be posiধvely or negaধvely corre-
lated. The data in the boħom-right graph demonstrates the impact outliers have on this
measure of correlaধon—moving a single point to the top-leđ part of the graph reduces
the correlaধon from 1.0 to 0.8.
These visuals reinforce the importance of looking at our data as we conduct our
analysis. Data visualizaধons help us not only communicate our work to our readers,
but also enable us to explore our data. They can reveal paħerns and relaধonships we
wouldn’t otherwise see. It’s important not to leave this to the end of the work*ow.
Perfectly Negatively Correlated, =-1 Strong Negative Correlation, =0.8 Positively Correlated with an Outlier, =0.8
10 10 10
8 8 8
6 6 6
4 4 4
2 2 2
0 0 0
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
256 CHART TYPES
In 2016, University of Miami journalism professor Alberto Cairo drew a dinosaur with
points in a scaħerplot and dubbed it the “Datasaurus.” His goal was to show the impor-
tance of visualizing your data in the exploratory phase. Imagine if you were teaching a
data visualizaধon class and asked your students to draw a scaħerplot with 142 points, an
average x value of 54.26, an average y value of 47.83, accompanying standard deviaধons,
and a Pearson correlaধon of Ȅ0.06. Do you think anyone would draw a dinosaur?
In a 2017 paper, researchers Jusধn Matejka and George Mitzfaurice took the
“Datasaurus” one step further and generated twelve alternaধves that maintained the
same summary staধsধcs (mean, standard deviaধon, and correlaধon). The message of Cai-
ro’s “Datasaurus,” Matejka and Fitzmaurice’s paper, and Anscombe’s quartet from Chapter
1, is that we should never rely on summary staধsধcs alone but also on visuals of the data.
BUBBLE PLOT
The scatterplot can be transformed into a bubble plot (or bubble scatterplot) by varying the sizes
of the circles according to a third variable. The data points don’t have to be circles, they can be
any other shape that doesn’t distort our perception of the data. As mentioned in the section on
RELATIONSHIPփ 25 7
bubble charts, the circles should be sized by area, not radius (see page 123). Color can help group
or highlight certain points or direct the reader’s attention to different parts of the graph.
The circles in this bubble plot are scaled according to the population of each region. The
same positive relationship is still evident, but you can now see the relative size of each area.
Because it is a more uncommon graph, be especially mindful of how labeling and annota-
tion can guide readers through the chart and its content. One strategy is to label each axis
and the direction of the change along the axis. The bubble plot on the next page has a cen-
trally located horizontal axis that reads “Net Immigration” and two other labels, “Higher Net
Immigration” and “Lower Net Immigration.”
To further guide the reader, we could add a 45-degree line, on which the values are equal.
We could also highlight specific points with color or outlines, or we could add text to explain
what a point or set of points means. Properly labeled, these elements can lead the reader
through the graph and content. Even people who know how to read scatterplots can struggle
for a moment to understand what is going on when there are lots of points.
A bubble chart adds a third variable to the typical scatterplot. Here, the size of the circles
corresponds to the population in each region.
258 CHART TYPES
As before, more colors can be added to denote another variable, such as region of the world.
3RVLWLYHUHODWLRQVKLSEHWZHHQWKHLPPLǨUDWLRQUDWHDQGSHUFDSLWD*'3
(Per capita GDP)
160,000
140,000 Monaco
Liechtenstein
120,000
Luxembourg
100,000
Norway
80,000 Isle of Man
Bermuda Switzerland
Ireland
Qatar
60,000 Denmark
Sweden Australia
Singapore
Netherlands
San Marino United States
Canada
Greenland Iceland Austria
Japan
Finland BelgiumGermany
40,000 United Kingdom Andorra
France United Arab Emirates
New Zealand
Italy Israel
Spain Kuwait
Brunei Darussalam
Cyprus Guam
Puerto Rico Malta Slovenia
20,000 Portugal Greece Aruba Bahrain
Estonia Saudi Arabia
Northern Mariana Islands
Trinidad and Poland Lithuania
Barbados
Tobago Latvia
Hungary Oman
Equatorial Guinea
Romania
Uruguay Antigua and Croatia
Seychelles Barbuda Chile
Russian
Turkey
Malaysia
Panama
Federation
Grenada Mauritius
Mexico
Bulgaria Brazil Kazakhstan Nauru BotswanaArgentina
PalauCosta Rica American
Gabon Samoa
Maldives
Cuba
Bosnia Suriname
Colombia
China
Dominica
Peru
and Dominican
Azerbaijan
Iraq
Herzegovina Ecuador Montenegro
Republic Namibia Turkmenistan
Belarus
Serbia South
Lebanon Africa Libya
Thailand
Jamaica
Sri
El Albania
Tonga
Samoa
Lanka
Guyana
Tuvalu
Cabo Fiji Algeria
Tunisia
Georgia
Indonesia
Salvador
Morocco Verde
Guatemala
Philippines Mongolia
Armenia
Paraguay Marshall
Vanuatu
Timor-Leste Islands Bhutan Belize
Angola
Ukraine
Nigeria Jordan
Sao
0K
Honduras
Nicaragua
Myanmar
Lesotho
Tome
CambodiaandGuinea-Bissau
Bangladesh
Haiti
Comoros
Eritrea
Afghanistan
Central Principe
Madagascar
African Republic India
Mali
Guinea
Nepal
Zimbabwe
Mozambique Sudan
Benin
Burkina
LiberiaUzbekistan
Ghana
Solomon Kiribati
Zambia
Islands
Pakistan
Tajikistan
Senegal Malawi
Faso
Sierra Togo Rwanda
Niger
Leone
Burundi Cameroon
Mauritania
Ethiopia Uganda Papua
Kenya
Chad New Guinea
Djibouti
0 20 40 60 80 100
Net Immigration Rate
Source: United Nations and World Bank
There’s rarely a case where labeling all of the data points is necessary. Reduce clutter like
this so your readers can better see the data.
Second, there may be times when normalizing your data, calculating percent change,
or taking the logarithm of your data may improve the visual clarity of your graph. This is
especially true when our data are clustered too densely in a visual. The logarithm (or log for
short), is, simply put, an exponent written in a different way. Using mathematical laws of
exponents, the log transformation shows relative values instead of absolute ones. Visualizing
log data can make highly skewed distributions appear less so.
In a log scale, the fact that 101 minus 100 and 2 minus 1 are the same doesn’t matter.
Instead, what matters is that going from 100 to 101 is a 1 percent increase and from 1 to 2 is
a 100 percent increase. Thus, on a log scale, going from 100 to 101 is about 1 percent of the
distance as going from 1 to 2.
Another way to understand the differences between absolute (level) and relative (log) val-
ues is to graph some sample data. The graph on the left shows a simple doubling of each
RELATIONSHIPփ 261
2 3 4 … 20 30 40 …
1 10 100
On a log scale, going from 100 to 101 is about 1 percent of the distance as going from 1 to 2.
Based on Lisa Charlotte Rost (2018).
number—2, 4, 8, 16, 32, and so on. In this linear scale, we see what is called an “exponential”
curve as the difference between each sequential value gets farther and farther apart. By con-
trast, in a logarithmic plot (the graph on the right), each gridline represents a tenfold increase
over the previous one: 1, 10, 100, and 1,000. In this representation, the same numbers appear
as a straight line as opposed to a curved one, even though the growth rate is the same.
In the GDP-immigration scatterplot, there are many countries clustered around the ori-
gin, with both low per-capita GDP and net immigration rates. By taking the logs of both
variables, the data are more spread out across the graphic space. The tradeoff is that logs
(or other transformations, for that matter) are not immediately intuitive. Knowing that per
Logarithmic values are useful to show relative values rather than absolute values.
262 CHART TYPES
3RVLWLYHUHODWLRQVKLSEHWZHHQWKHLPPLǨUDWLRQUDWHDQGSHUFDSLWD*'3
(Per capita GDP, log)
Africa Asia Europe Latin America North America Oceania
12
11
10
5
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
Net Immigration Rate (log)
Source: United Nations and World Bank
By taking the logs of both variables, the data are more spread out across the graph.
capita GDP in Luxembourg is $107,865 (in U.S. dollars) is a number we can all understand,
but we don’t as easily grasp that if we write it as the log per capita GDP: $11.59.
Here’s another example of how to use a log scale. This one uses time series data so we can
see relative changes over time. Lisa Charlotte Rost, a designer and blogger at the online data
New Zealand Tourists New Zealand Tourists
Number of overseas visitors whose intended length of stay is less than 12 months, per month, 1921-2018 Log of the number of overseas visitors whose intended length of stay is less than 12 months, per month, 1921-2018
600,000 1,000,000
500,000 100,000
400,000 10,000
300,000 1,000
200,000 100
100,000 10
0 1
1921 1931 1941 1951 1961 1971 1981 1991 2001 2011 1921 1931 1941 1951 1961 1971 1981 1991 2001 2011
We can’t see the drop in the number of visitors to New Zealand during World War II, but when
we convert the data to logs, the change becomes apparent. Based on Lisa Charloħe Rost (2018).
RELATIONSHIPփ 263
visualization tool Datawrapper, used New Zealand tourism data to demonstrate how the
log transformation works and how it can affect our view and understanding of data show-
ing changes over time. In the line chart on the left, she’s plotted the number of tourists per
month from 1921 to 2018. We can see the relatively flat pattern from the beginning of the
period to about 1970, when the number of visitors starts to rise. The version on the right
uses log values and thus shows the relative number of visitors. Here, we can see a clear drop
in the early 1940s during World War II. In the graph on the left, there isn’t as clear a decline
in absolute numbers (the number of tourists fell from about 2,000 in early 1939 to fewer
than 100 in 1942) but there was a sharp decline in relative numbers.
Whether it’s appropriate to transform your data is largely a function of the question you
want to answer. Are you after relative or absolute values? Percent changes or levels? There is
no right or wrong answer to this question, but each has their tradeoffs.
A scatterplot has data along two variables represented by a horizontal axis and vertical axis.
Sometimes, though, we have more variables to visualize. That’s where the parallel coordi-
nates plot comes in.
In these charts, the data values are plotted along multiple vertical axes and connected by
lines. As in the scatterplot, the axes can have different units of measurement, or they can
be normalized—for example, as percentages—to keep the scales uniform. Thus, instead of
visualizing a single correlation between two variables, the parallel coordinates plot permits
multiple correlations within a single view.
As an example, the parallel coordinates plot on the next page shows correlations across
six different variables related to migration for thirty-two countries around the world. Each
vertical axis represents a different variable—like educational attainment, employment rate,
and life expectancy—and each line represents a different country.
But this graph is really, really hard to read! There are too many lines, all different colors,
and all crossing at different places. But before we try to address the challenges of the full
parallel coordinates plot, let’s try simplifying.
If we zoom in for a moment on the first two axes, we see perhaps the simplest parallel
coordinates plot, one that resembles a slope chart. (I differentiate this from the slope chart
in Chapter 5 because slope charts show changes over time, while parallel coordinate plots
264 CHART TYPES
(FRQRPLFZHOOEHLQǨLQWKH2(&'
100
80
60
40
20
0
Educational Employment Life Immigration Voter Water
attainment rate expectancy rate turnout quality
The parallel coordinates plot shows correlations between two or more variables across
multiple vertical axes.
are used to compare different variables.) Here, I plot the relationship between educational
attainment and the employment rate. Because the lines (countries) at the top of the left axis
(education) are also near the top of the right axis (employment rate), these two variables are
positively correlated. (It’s not that the lines slope down, but the relative position of the points
on each axis.) This is also clear in the scatterplot shown to the right. As always, which chart
you use depends on your purpose and audience.
(FRQRPLFZHOOEHLQǨLQWKH2(&' (FRQRPLFZHOOEHLQǨLQWKH2(&'
90
85
80
Employment rate
75
70
65
60
55
50
25 35 45 55 65 75 85 95
Educational Employment
attainment rate Educational attainment
Source: Organisation for Economic Co-Operation and Development Source: Organisation for Economic Co-Operation and Development
It’s much easier to see the relationship between two variables in a parallel coordinates plot
with just two axes (similar to a slope chart). An alternative visual approach is the scatterplot.
RELATIONSHIPփ 265
Canada 91
United States 90
86
Iceland 78
73
69
65
61 61
60
Spain 58
51
Portugal 47
Turkey 39
Mexico 37 Mexico 37
Source: Organisation for Economic Co-Operation and Development Source: Organisation for Economic Co-Operation and Development
As with the slope chart, you can use different colors, line thicknesses, or other visual ele-
ments to highlight areas or values.
As with the slope chart, you can use different colors, line thicknesses, or other visual or tex-
tual elements to highlight certain areas or values. You could, for example, highlight the North
American countries (left graph) or maybe just those lines that are upward sloping (right graph).
Back to the parallel coordinates plot with all six variables shown at the top of the next
page. Now that you understand how to read the chart, you can see the positive correlation
between education and the employment rate in the first two axes. You can also see the posi-
tive correlation between the employment rate and life expectancy in the second and third
axes. Our view of the data and the specific correlations we can most clearly identify are a
function of how we organize the axes. The plot on the right changes the order of the vertical
axes so we can now see the positive correlation between voter turnout and life expectancy in
the first two axes, which we could not see in the original plot.
Placing all six metrics on the same vertical range also has the effect of suppressing the range
(or variance) in some of these measures. For example, life expectancy varies only slightly,
from 74.6 years to 83.9 years, while net migration varies more widely, from 8.6 percent to 93.9
percent. Allowing the ranges along the axes to fluctuate from one to the next (as in the middle
chart on the next page) requires more labeling along each axis, but it also gives a better view
of the data. The advantage of the two plots at the top is that you don’t need to label every line.
The disadvantage is that you suppress the variation within each variable. Notice, however, that
this parallel coordinate plot—in which the range of each axis differs—looks more variable.
In sum, the challenge with many parallel coordinates plots is that they quickly become clut-
tered. With lots of observations (lines) and multiple axes, readers may have trouble finding the
(FRQRPLFZHOOEHLQǨLQWKH2(&' (FRQRPLFZHOOEHLQǨLQWKH2(&'
100 100
80 80
60 60
40 40
20 20
0 0
Educational Employment Life Immigration Voter Water Voter Life Immigration Educational Water Employment
attainment rate expectancy rate turnout quality turnout expectancy rate attainment quality rate
Source: Organisation for Economic Co-Operation and Development Source: Organisation for Economic Co-Operation and Development
Parallel coordinates plots with too many observations quickly become cluttered.
(FRQRPLFZHOOEHLQǨLQWKH2(&'
94% 86% 84 years 94% 91% 99%
37 51 75 9 49 63
Educational Employment Life Immigration Voter Water
attainment rate expectancy rate turnout quality
Source: Organisation for Economic Co-Operation and Development
The axes in parallel coordinate plots can diøer based on the minimum and maximum of the metric.
(FRQRPLFZHOOEHLQǨLQWKH2(&'
100
Canada
United States
80
Mexico
60
40
20
0
Educational Employment Life Immigration Voter Water
attainment rate expectancy rate turnout quality
Source: Organisation for Economic Co-Operation and Development
As we’ve seen previously, one way to simplify dense graphs like these is to use the “start
with gray” strategy and add color to only a select number of observations.
RELATIONSHIPփ 267
correlations and picking out specific values. One way to alleviate this difficulty is to remember
the “Start with Gray” guideline: Color a group of lines gray and highlight just a subset of the data.
RADAR CHARTS
Radar charts are like parallel coordinate plots, but the lines wrap around a circle instead of
being arranged parallel to one another. These are also sometimes called spider charts or star
charts, and they’re a good way to show multiple comparisons within a relatively compact
space. Data values are plotted along separate axes that radiate from the center (the axes
themselves may or may not be shown) and are connected by lines or areas to show the rela-
tionships between the different variables.
The radar chart on the left shows the same six variables used above—the line for the
United States and the gray area behind it the average for the thirty-two countries shown
earlier. The version on the right shows the same variables for those six countries as well as
the overall average in the gray. Both charts are compact and especially good at highlight-
ing outliers. You can quickly and easily see the shape for Turkey (the pink line) is markedly
(FRQRPLFZHOOEHLQǨLQWKH8QLWHG6WDWHV (FRQRPLFZHOOEHLQǨLQWKH2(&'
(Gray area denotes average among the OECD) (Gray area denotes overall average)
Employment rate Australia Chile Germany
Japan Turkey United States
Employment rate
Educational
Net migration attainment
Radar charts are like parallel coordinate plots, but the lines wrap around a circle instead of
being arranged parallel to one another. The gray interior area represents the overall aver-
age for each metric.
268 CHART TYPES
(FRQRPLFZHOOEHLQǨLQWKH2(&'
Employment rate Educational attainment Water quality
Voter turnout Life expectancy Net migration
100
80
60
40
20
0
Australia Chile Germany Japan Turkey United
States
Too many bars in a bar chart like this make it hard to pick out specific
observations or patterns.
different than the other countries. It’s much harder to make that observation when the data
are arrayed in the paired bar chart above.
As with many charts, the radar chart gets more complicated as more lines are added, and
the crossing pattern around the circle can make perception even more difficult. As was the
case with the parallel coordinates plot, plotting different metrics can also make perception
difficult because it requires some normalizing or other modification of the data values—in
the multi-country radar chart above, again notice how the values for life expectancy bunch
together.
Another strategy is to use the small multiples approach, in which a separate radar
chart is created for each country or group. In this case, the small multiples version takes
up more space than the original and doesn’t necessarily allow easy comparison across
specific countries. But it is easier to see the values for each country relative to the overall
average.
RELATIONSHIPփ 26 9
(FRQRPLFZHOOEHLQǨLQWKH2(&'
(Gray area denotes overall average)
$XVWUDOLD &KLOH
Employment rate Employment rate
Voter turnout
Voter turnout
*HUPDQ\ -DSDQ
Employment rate Employment rate
7XUNH\ 8QLWHG6WDWHV
Employment rate Employment rate
A small-mulধples approach lets us see the values for each country relaধve to the overall average.
CHORD DIAGRAM
Like the radar chart, the chord diagram is another way to show associations or relationships
between observations arrayed in a circle. It is perhaps best used to show how observations
270 CHART TYPES
have shared characteristics. In chord diagrams, observations (called nodes) are located
around the circumference of the circle and connected by arcs within the circle to illustrate
connections. The thickness of the arcs—often also differentiated by color or the transpar-
ency of the color—represent the degree of the connection between the different groups.
This chord diagram uses the same migration data used so far in this chapter to show
migration flows between major regions of the world in 2017. Each region is placed along the
circumference of the circle and the bands emanating from each correspond to the number
of migrants entering or leaving each region. There are more than 110 values plotted in this
0LǨUDWLRQDURXQGWKHZRUOG
Africa
Asia
Australia/New Zealand
Europe
Latin America and the Caribbean
North America
North America
a
Sout
Eastern Afric
Oceania
rica
ica
h Am
le Af
Afr
Ce
a
ric
ern
nt
Midd
erica
Af
ral
rth
a
ric
ern
Am
Af
No
uth
er
rn
ica
So
te
Ca es
rib W
bea
n sia
Wes te rn A
tern
Euro Eas
pe
Southern Europe
South-Eastern Asia
Northern America
pe
rn Euro
Northe
pe So
ro ut
he
Eu rn
rn As
ste ia
Ea
Asia
Western Asia
tral
Cen
In chord diagrams, the observations are located on the circumference of the circle and arcs
within illustrate the connections.
RELATIONSHIPփ 27 1
single graph—though you could pick out specific values in a (very large) table, the chord
diagram is clearly more visual and spatially efficient.
In the first chord diagram, you can see the large migrant flows within Asia (the red areas),
and the movement between Central and North America (the thick green line to the blue seg-
ment near twelve o’clock in the circle). One danger is that the graph can quickly become clut-
tered and hard for the reader to easily see relationships. Again, we can use the strategy of
highlighting specific groups with colors or lines. I’ve done that in this chord diagram in which
the Asian region is red with all other regions in gray. The complexity of the chord diagram (and
its relative compactness), make them visually intriguing and invites the reader to explore the
data in more depth.
ARC CHART
Stretch a chord diagram out along a single horizontal axis and you have an arc chart. In this
case, the nodes are placed along a line and are connected by arcing lines. The lines can vary in
height, thickness, and color to illustrate the strength of the relationship or correlation. This arc
chart shows the same migration flows between regions of the world as in the chord diagram.
A major consideration of the arc chart—and which also applies to many charts in this
section—is that the order of the data can influence our perception of the results. Notice the
high, wide green arcs stretching from North America to countries in the Caribbean, and
Central and South America. If, by contrast, North America is placed to the far-right side of
the graph and next to the other countries in the Western Hemisphere, the visualization is
0LǨUDWLRQDURXQGWKHZRUOG
Central Asia
Eastern Asia
South-Eastern Asia
Southern Asia
Western Asia
North America
Eastern Africa
Middle Africa
Northern Africa
Southern Africa
Western Africa
Eastern Europe
Northern Europe
Southern Europe
Western Europe
Caribbean
Central America
South America
Australia/New Zealand
Melanesia
The arc chart is like a chord diagram stretched along a single horizontal axis.
RELATIONSHIPփ 27 3
0LǨUDWLRQDURXQGWKHZRUOG
Central Asia
Eastern Asia
South-Eastern Asia
Southern Asia
Western Asia
Eastern Africa
Middle Africa
Northern Africa
Southern Africa
Western Africa
Eastern Europe
Northern Europe
Southern Europe
Western Europe
Caribbean
Central America
South America
Australia/New Zealand
Melanesia
North America
Source: Organisation for Economic Co-Operation and Development
Note: Data limited to a minimum of 200,000 immigrants or emigrants
Like many graphs in this chapter, the organization of the data along the horizontal axis can
affect our perception of the data. Compare this arc chart to the previous arc chart—same
data, different shape.
substantially changed. There is no longer a tall green arc dominating the view, but a series of
red bands that reach across the entire space between North America and countries in Asia.
Some of this is obviously a function of the colors used, but the arrangement of the countries
also matters. It is worth taking time to experiment with color and node placement to arrange
the arc chart in the way that best communicates your argument.
A variation on the arc chart is an arc-time chart or arc-connection chart, in which connec-
tions over time are plotted in the same way. Instead of illustrating the correlation or relation-
ship between two distinct variables, the nodes denote time. The arc-connection chart can
also be thought of as an alternative to a timeline or flow chart, which we saw in Chapter 5.
On the next page, the arc chart from Adam McCann shows the tenure of all Supreme Court
justices in the United States since 1804. The origin (the left-most point) shows when each
judge started his or her tenure, and the arc stretches to their retirement age. The height of
each arc represents the age at the time of appointment (taller is younger) and the color repre-
sents their political party and year they were appointed (lighter shades are earlier years). An
Adam McCann used an arc chart to show the tenure of all US Supreme
Court justices since 1804.
Arc charts can also show geographic data, as in this one from the Guardian showing where
New York City sends their homeless population.
RELATIONSHIPփ 27 5
alternative chart type, like a bar chart or a heatmap, could be used to show the same changes,
but there is something arresting about the shapes in this view.
Another way to use the arc chart is to plot distances. The arc chart from the Guardian
shows where New York City sends their homeless population. The cities are organized by
distance from New York—Richmond, Virginia, on the left and San Francisco, California, on
right. The height and thickness of the arcs and the size of the circles show how many people
go to each city.
A correlation matrix is a table with the variables listed along the horizontal and vertical axes.
Numbers in each cell represent the strength of that relationship, often as a Pearson’s correla-
tion coefficient (see Box on page 254).
The correlation matrix graph uses the same layout but instead of numbers it uses shapes—
often circles—to show the strength of the correlation, and sometimes color and shades to
organize the table. The correlation matrix is a cousin of the heatmap, and we can also think
of it as a way to add a visual element to a standard table, a topic we will visit in Chapter 11.
World Migration
Latin American
Africa Asia Europe and Caribbean Oceania
Australia/New Zealan
South America
Central America
South-Eastern
North America
Micronesia
Caribbean
Melanesia
Southern
Southern
Southern
Polynesia
Northern
Northern
Western
Western
Western
Eastern
Eastern
Eastern
Central
Middle
Africa Eastern 49.0 10.0 7.1 0.6 0.1 0.0 0.1 0.6 0.0 0.1 0.0 0.1 0.1 1.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Middle 3.9 17.0 3.7 0.5 4.3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Northern 7.2 1.1 3.2 0.0 0.4 0.0 0.1 0.1 0.4 9.0 0.2 0.2 0.3 0.8 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0
Southern 14.0 2.0 0.2 7.1 0.5 0.0 0.5 0.8 0.1 0.2 0.4 1.7 1.1 1.8 0.0 0.0 0.1 0.3 0.2 0.0 0.0 0.0
Western 0.0 1.5 0.6 0.0 58.0 0.0 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.4 0.1 0.0 0.0 0.1 0.0 0.0 0.0 0.0
Asia Central 0.0 0.0 0.0 0.0 0.0 4.9 1.0 0.1 0.0 1.6 44.0 0.1 0.1 2.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Eastern 0.0 0.0 0.0 0.0 0.1 0.3 53.0 1.8 12.0 0.0 0.2 0.6 0.1 0.3 0.0 0.0 3.6 2.2 0.4 0.0 0.0 0.0
South-Eastern 0.0 0.0 0.0 0.0 0.0 0.0 9.2 13.0 68.0 0.2 0.0 0.6 0.1 0.2 0.0 0.0 0.0 0.9 0.4 0.0 0.0 0.0
Southern 0.0 0.0 0.0 0.0 0.0 0.1 2.1 110.0 8.6 1.6 0.0 0.6 0.1 0.1 0.0 0.0 0.2 0.5 0.0 0.0 0.0 0.0
Western 6.0 0.1 38.0 0.2 0.3 0.9 0.3 170.0 40.0 130.0 12.0 1.7 2.9 5.4 0.0 0.0 0.6 1.4 0.1 0.0 0.0 0.0
Europe Eastern 0.0 0.0 0.2 0.0 0.1 56.0 1.5 0.4 1.0 21.0 100.0 4.7 2.5 3.9 0.0 0.0 0.1 0.5 0.0 0.0 0.0 0.0
Northern 8.5 0.7 2.0 2.4 4.5 0.4 5.0 23.0 5.9 8.8 26.0 20.0 8.9 8.8 2.4 0.3 2.7 4.2 2.2 0.1 0.0 0.0
Southern 2.0 2.4 15.0 0.3 5.3 0.5 4.1 6.2 2.3 3.2 32.0 5.8 31.0 16.0 3.8 1.7 25.0 1.9 0.6 0.0 0.0 0.0
Western 4.2 4.0 34.0 0.5 6.5 12.0 4.8 8.7 8.5 31.0 60.0 7.7 49.0 29.0 2.0 0.7 7.3 3.6 0.6 0.0 0.0 0.0
Latin American and Caribbean Caribbean 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.1 0.1 0.0 0.0 0.2 0.3 1.4 7.1 0.2 1.0 2.6 0.0 0.0 0.0 0.0
Central America 0.0 0.0 0.0 0.0 0.0 0.0 0.5 0.1 0.0 0.1 0.1 0.1 0.5 0.4 0.5 6.5 2.1 9.8 0.0 0.0 0.0 0.0
South America 0.0 0.1 0.1 0.2 0.1 0.0 1.8 0.1 0.1 0.6 0.5 0.5 8.0 1.6 0.8 0.5 42.0 1.3 0.0 0.0 0.0 0.0
North America North America 8.2 1.3 6.7 1.5 8.1 1.2 53.0 47.0 53.0 17.0 23.0 18.0 19.0 15.0 66.0 160.0 34.0 12.0 1.6 0.7 0.2 0.4
Oceania Australia/New Zealand 1.6 0.1 0.8 2.5 0.2 0.0 8.8 8.3 10.0 3.1 1.9 18.0 6.8 3.7 0.1 0.2 1.3 2.1 7.4 1.4 0.0 1.5
Melanesia 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.4 0.0 0.0 0.0 0.0 0.2 0.1 0.0 0.1
Micronesia 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.2 0.0
Polynesia 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.1 0.1 0.0 0.2
The basic correlation matrix is a table with numbers that show the strength of the
relationship between observations.
27 6 CHART TYPES
World Migration
Latin
American
and
Africa Asia Europe Caribbean Oceania
Australia/New Zealand
South America
Central America
South-Eastern
North America
Micronesia
Caribbean
Melanesia
Southern
Southern
Southern
Polynesia
Northern
Northern
Western
Western
Western
Eastern
Eastern
Eastern
Central
Middle
Africa Eastern
Middle
Northern
Southern
Western
Asia Central
Eastern
South-Eastern
Southern
Western
Europe Eastern
Northern
Southern
Western
Latin American and Caribbean Caribbean
Central America
South America
North America North America
Oceania Australia/New Zealand
Melanesia
Micronesia
Polynesia
A heatmap approach to the matrix makes the patterns clear without (in this case)
showing the numbers.
These two matrices show the relationship between immigrants (those entering each
region) and emigrants (those leaving each region) across the world in 2017. The view on the
left is the standard correlation matrix shown as a table. This gives us all the detail we would
need to understand the exact correlation between different variables, but it’s difficult to
navigate. There are a lot of numbers, and the important values don’t stand out. The matrix
on the right is a heatmap (see page 112), which loses the detail of the table version in favor
of highlighting the stronger (positive) correlations, especially within Asia. We could include
both the colors and numbers, but the view might end up looking cluttered and busy.
The next two visuals display the data as standard correlation matrix graphs. Circles
represent the strength of the relationship, and color (in the version on the right) helps
organize each area, though in this case there may be too many colors. It can be hard for
the reader to clearly see differences because humans are not very good at assessing quan-
tities from the sizes of circles. In both cases, the circles are sized to fit within each cell,
but that doesn’t necessarily need to be the case. We could make the circles larger to fill
the entire space and use transparent colors when they overlap.
One final consideration with any correlation matrix or table is that the values along the
diagonal are, by definition, equal to one. That is, migration between Eastern Africa and East-
ern Africa is the same. This means they are often left out because they can visually dominate
or clutter the visual.
RELATIONSHIPփ 27 7
Australia/New Ze..
Africa Asia Europe Latin America a..
Nor.. Oceania Africa Asia Europe Latin America a.. Nor.. Oceania
Central America
Australia/New Ze..
South America
North America
Central America
South-Eastern
South America
North America
South-Eastern
Micronesia
Caribbean
Melanesia
Polynesia
Micronesia
Southern
Southern
Southern
Northern
Northern
Caribbean
Melanesia
Western
Western
Western
Polynesia
Eastern
Eastern
Eastern
Southern
Southern
Southern
Northern
Northern
Central
Middle
Western
Western
Western
Eastern
Eastern
Eastern
Central
Middle
Africa Eastern Africa Eastern
Middle Middle
Northern Northern
Southern
Southern
Western
Western
Asia Central
Asia Central
Eastern
Eastern
South-Eastern
South-Eastern
Southern
Southern
Western
Western
Europe Eastern
Europe Eastern
Northern
Northern
Southern
Southern
Western
Western
Latin America Caribbean
and Caribbean Latin America Caribbean
Central America and Caribbean
Central America
South America
South America
North America North America
North America North America
Oceania Australia/New Zealand
Oceania Australia/New Zealand
Melanesia
Melanesia
Micronesia
Micronesia
Polynesia
Polynesia
Source: Organisation for Economic Co-Operation and Development
Source: Organisation for Economic Co-Operation and Development
An alternative to the correlation matrix table is to use circles or other shapes, to which
color can be added to visually organize the space.
NETWORK DIAGRAMS
We now enter a class of graphs for which I use the term diagram instead of graph, plot, or
chart, largely because some of the decisions about layout and structure are not always deter-
mined by math or the data but by what looks best and is most clear. These diagrams are used
to show hierarchies and connections within and across groups and systems. The thickness
of the lines and size of the points can be sized according to data values to signal the strength
of those relationships, and arrows can visualize movements inside groups and communities.
Consider a family tree: the lines show links between parents, siblings, spouses, and children,
but the connecting lines and the pictures or names of family members are not scaled accord-
ing to a data value.
We start with the standard network diagram, which shows connections between people,
groups, or other units. Generally speaking, the points in a network diagram (called nodes or
vertices) denote the individual person or observation, and the lines (called edges) link them
together and show the relationship. The position of the nodes and the length (and some-
times thickness) of the linking lines illustrates the strength of the relationship. While nodes
are often depicted as circles, you could also use icons, symbols, or pictures.
27 8 CHART TYPES
The ultimate appearance and organization of a network diagram depends on the kind of
network we want to visualize and the method with which we arrange the nodes and edges.
When creating a network diagram, we must be careful about how edges cross and nodes over-
lap. In general, we want to achieve some kind of visual harmony in the visualization by find-
ing a uniform and meaningful length of the edges and some symmetry for the entire graph.
To start, we can distinguish between four different kinds of network diagrams:
Four kinds of network diagrams, clockwise from top-left: Undirected and Unweighted;
Undirected and Weighted; Directed and Unweighted; Directed and Weighted.
RELATIONSHIPփ 27 9
There are many algorithms we can choose from to lay out the nodes and edges in a net-
work diagram. Usually, network algorithms try to minimize how often the edges cross one
another and prevent overlap of the nodes. Generally, we want the edges in a network diagram
to be of roughly uniform length and the vertices to be distributed evenly. Using example data
Sphere Circle
5DQGRP )UXFKWHUPDQƂ5HLQJROG
around twenty points, these four network diagrams demonstrate how select organizing algo-
rithms will generate different views of the network and the relationship between the points.
This network diagram shows the relationship of the seventy-five or so most populous
countries in the world (those with more than one million people) within their different
geographic regions. I’m not arguing that this network diagram is a better visualization than
a standard geographic map, but I show it here because you can easily understand the content
A simple network diagram that shows the arrangement of countries with more than one
million people.
RELATIONSHIPփ 2 81
and see how the diagram works. Imagine showing the links between people in your Twitter
or Facebook network, grouped by family, friends, and coworkers.
Network diagrams are ideal for showing the structure and relationship between differ-
ent agents in a system. In some cases, groupings or concentrations become clear as specific
nodes cluster near one another. We can use color or other shapes to highlight specific groups
within the larger network.
This small multiples set of network diagrams from Andris et al. (2015) shows voting behav-
ior in the United States Congress.
2 82 CHART TYPES
As with all other charts, including too much data in a network diagram can clutter it and
make them difficult to read. But unlike many charts, sometimes the goal of a network diagram
is to show the dense clustering. The set of thirty-two network diagrams from Clio Andris and
her collaborators shows the polarization of voting behavior in the United States Congress. The
authors created network diagrams for each U.S. House of Representatives from 1949 (top-left)
to 2011 (bottom-right). Republican members are represented by red dots, and Democrats by
blue dots. The lines denote how often members of Congress voted with one another. Even
though each network diagram is very dense, the use of these small multiples makes it clear that
the two parties were much less likely to vote together in 2011 than in the past.
By comparison, the line chart below shows the measure of disagreement between the par-
ties in a simple, straightforward way. Though it’s immediately informative, it’s not as visually
stunning as the network view.
We can make a similar case about voting behavior in the United States Congress using
summary data from Andris et al. (2015), but the line chart probably doesn’t grab you the
way the small multiples network diagrams did.
RELATIONSHIPփ 2 83
Because network diagrams can look like hairballs with too many edges and nodes to make
the visualization readable, some researchers have developed alternative visualization types.
The hive plot, for example, first organizes the space along linear axes emanating outward
from a single central point. Nodes are placed along three or more axes (possibly divided into
Population: 1000
25% 75%
40% 60%
Population: 1000
segments) and edges are drawn as curved links connecting the points. Martin Krzywinski,
inventor of this visualization, writes that, “The hive plot is itself founded on a layout algo-
rithm. However, its output is not based on aesthetics but network structure. In this sense, the
layout is rational—it depends on networks features that you care about.”
It’s also worth noting that some fields refer to network diagrams in different ways. Instead
of plotting how individual values correlate to one another, some network diagrams show flows
or processes, similar to a flow chart or timeline. Examples might include how a computer net-
work, a staff directory, or even a logic model of probabilities, as shown on the previous page.
TREE DIAGRAMS
Like the flow chart in Chapter 5, tree diagrams show levels of a hierarchy in a system or
group. To imagine the basic tree diagram, think of a hierarchical organizational (or “org”)
chart. Nodes branch outward from an initial root connected by lines called links, link lines,
or branches. The initial node is called the root and is the parent to all other nodes, some of
which have child nodes of their own. Nodes who are not parent nodes are called leaf nodes.
Tree diagrams describe hierarchies. These two org charts could be used for different pur-
poses, depending on whether we think the reader would like a more designed version.
RELATIONSHIPփ 2 85
As with many of the visuals we have discussed so far, design is especially important here
because there may not be much or even any data to define the elements in the diagram. Take
these two imaginary org charts, for example. The chart on the left consists only of names,
while the one on the right includes icons. Which one you would use depends entirely on
your purpose. The one on the left might work well for the company board meeting or formal
presentation; the one on the right might work better in a marketing campaign or website.
While the basic tree diagram will often branch downward, starting with the CEO or presi-
dent at the top, there are lots of other ways to show these kinds of relationships. We could
create a family tree that branches upward instead of downward, or a horizontal arrangement
to show a different kind of hierarchy or taxonomy.
Another type of tree diagram is the word tree. Developed by Martin Wattenberg and
Fernanda Viegas in 2007, the word tree is a visual representation of text in a book, article,
or other passage (also see Chapter 10 on qualitative data visualization). The visualization is
This word tree is a visual representation of the text of Dr. Seuss’s book, The Cat in the Hat.
Source: Jason Davies
A simple tree diagram that shows the breakdown of regions into countries.
RELATIONSHIPփ 2 87
typically ordered horizontally with a word on the left or right that branches out to show the
different contexts in which it appears. These contexts are arranged in a treelike structure so
the reader can uncover themes and phrases. Individual words, which act here as the nodes,
are often sized by the frequency in which they appear.
There are lots of different kinds of trees to visualize quantitative or qualitative data. They
can show complex data, like the human genome, or simple data, like the breakdown of
countries into regions. What kind of tree you create and the design touches you include
will—as always—depend on your purpose and audience. The tree on the previous page,
for example, shows the same data as the network diagram shown earlier. On the one hand,
it’s not particularly efficient, but on the other hand, it might be easier to navigate than the
network diagram.
CONCLUSION
In this chapter, we surveyed charts and diagrams that visualize relationships between vari-
ables, individuals, or groups. We often want to understand how two or more things are
related, but remember that because two variables might be correlated does not mean there
is a causal relationship. Clearly understanding how elements in your data are related before
presenting them to your reader or audience is of utmost importance.
This class of graphs uses different strategies and shapes to communicate these relation-
ships, and there are advantages and disadvantages to each approach as they trade off between
clarity, order, and compactness. Scatterplots have a single horizontal and vertical axis; bubble
plots add a third variable. Parallel coordinate plots are defined by using two or more vertical
axes. Radar charts pull the axes together and radiate outward from a center point of a circle
while a chord diagram wraps everything around the circumference of a circle. An arc chart
then stretches everything out along a single horizontal axis and a correlation matrix uses a
square or rectangular format. Network and tree diagrams can be used to show relationships
between individuals or groups or passages of text.
As with the graphs in previous chapters, some of the graphs in this chapter may be unfa-
miliar, even difficult for you or your readers to understand. This doesn’t require you to dumb
things down or leave things out, but it should prompt you to consider how to best communi-
cate the content of nonstandard graphs. Use labels, annotations, active titles, and helpful point-
ers to teach them how to read the graph so that they can more easily understand the content.
ҺҺ 9
T his class of charts shows how the shares of some amount relate to the total. The most
popular and familiar graph in this class is the pie chart, which introduces a variety of
perceptual challenges, as we’ll see. Other charts in this class like the treemap and sunburst
have different perceptual issues and, as always, we must ask ourselves whether we must show
all of the components and how they sum to the total. Graphs in this chapter can also be used
to visualize hierarchical data—data that can be grouped into layers where natural groups
exist—which we have already seen in cases such as the tree diagram.
Graphs in this chapter are based on the online style guide from the Texas Tribune, a
digital-first publication based in Austin, Texas. In addition to the basic styles outlining col-
ors and fonts, the Tribune’s style guide also includes instructions for the online elements of
their website.
PIE CHARTS
Disdain for pie charts pervades the entire field of data visualization. The most often-cited
reason is that pie charts are a poor visualization choice because we have a hard time discern-
ing exact quantities when they are visualized as slices of the pie. If we return to the percep-
tual ranking chart, we’ll see that pie charts fall below the middle of the ranking. While the
pie chart gets its fair share of complaints, it’s also a very familiar chart type for many people,
290 փ CHART TYPES
and familiarity can be useful. Research has also shown that people are more attracted to
curves than to objects with sharp points, and author Manuel Lima has presented evidence
that affinity for the circle shape dates back millennia in human evolution.
These pie charts show the distribution of imports (the dollar value of goods flowing into a
country or region) in seven areas around the world. Notice that the version on the left shows
the percentages, while the version on the right shows the dollar amounts (in billions). Either
approach works, just be sure to use correct labeling and note it in the title.
The most important rule for pie charts is that the slices must sum to 100 percent or at
least some sort of total. You cannot leave segments out or—unfortunately more common—
include segments that sum to more than 100 percent. When you arrange the slices of a
pie chart, a good rule of thumb is to order them from largest to smallest beginning at the
12-o’clock position. This is often the best way, but sometime it isn’t always possible or natural.
For example, if you were visualizing shares of a total by age group, it would be better to start
with the youngest group at 12 o’clock, then the next oldest group, and so on. In this case, it
is more natural for our reader to comprehend the data when they are ordered by category
rather than by value.
In situations where you might use more than one pie chart—something I would only
recommend in very specific cases—always order the slices in the same way in all the charts.
It becomes far too difficult to compare the slices in multiple pie charts when the slices are in
different positions.
Pie charts show part-to-whole relationships. These two show the distribution of imported
goods to the United States, either in dollars or percentages.
PARTҺTOҺWHOLEփ 291
The trouble with pie charts is that humans cannot easily compare differently sized slices.
Here we have the same data arrayed as a pie chart and as a bar chart. See how it’s easier to
rank the values in the bar chart and how small differences are imperceptible in a pie chart?
Even if you may have correctly ranked the values, we can agree that it is easier and faster
to rank them in the bar chart version. It’s also easier to see how much the values differ from
each other. The pie chart simply does not let us accurately discern the values, and if your
goal is to help your reader make clear and accurate determinations about the data, the pie
chart is not the best choice.
It’s also worth noting that it’s not really clear how we perceive the quantities in the pie
chart. Is it the angle at the center? The area of each slice? The arc length? A pair of research
papers in 2016 suggested that the angle of slice meeting at the center is not the main way we
read pie charts and that area and arc length—the segment of the circumference—were better
predictors of people’s ability to read values than angle alone.
25
20
5 1
15
10
2 5
4
0
3 1 5 4 3 2
Comparing the values is easier to do in the bar chart on the right than in the pie chart on the leđ.
292 փ CHART TYPES
The donut chart punches the middle out of a pie chart, which leaves
room for an additional title or text.
If we don’t perceive the values in a pie chart from the angle at the center, it would mean
that a donut chart—a pie chart with the center removed—is a viable, even preferable alterna-
tive. But if it’s the case that we discern the quantities in the pie chart by the angle of the slices,
then the donut chart is an even worse choice than the pie chart because the angles are held in
the center, which is missing from the donut chart. No one has decisively proved it either way.
Still, the primary advantage of the donut chart is that you can include a number or statement
in the center of the chart.
I recommend avoiding using pairs of pie charts, even when they show only a few slices.
The reader must look back and forth between the two pie charts to see whether the values of
the different groups changed. That task is made faster and easier with a stacked bar chart or
slope chart, though with the slope chart, we move away from the part-to-whole comparison
and instead focus on the change over time.
Especially avoid pie charts that show too many slices. Five is already too many—more
becomes incomprehensible. Also avoid the “breakout” pie chart that removes a single slice
Distribution of imported goods to the United States in 1996
(Billions of dollars)
South Asia Sub-Saharan Africa
$11 $16
Middle East & North Africa $23
North America
$162
Source: The World Bank
Pairs of pie charts are rarely useful to show changes over time. That task is made faster
and easier with other visualization types, like a stacked bar chart or slope chart.
294 փ CHART TYPES
and breaks it down even further. These charts are hard to read and there are better ways to
plot such data: in a bar chart or even a Sankey diagram (see page 126).
Let me now make the case for using pie charts. These two pie charts show the same data but
arranged differently. In the version on the left, the value of Group B (the purple slice) is not
immediately clear. In the version on the right, the rotation of the chart creates a familiar
right angle in the center of the chart, and now Group A’s value (25 percent) is clear. A pie
chart is therefore perhaps best used when the value(s) of the slice or slices sums to one of
these round numbers (25 percent, 50 percent, and 75 percent), for which the angle is familiar.
These cases let you you easily focus your reader’s attention on three or maybe four slices.
Let’s look at a practical example of when a pie chart can be useful. Imagine giving a
presentation to report the fundraising efforts you and your colleagues have accomplished
for your nonprofit organization. Let’s say you have received one hundred donations, and
a bit more than half of all money you have raised has come from a few, very large gifts of
$1 million or more. You break down the distribution of gifts into seven categories and break
out the top categories into a group of $1 million to $2 million donations and a group of
$2 million donations and above. The perceptual rankings chart may guide you to use the
most perceptually accurate graphs, like a bar chart or a scatterplot.
But for your audience, a pie chart might work best. They can read and understand a pie
chart almost instantaneously, and because the share you want your audience to focus on
It’s easier to guess the value of the purple slice for Company B when the chart is arranged
with right angles.
PARTҺTOҺWHOLEփ 295
Most donations come from a small number of large gifts Most donations come from a small number of large gifts
32%
Share of all dollars donated
35%
26% $2m+
30% Donor Share: 5%
Dollar Share: 32%
25%
$1m-$2m
Donor Share: 2%
20%
Dollar Share: 26%
12% 15%
11%
9%
10%
6%
4% 5%
0%
0% 10% 20% 30% 40% 50% 60%
$0-$50k $50k- $100k- $250k- $500k- $1m-$2m $2m+
$100k $250k $500k $1m Share of all donors by class of donation
These two charts show that most donations to this nonprofit came from a small number
of large gifts. But if the point is to show that slightly more than half came from these two
groups, a pie chart is actually a good choice.
accounts for just more than half, it’s easy for your audience to mentally draw the vertical line
through the circle. Imagine explaining how a Marimekko chart or a scatterplot works to an
audience who just wants the basic information as quickly as possible.
Notice how all seven values are plotted and labeled in the pie chart on the next page, but
only the two slices of interest are colored, and together they are clearly larger than 50 percent.
(Also notice how we are not following the start-at-12-o’clock rule, because arranging the slices
by category value makes more sense here.) Because they are not the primary focus of the
graph, the other slices are in gray and move to the background. For a presentation, we might
even collapse our two groups to one and delete many of the labels to reduce the amount of
information on the screen that would only distract the audience’s attention from the speaker.
This example leads to another scenario in which a pie chart can be useful, and that is
when you are focusing your reader’s attention on a single value—the true part-to-whole
relationship. In these cases, consider whether you actually need a chart at all. Do you need
a visual to back up the statement that the U.S. poverty rate is 12.3 percent? Maybe on social
media or on a presentation slide, but in a report or article you might do just as well leaving
the number in the text.
A bad pie chart is still a bad chart. If you choose to use pie charts, do so strategically and
thoughtfully. It will almost always be difficult for your reader to discern specific quantities
or compare slices, but if you want them to understand a general difference in magnitude or
focus on a single slice, the pie chart is appropriate.
Most donations come from a small number of large gifts
<$50k
6%
$50k-$100k
9% $2m+
26%
$100k-$200k
4%
$250k-$500k
13%
$500k-$1m
12% $1m-$2m
32%
This pie chart does a good job of highlighting the two groups of interest because (a) the
groups sum to slightly more than 50 percent and (b) the other groups are gray.
Most donations come from a small number of large gifts Most donations come from a small number of large gifts
$2m+
26%
$2m+
<$1m 7 gifts: 57%
43%
All other
93 gifts: 43%
$1m-$2m
32%
Basic simplifications to these pie charts can focus attention on the groups of interest.
PARTҺTOҺWHOLEփ 297
TREEMAP
Originally developed by Ben Shneiderman at the University of Maryland, the treemap di-
vides sections of a square or rectangles into groups to illustrate a hierarchy or part-to-whole
relationship. In other words, the treemap is a squarified version of a pie chart.
Distribution of imported goods to the United States in 2016 Distribution of imported goods to the United States in 2016
Switzerland
Ireland 2% 2%
France 2%
Treemaps are an alternative to the pie chart and can show part-to-whole or hierarchical
relationships. Notice how it is essentially impossible to read the pie chart.
298 փ CHART TYPES
The two treemaps on the previous page show the breakdown of total imports from spe-
cific countries to the United States in 2016. You may find this easier to read than a pie chart
because rectangles are more easily compared. Or, because it’s an unfamiliar graph type, you
might find it slower to navigate or more difficult.
The treemap on the right breaks down the regional categories further into individual
countries. Not every country in the world is labeled in this chart, but you do get a clear sense
of which countries send the most goods to the United States.
Try to read the pie chart that shows the same data—it’s impossible.
Shneiderman originally developed the treemap to concisely view the file directory on his
computer. The compactness of the treemap is one of its biggest advantages because it can
include many different groups and variables. Hierarchies are easy to visualize in the treemap
because subsections can be labeled and embedded within the parent group. You can also
add other encodings to the treemap—for example, this treemap shows the 2016 distribution
Color can add another dimension to a treemap. In this case, the change in imports
between 1996 and 2016.
PARTҺTOҺWHOLEփ 299
of the value of imports to the United States and the colors denote the shares that increased
(blue) and decreased (red) between 1996 and 2016.
SUNBURST DIAGRAM
If you want to show the proportions of parts to a whole at several levels in a hierarchy,
you might use what is called a sunburst diagram. Like a treemap, the sunburst can show
part-to-whole relationships or hierarchical relationships. This sunburst diagram, for exam-
ple, shows the same data as the detailed treemap above, with an additional ring that breaks
the major regions into their components.
Each ring in a sunburst corresponds to a different level in the hierarchy, and the slices of
each ring—sometimes called nodes—refer to the different subgroups. The central ring shows
India
'#$".
di A
#
Isra
K4
rabia
el
<'H
4
a
in
Br
South Asi
Ch
az
Mid
il
dle
So
East
a
ut
hA
Ea
m
er
st
er
ica
n
Eu
South Asi
ro
La Carib
pe
tin
Swit
zerl
Am ean
and
a
eric
b
a&
Irelan
d
Asia
East
Italy ific
& Pac
Europe Asia
Europe & Centr East
al
Asia Japan
France
gdom
United Kin
Korea
, Rep
.
ica
any
Germ
Amer
Vie
tna
m
Ot
North
he
C ?" rA
sia
39 ,n
K4 es
M
ala
ys
ia
ica
Th
ai
Amer
la
nd
In
do por
Si
North
ne
ng
si
a
a
da
!"#
na
>$2
e
Ca
%
.B>
" '9
23
Mexico
the top level, sometimes referred to as the root. Outer rings show how the groups break into
subgroups. You can use color to highlight different rings, groups, or hierarchies. The sun-
burst graph above uses color to differentiate the seven major areas of the world.
As with all circular visualizations, it’s difficult to clearly and quickly see patterns. Sunburst
diagrams with too many categories are likely too cluttered for a reader to pick out patterns
even after a close look. Strategic use of color, labels, and annotation can guide the reader to
the most important parts of the visualization.
NIGHTINGALE CHART
This type of chart is sometimes called the coxcomb or rose diagram but is most often known
as the Nightingale chart. The first and most famous of these was created by Florence Night-
ingale to visualize soldier casualties during the Crimean War.
Nightingale was born in the early nineteenth century to wealthy parents who provided
her with a comprehensive education in liberal arts and mathematics. Deciding early on that
she wanted to dedicate her life to health care and helping the poor and needy, Nightingale
became a nurse in the early 1850s. With experience in organizing hospital supplies, she even-
tually took care of British casualties at the Barrack Hospital of Scutari in Turkey during the
Crimean War. For two years, she and her team of nurses helped care for the wounded and
ill, all the while keeping careful records of their patients.
Convinced that cleanliness was a major reason for the high death rates at Scutari, Night-
ingale wrote hundreds of publications showing her data in dozens of visualizations. Ulti-
mately, the British government created the Sanitary Commission to investigate the poor
outcomes at the hospital and soon implemented improvements to sanitation, ventilation,
and cleanliness.
Her most famous chart—and the one that bears her name—is drawn around a circle
with values segmented into different time periods. Each slice represents the total deaths in
each month from April 1854 (in the 9 o’clock position in the right diagram) to March 1856 (in
the 9 o’clock position in the left diagram) during the Crimean War. Cause of death is broken
down into three categories: deaths from wounds in battle (pink); deaths from “other causes”
(black); and deaths from diseases (blue). The diagram on the right shows patterns in deaths
over the first twelve months of the war. The one on the left shows the next twelve months after
the Sanitation Commission implemented its reforms in March 1856.
PARTҺTOҺWHOLEփ 301
Florence Nighধngale created her legendary chart to show the disproporধonate share of
deaths from disease like cholera, typhus, and dysentery. Image from the Wellcome Collecধon.
We can think of the Nightingale as a pie chart in which the slices have been expanded in
different directions. The area of each slice represents its value relative to the whole, and the
slices are arranged along the time dimension. The Nightingale chart therefore shows both
changes over time and a part-to-whole relationship.
The visualization demonstrates two major points. First, deaths in battle were only a small
portion of the total number deaths. Deaths from disease—in this case, cholera, typhus, and
dysentery—made up a disproportionate share of deaths. Second, it illustrates how the Sani-
tary Commission, which started its work in the middle of the war, dramatically reduced the
overall number of deaths.
Already in this book we’ve seen variations on the Nightingale chart. The circular column
chart, for example, arranges columns around a central circle radiating outward. The radar
chart also uses a circular layout to show relationships across variables aligned along the
radii. As with many circular charts, the Nightingale can be difficult to read, and comparisons
across the slices are hard to gauge. The major problem with the Nightingale chart is that the
302 փ CHART TYPES
This stacked bar chart shows the same data as in the original Nightingale graph—but is it
as memorable?
outer segments are necessarily larger and therefore are emphasized disproportionately. This
distortion increases as we move farther from the center.
The stacked bar chart above uses the same Crimean War death data as the Nightingale
chart on the previous page. I’ve added a vertical line between March and April 1855 to denote
the break in the two circles. While the bar chart shows the totals more clearly, it does not
make the break between the two years as clear as the original, nor perhaps is it necessarily
as engaging.
The natural question, of course, is whether Nightingale’s charts would be better as some
kind of more standard chart, like this stacked bar chart. Data visualization expert R. J.
Andrews addressed this debate head-on:
Critics suggest that her mortality data is better shown in something more straightforward like a
bar chart. But this is not true: Florence Nightingale made lots of bar charts. No one cares about
them! Her roses gripped 1858 readers and they still hold our attention today.
PARTҺTOҺWHOLEփ 303
In this case, as in many others, we must strike the balance between what is perceptually
accurate and what will engage people and stick in their minds. Or, as author Alberto Cairo
wrote about Nightingale, “I believe her goal wasn’t just to inform but also to persuade with
an intriguing, unusual, and beautiful picture. A bar graph conveys the same messages effec-
tively, but it may not be as attractive to the eye.”
Let’s use the import data from early in this chapter and lay it out as a Nightingale chart.
Again, this chart shows total imports to the United States from each of the seven regions
Nightingale charts can visualize all sorts of data, though they introduce
their own perceptual hurdles.
304 CHART TYPES
50%
Europe & Central Asia
40%
30%
10%
0%
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
A stacked area chart or a line chart are both simple alternatives to the Nightingale chart.
around the world from 1996 through 2016. Each region is stacked on top of each other, and
the year data is arranged from clockwise from earliest to latest, beginning in 1996 at the top.
We could also use a stacked area chart or line chart to show these same data. Here, notice
that the smaller values appear slightly larger in the Nightingale chart because, by definition,
those areas become larger the further they are from the center of the circle. But as we’ve seen
throughout, there might be cases where the Nightingale is useful over these standard chart
types—it’s more compact and looks very different from these basic chart types. And that
may be a goal in itself.
VO RO NOI DIAGRAM
Named after Georgy Voronoi, a Russian mathematician who lived around the turn of
the twentieth century, the Voronoi diagram divides a space of some number of points—
called sites—into a corresponding number of shapes (polygons), which meet at bound-
aries that do not overlap. If we drop a new point in the space, it will be closer to the site
in its area (polygon) than to any other in the space. Voronoi diagrams are sometimes
used for geographic data, but I’ve placed them in this chapter because they can also show
part-to-whole relationships. You can often find these diagrams in the biology literature
(for cell structures), ecology (for the study of forest growth), and chemistry (for molecu-
lar positions).
PARTҺTOҺWHOLEփ 305
The interesting property of Voronoi diagrams is that the border of each polygon is the same
distance to the two nearest sites. In other words, each polygon is defined so that the distance
from the edge of the polygon to its site is the shortest it can possibly be. When three borders
meet, they result in a point—called a vertex—which is equidistant to the three nearest sites.
There are a variety of algorithms to determine the shape and position of the various polygons.
That explanation is complex, so let’s take a simple example. Say you live in a city with
nine fire stations, and a fire breaks out at a building somewhere in the city. Which fire sta-
tion should respond? The locations of the nine fire stations are the generating points, and the
polygons that surround each tells you which fire station should respond to the fire on the prin-
ciple that the closest fire station should respond. (Obviously, I’m ignoring issues of roads and
bridges and other obstacles to actually get to the fire, but more complex algorithms can factor
in these kinds of obstacles.) As you can see on the next page, the Voronoi diagram splits up the
city into its different components, effectively letting us visualize a part-to-whole relationship.
To demonstrate how a Voronoi diagram works, imagine there are nine fire stations located
across the city.
Source: Map from the City Roads project by Andrei Kashcha
Using a Voronoi map, we can divide the city into nine separate areas.
If a fire breaks out, we can identify which fire station should respond based on its proximity.
PARTҺTOҺWHOLEփ 307
Maybe a more common use of the Voronoi chart is to show part-to-whole relationships,
like this graph from Will Chase that shows the populations of countries around the world.
More generally, you may see Voronoi diagrams like this one, which shows the population
of countries around the world categorized by region.
Many people probably don’t know that one of the most famous maps in the history of
data visualization is also a Voronoi diagram. In mid-1854, John Snow, an English physician
and founder of the modern science of epidemiology, plotted each death from an outbreak
of cholera as a small dash on a map of Soho in London. More than six hundred people died
during the outbreak in a little more than a month.
Source: John Snow
John Snow’s famous cholera map is actually a Voronoi diagram. There is a clear clustering
of deaths around the Broad Street pump—seen as a small dot in the center of the map just
to the right of the thickest black bar. I’ve highlighted the dashed line that Snow included to
show the distance from the Broad Street pump.
If we put a purple dot at each water pump in the map, we can create a Voronoi diagram
from Snow’s original map.
PARTҺTOҺWHOLEփ 309
Snow’s map showed a clustering around a single water pump on Broad Street. You can see
a large number of dashes just to the left of the infected pump (the dot with the “PUMP” label
directly in the center of the image) as well as other deaths up and down the street. In this
version of the map (published slightly later than his original), a small dashed line is drawn
around the affected area (which I’ve highlighted in purple). According to Snow, the dotted line
“shows the various points which have been found by careful measurement to be at an equal
distance by the nearest road from the pump in Broad Street and the surrounding pumps.”
We can also apply a direct Voronoi approach by marking each water pump on the map as
a site (purple dots in this map) and construct the polygons around each (purple lines on the
map). Thus, the “Snow Map” is actually the “Snow Voronoi.”
CONCLUSION
Pie charts are the default visualization of part-to-whole relationships because they’re famil-
iar, easy to make, and easy to read. But pie charts are rife with perceptual difficulties and
should be used with care. Limit the number of slices in your pie charts and remember that
right angles are familiar to our eyes and brains, so 25 percent increments are useful markers.
It’s difficult for readers to compare values across pairs of pie charts, but if we use more than
one pie chart, we should at least keep the slices in the same order.
There are alternatives to the pie chart. Treemaps are essentially squarified pie charts,
which can hold more annotation and show hierarchical relationships. Sunburst diagrams
also show hierarchical relationships, though a sunburst diagram with many series can look
complex and cluttered. Voronoi charts show part-to-whole relationships in a different lay-
out. They can also be used to visualize geospatial data, cell structures, or ecological data.
Some of these other graph types may be less familiar to your readers, but they also help
them more accurately discern the data values. Many of the graphs we have covered in ear-
lier chapters—like the bar chart, stacked bar chart, and slope chart—can also be used to
visualize part-to-whole relationships, but they often require explanatory text to make those
relationships clear.
QUALITATIVE
10
U
ntil now, we’ve explored charts that mostly communicate quantitative data. But
there are also graphs that communicate qualitative data, non-numerical informa-
tion collected through means of observations, interviews, focus groups, surveys, and other
methods. The charts in this chapter primarily communicate words and phrases.
We can build narratives and tell stories around qualitative results in ways that can be
more difficult with quantitative data. Downloading a big data set, running regressions, and
creating tables may provide more generalizable results, but readers don’t connect to them
the way they connect to a story. Qualitative data can help tell those stories.
As with quantitative data, sometimes you may want to summarize your qualitative results.
Images in this chapter are drawn from a variety of news and research organizations from
around the world. Unlike previous chapters, they do not use the same style guide, which
helps demonstrate some of the variety in layout, design, and approach. Some of the graphs
in this chapter show an overall view of your data, while others ask the reader to review the
detailed results, text, or quotations. Which you use will depend on your intended purpose
and where you are publishing your results.
ICONS
Data visualization design can go a long way to deliver more content and captivate your
reader. We’ve seen in previous chapters how color, layout, and different shapes can engage
312 փ CHART TYPES
readers and bring them into a graphic. Icons, images, and photographs can similarly draw
your reader in and categorize your qualitative data for their ease.
Iconography, especially, can help visualize your qualitative data. Icons can be purely
decorative, they can represent data (as in a unit or Isotype chart, see page 106), or they
can guide the reader from one phase of a visual to the next. Icons (including emojis) are
themselves a visual language, so they can simplify and communicate ideas and feelings
that are otherwise difficult to express. They may also help readers with certain intellectual
or cognitive disabilities engage with your work. A body of research has shown that these
kinds of visual-graph forms of communication have helped advance successful language
development.
As an example of how icons can be used to support research and analysis, this short
graphic from the Center on Budget and Policy Priorities uses icons to offset the five
ways in which the Earned Income Tax Credit and the Child Tax Credit help families.
The content is inherently qualitative, and the icons organize the text to make it easier for
the reader.
Word clouds are perhaps the most popular and familiar way to visualize qualitative data, but
they are really a way to display quantitative data: the number of times a word appears in a
text. In a word cloud, the size of each word is adjusted according to its frequency in a passage.
You can use icons to support research and analysis.
Source: Center on Budget and Policy Priorities.
31 4 փ CHART TYPES
These graphs are probably best used to visualize overall patterns or where a single value is
obvious and stands out. It is less appropriate in cases where finding the specific values is
especially important.
Word clouds are visually engaging, but they present two primary challenges. First, it is
unclear what the specific frequency of each word is in the text. This word cloud uses the text
from President Barack Obama’s 2016 State of the Union address. You can see that he used
the words America, world, and people frequently. But how much more frequently? It’s hard
to tell. If understanding the exact frequencies of words in the text is important, then this
visual is insufficient.
What about the word Americans in this word cloud? It was also used frequently but its
vertical orientation may have obscured it from your view. This is the second challenge with
word clouds: some words may appear larger (and therefore more significant) than others
because of their length, orientation, font, or color.
Remember, to create a word cloud you must calculate the frequency of each word. After
you have quantified the text, you could use many different visuals—even a bar chart. It’s
much easier to see the most commonly used words in this bar chart than in the word cloud.
It’s worth noting that word clouds generally exclude the most common words, called stop-
words, such as “the” and “at.”
This word cloud shows the frequency of words spoken by President Barack
Obama in his 2016 State of the Union address.
Data Source: The White House.
QUALITATIVEփ 315
America 26
world 23
people 21
work 20
years 18
American 16
Americans 15
economy 15
change 15
better 13
A standard bar chart that shows the ten most frequently used words in Obama’s 2016 speech.
Data Source: The White House.
Another approach to creating effective word clouds may be to separate the overall body of
text into semantic groupings—words grouped by their meaning. Following Obama’s address,
USA Today summarized his speech as follows:
Obama defended the progress made over the last seven years and set out an agenda that will
likely remain unfinished long after his presidency ends: turning back the effects of climate
change, launching a “moonshot” to cure cancer, and a grassroots movement to demand changes
in the political system.
Those important policy aspects of the speech don’t come through in the basic word
clouds, but organizing the text into semantic groups and then creating a set of smaller
multiple word clouds might be a better way. In the next version, for example, you can
better see the most important topics and words in the speech. This obviously requires
316 փ CHART TYPES
Instead of a single word cloud, Hearst et al. (2019) have suggested breaking up
the text into semantic groups.
Data Source: The White House.
more work on the part of the analyst to determine the groups and clusters, but it also
results in a better visualization.
WORD TREES
Another way to visualize text data is to examine the contexts in which different words appear.
Word trees, a version of which was developed by Martin Wattenberg and Fernanda Viegas in
2007, show all of the ways in which specific words are used within a text. The tree structure
shows the combinations in which each word appears within the text, and the words are sized
according to their frequency.
QUALITATIVEփ 317
Word trees show all of the ways in which specific words are used within a text or, in this
case, the Beatles’ song “Come Together.”
Source: AnyChart.
We’ve already seen one word tree in Chapter 8 that depicted how to show hierarchical
relationships. Similarly, this word tree of the lyrics from the Beatles’ song “Come Together”
shows the elements of the song laid out in a hierarchy.
Trees can be used in other ways to show qualitative data. The next tree is based on inter-
views with fifty science bloggers and classifies their interview responses into their goals and
318 փ CHART TYPES
Paige Jarreau used a word tree to classify interview responses into different groups.
motivations. You can see the branching from each of nine different categories into specific
comments and quotes. Word trees suffer the same pitfalls as word clouds: it is difficult to see
the exact frequency of the words. But they do provide an engaging, interesting view of the
text within its context.
SPECIFIC WORDS
Another way to visualize qualitative text data is to combine individual words with a quanti-
tative metric. This 2019 visualization from Matt Daniels at the digital publication The Pud-
ding is a histogram (recall page 179) of the number of unique words used by individual rap
artists. Instead of a set of bars, the author plotted the artist’s name, using color to distinguish
the decade of album release. We see an overall view of the data, for example, that the most
QUALITATIVEփ 319
Just like specific data points can be shown in a scatterplot, specific words can be included
in data visualizations. Matt Daniels from The Pudding used such an approach in his visual-
ization of “The Largest Vocabulary in Hip Hop.”
common number of words is around 4,000. We see a shift over time toward fewer words
(more names in red to the left side of the graph). And we can focus on the details of the
names and eras of specific artists if we want to.
As with other graph forms we have explored—for example, the beeswarm chart (page
206) and network diagram (page 277)—there are a many ways to visualize individual words
in your data. The key is to organize that information logically, so the reader can sense the
pattern immediately and engage with the graph.
QUOTES
If context is important, visualizing individual words is not the most useful way to show your
qualitative data. Sometimes you need to show full quotations.
Following parliamentarian elections in 2016, the Berliner Morgenpost published a
story about how far some of the elected officials lived from their constituencies. Along
with a series of maps that showed the residence of each of six officials relative to the
32 0 փ CHART TYPES
To communicate qualitaধve data, someধmes you need to just show the enধre quote.
The ধtle of this story from the Berliner Morgenpost translates to “Why are you running for elec-
ধon from far away?” The quote from Ero ¿zkaraca, the image on the leđ, translates to,
“The Social Democrat is sধll involved in NeukØlln, although he moved to Frohnau years ago.
The family had wanted a house with a garden, but that was not available in the north of
NeukØlln. YI’m only there to sleep, I hardly know my way around.’ But he is aware that this can
certainly trigger social envy in the consধtuency.”
district they represent, the newspaper also presented a series of direct quotes. Read-
ers could see the crux of the story from the maps and find more detail by reading
the quotes.
As an analyst working with survey data, this might be a difficult task. You may not have
the exact names or pictures of your survey respondents, and even if you did, you might not
be allowed to publish them because of concerns about privacy and security. But it is worth
considering whether a generic quotation with a picture of a person would be sufficient in
making the content more visual and personalized. As the example from an Urban Institute
research project on unplanned pregnancies shows on the facing page, sometimes the quotes
alone are sufficient to help communicate the content.
QUALITATIVEփ 32 1
Again, using speci)c quotes or phrases can be a powerful way to communicate your message.
Source: The Urban Institute
COLORING PHRASES
Passages of text build up from words to phrases to sentences and then to paragraphs.
Depending on the goal of your visual, highlighting specific portions of text may be useful
to summarize your analysis. You can do so by highlighting with color or boldface to make
important passages visible to your reader.
32 2 փ CHART TYPES
The next image is from a Bloomberg News story that annotated and highlighted the tran-
scripts from the Democratic primary debates in 2019. Each color represents a different con-
tent area, which helps the reader easily navigate the piece and quickly see which topics were
discussed most frequently.
The Preservation of Favoured Traces project on the next page depicts the sequence of edits
in Charles Darwin’s On the Origin of Species. Darwin’s final manuscript was actually the fifth
draft, so by color-coding each word in the final text by the edition in which it first appeared,
we can see the evolution of Darwin’s writing and thinking over several years. An interactive
version enables the user to zoom in, search, and explore the text in more detail.
This qualitative data visualization approach gives us a bird’s-eye view of the data, but it isn’t
quantified. You can also use this technique to highlight quotes or passages in text to tag them as,
say, “positive” or “negative,” which might be subjective but also potentially enlightening. Excit-
ing advances in natural-language processing, text analytics, and machine learning algorithms
over the past few years now let researchers more accurately measure tone and semantics.
Bloomberg News highlighted phrases in words in a news story about the Democratic pri-
mary debates in 2019.
Especially in cases where there is a lot of text, highlighting specific phrases draws the read-
er’s attention. Ben Fry used this very approach in his “Highlighting words in Darwin” project.
If showing an entire qualitative data set is impossible, we can simplify and organize the data
into groups or categories. Essentially, this approach creates a table of qualitative data so read-
ers can more easily see the arguments and narratives. With this method, we can use lists to
make it easier for our reader to navigate through what might be a very dense data set.
This graphic from the Texas Tribune organized the major issues in the platforms of the
Texas Republican and Democratic parties in 2018. Combining the qualitative information
The Texas Tribune combined quotes, text, icons, and quantitative data about poll results.
Source: Texas Tribune, texastribune.org
QUALITATIVEփ 325
from the published platforms with quantitative data from other surveys, the reporters pro-
vide a broader view of six major policy issues. In this example for school subsidies, we can
see how each section begins with a set of three icons, a sentence describing the issue, and a
bar chart of survey results from a 2017 poll. Below, separate quotes from each parties’ plat-
form documents and a summary from the reporters bring the entire issue together.
While in previous qualitative graph types, the analyst might need to make a judgment
about semantics or meaning, in this case, the decision is to choose into which category the
person falls. In other words, visualizing qualitative data doesn’t need to be a complex process
but simply a task of organizing the text into ways our reader can easily navigate and process.
CONCLUSION
Presenting qualitative data succinctly is a challenge. Taking a series of interviews and sum-
marizing the results so that our reader can quickly and easily understand our argument is
no small feat. But qualitative data can engage readers in ways that quantitative data may
not. With good design and organization, we can hook our reader with qualitative data and
encourage them to explore the words, quotes, and phrases.
Word clouds are one of the more common ways to visualize qualitative data, but they
introduce perceptual issues depending on the length of the words, font, and layout. They are
engaging, however, so they are useful in certain scenarios. Using recent research, we might
consider ways to split our text into semantic groups and create a small-multiple version of
the word cloud, which assuages some of these perceptual issues.
There is an array of alternatives to word clouds. Showing specific words or phrases, pair-
ing quotes with photographs, or simply using icons in and around text are all ways to illumi-
nate your qualitative data. As we’ve seen in the examples in this chapter, highlighting those
specific passages and phrase often rely on specific design, color, and layout choices. But, as
with many other chart types, practice and experimentation will lead to better visualizations
of our qualitative data.
TABLES
11
Y
es, tables are a form of data visualization. If you want to show the exact amount of
every value in your data, a table might be your best solution. They are not the best solu-
tion if you want to show a lot of data or if you want to show the data in a compact space—but
still, a well-designed table can help your reader find specific numbers and discover patterns
and outliers.
As we’ve seen with other graphs, gridlines, tick marks, and other clutter can crowd a
visualization and obscure takeaways. Tables are especially susceptible to clutter. The same
guiding principles of creating effective visualizations apply here as well—clearly show the
data so that our reader can find the most important patterns, trends, or values; reduce the
clutter of gridlines, extra spacing, and uneven alignment; and integrate the table and the text
by using concise active titles and subtitles, and including unit labels like percentage signs
and dollar signs with care.
In this chapter, we’ll cover ten steps for making better tables.
We must first understand the components of a table before we can understand how and
when to adjust them to improve our data presentation. This diagram shows the ten primary
components of a table. Many of these parallel the parts of a chart that you will see in the Style
Guides section in Chapter 12. As with chart style, some of the style decisions you choose for
32 8 փ CHART TYPES
your tables will be subjective and depend on nothing more than your preferences for shad-
ing colors, font size, and line width.
1. Title. Use concise, active titles. “Table 1. Regression Results” is not particularly infor-
mative. Instead, guide your reader to the conclusion with a title like “A one year
increase in work experience increases annual earnings by 2.8 percent.” Left-aligning
the title and subtitle will line it flush with the rest of the table, creating a grid, which
is easier to navigate.
2. Subtitle. This sits below the title, often in a smaller font size or set in a different color
to differentiate it. The subtitle should specify the units of the data in the table (like
“Percent” or “Thousands of dollars”) or make a secondary point (such as “The experi-
ence effect is greater for men than for women”).
3. Stubheads or Column Headers. These are the titles of your columns. Differentiate
these from the rest of the table cells with boldface or separate them with a line, also
called a “rule.”
TA BLESփ 329
4. Rules. The lines that separate the parts of the table from one another. At minimum,
place rules below the stubheads and between the bottom row and any sources or
notes.
5. Border. The set of lines that surround the table. Including a border around the whole
table depends on how the table is arranged in the rest of the document. Sometimes
you need to add a visual differentiator to set the table apart, and in those cases a bor-
der is useful. If, however, too many lines and borders clutter the document, omit the
border altogether.
6. Columns, Rows, and Cells. Columns run vertically, and rows run horizontally. The
intersecting areas are called cells.
7. Spanner Header and Spanner Rule. The text and line that span multiple columns.
Text is usually centered over the multiple columns even if the specific column head-
ers are left- or right-aligned.
8. Gridlines. The intersecting lines within the table that separate the cells. As with
charts, take a light touch with your gridlines—heavy gridlines clutter the table.
9. Footer. The bottom area of a table where you might include a row for the total or
average. As with the stubhead, we should differentiate this row from the rest of the
table. We can do so by bolding the numbers, separating them with a line, or color
shading the cells.
10. Sources and Notes. The text below a table containing the citation or additional
details or notes to the table. Modern Language Association (MLA) style, for example,
suggests putting the sources first and the notes second.
These guidelines will take us from tables that have too much color, lines, and clutter to ones
in which readers can easily see the important numbers and patterns. On the next page, we
can see how these guidelines move us from the table on the left to the much more clear and
readable table on the right.
Make your column titles clear. Try using boldface text or lines to offset them from the num-
bers and text in the body of the table. It should be clear that the headers are not data values
330 փ CHART TYPES
r;u-ঞomv )-Ѵom -Ѵ|om A1873 May-11 5692.88 34.1 r;u-ঞomv )-Ѵom -Ѵ|om A1873 May-11 $5,693 34.1
r;u-ঞomv vঞm;;m7;uvom B56 -mŊƐƏ ƓƖƏƔĺƏƑ 43.522 vঞm;;m7;uvom B56 -mŊƐƏ ƓķƖƏƔ 43.5
!;v;-u1_ $_-Ѵb-o00 ƒƓѵ Apr-13 3179.49 -9 !;v;-u1_ $_-Ѵb-o00 ƒƓѵ Apr-13 3,179 ŊƖĺƏ
!;v;-u1_ -|_b-vbѴ; ƓƏƐ Mar-11 ƔƏѶƏĺƑѵ 3.2 -|_b-vbѴ; ƓƏƐ Mar-11 ƔķƏѶƏ 3.2
om|u-1|v m];Ѵ-)-Ѵh;u B31 Feb-18 ѵƑƏƕĺƔƒ -1.788 om|u-1|v m];Ѵ-)-Ѵh;u B31 Feb-18 ѵķƑƏѶ -1.8
om|u-1|v b-"_;Ѵ|om ѶƓƏ -mŊƐѵ ƐƏƕƏĺѵƐ 4.31 b-"_;Ѵ|om ѶƓƏ -mŊƐѵ ƐķƏƕƐ 4.3
but categories or headers. In this example, which uses growth in per capita GDP, the column
labels are boldface and separated from the data with a single line.
As with the basic principle to reduce clutter for graphs, you can lighten or even remove
much of the heavy borders and dividers in your tables. There is rarely a case when every
single cell border is necessary. For series that show the total, use shading, boldface, or subtle
line breakers to distinguish these.
Notice in the table on the left how the two columns that show the average (between
2007–2011 and 2012–2016) blend in with the other columns. At a quick glance you don’t even
TA BLESփ 331
Country 2007 2008 2009 2010 2011 Avg. 2012 2013 2014 2015 2016 Avg.
China 13.64 9.09 8.86 10.10 6.36 10.74 7.33 7.23 6.76 6.36 6.12 6.76
India 8.15 2.38 6.95 8.76 6.90 6.30 4.13 5.10 6.14 6.90 5.89 5.63
United States 0.82 -1.23 -3.62 1.68 2.09 -0.30 1.46 0.96 1.80 2.09 0.74 1.41
Indonesia 4.91 4.59 3.24 4.83 3.65 4.47 4.68 4.24 3.73 3.65 3.85 4.03
Mexico 0.70 -0.48 -6.80 3.49 1.90 -0.19 2.15 -0.06 1.45 1.90 1.68 1.41
Pakistan 2.72 -0.36 0.74 -0.48 2.61 0.64 1.34 2.21 2.51 2.61 3.44 2.42
Average 5.15 2.33 1.56 4.73 3.92 3.51 3.52 3.28 3.73 3.92 3.60 3.61
notice that there is a break in the annual series. In the version on the right, a light shade in
those columns sets them apart.
Right-align numbers along the decimal place or comma. You might need to add zeros to
maintain the alignment, but it’s worth it so the numbers are easier to read and scan. Here, for
example, it is much easier to compare the values in the far-right column where the numbers
are right-aligned than in either of the other two columns. To maintain the grid layout, the
column header is right-aligned with the numbers as well.
Along these lines, choose the fonts in your tables carefully. Some fonts use what are called
“oldstyle figures,” in which some numbers drop below the horizontal baseline, the same way
the letters p or g or q do. This is fine for cases where numbers are not a matter of data—like
the numbering of chapters in a novel. But in data tables, they can be distracting and more
difficult to read. Always use fonts that have “lining numbers,” where all the numerals hit the
baseline, and none drop below it.
Notice how the commas and decimal points in the table on the next page don’t line up
with custom fonts like Karla and Cabin. When choosing a font, be mindful that the numerals
are not always the same size. Also be aware that oldstyle figures of Georgia drop some of the
digits below the horizontal baseline (I’ve added an underline in each cell to make this clear).
Once we’ve right-aligned the numbers, we should left-align the text. The English language is
read from left to right, so lining up the entries in that way generates an even, vertical border
and is natural for the reader. Notice how much easier it is to read the country names in the
far-right column than in the other two columns.
Precision to the fifth-decimal place is almost never necessary. Strike a balance between neces-
sary precision and a clean, spare table. The per capita GDP growth rate, for example, is never
TA BLESփ 333
reported to five decimals—that would be unnecessary and suggest a level of precision that is
not supported by the data. This can also go the other way: Don’t report too few digits. Showing
per capita GDP growth as whole numbers masks important variation across countries.
Your use of space in and around the table can influence the direction in which your reader
reads the data. In the table on the left, for example, there is more space between the col-
umns than between the rows, so your eye is drawn to read the table top-to-bottom rather
than left-to-right. By comparison, the table on the right has more space between the rows
Rule 6. Guide your reader with space between rows and columns.
33 4 փ CHART TYPES
than between the columns, so your eye is more likely to track horizontally rather than
vertically. Use spacing strategically to match the order in which you want your reader to
take in the table.
Your reader knows that the values in your table are dollars because you told them in the title
or subtitle. Repeating the symbol throughout the table is overkill and cluttering. Use the title
or column title area to define the units, or place them in the first row only (remembering
to align the numbers along the decimal). If you are mixing units within the table, be sure to
make your labels clear.
Instead of showing just six countries and three years as in the previous example,
what if we need to show twenty countries and ten years of data? In this case, we
might want to highlight outlier values by making the text boldface, shading it with
color, or even shading the entire cell. Some readers will wade through all of the num-
bers in the table because they need specific information, but many readers are more
likely to look for only the most important values. Guiding them to those important
numbers lets them answer their own questions about the data or better comprehend
your argument.
TA BLESփ 335
Reduce repetition by grouping similar data or labels. Similar to eliminating dollars signs
on every number value, we can reduce some of the clutter in our tables by grouping like
terms or labels. In this next example, grouping the names of the country regions reduces the
amount of repetitive information in the first column. You can also use spanner headers and
rules to combine the same entry and reduce unnecessary repetition. Here, besides grouping
the country names, I’ve also applied some of the guidelines discussed so far such as left-
aligning text, right-aligning numbers, and using boldface headers and footers.
336 փ CHART TYPES
While grouping like elements does help reduce the amount of clutter on the page, be aware
that posting tables to the internet may require some concessions in this regard. If you post
tables to websites as images, users will be unable to copy and paste the data from the table to
another tool, and screen readers—which literally step through the table and read the values out
loud (see Chapter 12)—will be unable to read the data values. Instead, because of current con-
straints in web programming languages and formats, you might need to forgo spanner headers
and other special formatting decisions, depending on the tools you use to post it to the internet.
We can make larger changes to our tables by adding small visualizations. Just like highlighting
outliers with color or boldface, you might add sparklines (see page 152) to visualize some data
rather than showing every number. Or you can use small bar charts to visually illustrate a series
of numbers. Or you could use a heatmap (see page 112) and leave the numbers in the table or
hide them, which can help the reader focus on the overall patterns and ignore the details.
We can also embed a chart-type structure right into our table. If you want a full chart
embedded within the table, a dot plot (see Chapter 4) is succinct and can line up well within
the linear structure of a table. You can also use a modification on the standard dot plot to
place the numbers in their relative positions directly in a table.
Jan Feb Mar Apr May June Jul Aug Sept Oct Nov Dec Jan Feb Mar Apr May June Jul Aug Sept Oct Nov Dec
1 10 13 24 22 21 13 16 17 21 28 30 32 1
2 65 57 62 58 52 45 38 33 45 37 27 17 2
3 12 14 15 20 24 26 30 28 42 28 38 12 3
4 32 45 38 51 47 37 51 41 49 50 35 21 4
5 31 39 46 19 92 39 80 56 31 97 10 75 5
6 50 57 61 74 46 20 31 53 86 73 86 59 6
7 30 40 38 47 66 69 52 74 98 78 85 93 7
8 21 38 53 60 80 90 90 94 93 83 70 64 8
9 10 20 30 38 55 38 25 61 44 50 29 28 9
10 12 17 18 19 21 24 35 48 38 42 50 52 10
This table from the U.S. Department of Agriculture Food and Nutrition Service shows the
number of people who participate in the Food Distribution Programs on Indian Reser-
vations. The table presents participation estimates for twenty-four states over fiscal years
2013 through 2016, plus preliminary estimates for fiscal year 2017. Note the very dark, thick
This table from the U.S. Department of Agriculture Food and Nutrition Service
is cluttered and difficult to read.
TA BLESփ 339
gridlines, which make the table cluttered and difficult to read. As we zoom in, we can see that
the numbers are top-aligned in each cell, which cuts them off ever-so-slightly.
There is a better way. Instead of including all of the gridlines—and making them dark
and thick—we can remove them and keep only the line below the column header row. The
A simple redesign of the USDA table removes the clutter and lightens the view.
3 40 փ CHART TYPES
column header text is now bold to distinguish it from the numbers in the table. A line at the
bottom of the table separates it from the note, and the Total row is bolded to set it apart from
the body of the table.
Let’s take this a step further by adding some visuals and color to the table.
Adding a little color to the USDA table—such as a simple heatmap—makes it easier and
faster for our reader to pick out specific values or patterns.
TA BLESփ 341
Adding other visualizations—bar charts or icons denoting change—are other ways to add
visual elements to your tables.
The first example is a heatmap. Until I made this, I didn’t realize by how much program
participation in Oklahoma exceeded the rest of the states. It was only after the row appeared
in dark blue that the magnitude became clear.
Another approach is to maintain the core look of the original table but add additional
visual elements. In the tables above, the version on the left in the above pair adds a new data
point—the average between fiscal year 2013 and 2017—and a bar chart to its right. This small
graphic element gives the table a visual anchor and directs the eye to the states with more
participation. The chart on the right adds the percentage change between 2013 and 2017 and
a small up- or down-arrow to signal the change.
A typical regression table contains point estimates, standard errors, and some symbol (usually
asterisks) to denote the level of statistical significance, such as 1 percent, 5 percent, and 10 per-
cent. Such basic tables are especially useful when readers need the detailed numbers.
342փ CHART TYPES
* p < 0.05, ** p < 0.01, *** p < 0.001 * p < 0.05, ** p < 0.01, *** p < 0.001
A table of basic regression results can be improved following the ten rules
shown earlier.
We can make a table of regression estimates clearer and more visually engaging by follow-
ing the ten table rules and the visualization strategies from earlier in this chapter. We might
also consider putting the dense table in an appendix (in the paper itself or maybe online)
and using a graph in the main body of the paper instead.
Consider this relatively simple regression table that includes the coefficient estimates with
asterisks, standard errors with parentheses, and unreadable variable names in the first col-
umn. Don’t use variable names to list your results! Your reader—even a reader of an aca-
demic journal article—does not know what “_educ” or “expr” means. For our final table, let’s
use real words like “Education” and “Experience,” and use the rules above to make the table
cleaner and easier to read.
We can also convert these kinds of tables into data visualizations. A standard way is to use
a bar chart with error bars, though, as noted in Chapter 6, some research has shown that we
tend to discount the end of the error bar that sits within the bar itself.
Or we could try a dot plot approach (with or without error bars) and maybe use color to
further signify statistical significance. In the graph on the right, solid circles contain esti-
mates that are statistically significant and empty circles are estimates that are not statistically
significant.
TA BLESփ 3 43
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.00
-0.01
Age Gender 71-ঞom ouv Experience Married
)ouh;7
Age Age
Gender Gender
ĚƵĐĂƟŽŶ ĚƵĐĂƟŽŶ
Hours Hours
Worked Worked
Experience Experience
Married Married
-0.02 -0.01 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 -0.02 -0.01 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Ultimately, if you decide such visual elements are unnecessary or insufficient, stick with
the ten rules to make the table clearer and easier to read. Remember, the goal of our tables is
to let the reader to more easily find the important numbers and patterns in the data and not
ask them to wade through clutter.
344փ CHART TYPES
CO NCLUSION
Tables are themselves a form of data visualization, and the same rules apply. Many
researchers and scholars rely heavily on tables, likely because they don’t require much
creative thinking—filled with text and numbers, columns and rows intersect, and details
are left for the reader to navigate and decipher. And while tables are valuable and have
their place, we can use these ten strategies to elevate our tables and make them clearer
and easier to read.
A data visualization style guide does for graphs what the Chicago Manual of Style does
for English grammar. It defines the components of a graph and their proper, consistent
use. Like a writing style guide, a comprehensive data visualization style guide breaks down
the parts of graphs, charts, and tables to demonstrate best practices and strategies to design
and style your charts. Elements like font and color, the widths of lines and style gridlines, and
the use of tick marks are all choices that determine whether a graph is clear, engaging, and
consistent—or whether it isn’t.
The difference between a grammar guide and a data visualization guide is that many of
our data style decisions are subjective. While the word their is objectively different than
they’re, and the use of one in a particular case is either correct or incorrect, there is no objec-
tively correct or incorrect line thickness for a chart. There are, however, certain principles to
consider, many of which we have covered so far. But for the most part, the styles you choose
will reflect you and your organization’s preferences.
Second, it guides those who may not be familiar with (or care about) all the styl-
ing and branding guidelines the organization may value. Instead of asking research-
ers and analysts to compile the data, create the graph, and then worry about which
colors and fonts to use, a style guide makes those decisions easier. Building these
styles into software tools streamlines the process and automates the application of
graph styles.
Finally, a style guide sets the tone and expectations for people in the organization that
the style, look, and details about data visualization are as important as other branding
materials.
Even if you’re an individual working with data, a style guide can be worthwhile. A custom
style guide will make your work more consistent and efficient, and it will build your indi-
vidual brand so your work stands out. A good style guide handles the basic style decisions
for you, so you can focus on more important aspects of creating data visualizations.
As you build your style guide, test the components to make sure you or your team mem-
bers can use and implement them. The style needs of your charts may differ from those of
other branding materials. Colors that might look great in a logo may not work in a line chart
or bar chart. Also remember to treat your data visualization style guide as a living document,
just as you would a style guide for text or design. The guide should change as your personal
or organizational aesthetic changes and evolve alongside changes in publication types and
software tools.
Consider these Marimekko charts from the Economist and the Financial Times. Both
publications have a distinct look and feel, and even those who are not regular read-
ers may recognize the style. This branding is an important aspect of the organization’s
identity.
There can be many sections to a detailed style guide, which we’ll cover in detail in this
chapter, but here are the basics that any data visualization style guide should cover.
1. Graph Anatomy. Where should labels, titles, and other elements be placed? What is
the proper size of charts and should this size differ for different types of output?
2. Color Palette. What colors should be used across graph and data types? Does the
color palette vary across graph types? Does it vary for print and digital products?
3. Font. What font should be used and how should its size, boldness, and position vary?
Should there be one font style for the title and another for the text in the body of the
graph?
The color, fonts, and overall style of these Marimekko charts from the Economist (top) and
the Financial Times (bottom) make them easily identifiable.
352 փ DESIGNING AND REDESIGNING YOUR VISUA L
4. Graph Types. Are there special considerations for specific graphs? For example, are
pie charts forbidden in all situations? Is there a maximum number of series allowed
in a line chart?
5. Exporting Images. How should team members move graphs from their software
tool to the final report or website? Should they use PNG, JPEG, or some other image
format? How should people create those image formats if they are not native to their
software tool?
6. Accessibility, Diversity, and Inclusion. What steps do you and your organization
need to take to make your graphs accessible to people with vision impairments or
intellectual or other disabilities? Are you being mindful of how you’re presenting
results for different races, genders, and other groups?
To set graph styles, we should first define each part of a graph. To illustrate how this can be
done in practice, we can use the basic template in the style guide published by the Urban
Institute, a nonprofit research institution in Washington, DC.
1. OV ERALL DIMENSIONS
Specify the overall size of the chart. This may differ for different product types—an online
graph, for example, is often measured in pixels, while a chart for a print document is typically
measured in inches or centimeters. The dimensions may depend on the tools and workflow
your organization uses. In the Urban Institute guide, the horizontal dimensions for a print
graphic are specified at the top (6.5 inches) to fit on an 8.5 × 11 inch page.
2. FIGURE NUMBER
Figures and tables can either be numbered, lettered, or left alone. You can place it above the
chart title, centered or left-aligned, in a different font size and color. You could also leave it
in-line with the chart title like, “Figure 1. Chart Title.” As you can see in the image, the Urban
style is to put the figure number above the chart title in all capital letters in its standard
blue color. The note includes details for font (Lato Regular), font size (9 pt), capitalization
(Uppercase), and color (RGB: 22 150 210).
DEVELOPING A DATA VISUA LIZATION STYLE GUIDEփ 35 3
The Urban Institute style guide defines each part of the chart and style
choices that apply to them.
3. TITLE
Will the title be aligned with the left side of the chart or centered over the plot space? Left-
aligning the title (and other text objects) has the advantage of creating a vertical grid along
the left edge, which makes it appear more organized and therefore easier to read. You should
also lay out a basic style for the tone of the titles. Will they be purely descriptive or more
active, summarizing the important points of the chart? Will you use sentence case or title
case? The Urban style is to use title case, as detailed in the sample title. Notice also how the
35 4 փ DESIGNING AND REDESIGNING YOUR VISUA L
use of the title is a direction to the user—it should be title case, short as possible, and no
more than two lines.
4. SUBTITLE
If there are going to be subtitles, how will they be used in the chart? Is this a place to insert a
more active statement or is it a place to list the units in the chart? The subtitle is a good place
to include the vertical axis title because, when left-aligned, it’s located close to the top of the
axis. To offset it from the title, you might place it in parentheses, make the size smaller, or
even change the color. In the Urban example, the subtitle is written out in sentence case with
a smaller size and black color.
5. AXIS TITLES
Where will the vertical and horizontal axis titles go? In many software tools, the vertical
axis title is rotated and placed alongside the vertical axis. A better position is to have it
horizontally oriented and positioned above the vertical axis, aligned with the title and
subtitle (or, as mentioned, it might be the subtitle). For the horizontal axis title, you might
need to decide how far below the axis labels it will sit. There are cases—such as months
or years—where the units are obvious and a horizontal axis title can be omitted. Axis
titles can be differentiated by using smaller text or different colors. You should also decide
whether to spell out and capitalize units like “dollar” or “percent,” or to use a symbol. The
Urban style is to place the vertical (y-) axis title above the axis and to have units in paren-
theses; the horizontal axis label sits below the axis in 8.5 pt Lato Italic font, horizontal, and
centered.
6. AXIS LABELS
How these should be formatted? Boldface, italics, different font size? The vertical axis
labels (as distinct from the title) typically sit to the left of the chart, though they can also
be added to the right side if the chart is very wide. For the horizontal axis labels, are
there specific formats for certain units? For example, when using years along the axis,
would a series like 2000, ,01, ,02 . . . be acceptable or should each number be written
out in full?
DEVELOPING A DATA VISUA LIZATION STYLE GUIDEփ 35 5
What color and thickness will you use for the axis lines? Will the tick marks be inside or
outside the chart? Some organizations leave out the vertical axis line altogether, but the hori-
zontal axis line is typically included to give the chart a consistent anchor. I prefer to make
the zero-axis line slightly darker than the other gridlines because it acts as a baseline. This is
especially true in cases with negative values: We want to make it clear that the zero-axis line
is not at the bottom of the chart. Tick marks are likely not needed in the space between the
bars in a bar chart, but may be necessary in a line chart. In the Urban example, there is no
vertical axis line, but the horizontal axis line is a 1 pt black line with major tick marks that
are outside the chart.
8. GRIDLINES
Many charts include horizontal gridlines, though the exact formatting varies. Will they be
solid, dashed, or dotted? How thick will they be? And what color? At what increments will
they be added? Many charts do not include vertical gridlines, though the occasional scat-
terplot will include them to create a visible grid.
Data sources should be documented and note any important modeling or modifications.
A box for sources and notes is typically found at the bottom of the chart, left-aligned with
the vertical axis labels, title, and subtitle. In many cases, the word Source and Note are bold-
faced. The Chicago Manual of Style (section 3.20), for example, suggests placing the source
line above the note line. In the Urban style, the words Source and Notes are in bold face and
ordered in that way.
10. LOGO
If you want to include a logo on the graph, decide where it will go and what size it will be (and
be sure to use a high-resolution image). Logos are often placed in the bottom-right corner,
but sometimes in other places. The advantage of placing it in the bottom-right corner is that
356 փ DESIGNING AND REDESIGNING YOUR VISUA L
it is out of the way of the title/subtitle and sources/notes areas. Urban adds one of its logo for-
mats to the bottom-right area of the graph, with specific instructions for color and spacing.
11. LEGEND
Will a legend be used and if so, where will it go, what size will it be, and what markers will
be used? It is not labeled on this image, but the Urban style guide includes a separate section
that specifies font sizes for other elements of graphs, including the legend.
Will graphs, especially line graphs, include data markers, like circles or squares? Will the
markers be filled or hollow? When will data values be labeled? You may want to set rules
about using data markers for graphs with some number of values.
Determine when data points should be labeled and how they should be placed and for-
matted. The Urban guide has a separate table of font sizes that describes how these labels
should appear.
This will vary by chart type—thickness of lines, space between bars and columns, colors for
each element. You may need a separate section of the style guide to address issues of specific
chart types, depending on the complexity of the charts your organization uses.
As just two other examples of how published style guides define different parts of a chart, the
first image is from the data visualization style guide published by the London Datastore, an
effort by the city of London to open and share its data and resources. Similarly, the Sunlight
Foundation, a nonpartisan organization that advocates for open government, has a style
guide that defines each part of their charts to reflect their styling preferences.
Source: Mike Brondbjerg of the Greater London Authority, reproduced under the Open Government License
A data visualization style guide should lay out specific chart fonts, styles, colors, and sizes.
35 8 փ DESIGNING AND REDESIGNING YOUR VISUA L
CO LOR PALETTES
Color has unmistakable power in our visualizations. It may be the first thing people notice
about our graphs. Color can evoke emotions and draw attention. As Vincent van Gogh
wrote to his brother in 1885, “Color expresses something in itself. One can’t do without it;
one must make use of it.”
Successful brands have recognizable color palettes for everything from their logo
and letterhead to their data visualizations. But a palette that works for company let-
terhead or website may not necessarily work for a line chart with five lines. There are a
number of free online color tools to develop color palettes: Adobe Color, Color Brewer,
Colour Lovers, and Design Seeds are a few examples, and the Appendix contains a longer
list. Besides the basic colors, we will also need different shades and tints for each color
in the palette.
A style guide should contain different schemes to guide chart creators in their color
choices. The easier you make it for an analyst to apply the branding and design elements,
the more time they have to work with the data and develop the best graph for their purpose.
There are five primary color schemes you can apply to your data visualizations.
Binary. Nominal differences divided into two (binary) categories: urban-rural, Democrat-
Republican, agree-disagree.
Sequential. Data values that are logically arranged from low to high should use sequen-
tial color schemes. Low values are usually represented by light colors, and high values by
Low High
DEVELOPING A DATA VISUA LIZATION STYLE GUIDEփ 359
dark colors. Choropleth maps that show poverty rates or population, for example, would use
sequential color palettes.
Diverging. In this scheme, the colors progress outward, growing darker from a central
midpoint. A diverging color palette will share sequential schemes on two different colors and
diverge from a shared, lighter color, for example, deviations from zero or a central number.
Midpoint
Categorical. Color schemes that use separate colors to represent nominal differences, for
example, different race or gender groups.
Highlighting. This is a special case of the categorical color scheme. These color schemes
highlight a certain value or group within the visualization. For example, we could use this
palette to emphasize a single or small group of points in a scatterplot.
Transparency. Not so much a color scheme but a technique for using color, transpar-
ency in graph objects lets us (and our reader) see the object behind it. We’ve seen this
technique a few times already (see the area chart section on page 157 as an example). You
can use transparent colors—with or without a solid border—to make overlapping objects
visible.
360 փ DESIGNING AND REDESIGNING YOUR VISUA L
As an example of how color palettes are defined in practice, this section of the National
Cancer Institute style guide shows primary and secondary color palettes along with a full list
of tints and shades. The Consumer Finance Protection Bureau (CFPB) style guide includes
sets of colors to “maintain CFPB brand cohesion.”
We should also be mindful of readers with color vision deficiency (CVD) or color blind-
ness. About 300 million people around the world have some form of CVD, most of them
There are many ways to define branded colors and styles. The National Cancer Institute (left)
and the Consumer Finance Protection Bureau (right) are demonstrations of two such ways.
DEVELOPING A DATA VISUA LIZATION STYLE GUIDEփ 361
men, and most of them have difficulty discerning between reds and greens, though there are
other forms as well. There are a variety of online color and color contrast checking tools such
as Vischeck.com and WebAim that can be used to test colors.
When choosing a color palette, avoid the rainbow color palette. In most cases, the rainbow
palette is a poor choice for visualizing your data for at least three reasons. First, while a
color ramp from light blue (small data values) to dark blue (large data values) makes logical
sense, it isn’t really logical to say that “purple” means more than “orange.” Second, and more
importantly, the rainbow color palette does not map to our number system. Notice how
wide the green area is in the rainbow palette below compared with the thin light blue area. If
we were to show a unit change from, say, 1 to 2, we might not see a change in greens; but the
same unit change from, say, 9 to 10, in the blue spectrum might shift all the way from teal to
Avoid the rainbow color palette. It doesn’t map to our number system, isn’t logical when
mapped to data, is not comprehensible for people with color vision deficiency, and does
not translate to grayscale.
362 փ DESIGNING AND REDESIGNING YOUR VISUA L
navy. Finally, the rainbow palette is not consistent for people with CVD (the middle image)
or when printed in black and white (the last image).
Finally, be mindful that colors can reinforce stereotypes or hold different meanings in dif-
ferent cultures. For many years, pink and blue colors were used to differentiate data values
for women and men. But in modern-day western cultures, these colors come with gendered
stereotype baggage: pink suggests weakness and blue suggests strength. Interestingly, this
was not always the case—up until about the mid-twentieth century, it was the opposite. In
her book, The Secret Lives of Color, Kassia St. Clair writes, “Pink is, after all, just faded red,
which in the era of scarlet-jacketed soldiers and red-robed cardinals was the most masculine
color, while blue was the signature hue of the Virgin Mary.” Instead of the standard pink-
blue pairing, consider using other color combinations such as purples and greens (as in the
Telegraph newspaper) or blues and oranges (as in the Guardian).
More generally, also consider how different cultures use and perceive different colors. In
western cultures, for example, the color red may evoke emotions of passion and excitement
and has both positive and negative associations. In eastern cultures, however, the color red
represents happiness, joy, and celebration. In India, red relates to purity, and in Japan it is
associated with life, anger, and danger.
A data visualization style guide should define typefaces (or fonts) for each part of the chart.
You probably don’t need more than two different fonts, and one will usually suffice. Also
remember that you can vary the look of a single font by changing its thickness (thin, bold),
angle (italics), and color.
BE WA RY OF CUSTOM FONTS
A custom font can differentiate your style from the standard fonts found in, for instance,
the Microsoft Office package. But beware: using custom fonts requires that they be installed
on any machine that shows the graph. While custom fonts make your graphs stand out,
DEVELOPING A DATA VISUA LIZATION STYLE GUIDEփ 363
A data visualization style guide should also define appropriate fonts to use and when to
use them. The BBC (left) and US Department of Agriculture (right) are just two examples of
how to provide this guidance.
they can also raise trouble when sharing files or presenting from a laptop different from
your own.
Default fonts like Century Gothic, Tahoma, Trebuchet MS, and Verdana are examples of
effective fonts for data visualization that are available on most operating systems but are less
commonly used and therefore appear more novel.
The BBC style guide for data visualization in Tableau (on the left) includes a typography
section that demonstrates which fonts to use, where, and how to align them to the broader
chart space. Their Reith Sans font is not a default font type, and they must therefore make
sure that everyone in the organization has that font installed on their computers. The Visuals
Standards Guide from the U.S. Department of Agriculture (on the right) features a broad set
of fonts that are used across their publication types and at least two of the three (Arial and
Times New Roman) are typically default fonts.
364 փ DESIGNING AND REDESIGNING YOUR VISUA L
Another section you might want to include in your style guide is a set of examples or instruc-
tions about specific chart types. Your organization may want to specify certain styling or data
visualization best practices that differ chart to chart. You might also include examples of less
common chart types to broaden your organization’s data visualization toolbox, just like you
did while reading this book.
Start by constructing guidelines for the most common chart types your organization uses. For
example, you might specify that dual-axis line charts should never be used (see page 143) or that
pie charts should have some upper limit of series (see page 289). There are also more granular
specifications, such as where exactly labels should sit in a stacked bar chart or whether data points
on a line chart sit on or between the tick marks. Or you might specify never to include tick marks
on bar charts, or that whenever data labels are included, gridlines and tick marks must be omitted.
Another issue you might address in the specific chart area is how to manage the color pal-
ette among different data series. If the main colors in your palette are blue, red, and orange,
the order of those colors may change if you have two or three series or may vary for, say, a
paired bar chart versus a stacked bar chart.
You might also include data visualization tips and tricks. These five tips are listed at the
very top of the Urban Institute’s guide. You can develop your own rules and tips or borrow
those published by other groups and organizations.
EXPORTING IMAGES
Once a visualization is ready for external consumption, the chart creator must export it to a
usable file format. This is another opportunity for things to go wrong: improper exporting
might compress the resolution and pixelate the whole image. You can see the difference in
resolution between these two versions of the same chart from Chapter 4. You can spend all
the time you want creating a great, effective graph with clear colors and fonts, so don’t waste
all that effort with a blurry, hard-to-read final image.
Choosing the right file format for your visualizations is key. There are many file formats
from which to choose, each with its own advantages and disadvantages. The biggest difference
between image file formats is whether they are bitmap or vector. Images in bitmap format
(also called raster) are stored as a series of squares (called pixels), each assigned a specific
color. When you take a bitmap image and stretch it out, the pixels get larger and the resolution
falls. You may have seen something like this if you put a photograph in a document and then
tried to make it bigger—each pixel is now larger, and the crispness of the image deteriorates.
The other image format is vector. As opposed to bitmap images, vector images contain
information about the actual shape in the image. A vector image is recreated when you
Brazil
Russia
Japan
Mexico
Ethiopia
Vietnam
Germany
Turkey
France
Italy
When moving images from your data visualization tool to your final product—either a
report or web image—be sure the image has a sufficiently high resolution. The key is to
test the image before you publish or post it.
366 փ DESIGNING AND REDESIGNING YOUR VISUA L
stretch it, so it won’t lose resolution the way bitmap images do. Vector images are also called
“resolution-independent” because they can be stretched forever without ever losing sharp-
ness or detail. It might not surprise you to learn, then, that one of the biggest downsides of
vector images is that the file size can be surprisingly large.
How you and your organization guide the export of graphs from software tool to the final
product depends on a variety of factors, including the primary data visualization software tool,
operating system, and where the final output will appear: Will it be in a PDF report? A standalone
image on a website? Embedded in a tweet? The best strategy is to try a variety of approaches,
but double-check the final product to make sure you have the sharpest, clearest image possible.
Many people with vision impairments rely on screen readers to navigate the internet.
A screen reader reads the content on a screen aloud to the user, so if you post a graph with
the filename “Image1.png,” that is what the user will hear. People with other physical, cogni-
tive, or intellectual disabilities may have difficulty reading your work or using your website
if you have not taken into account their accessibility needs. Accessibility also extends to
whether people can access the internet and the speed of their Internet connection. It is worth
considering how your content (and website more generally) can be made more accessible by
users who may require different levels of assistance.
To create accessible content, you might follow the guidelines laid out in Section 508 of
the Rehabilitation Act of 1973. Section 508 requires U.S. federal government agencies to
DEVELOPING A DATA VISUA LIZATION STYLE GUIDEփ 367
develop, procure, maintain, and use information and communications technology (ICT)
that is accessible to people with disabilities. This means that federal agencies that fall under
Section 508 compliance rules must make their ICT—such as online training and websites—
accessible for everyone.
One 508 standard for images that we can all apply is to use “alternative text” (commonly
called “alt text”) in our images. Alt text succinctly describes the content in the image. For
data visualizations, this might be text communicating the general conclusion or message of
the chart. In other words, what is the single concise sentence that summarizes your chart?
You can find some basic issues to consider by following the recommendations of the Web
Content Accessibility Guidelines (WCAG), an international group charged with leading “the
Web to its full potential.” With respect to accessibility, WCAG defines four main areas:
1. Perceivable. Information must be presented in ways that users can perceive them.
This might mean making non-text content available to other forms people need such
as speech, symbols, or large print. Text should have sufficient color contrast with the
background and images should have information (“alt tags”) that will make them
readable by screen readers and other assistive technology.
2. Operable. Make all functionality available from a keyboard. This means, for exam-
ple, that using the tab, enter, and space bar keys enables the user to navigate the page
and each interaction can be triggered.
3. Understandable. Text content should be readable and understandable, and web
pages should operate in predictable ways. For example, significantly rearranging the
content on the page can make content more difficult to read and understand.
4. Robust. Online content should be robust enough to be compatible with current and
future users as well as assistive technologies. This might mean, for example, develop-
ing the website in such ways that screen readers and other assistive technologies can
accurately interpret the content.
To date, there are no concrete rules about how to make a website completely accessible,
though there are existing threads of research exploring how to use different assistive tech-
nologies to make visual content more accessible. Computer operating systems, browsers,
and programming languages change and evolve, and any accessibility guide would be trying
to hit a moving target. What we can do, however, is to consider how people with different
abilities can or cannot access our content. A lot of these strategies are just good practices we
368 փ DESIGNING AND REDESIGNING YOUR VISUA L
can use to more effectively communicate our work with text and explanations. Considering
better accessibility then leads to better usability for everyone.
Another issue to keep in mind in data visualization is how you refer to different groups.
You may have considered this when using terms like “Black,” “African American,” or “His-
panic” in your writing, tables, or graphs. Use the phrasing accepted and recognized by your
audience and the communities you are studying. Consider the lived experiences of the peo-
ple and groups you study and write about. Also consider using “people-first” language, such
as “people with disabilities” instead of “disabled people.” It is important to remember that
data are a reflection of the lives of real people.
This also applies to the layout of your graphs and the language you use. How do you order
the bars or lines in your tables and graphs? Is it alphabetical, based on sample size, or is it
based on some unknown, arbitrary decision? Again, there are not many answers to these
questions, but it is worth taking some time to consider approaches and strategies to make
your work more accessible and inclusive of different groups.
There are not necessarily right or wrong answers to some of these questions and style deci-
sions. Whether the thickness of your gridlines is 1 pt or 2 pt, one shade of gray or another—
these are primarily style decisions, but they are also functional decisions. As you saw in the
first chapter, the goal is to emphasize the data over the gridlines, tick marks, and markers.
An effective, comprehensive data visualization style guide is best developed at the organi-
zational level. If possible, bring your design and data teams together to determine branding
guidelines that meet the needs of your organization, including data visualization. If your
organization does not have these divisions, or if you are working to develop your own indi-
vidual style guide, you might reach out to experts or refer to other published style guides to
develop branding guidelines and styles.
Remember to treat your data visualization style guide as a living document. Revisit the
guide as technologies and trends change. And remember to be flexible to the different needs,
tools, and skills in your organization. Creating an instructive and clear guide that can be
accessed and implemented by everyone can serve you, your organization, and your reader.
REDESIGNS
13
B
y this chapter, your data visualization toolbox contains much more than it did when you
began this book. We’ve seen dozens of graphs, many of which may have been new to
you. As you develop your own eye for data visualization, you’ll find places where these new
graph types may be especially useful.
In this chapter, we’ll cover a handful of data visualization redesigns. The graphs I choose
to redesign here are not all especially bad graphs. Some are simply chosen because I believe
there are more effective ways to plot the data. My goal is not to criticize these chart creators
or their efforts but to demonstrate how the lessons we have learned can be applied to making
data visualizations cleaner, clearer, and more effective.
The changes made here are by no means the only ways to modify these graphs, but each
redesign follows the guidelines discussed throughout this book. In general, there is no “right”
or “wrong” approach, just different ways of making improvements. As you develop an eye
for better data visualization design, you will develop your own aesthetic and preferences.
Take a moment and examine this bar chart from the U.S. Department of Agriculture that
shows the number of harvested acres for five major crops in the United States for six differ-
ent years. What do you see first?
My guess is you saw what I first saw: The acreage for all five crops increased over time.
Your second observation, which quickly follows the first, is that cotton acreage (the second
370 փ DESIGNING AND REDESIGNING YOUR VISUA L
group) fell in the last year. Unlike the other groups, the last bar for cotton (the green bar) is
shorter than the bar for the preceding year. But it doesn’t jump out at you because there is so
much ink and color in the graph.
If the goal with this chart is to show relative trends in acreage among five crops, a bar
chart is a poor choice. The paired bar chart is good at showing exact values, but the relative
trends are not clear or immediately evident.
We could redesign this as a simple line chart.
Here, the drop in acreage for cotton is very clear, as are the relative sizes of the five crops. In
the bar chart, I couldn’t see immediately that rice acreage sits right in the middle of the five crops,
but here I can see that right away. I didn’t use a legend here, as might be the default approach, but
instead added the labels at the end of each line, using color to link them with the lines.
Midpoint acreage for major field crops, 1987-2012
(Midpoint acreages more than doubled for all five major field crops)
1,200
1,000 Wheat
Cotton
800 Rice
600 Corn
Soybeans
400
200
0
1987 1992 1997 2002 2007 2012
Source: U.S. Department of Agriculture
800
800
633
567
600
400 450
404
295
200 243
200
0
’87’92’97’02’07’12 ’87’92’97’02’07’12 ’87’92’97’02’07’12 ’87’92’97’02’07’12 ’87’92’97’02’07’12
Source: U.S. Department of Agriculture
Two ways to redesign the USDA bar chart: A line chart or a cycle chart.
372 փ DESIGNING AND REDESIGNING YOUR VISUA L
Another approach is a cycle chart. Instead of putting the lines together, this cycle chart is
essentially a small multiples line chart where each crop gets its own panel. The advantage is
that there’s more space for the graph and it’s perhaps a little more engaging because it’s dif-
ferent. The disadvantage is that relative patterns are slightly less clear than in the line chart.
Let’s go back to page 14 in Chapter 1 and consider the perceptual rankings diagram. At the
very top are graphs positioned along common scales—the bar chart or line chart with a
single horizontal axis, for example. One step below are those graphs that are not positioned
along common scales graphs. It is slightly harder to accurately assess the values in these.
This graph contains data from both sections of the ranking diagram. We can clearly dis-
cern the differences between the values of the blue series (Functional assignment) because
they all sit on the same vertical baseline. We are not as well equipped, however, to similarly
We can barely see that the value for Poliধcal Insধtuধons is larger for All countries than Zambia.
Source: Roth and Malik, 2016
REDESIGNSփ 37 3
assess the values for the other series, because they don’t share the same baseline. You can test
this yourself: Is the value for Political Institutions (the yellow series) larger for “All Coun-
tries” or Zambia (the first two series)?
Instead of packing all of the data onto a single chart, we can break it into five separate
charts. In this case, each series is given its own vertical baseline, so it’s easier to make com-
parisons across countries within each series. The important point with making a graph like
this is that the horizontal space for each series is the same. If we shrank the space for “Fiscal
Institutions,” for example, it might look like those values are larger than others.
This approach, however, doesn’t tell you much about the overall values between countries.
There’s nothing wrong with adding a series for the overall Total, again, as long as we use the same
horizontal spacing. In other words, the space between each gridline is the same. This approach
works even better in cases where the values sum to the same total or to 100 percent, because the
total length will be the same for all of the bars and thus a Total segment is unnecessary.
374 փ DESIGNING AND REDESIGNING YOUR VISUA L
When breaking up stacked bar charts, it is sometimes important to include the totals.
Each year, the Board of Trustees of the Federal Old-Age and Survivors Insurance and Fed-
eral Disability Insurance Trust Funds report on the current and projected status of the U.S.
Social Security program. The Trustees are responsible for estimating the current and future
financial picture of the program to communicate to the public and policymakers the chal-
lenges the program faces. The Social Security Technical Panel is an independent expert pan-
el responsible for reviewing the work of the Trustees, including the methodological details,
economic and demographic assumptions, and the Trustees communication efforts.
The 2019 Technical Panel placed an emphasis on this latter category: “The Panel believes
that trust in public institutions is enhanced by greater understanding . . . In this context,
we believe it is paramount for the Trustees to communicate clearly and effectively with the
general public about its finances.” The panel emphasized clear, plain language, a focus on the
core message, and better data visualizations in the Trustee’s work.
REDESIGNSփ 37 5
A CLEANUP
The first example is a relatively simple clean-up rather than a wholesale redesign. This
line chart—which has appeared in virtually every Trustees Report—shows the time series
of the basic finances of the Social Security system. System income (taxes paid into the
system) are set next to system costs (benefits paid to beneficiaries) for a short historical
period (here from 2000 to 2018) and out in the longer projection period (here from 2018
through 2092). Two sets of costs are shown: one that shows how many benefits are sched-
uled to be paid (the dashed line) and one that shows how many benefits can actually be
paid (the solid bold line).
The existing graph has annotation and labels to help the reader better understand the
content and the concepts. A small table near the bottom of the graph lists benefit shares in
The Social Security Administration (2019) has published this graph showing the basic
finances of the Social Security system for many years.
37 6 փ DESIGNING AND REDESIGNING YOUR VISUA L
25%
20%
Cost: Scheduled but not
fully payable benefits
Cost: Scheduled and
payable benefits
15%
0%
2000 2010 2020 2030 2040 2050 2060 2070 2080 2090
Some basic cleanup and annotation improves the clarity of the Social Security finances chart.
specific years. But there is also a lot of ink used on extraneous details: horizontal and vertical
gridlines and tick marks for every percentage point and year.
Let’s take a simple approach to redesigning this graph by removing some of these
extraneous details and markers. Here, I’ve removed the vertical gridlines and all of the
tick marks. I deleted the small table and instead directly labeled the years those num-
bers referenced. I used some slight color here—which is consistent with black-and-white
printing—and added a gray box to the projection period (after 2018) to draw attention to
the imbalance.
The next graph appeared in the 2011 Technical Panel Report. It shows the sensitivity of dif-
ferent assumptions of the Social Security model. Most of the six Technical Panel Reports
published since 1999 include this information as a series of tables. But in 2011, the Panel
REDESIGNSփ 37 7
Gridlines, rotated text, and general clutter make this graph hard to read.
Source: 2011 Technical Panel Report on Assumptions and Methods
37 8 փ DESIGNING AND REDESIGNING YOUR VISUA L
50-year horizon:
2011-2060
75-year horizon:
2011-2025
Rearranging the plot and removing some of the clutter makes the graph easier to read.
reduces the extra lines and dots in the graph. By using a box, one end can encode the low-
cost value; one end, the high-cost value; and a middle marker, the intermediate value. We
can also place the labels for each metric right next to the first set of boxes though we could
also repeat them if we thought it was necessary.
In late 2017, there was a runoff election for a U.S. Senate seat in Alabama. In a tense and com-
petitive election, journalist Sarah Slobin (then at Quartz) wrote a story about the relation-
ship between voting behavior and the distribution of enslaved people in 1860. Slobin wrote
that, “[W]hile correlation is not causation, there is a startling visual parallel when you zoom
in to Alabama . . . and compare it to how Alabama just voted this week.”
REDESIGNSփ 37 9
Using an 1860 map from the US Bureau of the Census and voting results in the 2017
Alabama Senate election, we can see similar bands of darker colors running through the
middle of the state.
Source: Author’s rendering based on original chart from Quartz. Map from the US Bureau of the
Census; voting data provided courtesy of Sarah Slobin
Using voting data from that election, there is a clear horizontal band of dark blue (repre-
senting more Democratic votes) in the map on the right. The map on the left, from an 1860
map published by the U.S. Bureau of the Census, shows a streak of darker colors represent-
ing counties with a larger proportion of people held in bondage. Slobin writes: “If you focus
on the ‘black belt’ moving horizontally across both maps, you can see that in areas with a
history of slavery, the vote went to [the Democratic candidate] Jones.”
Though visually striking, this pairing forces the reader to jump between the two maps to
see the similarity in the bands. Can we take a different approach?
The share of the vote going to the Democrat in Alabama’s sixty-seven counties ranged
from 16.1 percent to 88.1 percent. These roughly (though not perfectly) overlap with fifty-
one counties I could identify in the Census Bureau map, which range from 3.1 percent to
78.3 percent. (There are some data issues we will ignore here as counties have changed,
merged, or broken up between 1860 and 2017).
380փ DESIGNING AND REDESIGNING YOUR VISUA L
Plotting the two variables in a scatterplot lets us more easily see the positive relationship
with the (darker) circles in the top-right part of the graph marking those twelve counties
along the east-west corridor.
Slobin’s original maps are visually striking. For a news story, they may well be the best way
to show the data. It’s easy to see the basic band pattern running through each map. By com-
parison, the scatterplot may take some additional explanation for a casual reader who may be
less familiar with this graph type. We could publish both graphs to pair the visually-striking
maps—to draw readers in—with the more technical scatterplot for those who want to see the
detailed comparison. If I were publishing this in a peer-reviewed academic journal, I would
lean toward the scatterplot because it clearly shows the association between the two series.
In reviewing a report from the National Assessment of Educational Progress (NAEP) on the
achievement gaps between Black and white students, I came across this chart, which shows
REDESIGNSփ 381
differences in school achievement scores for Black and white students, arranged by scores
of Black students.
Let’s focus on the bars on the far-left side of the graph. There are three numbers here: 28
percent, 32 percent, and 60 percent. The numbers in the green boxes show test scores for
white (60 percent) and Black (28 percent) students, and the middle number shows the gap
between the two groups (32 percent). But the green boxes make it appear as if the 28 per-
cent represents a range of numbers, from, say, 22 percent to 28 percent. By using rectangles
instead of points or markers, it resembles a stacked chart rather than the dot plot, which was
likely the intention.
This graph from the National Center for Education Statistics shows the percentage of stu-
dents eligible for the National School Lunch Program.
3 82 փ DESIGNING AND REDESIGNING YOUR VISUA L
Changing shapes and removing some of the clutter makes the graph from the National
Center for Education Statistics easier to read.
As an alternative, let’s make it a true (vertical) dot plot. We can replace the green boxes
with green circles and connect them with a gray, vertical line. Now we perceive the green
circles as specific points rather than ranges or a stacked set of values.
You might also notice that I deleted the legend and labeled the three series directly on
the left chart. I didn’t repeat the labeling in the chart on the right for two reasons: First,
because the gaps are smaller, there is less space for the labels. And second, the reader
doesn’t need to be reminded of the definition of each dot and line at every single occur-
rence on the page.
Every quarter, the U.S. Bureau of Economic Analysis (BEA)—the federal agency responsible
for producing some of the most important measures of the U.S. economy—releases their
REDESIGNSփ 3 83
This bar chart from the Bureau of Economic Analysis’s quarterly report does not match
what’s written in the text.
report about changes in gross domestic product (GDP). And each quarter, they publish a
press release on changes over time and in specific industries.
The graph above is from the third quarter of 2014 press release. It shows the “Real
Value Added”—a measure of each industry’s contribution to GDP—for major indus-
tries in the country. Given what you’ve learned so far, there are a variety of things
you might change to make this graph more effective. You might directly label the
bars, rotate the vertical axis legend and place it near the title, and lighten some of
the gridlines.
More importantly, let’s take a look at what this graph is supposed to show. Here are the
six bullet points that surround the Real Value Added (RVA) by Industry graph in the BEA
document:
Mining increased 25.6 percent, after increasing 11.5 percent. This was the largest
increase since the fourth quarter of 2008.
Real estate and rental and leasing increased 4.4 percent, after increasing 0.9 percent.
Real value added for manufacturing increased 0.5 percent, after increasing 6.8 per-
cent. Durable-goods increased 7.0 percent following an increase of 8.0 percent, while
nondurable-goods decreased 6.6 percent, after increasing 5.4 percent.
Agriculture, forestry, fishing, and hunting increased 7.6 percent after increasing
14.2 percent.
Wholesale trade continued to show strong growth, increasing 7.3 percent, after
increasing 6.5 percent.
What do you notice about how these points are arranged? Each one details how RVA
changed between the first and last period for each industry. The structure of the graph, how-
ever, is formatted to compare across industries within each period.
A better approach for the graph would match the text and show the inverse: the change
within each industry across periods. A paired bar chart and dot plot are two effective ways
to do this. In the paired bar chart, I’ve sorted the data according to the most recent period
(2014:Q3) to subtly guide the reader to the best-performing industries. In the dot plot, the
data are sorted based on the change between the two periods—the largest positive changes
are at the top of the graph and the largest declines at the bottom.
Mining
20
Real estate and
rental and leasing
10
Wholesale trade
Durable goods
0
Agriculture,
forestry, fishing
and hunting
-10
Mining Finance and Agriculture, Wholesale Durable Real estate Nondurable Nondurable
goods
Insurance forestry, trade goods and rental goods
fishing and and leasing -10 -5 0 5 10 15 20 25 30
hunting
Source: Bureau of Economic Analysis Source: Bureau of Economic Analysis
In either case, these two graphs better illustrate the takeaways from the text. They are
sorted by industry so that now, for example, you can more easily see that, “Mining increased
25.6 percent, after increasing 11.5 percent.” Your data visualizations should not be intended to
break up long sections of text or to provide a “visual break.” They are there to support your
argument. Integrate them with your writing for a seamless reading experience.
As I mentioned in Chapter 5, I’m a big fan of line charts. They clearly show changes over
time, and everyone knows how to read them. Consider, however, this line chart in a 2012
Economic Policy Paper from the Federal Reserve Bank of Minneapolis.
16
Ireland
12
Spain
percent GDP
Greece
Portugal
4
Germany
Italy
-4
2005 2006 2007 2008 2009 2010 2011
This line chart, originally published by the Federal Reserve Bank of Minneapolis,
simply cuts off the data for Ireland.
Source: Author’s rendering based on original chart from Arellano, Conesa, and Kehoe (2012).
3 86 փ DESIGNING AND REDESIGNING YOUR VISUA L
Notice anything strange about this graph? Anything aside from the title split into three
parts (“Figure 2” in the top-left; “Net government borrowing” centered over the graph; and
“percent GDP” along the vertical axis)? Anything besides the equal-weighted gridlines even
though zero is not at the bottom? Or the mix of pastel and bright colors?
How about how the line for Ireland shoots off the top of the graph? You must have a very
good reason not to show all of the data on the graph, and putting the value in the footnote
does not count!
The chart creator here faced a problem: Ireland had a debt spike in 2011, far more than
the other countries. To show all the data in one chart would scrunch up and lose the detail
among the other countries.
But there is a better solution: Use two graphs. One that includes Ireland with a verti-
cal axis that ranges from −10 percent to 35 percent, and another with a vertical axis
from −4 percent to 18 percent that shows the detail among the other countries. I could leave
these as equal-sized charts, or even make the second one smaller in a sort of zoom-out/
zoom-in comparison. In either case, I use the subtitle to explain that one graph includes
Ireland and the other does not.
Do not be afraid to use more than one graph, if that’s what it takes to clearly communicate
your argument. We are now a digital-first society—using more space only requires more
computer memory, not more paper.
35
30
25
20
15
Ireland
10 Greece
Spain
5 Portugal
Italy
0 Germany
-5
-10
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
As we saw in Chapter 11, there are many ways to make our tables more visual. We can add
color, icons, bars, or other elements to highlight the important values for our reader instead
of asking them to sift through all the data values.
This table uses different shapes and shades of gray to show the share of firms that engage
in different business activities like design and market research. As the reader, we must
understand which shapes correspond to which percentages and then figure out the different
Author’s rendering of an original chart by Chappell and Jaffe (2018), which could
be improved by changing how the values are displayed.
n
io
t
or
g dia
at
pp
tr
Pr ing urin Me
e su
is
ts ur e/
in
ric stra s
in ct &
th ale n
Ag ini tion
et A l
Ar ult tiv
O es io
R lic na
R il m
El ti n
M ufa on
o
t
a d
a
uc cti
m da
b io
ec on
an ati
na ity
ho rt
Pu fess
W spo
H nce
Ed tru
Ad mo
Fi tric
M rm
Tr tal
In lth
a
er
l
ns
co
an
fo
en
ea
o
Co
Ac
Computerware
Business strategies
Restructuring
Design
Market research
Marketing strategies
Employee training
R&D
0% 50% 100%
n
io
t
or
g dia
at
pp
tr
Pr ing urin Me
e su
is
ts ur e/
in
ri str s
in ct &
th ale n
Ag ini tion
et A l
Ar cult ativ
O es io
R lic na
R il m
El ti n
M ufa on
o
t
a d
a
uc cti
m da
b io
ec on
an ati
na ity
ho rt
Pu fess
W o
H nce
Ed tru
Ad mo
Fi tric
M rm
Tr tal
p
In lth
a
er
s
l
ns
co
an
fo
en
ea
o
Co
Ac
Employee training
Computerware
Business strategies
Restructuring
Market research
Marketing strategies
Design
R&D
0% 50% 100%
This heatmap alternative to the Chappel and Jaffe (2018) chart sorts the data.
REDESIGNSփ 389
shading styles. Of course, triangles don’t necessarily mean “more” of something than circles,
so the rank-ordering of the values is hard to interpret.
Instead, what if we use a monochromatic color ramp moving from a light blue for the
lower percentages to darker blues for the higher values? In this heatmap approach, it’s much
easier to see that there is a lot of time spent on Employee training, the dark blue row towards
the bottom of the table.
We can take this a small step forward and sort the data, which will naturally focus the
reader’s visual attention. The longer labels either require us to use rotated labels, as I’ve done
here, or perhaps rotate the entire graph and change the spacing or cell size so all of the text
can fit.
CONCLUSION
With more graphs in your data visualization toolbox to choose from, and having seen more
graphs and best practices, I’m confident that you’re ready to improve upon your own graphs.
Finding and redesigning even the simplest graph—I find that mining the academic peer-
reviewed literature a good place to start—can help you refine your skills and develop your
own data visualization aesthetic. Like any other skill, practice makes better.
Two important caveats. First, if you critique a graph publicly, keep in mind that some-
one made that graph and that even your well-intentioned efforts to redesign it may not be
appreciated. The chart creator may have had time pressures, software limitations, or organi-
zational demands of which you are not aware. Reaching out to the person who created the
original graph may be worth your effort. Second, try to identify the central goal of the chart
and the possible challenges of the data series. This will help lead you to the best chart type
for the task at hand.
CONCLUSION
N
ow, at the end of this book, your data visualization toolbox has expanded considerably.
Instead of using whatever the default graphs are available in your favorite software
tool, you can now draw upon more examples to visualize your data in the ways that best
serve your reader, user, and audience.
The graphs presented in these chapters have been tested over and over again. Analysts,
researchers, reporters, and scholars have used them with different data sets and layers of
text, annotation, color, font, purpose, and platform. But the set of graphs available to you
is infinite. The bar chart did not exist before someone invented it. Maybe it’s you who will
invent the next great graph type.
Sources of inspiration for new and innovative visualizations are all around us, from pub-
lic and private organizations, the media, data scientists, designers, and artists. The images
on the next few pages are from some popular data visualization projects and stories. They
use different shapes, layouts, approaches, and techniques to visualize data in ways that are
unique to the field and form.
Freelance data designer Maarten Lambrechts maintains a special project website,
Xenographics, with the tagline, “Weird but (sometimes) useful charts.” It is a repository
of “novel, innovative and experimental visualizations to inspire you, to fight xenograph-
phobia and popularize new chart types.” There, you can find the different, unique, and
strange graphs that can be used to show data in different ways. By experimenting with such
forms, we can move the field forward and communicate data and information in new and
better ways.
392 փ CONCLUSION
As you go forth and create your visualizations, continue to explore. There are yet undis-
covered options, forms, shapes, and graphic types that can best show your data. Even if you
come up with a great idea but don’t know how to create it on the computer, you may find
colleagues or partners who can put your ideas in practice.
LESSONS LEARNED
Over the course of this book, we have covered lots of rules, principles, and guidelines. If I
were to sum these up into the absolute universal principles of data visualization, I would
leave you with just these four.
Source: Jan Diehm and Amber Thomas, “Women’s Pockets are Inferior,” The Pudding
People are reading your graph to learn something, and they do that best by seeing the data.
This doesn’t mean you need to show them all of the data, but that you should always high-
light the most important data.
REDUCE CLUTTER
Reduce and remove all of the clutter that distracts your audience from the data or distorts
the representation. Make it as easy as possible for your reader to see the most important
points in your graph.
Directly label your data, remove legends, use active titles, and employ good labels and
annotation. You may need to guide your reader through the graph before you help them
Source: Accurat. Portraits by Wendy MacNaughton; Design and research by Giorgia Lupi, Simone
Quadri, Gabriele Rossi, Davide Ciuffi, Federica Fragapane, Francesco Majno. 2013.
396 փ CONCLUSION
understand the content, but even that lesson can be accomplished with simple, smart labels
and callouts.
Always remember with whom you are communicating. Academic researchers are looking
for different things than practitioners, and practitioners are looking for different things than
managers and policymakers. Try to identify your likely audience—try to talk to them if
possible—and design your graphs to meet their needs. That way you can help them find
insights, make discoveries, and do their jobs better.
FINAL THOUGHTS
I first became interested in data visualization after seeing much of my and my colleagues’
work go unnoticed and unused. I did not come to the field with a degree in design or com-
puter science or data science. And because I did it, I believe you can too. In fact, anyone can
effectively communicate their data by thinking critically about their own work and the needs
of their audience, readers, and users.
APPENDIX 1
DATA VISUALIZATION TOOLS
T
his book is tool agnostic. My goal is not to show you how to create each of the eighty-
plus graphs we explored. There are far too many tools you can use and far too many
approaches within each tool. What matters is not which tool you use, but that it helps you
create the graphs you need to best serve the needs of your audience.
There are many, many data visualization tools available, for use on different platforms
and with different purchasing and subscription options. The number, types, and capabilities
of these tools is constantly changing to reflect updated underlying technologies and coding
languages. Which tool you use will depend on your personal preferences and skills, as well
as those of your colleagues and support within your organization.
Data visualization tools live along a spectrum. On one end are click or drag-and-drop
tools like Excel, which allow the user to click and insert a chart. On the other end are pro-
gramming languages like R and JavaScript that require written code to create data visuals.
The barrier to entry is much lower in Excel—virtually anyone can create a line chart or
bar chart in seconds. It’s much different on the other end, where programming languages
require an understanding of how to write code and the different syntax across the different
languages. Yet programming languages give you substantially more flexibility, while tools
like Excel can box you in within a small subset of graphs.
How “difficult” these tools are depends on the person and sometimes the organiza-
tion. You may have an affinity for computer programming languages, in which case Java-
Script, Python, or R may be appropriate, but for a variety of reasons, your organization
may not allow you to use these open source tools. Many social science researchers use
398 APPENDIX Ǣ
Lower Higher
(Primarily drag- and-drop Barrier to Entry (Primarily programming
or click-based) languages)
statistical languages like SAS, SPSS, and Stata, but, in my opinion, the graphing capabili-
ties of these tools lag behind those of other languages. Some of the drag-and-drop tools
like Excel and Tableau are easier to start with but designing more bespoke visualiza-
tions is either impossible or may require some coding (for example, Calculated Fields
in Tableau).
I used a variety of tools to create the graphs in this book. Many, if not all, could probably
be created in programming language tools like R or JavaScript; a sizable proportion could
be created in drag-and-drop tools like Excel and Tableau. By utilizing a variety of tools,
I found that some were easier to create the graph, but harder to style the way I liked and
some were more difficult to learn while others were more intuitive. This list of tools is not
exhaustive (by far) and I based my decision on which tools to include on my experience in
the field, not on any formal survey or data set. Before you invest substantial time or money
to learn any of these, you should explore the landscape of available tools and products.
Adobe Illustrator. This is primarily a design tool. Adobe Illustrator, and the rest of the
Adobe Creative Suite like Photoshop and InDesign, are the workhorse tools for designers.
The graphing library is actually pretty poor in Illustrator, but you can insert graphs made in
APPENDIX Ǣփ 399
other tools to add more styling, labels, and annotation. The Adobe Creative Cloud is now
primarily subscription-based, but is not a particularly cheap tool to purchase.
Charticulator. Launched in 2018, Microsoft’s Charticulator is an online tool that enables
you to create a custom chart layout. It differs from some other tools in that you don’t select
from predefined charts but instead the creators have transformed chart specifications into
mathematical layout parameters, such as marks (e.g., rectangle, line, or text) and axes (e.g.,
property of one direction in a plot). Charticulator charts are primarily static, but can be
integrated with Microsoft’s PowerBI tool to create interactive visualizations. At the time of
this writing, Charticulator is free to use.
Datawrapper. An online tool from a team based in Germany where you can upload your
data, select a graph template, refine and style, and publish or download. Created charts can
be embedded in websites and can be made interactive. Datawrapper is free for most pur-
poses, and paid versions allow for custom themes and additional exporting options. There
are many tools in this same vein (Flourish and RAW below are two other examples)—some
are better than others with different default options and user experience.
Excel. Likely the primary data and data visualization tool for many people around the
world. As part of the Microsoft Office suite, Excel is not free, but it is your basic click-based
tool. At the time of this writing, you can create more than 16 basic chart types in Excel,
with variations within most of them. There is a basic library of charts that you can expand
with clever “hacking” or additional coding using the Visual Basic for Applications (VBA)
programming language that sits behind all Microsoft Office tools.
Flourish. Launched in 2016, Flourish is an online tool primarily aimed at newsrooms
to help journalists create both static and interactive data visualizations in a drag-and-drop
framework. There are options to customize and further develop Flourish graphs using the
underlying JavaScript framework. Pricing options range from the free Public version to the
paid Personal version (which gives you additional features) and the paid Business version
(targeted towards large teams and organizations). Through a partnership with the Google
News Lab, Flourish provides newsrooms free premium accounts.
Google Sheets. Akin to Excel, Google Sheets is part of the Google suite of tools. It works
very similarly to Excel, though without some of the more sophisticated options. Because it
is based online, the sharing capabilities are somewhat better than Excel (with the side effect
being you need to have internet access to use it).
PowerBI. Microsoft’s business intelligence tool that allows you to create interactive
dashboards and visualizations. It directly links with the rest of the Microsoft Office suite
400 փ APPENDIX Ǣ
(especially Excel) and can be modified and customized in ways similar to Tableau. There is
the free Power BI Desktop version, the paid Power BI Pro, and the Power BI Premium pack-
age for organizations.
RAW. Created by DensityDesign Research Lab in Italy in 2013, RAW was an early project
designed to help creators link spreadsheet tools like Excel to graphic editing tools like Adobe
Illustrator. It is an open source tool, which means you can download the code to further
customize the visualization options. There is also an online platform in which, like other
tools, you upload your data, select the graph, and customize. RAW has a variety of options
for certain non-standard graphs (like streamgraphs and bump charts) that are not typically
available in other tools. RAW is free to use.
Tableau. Perhaps the most popular business intelligence dashboarding tool, Tableau’s
drag-and-drop interface enables you to create interactive dashboards and visualizations.
Like Excel, users have customized their Tableau work to create an array of visualizations
outside the basic graph menu. There are a number of versions of Tableau, from the free Tab-
leau Public (but which means you save your work to the Tableau website) to paid versions
like Tableau Desktop and Tableau Server (for large organizations).
Infogram, Venngage, and Vizzlo. These are just three of the many click-based online tools
that are aimed more for people who want to quickly create infographics and reports. In my
experience, these tools sometimes have more graph options than other online tools, but they
are not always based on best practices. Pricing varies from free packages that usually mean
your data and visualizations can be viewed by anyone, to enterprise packages for large teams
and organizations.
PROGRAMMING LANGUAGES
D3. We first need to define JavaScript. JavaScript is programming language that allows you
to implement information onto a webpage. Every time a web page does something, like dis-
play updates, animate graphics, or play videos, JavaScript is probably involved. D3 is a Java-
Script library for manipulating objects based on data and was developed by Mike Bostock
along with Jeff Heer and Vadim Ogievetsky at Stanford University in the early 2010s. Most
of the interactive data visualizations we currently see on the web are run on D3—virtually
APPENDIX Ǣփ 401
every interactive graph you play with on the New York Times, Washington Post, and Guard-
ian websites is built with D3. Like other programming languages, there is a steep learning
curve to using D3, but you are basically unlimited in the kinds of visualizations you can cre-
ate. D3 is an open source language, which means it is free to use.
Highcharts. Launched in 2009 by a team in Norway, Highcharts—and its cousin tools
Highstock, Highmaps, Highcharts Cloud, and Highslide—is a suite of interactive data visu-
alization tools rooted in JavaScript. You do need to know a bit of coding to use Highcharts,
but templates and libraries help you create the basics of a graph and then add additional
styling and formats. Highcharts is free for personal use and nonprofit organizations; pricing
then varies by number of licenses and package.
Python. Python is used in a wide number of fields and industries from basic and complex
data analysis to web application to artificial intelligence. Like D3 and R, the language is open
source, which means there are a lot of open, free libraries to help you create data visualiza-
tions including, for example, Matplotlib, Seaborn, Bokeh, and ggplot.
R. Conceived in 1992 and initially released in 1995, R is a free, open source programming
language for statistical computing and graphics. R is becoming more and more widely used
for data visualization, especially since the launch of the “ggplot2” package by Hadley Wick-
ham in 2005 (based on the “Grammar of Graphics” by Leland Wilkinson). R allows you to
conduct statistical analyses and create customizable data visualizations; additional tools and
packages can enable you to create interactive visualizations.
APPENDIX 2
FURTHER READINGS AND RESOURCES
SCOT T BERINATO. Berinato’s book Good Charts: The HBR Guide to Making Smarter, More
Persuasive Data Visualizations focuses on graphs for the business community. The book is
wide-ranging and discusses more of the differences between explanatory and exploratory
charts than some of the other leading data visualization books. He has a follow-up work-
book that provides hands-on examples and tutorials.
ALBERTO CAIRO. Author of several books on data visualization including The Functional
Art, The Truthful Art, and How Charts Lie. Cairo is a journalism professor, so his books focus
primarily on creating data visualizations for telling stories to a wide audience. His books
provide fundamental overviews of data, data visualization, introductory statistics, and how
to create visualizations. His most recent book How Charts Lie, helps readers spot lies in
deceptive graphs and how to become better consumers of data visualization.
JORGE CAMÕES. His book Data at Work covers a wide range of data visualization prin-
ciples and strategies, ranging from rules of visual perception to design considerations to data
preparation and visualization.
STEPHEN FEW. Author of several books on data visualization, his Show Me the Numbers
and Now You See It: Simple Visualization Techniques for Quantitative Analysis are compre-
hensive overviews of how to present data effectively and strategically.
404 փ APPENDIX ǣ
ANDY KIRK. Author of two books on data visualization, his Data Visualisation: A Hand-
book for Data Driven Design provides readers with a system to conceptualize and develop
data visualizations, and a process to help readers make design choices that result in clear and
effective visualizations.
JUSSO KOPONEN AND JONATAN HILDÉN. Their Data Visualization Handbook (trans-
lated from Finnish) is a practical guide to data visualization and contains lots of examples of
data visualizations and information graphics.
COLE NUSSBAUMER KNAFLIC. Knaflic’s book Storytelling with Data, and blog of the same
name, provides an introductory treatment of data visualization, and how to pair text with graphs
to tell effective, compelling stories. Her follow-up book, Storytelling with Data: Let’s Practice!,
takes the reader through a series of hands-on exercises to practice their data visualization skills.
ISABEL MEIRELLES. A professor of design, Meirelles’ Design for Information surveys cur-
rent examples of data visualizations for both elements of content and design. Numerous
examples provide a library of visualization types and approaches.
TAMARA MUNZNER. Her Visualization Analysis and Design book approaches data visu-
alization from a more systematic academic-based perspective. It features a unified approach
to reading and creating visualizations, all rooted in the academic literature. Munzner’s book
is more along the lines of a true data visualization textbook.
STEVE WEXLER, JEFF SHAFFER, AND ANDY COTGREAVE. Their Big Book of Dash-
boards is one of the more comprehensive treatments of creating interactive dashboards.
Mostly rooted in the Tableau software tool, the authors review nearly thirty examples to help
the reader determine good design and interactive experiences.
DONA WONG. Wong’s Guide to Information Graphics dedicates individual pages to spe-
cific graph types, how and why to choose the best chart to fit the data, the most effective way
to communicate data, and what to include and not include in different graphs.
APPENDIX ǣփ 405
R. J. ANDREWS. Andrews’ Info We Trust: How to Inspire the World with Data takes a differ-
ent perspective on data visualization techniques, drawing heavily on historical visualizations
and design approaches by luminaries in the field over the past century.
W HI TNE Y BAT T L EҺBA PTISTE AND BRITT RUSERT. The first full exploration of
W.E.B. DuBois’s data visualizations from the 1900 Paris Exposition, W.E.B. Du Bois’s Data
Portraits: Visualizing Black America, is an image-by-image description of these early data
visualizations.
BRUCE BERKOWITZ. His biography of William Playfair shows the real life of the man
who invented “statistical graphics.” Berkowitz goes beyond the statistics and graphics that
Playfair is best known for to tell a detailed story of his life.
MANU E L LI MA . In his two books, The Book of Trees: Visualizing Branches of Knowledge
and The Book of Circles: Visualizing Spheres of Knowledge, Lima explores the long history of
the tree diagram and circular information design. Both take you on a tour of the long history
of both kinds of visualization methods.
SANDRA RENDGEN. Similar to the DuBois book, in Rendgen’s The Minard System: The
Complete Statistical Graphics of Charles-Joseph Minard, she explores the career and story
behind Minard’s graphs, maps, and tables.
The list in Appendix 1 includes just a selection of tools I’ve used for this book. Which tool is
best for you depends on your existing expertise, your needs, and the needs of your audience.
In my experience, these books have the best tool-specific data visualization tutorials. There
are countless online blogs and resources you can also explore.
KEIRAN HEALY. Maybe more of an introduction to basic core principles of data visualiza-
tion, Healy’s Data Visualization: A Practical Introduction mixes in a how-to in R. There is an
online companion with code snippets.
NORMAN MATLOFF. At more than 400 pages, Matloff ’s The Art of R Programming is
a beast—and that’s because it covers everything in R, an essential reference book in your
library.
ERIC MATTHES. Every Python programmer I know loves Matthes’ Python Crash Course:
A Hands-On, Project-Based Introduction to Programming. Beginners and more experienced
coders are really fond of how he leads the reader through the language and the useful exer-
cises in each chapter.
RYAN SLEEPER. One of the more recent (and better) books on creating calculations and
custom visualizations in the Tableau software tool, Sleeper’s Practical Tableau is a great start
to using Tableau.
AMELIA WATTENBERGER. One of the few books about the D3 JavaScript library, likely
because coding examples and snippets are really important, Fullstack D3 and Data Visualiza-
tion (and the companion website) is a comprehensive introduction to D3. This is the most
recent book about D3 (and perhaps the best) because it walks the reader through the entire
process and includes a large amount of code snippets and examples on the book’s website.
There is no one-to-one mapping between data types and graph types. A column chart, for
example, can be used to show changes over time or to compare differences between catego-
ries. There are a variety of resources you can use to help you select a graph, but the ultimate
decision will rest with you, your data, your audience, and your creativity.
THE DATA VISUA LISATION CATA LOGUE. Online graph resource with descriptions,
anatomy, and some video guides to each chart type. https://datavizcatalogue.com/
THE DATA VIZ PROJECT. An interactive website with more than 100 graph types includ-
ing select examples of each. http://datavizproject.com/
THE GRAPHIC CONTINUUM. A poster, smaller sheet, flash cards, and card game I pro-
duced with a friend, The Graphic Continuum shows more than 90 different graph types
grouped into six categories. https://policyviz.com/product/graphic-continuum-poster/
INTERACTIVE CHART CHOOSER. An interactive chart chooser that lets you sort and
filter more than 30 graph types based on the data you are trying to visualize. Links to real
examples help you see these chart types in action. https://depictdatastudio.com/charts/
THE R GRAPH GA LLERY AND PYTHON GRAPH GALLERY. A pair of webpages that
contain hundreds of charts made with the R and Python programming languages. Each
chart contains reproducible code. https://www.r-graph-gallery.com/ and https://python
-graph-gallery.com/
TEXT VISUA LIZATION BROWSER. Nearly 450 examples of ways to visualize text data.
https://textvis.lnu.se/
THE VISUAL VOCABULARY. A poster project from the graphics desk at the Financial
Times, the Visual Vocabulary shows more than 70 different graph types in 9 categories. https://
github.com/ft-interactive/chart-doctor/blob/master/visual-vocabulary/Visual-vocabulary.pdf
XENO.GRAPHICS. A collection of unusual charts and maps to help you expand your
visualization repertoire even further. https://xeno.graphics/
WHERE TO PRACTICE
Maybe the best way to improve and refine your data visualization technique is to create
visualizations. Exploring different data sets, visualization types, and tools can help you refine
your aesthetic, and play with different techniques and forms. There are a few community
projects that you may want to explore to help you on your way.
MAKEOVER MONDAY. A weekly learning project in which participants work with a sample
data set to try to create better, more effective visualizations. It tends to be focused in the
408 փ APPENDIX ǣ
Tableau community, but visualizations can be created in any tool. Visualizations are created
and posted publicly with review and feedback on social media and a mid-week webinar to
enable people to give feedback and refine their ideas.
STORYTELLING WITH DATA . The Storytelling with Data Community is a place to prac-
tice your data visualization skills, get and give feedback, and discuss topics related to effec-
tively communicating data. There is a monthly “SWDchallenge,” which asks participants to
address a specific data visualization challenge and a periodic “SWDexercise” that is based on
real-world scenarios with actual data.
TIDY TUESDAY. A weekly project aimed at R users, project managers post a dataset and
a published chart or article related to the dataset. Participants are asked to explore the data
and publish their results (and code) for others to use and adapt.
OTHERS. There are other similar projects, but tend to have dedicated audiences in different
ways. The Data Visualization Society, a community of data visualization practitioners, hosts
Slack channels revolving around practice and critique. There are also various forums on
HelpMeViz, Reddit, Stack Exchange, Twitter, and other social media platforms where you
can practice and publish your work.
ACKNOWLEDGMENTS
I
n 2014, I wrote an introductory article on data visualization for the Journal of Economic
Perspectives (JEP). I had always wanted to publish in the JEP, but never imagined that
data visualization—rather than public policy or economics—would be the topic of inter-
est. Shortly after the article was published, I received a call from Bridget Flannery-McCoy
at Columbia University Press asking to see if I was interested in writing a full book on data
visualization. At that time, I wasn’t ready to write this book—there were a lot of books just
out or coming out and to do it right, I would need to dedicate a lot more time and resources
than I had available. Bridget’s call provide fruitful nonetheless, ultimately resulting in my
first book, Better Presentations.
Fast-forward five years and Bridget called again to see if I was ready for a follow-up book.
Now, having changed jobs and focusing more on data communication, I found myself with
more to say and an idea for a book that would place it in a unique space in the field.
I wrote the first draft of the entire manuscript on two train rides between Washington,
DC and New York City. But it would take another two-and-a-half years to fully develop
it, reorganize it, expand it, and create the graphs. The result is the book you are now
holding.
For getting me to this point, my gratitude belongs to several important people: Brittany
Fong, Ajjit Narayanan, Jon Peltier, Anthea Piong, and Aaron Williams helped me with vari-
ous Tableau, Excel, R, and JavaScript challenges and tasks. RJ Andrews, John Burn-Mur-
doch, Alberto Cairo, Jennifer Christiansen, Alice Feng, John Grimwade, Steve Haroz, Robert
Kosara, and Severino Ribecca were generous in their time to have discussions, provide
4 10 փ AC KNOWLEDGMENTS
feedback, and assist with a number of requests to find research and images, both modern
and historical.
More thanks are due to the people and organizations who have granted me permission
to include their work in this book. Their work is truly special and I am grateful they allowed
me to include it in these pages.
Additional thanks to Ken Skaggs for helping to manage the PolicyViz Podcast and to the
more than 200 guests who have appeared on the show.
A special thanks to Alberto Cairo, Nigel Holmes, Jessica Hullman, David Napoli, and
Chad Skelton for reviewing parts or all of the manuscript and providing invaluable feedback.
Very special thanks are due to Kenneth Field for creating many of the maps and for keep-
ing up with my “feature creep.” Special thanks are also due to Hiram Henriquez for creating
original illustrations and making all of the graphs ready for print.
This book could not have come together without the help of my Columbia University
Press editor, Stephen Wesley, who kept up with my incessant emails and questions. Addi-
tional thanks to Christian Winting, Ben Kolstad, and the rest of the Columbia University
Press team for helping to bring this project to life.
I am grateful for my Urban Institute colleagues whose dedication to fact-based research
has the power to improve public policy and practice, strengthen communities, and trans-
form people’s lives for the better. They have helped create a special place that not only values
in-depth scholarship, but also innovative ways to communicate that work.
I am also indebted to the many, many friends and strangers with whom I have discussed
and debated aspects of data visualization and data communication over the years. The data
visualization community is a special place—the warm welcome it gave me about 10 years
ago changed my career in unimaginable ways, and I hope to pass it forward to another crop
of data communicators. I am also especially thankful to people in the field whose creativity
continues to inspire.
Writing a third book is simultaneously easier and harder than writing the first and sec-
ond. I could not have summoned the courage to go through another project without the love
and support of friends and family who have cheered me on and taken interest in even the
most mundane details of my work. Special thanks are due to the ongoing support and love
of my entire family.
Finally, my most special thanks are due to the three most important people in my life: my
wife and kids. My kids, Ellie and Jack, are my favorite sources of happiness, pride, and fun.
They have watched their dad struggle with code, complain about color palettes and fonts,
AC KNOWLEDGMENTSփ 4 11
and watched me stay up late into the night, reading and writing. Through it all, they sup-
plied me with steady encouragement, love, and support. As a dad, you know you’re doing
something right when they lean over your shoulder and say, “How’s it coming with that
heatmap?”
My deepest thanks goes to my wife, Lauren, who has edited every page of this book, cut-
ting out repetitive text and keeping the language clear and descriptive, always working with
my reader in mind. She keeps all of us moving forward, especially when I’m writing, speak-
ing, or traveling demands take me far and wide. She routinely reminds me she’s the best
thing that’s ever happened to me. A truer statement has never been spoken.
REFERENCES
Ahmed, Naema, Cassi Pollock, and Alex Samuels. “How the Texas Democratic and Republic Party
Platforms Compare.” Texas Tribune, July 5, 2018, https://apps.texastribune.org/features/2018
/party-platforms/?_ga=2.129478090.770685496.1576106215–1729798919.1576004948.
Alda, Alan. If I Understood You, Would I Have This Look on My Face?: My Adventures in the Art and
Science of Relating and Communicating. New York: Random House, 2018.
Andrews, R. J. “Florence Nightingale Is a Design Hero.” Medium, July 15, 2019, https://medium.com
/nightingale/florence-nightingale-is-a-design-hero-8bf6e5f2147.
Andrews, R. J. Info We Trust: How to Inspire the World with Data. New York: Wiley, 2019.
Andris, Clio, David Lee, Marcus J. Hamilton, Mauro Martino, Christian E. Gunning, and John
Armistead Selden. “The Rise of Partisanship and Super-Cooperators in the US House of Represen-
tatives.” PloS one 10, no. 4 (2015): e0123507.
Anscombe, Francis J. “Graphs in Statistical Analysis.” The American Statistician 27, no. 1 (1973): 17–21.
Arellano, Cristina, Juan Carlos Conesa, and Timothy J. Kehoe. “Chronic Sovereign Debt Crises in
the Eurozone, 2010–2012.” Federal Reserve Bank of Minneapolis Economic Policy paper 12–4,
May 2012, https://www.minneapolisfed.org/~/media/files/pubs/eppapers/12-4/epp_12-4_chronic
_sovereign_debt_crisis_eurozone.pdf.
Avery, Beth. “Ban the Box: U.S. Cities, Counties, and States Adopt Fair Hiring Policies.” National Employ-
ment Law Project. July 1, 2019. https://www.nelp.org/publication/ban-the-box-fair-chance-hiring
-state-and-local-guide/.
AxisMaps. “Cartography Guide.” https://www.axismaps.com/guide/general/map-projections/, accessed
November 2019.
Bard Graduate Gallery. “Marimekko: Fabrics, Fashion, Architecture.” https://www.bgc.bard.edu/gallery
/exhibitions/43/marimekko.
414փ REFERENCES
Battle-Baptiste, Whitney, and Britt Rusert. “WEB Du Bois’s Data Portraits: Visualizing Black America:
The Color Line at the Turn of the Twentieth Century.” Amherst, Massachusetts: WEB Du Bois
Center at the University of Massachusetts Amherst (2018).
BBC. “BBC Audiences Tableau Style Guide.” https://public.tableau.com/profile/bbc.audiences#!/vizhome
/BBCAudiencesTableauStyleGuide/Hello
Béland, Antoine, and Thomas Hurtut. “Unit Visualizations for Visual Storytelling.” OSF Preprints,
https://osf.io/bshpc/.
Bennet, Tim. “How 52 Ninth-Graders Spell ‘Camouflage’, Sankey Diagram.” https://www.reddit
.com/r/dataisbeautiful/comments/6a4pb8/how_52_ninthgraders_spell_camouflage_sankey/?st
=J2NBTE0Q&sh=ddd5c5e1
Berinato, Scott. “The Power of Visualizations’ ‘Aha!’ Moments.” Harvard Business Review, March 19,
2013, https://hbr.org/2013/03/power-of-visualizations-aha-moment.
Berman, Jacob. “Washington Metro Map Print Original Art.” Etsy listing, 2019, https://www.etsy.com
/listing/647623499/washington-metro-map-print-original-art
Bontemps, Xtophe. “Why You Should Never Use Radar Plots.”data.visualisation.free.fr, March 2017,
https://rpubs.com/Xtophe/268920.
Borkin, Michelle A., Zoya Bylinskii, Nam Wook Kim, Constance May Bainbridge, Chelsea S. Yeh,
Daniel Borkin, Hanspeter Pfister, and Aude Oliva. “Beyond Memorability: Visualization Recog-
nition and Recall.” IEEE Transactions on Visualization and Computer Graphics 22, no. 1 (2015):
519–528.
Börner, Katy, Andreas Bueckle, and Michael Ginda. “Data Visualization Literacy: Definitions, Con-
ceptual Frameworks, Exercises, and Assessments.” Proceedings of the National Academy of Sciences
116, no. 6 (2019): 1857–1864.
Bortins, I., Demers S., & Clarke, K. (2002). Cartogram Types. http://www.ncgia.ucsb.edu/projects
/Cartogram_Central/types.html
Bostock, Michael, Vadim Ogievetsky, and Jeffrey Heer. “D³ Data-Driven Documents.” IEEE transac-
tions on visualization and computer graphics 17, no. 12 (2011): 2301–2309.
Brehmer, Matt. “Visualizing Ranges Over Time on Mobile Phones.” Medium, November 3, 2018,
https://medium.com/multiple-views-visualization-research-explained/ranges-900a04d7d32a.
Brehmer, Matthew, Bongshin Lee, Petra Isenberg, and Eun Kyoung Choe. “Visualizing ranges over
time on mobile phones: a task-based crowdsourced evaluation.” IEEE transactions on visualization
and computer graphics 25, no. 1 (2018): 619–629.
Bremer, Nadieh. “More Fun Data Visualizations with the Gooey Effect.” Visual Cinnamon (blog). June
20, 2016. https://www.visualcinnamon.com/2016/06/fun-data-visualizations-svg-gooey-effect.html.
Bremer, Nadieh. “Techniques for Data Visualization on Both Mobile & Desktop.” Visual Cinnamon
(blog). April 17, 2019. https://www.visualcinnamon.com/2019/04/mobile-vs-desktop-dataviz#recap.
Brinton, Willard Cope. Graphic Methods for Presenting Facts. Engineering magazine company, 1919.
REFERENCESփ 4 15
Broderick, Ryan. “Who Watches More Porn: Republicans or Democrats?” BuzzFeed News. April
2014. https://www.buzzfeednews.com/article/ryanhatesthis/who-watches-more-porn-republicans
-or-democrats.
Bureau of Labor Statistics. Unemployment rate. https://www.bls.gov/cps/
Burn-Murdoch, John. “Episode #155: John Burn-Murdoch.” The PolicyViz Podcast, June 18, 2019,
https://policyviz.com/podcast/episode-155-john-burn-murdoch/.
Burn-Murdoch, John. Twitter post. October 28, 2019, 4:04 a.m., https://twitter.com/jburnmurdoch
/status/1188728193945100288.
Bush, George W. “Address Before a Joint Session of the Congress on the State of the Union.” The
American Presidency Project, February 2, 2005, https://www.presidency.ucsb.edu/documents
/address-before-joint-session-the-congress-the-state-the-union-14.
Byron, Lee, and Martin Wattenberg. “Stacked Graphs–Geometry and Aesthetics.” IEEE Transactions
on Visualization and Computer Graphics 14, no. 6 (2008): 1245–1252.
Cairo, Alberto. “Annotation, Narrative, and Storytelling in Infographics and Visualization.” The Func-
tional Art (blog), April 16, 2014, http://www.thefunctionalart.com/2014/04/annotation-narrative
-and-storytelling.html.
Cairo, Alberto. “Download the Datasaurus: Never Trust Summary Statistics Alone; Always Visualize
Your Data.” The Functional Art (blog), August 29, 2016, http://www.thefunctionalart.com/2016/08
/download-datasaurus-never-trust-summary.html.
Cairo, Alberto. The Functional Art: An Introduction to Information Graphics and Visualization. Berke-
ley, CA: New Riders, 2012.
Cairo, Alberto. “How Charts Lie.” (2019).
Cairo, Alberto. “Visual Storytelling w/ Alberto Cairo and Robert Kosara.” The Data Stories Pod-
cast, Episode #35, April 16, 2014, https://datastori.es/data-stories-35-visual-storytelling-w-alberto
-cairo-and-robert-kosara/.
Camões, Jorge. “Perception: Gestalt Laws.” excelcharts.com (blog), January 10, 2012, https://excelcharts
.com/data-visualization-excel-users/gestalt-laws/.
Carpenter, Patricia A., and Priti Shah. “A Model of the Perceptual and Conceptual Processes in Graph
Comprehension.” Journal of Experimental Psychology: Applied 4, no. 2 (1998): 75.
Center on Budget and Policy Priorities. “Working-Family Tax Credits Help at Every Stage of Life.”
Graphic, https://www.cbpp.org/working-family-tax-credits-help-at-every-stage-of-life-0.
Centers for Disease Control and Prevention. “CDC Wonder.” https://wonder.cdc.gov/.
Centers for Disease Control and Prevention. “Community Mitigation Guidelines to Prevent Pan-
demic Influenza—United States, 2017.” Morbidity and Mortality Weekly Report 66, No. 1, April 21,
2017. https://stacks.cdc.gov/view/cdc/45220.
Centers for Disease Control and Prevention. “National, Regional, and State Level Outpatient Illness
and Viral Surveillance.” https://gis.cdc.gov/grasp/fluview/fluportaldashboard.html.
4 16 փ REFERENCES
Cesal, Amy. “The Sunlight Foundation’s Data Visualization Style Guidelines.” Sunlight Foundation,
March 12, 2014, https://sunlightfoundation.com/2014/03/12/datavizguide/.
Chalabi, Mona. Presentation, Tapestry Conference, University of Miami, November 29, 2018.
Chang, Kenneth. “A Different Way to Chart the Spread of Coronavirus.” New York Times, March 20,
2020. https://www.nytimes.com/2020/03/20/health/coronavirus-data-logarithm-chart.html.
Chappell, Nathan and Adam B. Jaffe. “Intangible Investment and Firm Performance.” National Bureau
of Economic Research Working Paper No. 24363, March 2018, https://www.nber.org/papers/w24363.
Chase, Will. “Voronoi Treemap” (website). May 2, 2019, https://observablehq.com/@will-r-chase
/voronoi-treemap.
Chase, Will. “Why I’m Not Making COVID19 Visualizations, and Why You (Probably) Shouldn’t
Either” (blog post). March 31, 2020, https://www.williamrchase.com/post/why-i-m-not-making
-covid19-visualizations-and-why-you-probably-shouldn-t-either/.
Cherdarchuk, Joey. “Clear off the Table.” Dark Horse Analytics (blog). March 27, 2014, https://www
.darkhorseanalytics.com/blog/clear-off-the-table.
Choi, Jinho, Sanghun Jung, Deok Gun Park, Jaegul Choo, and Niklas Elmqvist. “Visualizing for the
Non‐Visual: Enabling the Visually Impaired to Use Visualization.” Computer Graphics Forum 38,
no. 3 (2019): 249–260.
Christiansen, Jen. “Pop Culture Pulsar: Origin Story of Joy Division’s Unknown Pleasures Album
Cover.” Scientific American, February 18, 2015, https://blogs.scientificamerican.com/sa-visual
/pop-culture-pulsar-origin-story-of-joy-division-s-unknown-pleasures-album-cover-video/.
Cleveland, William S. The Elements of Graphing Data. Belmont, CA: Wadsworth, 1985.
Cleveland, William S. Visualizing data. Hobart Press, 1993.
Cleveland, William S., and Robert McGill. “Graphical Perception: Theory, Experimentation, and
Application to the Development of Graphical Methods.” Journal of the American Statistical Asso-
ciation 79, no. 387 (1984): 531–554.
Colour Blindness Awareness. “Colour Blindness.” http://www.colourblindawareness.org/colour-blindness
/, accessed November 2019.
Congressional Budget Office. “CBO’s August 2018 report An Update to the Economic Outlook: 2018
to 2028.” www.cbo.gov/publication/54318.
Congressional Budget Office. “Changes in the Distribution of Workers’ Annual Earnings Between
1979 and 2007.” Congressional Budget Office, October 2009, https://www.cbo.gov/sites/default
/files/111th-congress-2009-2010/reports/10-02-workers.pdf.
Congressional Budget Office. “Social Security Policy Options.” Congressional Budget Office, July
2010, https://www.cbo.gov/sites/default/files/111th-congress-2009-2010/reports/07-01-ssoptions
_forweb.pdf
Congressional Budget Office. “The 2012 Long-Term Budget Outlook | House Budget Committee.” Con-
gressional Budget Office, September 12, 2014, https://www.youtube.com/watch?v=dqhqZYYGNnA
REFERENCESփ 417
Congressional Budget Office. “The Budget and Economic Outlook: Fiscal Years 2012 to 2022.” Con-
gressional Budget Office.” January 31, 2012, https://www.cbo.gov/publication/42905?index=12699
Congressional Budget Office. “The Budget and Economic Outlook: Infographic.” Congressional Bud-
get Office.” June 5, 2012, https://www.cbo.gov/publication/43289
Congressional Budget Office. “The Social Security Disability Insurance Program.” Infographic. https://
www.cbo.gov/publication/43432
Congressional Budget Office. “The Supplemental Nutrition Assistance Program.” Infographic. https://
www.cbo.gov/publication/43174.
Cook, Albert M., and Janice Miller Polgar. Assistive Technologies-E-Book: Principles and Practice. St.
Louis, MO: Elsevier, 2014.
Correll, Michael, and Michael Gleicher. “Error Bars Considered Harmful: Exploring Alternate Encod-
ings for Mean and Error.” IEEE Transactions on Visualization and Computer Graphics 20, no. 12
(2014): 2142–2151.
Correll, Michael, Enrico Bertini, and Steven Franconeri. “Truncating the Y-Axis: Threat or Menace?”
In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, ed. Regina
Bernhaupt et al., 1–12. New York: Association for Computing Machinery. 2020.
Correll, Michael. “Ethical Dimensions of Visualization Research.” Tableau Research, CHI 2019, 2019.
Correll, Michael. “Visualization Design Principles for the Pandemic.” Medium, April 3, 2020, https://
medium.com/@mcorrell/visualization-design-principles-for-the-pandemic-e65388280d16.
Cousins, Carrie. “Color and Cultural Design Considerations.” WebDesignerDepot (blog), June 11,
2012, https://www.webdesignerdepot.com/2012/06/color-and-cultural-design-considerations/.
Cox, Amanda. “Amanda Cox—Eyeo Festival 2011.” https://vimeo.com/29391942
Cruz, Pedro de, John Wihbey, and Felipe Shibuya. Simulated Dendrochronology of the United States.
2018.
D’Ignazio, Catherine and Lauren F. Klein. Data Feminism. Boston, MA: MIT Press. 2020.
Dallas Morning News. Dallas Morning News Graphics Stylebook. https://knightcenter.utexas.edu
/mooc/file/tdmn_graphics.pdf.
Daniels, Matt. “The Largest Vocabulary In Hip Hop.” The Pudding, January 21, 2019, https://pudding
.cool/projects/vocabulary/index.html.
DeBelius, Danny. “Let’s Tesselate: Hexagons For Tile Grid Maps.” NPR visuals team, May 11, 2015,
https://blog.apps.npr.org/2015/05/11/hex-tile-maps.html.
DeBold, Tynan and Dov Friedman. “Battling Infectious Diseases in the 20th Century: The
Impact of Vaccines.” Wall Street Journal, February 11, 2015, http://graphics.wsj.com/infectious
-diseases-and-vaccines/.
Desilver, Drew. “A Record Number of Women Will Be Serving in the New Congress.” Pew
Research Center, December 18, 2018. https://www.pewresearch.org/fact-tank/2018/12/18/record
-number-women-in-congress/
4 18 փ REFERENCES
Dorling, Daniel. “Area cartograms: their use and creation.” The map reader: Theories of mapping prac-
tice and cartographic representation (2011): 252–260.
DuBois, William Edward Burghardt, ed. The College-bred Negro: Report of a Social Study Made Under the
Direction of Atlanta University; Together with the Proceedings of the Fifth Conference for the Study of
the Negro Problems, Held at Atlanta University, May 29–30, 1900. No. 5. Atlanta University Press, 1900.
Economist. “The Climate Issue.” https://www.economist.com/leaders/2019/09/19/the-climate-issue.
Economist. “The Dragon Takes Wing.” May 1, 2014, https://www.economist.com/news/finance-and
-economics/21601568-new-data-suggest-chinese-economy-bigger-previously-thought-dragon.
Economist. “Flattening the Curve: Covid-19 Is Now in 50 Countries, and Things Will Get Worse.” The
Economist, February 29, 2020. https://www.economist.com/briefing/2020/02/29/covid-19-is-now
-in-50-countries-and-things-will-get-worse.
Eurostat, “Graphical Style Guide—A Practical Layout Guide for Eurostat Publications—2016 Edi-
tion.” Eurostat News, February 25, 2016, https://ec.europa.eu/eurostat/web/products-eurostat-news
/-/STYLE-GUIDE_2016.
Federal Home Loan Bank Board. Home Owners’ Loan Corporation. 1933-7/1/1939. “City of Rich-
mond, Virginia and Environs.” U.S. National Archives and Records Administration. https://catalog
.archives.gov/id/85713737
Federal Reserve Board of Governors. “Household Debt Services and Financial Obligations Ratios.”
Accessed January 2020. https://www.federalreserve.gov/releases/housedebt/
Few, Stephen. “Bullet Graph Design Specification.” Perceptual Edge, Visual Business Intelligence News-
letter, October 10, 2013, https://www.perceptualedge.com/articles/misc/Bullet_Graph_Design
_Spec.pdf.
Few, Stephen. “The DataVis Jitterbug: Let’s Improve an Old Dance.” Perceptual Edge, Visual Busi-
ness Intelligence Newsletter, April/May/June 2017, https://www.perceptualedge.com/articles/visual
_business_intelligence/the_datavis_jitterbug.pdf.
Few, Stephen. Show Me the Numbers: Designing Tables and Graphs to Enlighten. Oakland, CA: Analyt-
ics Press, 2004.
Few, Stephen. “Unit Charts Are For Kids.” Perceptual Edge, Visual Business Intelligence Newsletter,
October, November, December 2010, https://www.perceptualedge.com/articles/visual_business
_intelligence/unit_charts_are_for_kids.pdf.
Field, Kenneth. Cartography. Redlands, CA: Esri, 2018.
FiveThirtyEight. “50 Years Of World Cup Doppelgangers.” July 15, 2018, https://projects.fivethirtyeight
.com/world-cup-comparisons/romelu-lukaku-2018/
Florence, Philip Sargant. Only an Ocean Between. New York: Essential, 1946.
Fontenot, Kayla, Jessica Semega, and Melissa Kollar. “Income and Poverty in the United States: 2017.”
U.S. Bureau of the Census, September 2008, https://www.census.gov/content/dam/Census/library
/publications/2018/demo/p60-263.pdf.
REFERENCESփ 4 19
Friendly, Michael. “Playfair Balance of Trade Data.” September 13, 2018, http://www.datavis.ca
/courses/RGraphics/R/playfair-east-indies.html.
Fruchterman, Thomas MJ, and Edward M. Reingold. “Graph Drawing by Force‐Directed Placement.”
Software: Practice and experience 21, no. 11 (1991): 1129–1164.
Fry, Ben. “Tracing the Origin of Species.” https://fathom.info/traces/.
Gallo, Carmine. The Presentation Secrets of Steve Jobs: How to Be Insanely Great in Front of Any Audi-
ence. Upper Saddle River, NJ: Prentice Hall, 2010.
Gambino, Megan. “Do Our Brains Find Certain Shapes More Attractive Than Others?” Smithsonian Mag-
azine. November 14, 2013. https://www.smithsonianmag.com/science-nature/do-our-brains-find
-certain-shapes-more-attractive-than-others-180947692/.
Gantt, Henry Laurence. “A Graphical Daily Balance in Manufacture.” No. 1002 in American Society
of Mechanical Engineers. Transactions of the American Society of Mechanical Engineers. New York
City: The Society, 1880. https://babel.hathitrust.org/cgi/pt?id=mdp.39015023119541&view=1up&
seq=1363
Gee, Alastair, Julia Carrie Wong, Paul Lewis, Adithya Sambamurthy, Charlotte Simmonds, Nadieh
Bremer, and Shirley Wu. “Bussed Out: How America Moves Its Homeless.” The Guardian, December
20, 2017, https://www.theguardian.com/us-news/ng-interactive/2017/dec/20/bussed-out-america
-moves-homeless-people-country-study
Google Finance. http://finance.google.com. Accessed January 2020.
Gramacki, Artur. Nonparametric Kernel Density Estimation and Its Computational Aspects. Cham,
Switzerland: Springer International, 2018.
Groeger, Lena V. “A Big Article About Wee Things.” ProPublica (blog), September 25, 2014, https://
www.propublica.org/nerds/a-big-article-about-wee-things.
Guardian. “EU Referendum: Full Results and Analysis.” The Guardian, https://www.theguardian.com
/politics/ng-interactive/2016/jun/23/eu-referendum-live-results-and-analysis.
Guardian. “Full US 2012 Election County-Level Results to Download.” https://www.theguardian.com
/news/datablog/2012/nov/07/us-2012-election-county-results-download#data.
Hansen, Wallace R. Suggestions to Authors of the Reports of the United States Geological Survey, US
GPO, January 1991.
Haroz, Steve, Robert Kosara, and Steven L. Franconeri. “Isotype Visualization: Working Memory, Per-
formance, and Engagement with Pictographs.” In Proceedings of the 33rd annual ACM conference
on human factors in computing systems, 1191–1200. ACM, 2015.
Hawkins, Ed. “147 | Iconic Climate Visuals with Ed Hawkins.” Data Stories podcast, September 10,
2019, https://datastori.es/147-iconic-climate-visuals-with-ed-hawkins/.
Hawkins, Ed. ShowYourStripes.com. https://showyourstripes.info/
Healey, Christopher, and James Enns. “Attention and Visual Memory in Visualization and Computer
Graphics.” IEEE transactions on visualization and computer graphics 18, no. 7 (2011): 1170–1188.
420փ REFERENCES
Hearst, Marti, Emily Pedersen, Lekha Priya Patil, Elsie Lee, Paul Laskowski, and Steven Franconeri.
“An Evaluation of Semantically Grouped Word Cloud Designs.” IEEE transactions on visualization
and computer graphics (2019).
Heer, Jeffrey, Michael Bostock, and Vadim Ogievetsky. “A Tour Through the Visualization Zoo.” Com-
mun. Acm 53, no. 6 (2010): 59–67.
Hermann, E.P. “Maps and Sales Visualization” Personal Efficiency: The How and Why Magazine, Pub-
lished by the LaSalle Extension University, US Department of Education, July 1922, pages 6–7.
Hofman, Jake M., Daniel G. Goldstein, and Jessica Hullman. “How visualizing inferential uncertainty
can mislead readers about treatment effects in scientific results.” Proceedings of the 2020 CHI Con-
ference on Human Factors in Computing Systems.
Hofmann, Heike, and Marie Vendettuoli. “Common Angle Plots as Perception-True Visualizations of
Categorical Associations.” IEEE Transactions on Visualization and Computer Graphics 19.12 (2013):
2297–2305.
Horowitz, Juliana Mesace, Kim Parker, and Renee Stepler. “Wide Partisan Gaps in U.S. Over How
Far the Country Has Come on Gender Equality.” Pew Research Center, October 18, 2017, https://
www.pewsocialtrends.org/2017/10/18/wide-partisan-gaps-in-u-s-over-how-far-the-country
-has-come-on-gender-equality/.
Hullman, Jessica and Matthew Kay. “Uncertainty + Visualization, Explained.” Medium, Visualization
Research Explained (blog), https://medium.com/multiple-views-visualization-research-explained
/uncertainty-visualization-explained-67e7a73f031b.
Hullman, Jessica, Eytan Adar, and Priti Shah. “Benefitting Infovis with Visual Difficulties.” IEEE
Transactions on Visualization and Computer Graphics 17, no. 12 (2011): 2213–2222.
Hullman, Jessica, Paul Resnick, and Eytan Adar. “Hypothetical Outcome Plots Outperform Error Bars
and Violin Plots for Inferences About Reliability of Variable Ordering.” PloS one 10, no. 11 (2015).
Hullman, Jessica. “Leading with the Unknowns in COVID-19 Models.” Scientific American, April 11,
2020. https://blogs.scientificamerican.com/observations/leading-with-the-unknowns-in-covid-19
-models/.
Ingraham, Christopher. “Kansas Is the Nation’s Porn Capital, According to Pornhub.” Wonkviz, no date,
https://wonkviz.tumblr.com/post/82488570278/kansas-is-the-nations-porn-capital-according-to
Institute for Health Metrics and Evaluation. “Causes of Death (COD) Visualization | Viz Hub.”
Accessed January 2020. https://vizhub.healthdata.org/cod/.
International Social Survey Programme (ISSP). “ISSP 2009 ‘Social Inequality IV’—ZA No. 5400.”
Accessed January 2020. https://www.gesis.org/issp/modules/issp-modules-by-topic/social-inequality
/2009/
Irvin-Erickson, Yasemin, Jonathan Schwabish, and Nicole Weissman. “What We Know About Gun
Violence in the United States: Who’s Affected?” Urban Wire Urban Institute, October 4, 2016, https://
www.urban.org/urban-wire/what-we-know-about-gun-violence-united-states-whos-affected.
REFERENCESփ 421
Japanese: https://seibu.ekitan.com/pdf/20180310/235-16-1-0.pdf
Jarreau, Paige. “#MySciBlog Interviewee Motivations to Blog about Science.” Figshare.com (blog),
March 21, 2015, https://figshare.com/articles/_MySciBlog_Interviewee_Motivations_to_Blog_about
_Science/1345026/2.
Johnson, Brian, and Ben Shneiderman. “Tree-Maps: A Space-Filling Approach to the Visualization
of Hierarchical Information Structures.” VIS ’91: Proceedings of the 2nd conference on Visualization
’91. IEEE, 1991.
Johnson, Matt. “Which Metro Stations Are the Most Balanced?” Greater Greater Washington, Novem-
ber 29, 2012, https://ggwash.org/view/29468/which-metro-stations-are-the-most-balanced.
Joint Committee on Taxation. “A Distribution of Returns by the Size of the Tax Change for the ‘Tax
Cuts and Jobs Act,’ As Ordered Reported by the Committee on Finance on November 16, 2017.”
November 27, 2017.
Joint Committee on Taxation. “Distributional Effects of Public Law 115–97.” Scheduled for a Public
Hearing Before the House Committee on Ways and Means on March 27, 2019, March 25, 2019,
JCX-10-19.
Kaiser Family Foundation. “Status of State Action on the Medicaid Expansion Decision.” Accessed
January 2020. https://www.kff.org/health-reform/state-indicator/state-activity-around-expanding
-medicaid-under-the-affordable-care-act/.
Kastellec, Jonathan P., and Eduardo L. Leoni. “Using Graphs Instead of Tables in Political Science.”
Perspectives on Politics 5, no. 4 (2007): 755–771.
Katz, Josh. “Who Will Be President?” New York Times, November 8, 2016, https://www.nytimes.com
/interactive/2016/upshot/presidential-polls-forecast.html
Kiefer, Len. Twitter post. September 26, 2019, 4:44 p.m., https://twitter.com/lenkiefer/status
/1177323018143580160.
Kight, Stef W. and Harry Stevens. “By the Numbers: How Trump Properties Profited from His Presi-
dency.” Axios, June 28, 2018, https://www.axios.com/donald-trump-properties-taxpayer-campaigns
-presidency-91e3755d-23cd-42d1-897d-c0c81ac509dd.html.
Kirk, Andy. “The Problems with B’Arc Charts.” Visualisingdata.com, September 1, 2017, https://www
.visualisingdata.com/2017/09/problems-barc-charts/.
Klein, Scott. “Infographics in the Time of Cholera.” Pro Publica, March 16, 2016, https://www.propublica
.org/nerds/infographics-in-the-time-of-cholera.
Koblin, Aaron. “Flight Patterns.” http://www.aaronkoblin.com/work/flightpatterns/.
Koblin, Aaron. “Visualizing Ourselves . . . with Crowd-Sourced data.” TED2011, March 2011, https://
www.ted.com/talks/aaron_koblin_visualizing_ourselves_with_crowd_sourced_data.
Kommenda, Niko, Caelainn Barr, and Josh Holder. “Gender Pay Gap: What We Learned and How
to Fix It.” The Guardian, April , 2018, https://www.theguardian.com/news/ng-interactive/2018
/apr/05/women-are-paid-less-than-men-heres-how-to-fix-it.
422փ REFERENCES
Krzywinski, Martin, Inanc Birol, Steven JM Jones, and Marco A. Marra. “Hive plots—rational
approach to visualizing networks.” Briefings in bioinformatics 13, no. 5 (2011): 627–644.
Krzywinski, Martin. “Hive plots—rational approach to visualizing networks.” http://www.hiveplot
.com/.
Lekovic, Jovan. “How We Made the BBC Audiences Tableau Style Guide.” Medium, August 29, 2018,
https://medium.com/bbc-data-science/how-we-made-the-bbc-audiences-tableau-style-guide
-4f0a6b7525ce.
Likert, Rensis. “A Technique for the Measurement of Attitudes.” Archives of psychology (1932).
Lima, Manuel. “Why Humans Love Pie Charts.” Medium (blog). July 23, 2018. https://blog.usejournal
.com/why-humans-love-pie-charts-9cd346000bdc?.
Lima, Manuel. The Book of Circles: Visualizing Spheres of Knowledge. New York: Princeton Architec-
tural Press, 2017.
Liu, Yang, and Jeffrey Heer. “Somewhere Over the Rainbow: An Empirical Assessment of Quantitative
Colormaps.” In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems,
pp. 1–12. 2018.
Luxembourg Income Study. https://www.lisdatacenter.org/.
Malkulec, Amanda. “Leading with the Unknowns in COVID-19 Models.” Medium, March 11, 2020, https://
medium.com/nightingale/ten-considerations-before-you-create-another-chart-about-covid
-19-27d3bd691be8.
Manski, Charles F., and Francesca Molinari. “Rounding Probabilistic Expectations in Surveys.” Jour-
nal of Business & Economic Statistics 28, no. 2 (2010): 219–231.
Matejka, Justin, and George Fitzmaurice. “Same Stats, Different Graphs: Generating Datasets with
Varied Appearance and Identical Statistics Through Simulated Annealing.” In Proceedings of the
2017 CHI Conference on Human Factors in Computing Systems, pp. 1290–1294. ACM, 2017.
Mayr, Eva and Günther Schreder. “Isotype Visualizations: A Chance for Participation and Civic Edu-
cation.” Journal of Democracy 6(2): 136–150, 2014.
McCann, Adam. “Arc Chart in Tableau.” August 27, 2015, http://duelingdata.blogspot.com/2015/08
/arc-chart-in-tableau.html.
McKee, Robert. Substance, Structure, Style, and the Principles of Screenwriting. New York: HarperCollins,
1997.
Medina, John. Brain Rules: 12 Principles for Surviving and Thriving at Work, Home, and School. Read-
HowYouWant.com, 2011.
Meeks, Elijah. “Exploratory Design in Data Visualization: Understanding and leveraging chart
similarity.” Medium, January 15, 2019, https://medium.com/@Elijah_Meeks/exploratory-design
-in-data-visualization-87bc60ce7f04.
Meeks, Elijah. “Sketchy Data Visualization in Semiotic.” Medium (blog), September 11, 2017, https://
medium.com/@Elijah_Meeks/sketchy-data-visualization-in-semiotic-5811a52f59bc.
REFERENCESփ 423
Meyer, Bruce D., Nikolas Mittag, and Robert M. Goerge. Errors in Survey Reporting and Imputation
and their Effects on Estimates of Food Stamp Program Participation. No. w25143. National Bureau of
Economic Research, 2018.
Meyer, Bruce D., Wallace KC Mok, and James X. Sullivan. “Household Surveys in Crisis.” Journal of
Economic Perspectives 29, no. 4 (2015): 199–226.
Miller, Matthew, Steven J. Lippmann, Deborah Azrael, and David Hemenway. “Household firearm
ownership and rates of suicide across the 50 United States.” Journal of Trauma and Acute Care
Surgery 62, no. 4 (2007): 1029–1035.
Minard, Charles-Joseph. “Carte figurative des pertes successives en hommes de l’Armée Française dans la
campagne de Russie 1812–1813.” 1861, Photo courtesy of Ecole nationale des ponts et chaussées
Monmonier, Mark. How to Lie with Maps. Chicago: University of Chicago Press, 2018.
Munzner, Tamara. Visualization Analysis and Design. AK Peters, 2014.
National Conference of State Legislatures. “State Family and Medical Leave Laws.” July 2016. http://
www.ncsl.org/research/labor-and-employment/state-family-and-medical-leave-laws.aspx
National Conference of State Legislatures. “State Minimum Wages | 2020 Minimum Wage by State.”
January 2020. http://www.ncsl.org/research/labor-and-employment/state-minimum-wage-chart
.aspx#Table.
National Highway Traffic Safety Administration. Fatality Analysis Reporting System (FARS). https://
one.nhtsa.gov/Data/Fatality-Analysis-Reporting-System-(FARS).
National Institute on Drug Abuse. “Overdose Death Rates.” January 2019. https://www.drugabuse.gov
/related-topics/trends-statistics/overdose-death-rates.
NBC News. “Presidential Election Results.” http://elections.nbcnews.com/ns/politics/2012/all/president
/#.XiZR4lNKhTZ.
Nediger, Midori. “How to Use an Icon Story to Take Your Infographic to the Next Level.” Venngage
(blog), April 27, 2018, https://venngage.com/blog/icon-story/.
Neurath, O. International Picture Language. The First Rules of Isotype. London: Kegan Paul, 1936.
Neurath, O. Modern Man in the Making. New York: Knopf, 1939.
New York Magazine. “Approval Matrix.” provided via email.
New York Times. Paths to the White House.
New York Times. Twitter post. August 7, 2016, 10:52 p.m. https://twitter.com/nytgraphics/status
/762481520565030919?ref_src=twsrc%5Etfw
Newman, George E., and Brian J. Scholl. “Bar Graphs Depicting Averages are Perceptually Misinter-
preted: the Within-the-Bar Bias.” Psychonomic Bulletin and Review 19, no. 4 (2012): 601–607.
Noble, Safiya Umoja. Algorithms of Oppression: How Search Engines Reinforce Racism. New York: New
York University Press. 2018.
O’Brady, Kevin. “Fight to Repeal.” Joint Economic Committee Republicans, 2010, https://kevinbrady
.house.gov/obamacare/fighting-to-repeal-obamacare.htm.
424փ REFERENCES
O’Neill, Catherine. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens
Democracy. New York: Broadway, 2016.
Obama, Barack. “Remarks of President Barack Obama—State of the Union Address As Delivered.”
State of the Union Address, Washington, DC, January 13, 2016, https://obamawhitehouse.archives
.gov/the-press-office/2016/01/12/remarks-president-barack-obama-%E2%80%93-prepared-delivery
-state-union-address.
Olson, J. M. “Noncontiguous Area Cartograms.” The Professional Geographer, 28, no. 4 (1976):
371–380.
Ondov, Brian, Nicole Jardine, Niklas Elmqvist, and Steven Franconeri. “Face to Face: Evaluat-
ing Visual Comparison.” IEEE Transactions on Visualization and Computer Graphics 25, no. 1
(2018): 861–871.
Organization for Economic Cooperation and Development (OECD). “Better Life Index.” Accessed
December 2019. http://www.oecdbetterlifeindex.org/.
Organization for Economic Cooperation and Development (OECD). Program for International Stu-
dent Assessment (PISA), 2015 Reading, Mathematics and Science Assessment.
Organization for Economic Cooperation and Development (OECD). Social Expenditure Database.
Accessed January 2019. https://www.oecd.org/social/expenditure.htm.
Padilla, Lace, Matthew Kay, and Jessica Hullman. “Uncertainty Visualization.” PsyArXiv. April 27,
2020. doi:10.31234/osf.io/ebd6r.
Pätzold, André, Julius Tröger, Joachim Fahrun, Christopher Möller, David Wendler and Marie-Louise
Timcke. “These Candidates Live the Furthest Away from Their Voters.” Berliner Morgenpost, Sep-
tember 8, 2016, https://interaktiv.morgenpost.de/waehlernaehe-berlin/.
Perez, Caroline Criado. Invisible Women: Exposing Data Bias in a World Designed for Men. New York:
Random House, 2019.
Playfair, William. Playfair’s Commercial and Political Atlas and Statistical Breviary. Cambridge: Cam-
bridge University Press, 2005.
Porter, Mark and Rhys Blakely. 2020. “Coronavirus: How Safe Are You? All Your Health Questions
Answered.” Australian, March 31, 2020, https://www.theaustralian.com.au/world/the-times
/coronavirus-how-safe-are-you-all-your-health-questions-answered/news-story/c40ec5841763c6d
57f986c3ea3e19e4f
Presentitude. “A Guide to 44 Safe Fonts & 74 Safe Combos for PowerPoint.” Presentitude (blog) no
date, access January 2020, http://presentitude.com/fonts/.
Priestley, Joseph. “A Specimen of a Chart of Biography.” 1765.
Project Tycho. https://www.tycho.pitt.edu/data.
Rendgen, Sandra. The Minard System: The Complete Statistical Graphics of Charles-Joseph Minard.
New York: Princeton Architectural Press, 2018.
R Graph Gallery. “Network Graph.” https://www.r-graph-gallery.com/network.html.
REFERENCESփ 425
Richards, Neil. “Is White Space Always Your Friend?” blog post, January 6, 2018, https://questionsin
dataviz.com/2018/01/06/is-white-space-always-your-friend/.
Robbins, Naomi B. Creating More Effective Graphs. New York: Wiley, 2012.
Rogers, Simon. “Jon Snow’s Data Journalism: The Cholera Map That Changed the World.” Guardian
Datablog, March 15, 2013, https://www.theguardian.com/news/datablog/2013/mar/15/john-snow
-cholera-map.
Rosenberg, Daniel. “Joseph Priestley and the Graphic Invention of Modern Time.” Studies in Eighteenth-
Century Culture 36, no. 1 (2007): 55–103.
Roser, Max and Esteban Ortiz-Ospina. “Global Education.” Our World in Data. Accessed January
2020. https://ourworldindata.org/global-education. 2020.
Roser, Max and Esteban Ortiz-Ospina. “World Development Indicators.” Our World in Data. Accessed
January 2020. https://ourworldindata.org/global-education.
Roser, Max, Hannah Ritchie, and Esteban Ortiz-Ospina. “Coronavirus Disease (COVID-19)—Statistics
and Research.” Our World in Data, https://ourworldindata.org/coronavirus#the-growth-rate-of
-covid-19-deaths, accessed April 2020.
Rosling, Ted. “The Best Stats You’ve Ever Seen.” TED2006, TED Talk, February 2006, https://www.ted
.com/talks/hans_rosling_the_best_stats_you_ve_ever_seen.
Rost, Lisa Charlotte. “An Alternative to Pink and Blue: Colors for Gender Data.” Chartable, a blog by
Datawrapper, July 10, 2018, https://blog.datawrapper.de/gendercolor/.
Rost, Lisa Charlotte. “How to Read a Log Scale.”Chartable, (a blog by Datawrapper). May 31, 2018.
https://blog.datawrapper.de/weeklychart-logscale/
Rost, Lisa Charlotte. “What I Learned Recreating One Chart Using 24 Tools.” Source (blog), December 8,
2016, https://source.opennews.org/articles/what-i-learned-recreating-one-chart-using-24-tools/.
Roth, Madeline and Amma Malik. “What Keeps Cities in Asia and Africa from Effective Public Service
Delivery?” Urban Wire blog, Urban Institute, August 2, 2016, https://www.urban.org/urban-wire
/what-keeps-cities-asia-and-africa-effective-public-service-delivery.
Rothstein, Richard. The Color of Law: A Forgotten History of How Our Government Segregated America.
New York: Liveright, 2017.
Rousseeuw, Peter J., Ida Ruts, and John W. Tukey. “The Bagplot: a Bivariate Boxplot.” American Statis-
tician 53, no. 4 (1999): 382–387.
Ruggles, Steven, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer, Jose Pacas and Matthew
Sobek. IPUMS USA: Version 9.0 [American Community Survey 2017]. Minneapolis, MN: IPUMS,
2019. https://doi.org/10.18128/D010.V9.0
Sankey, M.H. “Minutes of Proceedings of The Institution of Civil Engineers.” Vol. CXXXIV, Session
1897–98. Part IV.
Schleuss, Jon and Rong-Gong Lin II. “California Crime 2013.” Los Angeles Times, https://graphics
.latimes.com/california-crime-2013/
426փ REFERENCES
Schuetz, Jenny. “Who Is the New Face of American Homeownership?” Brookings Institution,
October 9, 2017, https://www.brookings.edu/blog/the-avenue/2017/10/09/who-is-the-new-face-of
-american-homeownership/.
Schwabish, Jonathan A. “Take a Penny, Leave a Penny: The Propensity to Round Earnings in Survey
Data.” Journal of Economic and Social Measurement 32, no. 2–3 (2007): 93–111.
Schwabish, Jonathan, and Alice Feng. “Applying Racial Equity Awareness in Data Visualization.”
(2020). http://osf.io/x8tbw/
Schwabish, Jonathan. “How a Map’s ‘Bins’ Can Influence Your Perception of an Important Policy
Issue.” Urban Wire Urban Institute, November 14, 2017, https://www.urban.org/urban-wire/how
-maps-bins-can-influence-your-perception-important-policy-issue.
Shneiderman, Ben. “A Grander Goal: A Thousand-Fold Increase in Human Capabilities.” Edu-
com Review, 32, 6, Nov/Dec 1997, http://www.ifp.illinois.edu/nabhcs/abstracts/shneiderman
.html
Shneiderman, Ben. “Tree Visualization with Tree-Maps: 2-d Space-Filling Approach.” ACM Transac-
tions on Graphics (TOG) 11, no. 1 (1992): 92–99.
Skau, Drew, and Robert Kosara. “Arcs, angles, or areas: Individual data encodings in pie and donut
charts.” Computer Graphics Forum 35, no. 3. 2016.
Skau, Drew, and Robert Kosara. “Judgment Error in Pie Chart Variations.” In Short Paper Proceedings
of the Eurographics/IEEE VGTC Symposium on Visualization (EuroVis), 91–95. 2016.
Slobin, Sarah. “Two Maps Explain the Racial History Behind Alabama’s Senate Vote.” Quartz, December
13, 2017, https://qz.com/1155837/how-doug-jones-beat-roy-moore-maps-of-alabama-in-1860-and
-2017-offer-a-startling-comparison/.
Social Security Administration. “The 2019 Annual Report of the Board of Trustees of the Federal Old-
Age and Survivors Insurance and Federal Disability Insurance Trust Funds.” 2019, https://www.ssa
.gov/OACT/TR/2019/.
Social Security Advisory Board, Aspects of Disability Decision Making: Data and Materials, February
2012. https://www.ssab.gov/Portals/0/OUR_WORK/REPORTS/Chartbook_Aspects%20of%20
Disability%20Decision%20Making_2012.pdf
Social Security Advisory Board. “2011 Technical Panel Report on Assumptions and Methods.” 2011,
https://www.ssab.gov/Portals/0/OUR_WORK/REPORTS%20TO%20THE%20BOARD/TPAM
_Report_2011.pdf.
Soffen, Kim. “To Reduce Suicides, Look at Guns.” Washington Post, July 13, 2016, https://www.washington
post.com/graphics/business/wonkblog/suicide-rates/?tid=a_inl
Spear, Mary Eleanor. Charting Statistics. New York: McGraw-Hill, 1952.
Spence, Ian, and Howard Wainer. “William Playfair: A Daring Worthless Fellow.” Chance 10, no. 1
(1997): 31–34.
St. Clair, Kassia. The Secret Lives of Color. New York: Penguin, 2017.
REFERENCESփ 427
Strunk Jr, William, and E. B. White. The Elements of Style. New York: Macmillan, 1959.
Szűcs, Krisztina. “Spotlight on Profitability.” http://krisztinaszucs.com/my-product/hollywood/
Texas Tribune News Apps Style Guide. https://apps.texastribune.org/styles/
Texas Tribune. “What the Texas delegation says about impeachment.” Texas Tribune, accessed
September 26, 2019, https://graphics.texastribune.org/graphics/impeachment-delegation-tracker
-2019-09/.
Torres, Nicole. “Why It’s So hard for Us to Visualize Uncertainty.” Harvard Business Review, November
11, 2016, https://hbr.org/2016/11/why-its-so-hard-for-us-to-visualize-uncertainty.
Treisman, Anne. “Preattentive Processing in Vision.” Computer Vision, Graphics, and Image Processing
31, no. 2 (1985): 156–177.
Tse, Archie. “Why We Are Doing Fewer Interactives.” presentation at the 2016 Malofiej Conference,
https://github.com/archietse/malofiej-2016/blob/master/tse-malofiej-2016-slides.pdf
Tufte, Edward R. Beautiful Evidence, vol. 1. Cheshire, CT: Graphics Press, 2006.
Tukey, John W. Exploratory Data Analysis, vol. 2. Reading, MA: Addison-Wesley, 1977.
Turner, Cory. “Is There A Better Way To Pay For America’s Schools?” NPR, May 1, 2016, https://www
.npr.org/2016/05/01/476224759/is-there-a-better-way-to-pay-for-americas-schools.
Uncertainty Toolkit for Analysts in Government. Website. Accessed February 2020. https://analysts
uncertaintytoolkit.github.io/UncertaintyWeb/index.html.
United Nations. “Population by Age, Sex, and Urban/Rural Residence.” http://data.un.org/Data
.aspx?d=POP&f=tableCode%3A22.
Urban Institute. “Urban Institute Data Visualization Style Guide.” No date, http://urbaninstitute
.github.io/graphics-styleguide/, accessed January 2020.
US Bureau of Economic Analysis. “Gross Domestic Product by Industry and Input-Output Statistics.”
January 22, 2015, https://apps.bea.gov/histdata/fileStructDisplay.cfm?HMI=8&DY=2014&DQ=Q3
&DV=Quarter&dNRD=January-22-2015.
US Bureau of the Census. “Map Showing the Distribution of the Slave Population of the United States.”
https://www.census.gov/history/pdf/1860_slave_distribution.pdf
US Department of Agriculture, Economic Research Service. “Midpoint acreage more than doubled
for all five major field crops.” May 10, 2018, https://www.ers.usda.gov/data-products/chart-gallery
/gallery/chart-detail/?chartId=88868.
US Department of Agriculture. “Food Distribution Program on Indian Reservations.” https://www
.fns.usda.gov/fdpir/food-distribution-program-indian-reservations.
US Department of Agriculture. “Visual Standards Guide.” January 2013, https://www.usda.gov/sites
/default/files/documents/visual-standards-guide-january-2013.pdf.
US Department of Education. “School Composition and the Black-White Achievement Gap.” National
Center for Education Statistics, June 2015, https://nces.ed.gov/nationsreportcard/subject/studies
/pdf/school_composition_and_the_bw_achievement_gap_2015.pdf.
428փ REFERENCES
USA Today. “All of Obama’s State of the Union Speeches in Word Clouds.” USA Today, January 12,
2016, https://www.usatoday.com/story/news/politics/onpolitics/2016/01/12/obama-state-of-the-union
-word-clouds/78712780/.
van der Bles, Anne Marthe, Sander van der Linden, Alexandra LJ Freeman, James Mitchell, Ana B.
Galvao, Lisa Zaval, and David J. Spiegelhalter. “Communicating Uncertainty About Facts, Num-
bers and Science.” Royal Society open science 6, no. 5 (2019): 181870.
van Gogh, Vincent. Vincent van Gogh to Theo van Gogh, Amsterdam, October 28, 1885, http://www
.vangoghletters.org/vg/letters/let537/letter.html
Vessey, Iris. “Cognitive Fit: A Theory‐Based Analysis of the Graphs Versus Tables Literature.” Decision
Sciences 22, no. 2 (1991): 219–240.
Vigen, Tyler. Spurious Correlations. Website. http://www.tylervigen.com/spurious-correlations.
Wagemans, Johan, James H. Elder, Michael Kubovy, Stephen E. Palmer, Mary A. Peterson, Manish
Singh, and Rüdiger von der Heydt. “A Century of Gestalt Psychology in Visual Perception: I. Per-
ceptual grouping and figure–ground organization.” Psychological Bulletin 138, no. 6 (2012): 1172.
Wagemans, Johan, Jacob Feldman, Sergei Gepshtein, Ruth Kimchi, James R. Pomerantz, Peter A. Van
der Helm, and Cees Van Leeuwen. “A Century of Gestalt Psychology in Visual Perception: II. Con-
ceptual and Theoretical Foundations.” Psychological Bulletin 138, no. 6 (2012): 1218.
Wainer, Howard, and Ian Spence. Playfair’s Commercial and Political Atlas and Statistical Breviary.
Cambridge: Cambridge University Pres, 2005.
Walker, Mason. “Americans Favor Mobile Devices Over Desktops and Laptops for Getting News.” Pew
Research Center, November 19, 2019, https://www.pewresearch.org/fact-tank/2019/11/19/americans
-favor-mobile-devices-over-desktops-and-laptops-for-getting-news/.
Ware, Colin. Information Visualization: Perception for Design, 3rd ed. Waltham, MA: Morgan Kaufmann,
2020.
Washington Metropolitan Area Transit Authority. Public Access to Records Policy request. October 2019.
Watson, Cecelia. Semicolon: How a Misunderstood Punctuation mark Can Improve Your writing, Enrich
Your Reading and Even Change Your Life. HarperCollins UK, 2019.
Wattenberg, Martin, and Fernanda B. Viégas. “The Word Tree, an Interactive Visual Concordance.”
IEEE Transactions on Visualization and Computer Graphics 14, no. 6 (2008): 1221–1228.
Wattenberg, Martin. “Arc Diagrams: Visualizing Structure in Strings.” In IEEE Symposium on Informa-
tion Visualization, 2002. INFOVIS 2002., pp. 110–116. IEEE, 2002.
Web Content Accessibility Guidelines (WCAG) 2.0. https://www.w3.org/TR/WCAG20/#understandable,
accessed November 2019.
Weissgerber, Tracey L., Natasa M. Milic, Stacey J. Winham, and Vesna D. Garovic. “Beyond Bar and
Line Graphs: Time for a New Data Presentation Paradigm.” PLoS biology 13, no. 4 (2015): e1002128.
Wellcome Collection. “Diagram of the Causes of Mortality in the Army.” https://wellcomecollection
.org/works/sz9sms2m
REFERENCESփ 4 29
Wexler, Steve, Jeffrey Shaffer, and Andy Cotgreave. The Big Book of Dashboards: Visualizing Your Data
Using Real-World Business Scenarios. New York: Wiley, 2017.
The White House. “President Obama’s Final State of the Union.” February 12, 2016. https://obama
whitehouse.archives.gov/sotu.
Wickham, Hadley, and Lisa Stryjewski. “40 Years of Boxplots.” American Statistician (2011).
Wikipedia. “Right-to-Work Law.” Accessed January 2020. https://en.wikipedia.org/wiki/Right-to-work
_law.
Wikipedia. “State Income Tax.” Accessed January 2020. https://en.wikipedia.org/wiki/State_income
_tax.
Wilke, Claus O. Fundamentals of Data Visualization: A Primer on Making Informative and Compelling
Figures. Sebastopol, CA: O’Reilly, 2019.
Wilkinson, Krista M., and William J. McIlvane. “Perceptual Factors Influence Visual Search for Mean-
ingful Symbols in Individuals with Intellectual Disabilities and Down Syndrome or Autism Spec-
trum Disorders.” American Journal on Intellectual and Developmental Disabilities 118, no. 5 (2013):
353–364.
Wilkinson, Leland. “Dot Plots.” American Statistician 53, no. 3 (1999): 276–281.
Wilkinson, Leland. “The Grammar of Graphics.” In Handbook of Computational Statistics, 375–414.
Berlin: Springer, 2012.
Wilkinson, Leland. The Grammar of Graphics. Chicago: Springer Science & Business Media, 2013.
Witt, Jessica. “Graph Construction: An Empirical Investigation on Setting the Range of the Y-Axis.”
http://amplab.colostate.edu/reprints/Witt_Graphs_YaxisRange.pdf.
Wood, Jo, Petra Isenberg, Tobias Isenberg, Jason Dykes, Nadia Boukhelifa, and Aidan Slingsby.
“Sketchy rendering for information visualization.” IEEE Transactions on Visualization and Com-
puter Graphics 18, no. 12 (2012): 2749–2758.
World Bank. “World Development Indicators.” The World Bank Group. Accessed January 2020.
https://datacatalog.worldbank.org/dataset/world-development-indicators.
XKCD. “Correlation.” https://xkcd.com/552/.
Yau, Nathan. “Vehicles Involved in Fatal Crashes.” FlowingData blog, January 11, 2012, https://flowing
data.com/2012/01/11/vehicles-involved-in-fatal-crashes/.
INDEX
attention: engagement and, 159–160, 159–160; and, 133–134, 134; with maps, 229; mosaic
preattentive processing, 25–28, 25–28; of charts, 102–106, 103–106, 131; nested bubbles
readers, 26, 131; to series, 91 and, 121–122, 122–125, 125; outliers in, 74–75;
attributes: of colors, 27; preattentive, 25–28, paired, 369–370, 370–371, 372; paired bars,
25–28 84–86, 84–87; patterns in, 268; percentages
audiences, 61; charts for, 5, 5; communication in, 91–92; for perception, 68–69, 68–69; pie
with, 1, 62–63; comprehension of, 294–295, charts compared to, 291, 291, 293; radial,
294–296; designers and, 243, 396; engagement 81–83, 82–84, 117–118, 118, 149; for readers,
with, 2; highlighting for, 127, 128; perception 336, 337; sankey diagrams, 126–129, 126–129;
of, 301–304, 302–303; readers and, 17, 287 stacked, 302, 302, 372–373, 372–374; stacked
averages, 200 bars, 87–91, 88–91; tick marks for, 76, 76–78,
axis labels, 79, 79, 354 78–79; unit charts, 106–107, 107; variations
axis lines, 355 on, 80, 80–83, 82–84; waffle charts, 111,
axis settings: axis titles, 354; for bar charts, 111–112; waterfall charts, 129–130, 130
69–70, 70–71, 79, 79, 157; correlation and, basic data table redesign, 338–341, 338–341
253, 265–266, 266; data series and, 162–163, BEA. See Bureau of Economic Analysis
163–164; for dual-axis line charts, 143–147, Beatles (music), 317, 317
143–148, 149; label rotation in, 79, 79; in line beeswarm plots, 206–208, 206–208
charts, 136–138, 137–138; lines and, 249; for Bennett, Tim, 126
nodes, 283, 283–284; plotting and, 204; for Berliner Morgenpost, 320–321, 321
ranges, 137–138, 137–138; for ridgeline plots, Berman, Jacob, 241–242
201–202, 202; for scatterplots, 287; values and, Bertin, Willard Cope, 110
209, 209; variables and, 265 Big-Ass Numbers (BANs), 106–107, 107
azimuthal maps, 223–224, 223–224 Big Book of Dashboards, The (Wexler/Cotgreave/
Shaffer), 106–107, 107
background ranges, 120–121, 120–121 big data, 48
bad graphs, 369 bimodal distributions, 182, 182
balance, 27, 27 binary color schemes, 358, 358
BANS. See Big-Ass Numbers bins: alternatives for, 228, 228–229; arbitrary,
bar charts: axis settings for, 69–70, 70–71, 79, 228, 228; choosing, 224–225; for choropleth
79, 157; breaking the bar, 71–73, 72–73; bullet maps, 224–231, 225–231; data distribution, 227,
charts, 120–121, 120–121; for categorical 227; equal interval, 226, 226; for geospatial
comparisons, 130–131; circular, 81–83, data, 224–231, 225–231; labeling, 229–231, 230;
82–84; colors in, 13, 14; continuity in, 24, 24; no, 225, 225; for readers, 179–182, 180–181
diverging bars, 92–95, 92–97, 97; dot plots, biographies, 167, 168–169
97–100, 98–102; gauge charts, 118–120, 119; blogs, 18, 29, 53, 63
heatmaps, 112–118, 113–118; icons in, 341; Bloomberg News, 322
isotype charts, 107–111, 108–110; line charts borders, 328, 329
INDEXփ 433
318–319, 319; speeches, 314–315, 314–315; 254–256; Pearson correlation coefficient, 254,
vocabulary, 319–320, 320; word clouds, 312, 275. See also relationships
313–316, 314–316; word trees, 285, 285, 287, Cotgreave, Andy, 106–107, 107
316–318, 317–318 counting, 110
compactness, 298 COVID-19, 50–51
comparisons: bubble comparison, 121–122, Cox, Amanda, 33, 125
122–125, 125, 231, 232, 233; categories and, coxcomb diagrams, 300–304, 301–304
67, 97–100, 98–102, 126–129, 126–129, 200; Craft, Harold D., Jr., 202
details, 187; precision in, 149, 149, 186, 186; creativity: with data, 1–2; for designers, 19,
for readers, 84–86, 84–87; values, 291, 291; of 19–20, 344; for engagement, 17; learning and,
variables, 102–106, 103–106 3–8, 4–7
comprehension: accuracy and, 231; of analysis, crime, 39–40, 40
313; of audiences, 294–295, 294–296; Crimean War, 300–301, 301
categories for, 112–113, 113; circles for, 118, C-SPAN, 7, 7
118–120; colors for, 138, 150–151, 150–151, culture, 362
164–166, 165–166, 253; combinations for, curves, 139–140, 139–140, 290
120–121, 120–121; of data, 290; of differences, custom fonts, 362–363, 363
121–122, 122; encodings for, 111; engagement CVD. See color vision deficiency
and, 110–111; of geospatial data, 231, 232, 233; cycle graphs, 155–157, 156
of graphs, 287; heatmaps for, 336, 337; labels cylindrical maps, 222, 223
for, 257–258, 257–258, 260, 260; of readers,
52, 210, 211; of relationships, 134–135, 135; D3, 398, 400–401
similarity for, 30; width for, 106 Daniels, Matt, 319–320, 320
computers, 215, 367–368, 397–401, 398 DarkHorse Analytics, 330
confidence intervals, 191, 191–192 Darwin, Charles, 322, 323
Congressional Budge Office (CBO), 4–7, 4–8 data: basic data table redesign, 338–341, 338–341;
conic maps, 222, 223 budgeting and, 4–7, 4–8; charts with, 2, 2–3;
connected scatterplots, 175–177, 175–177 clutter from, 243, 282, 282; colors for, 226,
Consumer Finance Protection Bureau (CFPB), 245; communication of, 3; comprehension of,
360, 360 290; creativity with, 1–2; data-driven maps,
contiguous cartograms, 233–234, 235 217, 251, 253, 253; data-rich tables, 152, 152;
continuity, 24, 24 datasaurus, 256; Datawrapper, 262, 262–263,
continuous color palettes, 225 398, 399; for designers, 394; discrete, 46–47;
continuous data, 46–47 distribution bins, 227, 227; distribution of, 90,
Corona Virus, 50–51 90, 204–205, 204–206; editors, 33; elements,
correlation, 125, 275–276; axis settings and, 253, 120–121, 120–121; encodings of, 110, 110–111;
265–266, 266; correlation matrix graphs, equality, 47–49, 49–51; graphs and, 1, 391–392,
275–276, 275–277; nonlinear associations and, 392–396, 394, 396; independent data series,
436 փ INDEX
style guides for, 352–356, 353; sunburst, 299, Highcharts, 398, 401
299–300; symmetric distributions in, 278; highlighting: for audiences, 127, 128; with
white space in, 76, 76–77 coloring phrases, 321–322, 322–323; with
gray, 271; guidelines for, 29, 31, 43–45, 44–47; colors, 258, 259, 265; color schemes, 359, 359;
highlighting and, 267, 296; shapes and, 387, gray and, 267, 296; outliers, 334, 334–335;
387 for readers, 78, 78–79, 151; rings for, 299,
Greater London Authority, 357 299–300; values, 29–31, 31, 165–166, 165–166
grid charts, 42 “Highlighting Words in Darwin” (Fry), 323
gridded cartograms, 233, 237–240, 238–240 Hindustan Times, 163, 164
gridlines, 328, 329, 355; for clutter, 377; heavy, hinges, 197, 197
330–331, 331; for readers, 76, 76–78, 78–79; in hip hop, 319–320, 320
tables, 338–339, 338–339 histodot plots, 209, 209
growth, 177, 177, 382–385, 383–384 histograms, 179–183, 180–183, 200–202, 202, 212.
guidelines: clutter reduction, 31–33, 32–33, See also Wilkinson Dot Plots
394; for fonts, 350; for gray, 29, 31, 43–45, hive charts, 283, 283–284
44–47; integration, 29, 33–41, 34–37, 39–40; Home Owners’ Loan Corporation (HOLC),
perception and, 52; showing data, 29–31, 48–49, 49
30–31; spaghetti charts and, 29, 41, 41–42, 43; horizon charts, 164–166, 165–166
for tables, 329–336, 330–337, 344; WCAG, 367 horizontal spacing, 89, 89
How to Lie with Maps (Monmonier), 228–229,
hand-drawn looks, 195, 195–196 229
Hannibal, 246–248, 247 Hullman, Jessica, 189
Hawkins, Ed, 193
headers, 328, 328–332, 330–332 icons, 131, 311, 311–312, 325, 341
headlines, 36–38, 37 icon-scaling approach, 109–110, 110
heatmaps, 112–118, 113–118; for charts, 388; colors ICT. See information and communications
in, 340, 340–341; for comprehension, 336, 337; technology
correlation in, 275, 275–276; information in, images, 13, 352, 365–366, 365–366
129–130 inclusion, 352, 366–368
heavy gridlines, 330–331, 331 independent data series, 145, 145
Heer, Jeff, 400 India, 163, 164, 173
height, 81, 105, 109 individual points, 204, 204–205
Hermann, E. P., 217, 218 Infogram, 398, 400
hexagon grid map, 237–238, 238 information: calendars for, 116–117, 116–117; in
hierarchies, 170, 170–173, 172; in art, 317, 317; charts, 22; from data, 50–51; density of, 131;
nodes for, 299, 299–300; in relationships, 297, designers, 15, 15–16, 17; errors in, 188–189; in
297, 309 heatmaps, 129–130; information visualization
4 40 փ INDEX
radial bar charts, 81–83, 82–84, 117–118, 118, 149 outliers for, 207–208; patterns for, 128–129,
radial flow maps, 245 129; perception of, 15; pie charts for, 294–295,
“Radio Observations of the Pulse Profiles” 294–296; preattentive processing for, 25–28,
(Craft), 203 25–28; proximity for, 22, 22; relationships
rainbows, 361, 361–362 for, 144–145, 144–145; representation for,
rainclouds, 212–213, 212–213 159–162, 159–162; research on, 111; scales for,
ramps, 225 234, 235–236, 236; segments for, 87–91, 88–91;
randomness, 188–189 similarity for, 23, 23; space for, 333, 333–334;
random variability, 74 storytelling for, 137–138, 137–138; time for,
ranges: axis settings for, 137–138, 137–138; 133, 177–188; values for, 68–69, 68–69, 309,
background, 120–121, 120–121; IQR, 197, 197 372–373, 372–373; zooming for, 71–73, 72–73
ranking, 13, 14, 153–154, 153–155 real-time data, 51
Ratia, Armi, 102 Real Value Added (RVA), 383–385, 384
Ratia, Viljo, 102 recall, 37–38
ratio scales, 46–47 recognition, 37–38
Raw, 398, 400 redesigns, 36, 36; basic data table redesign,
readers: accuracy of, 291, 291; annotations for, 338–341, 338–341; for choropleth maps,
33–34, 34; attention of, 26, 131; audiences 378–380, 379–380; for dot plots, 380–385,
and, 17, 287; bar charts for, 336, 337; bins 381–384; for line charts, 370, 371, 374–378,
for, 179–182, 180–181; cartograms for, 375–378, 385–386, 385–386; for paired bar
233–234, 234; charts for, 295; circles for, charts, 369–370, 370–371, 372; regression
122–123, 125, 125, 301–304, 302–303; closure table redesign, 341–343, 342–343; for stacked
for, 24, 24; clutter for, 338–339, 338–339, bar charts, 372–373, 372–374; for tables, 387,
343, 343; comparisons for, 84–86, 84–87; 387–388, 389
comprehension of, 52, 210, 211; connection reduction, 31–33, 32–33, 394
for, 25, 25; content for, 172–173; data for, 75; regression table redesign, 341–343, 342–343
designers and, 391–392, 392; differences for, Rehabilitation Act, 366–367
85–86, 85–87; distribution for, 90, 90, 179, relationships: in arc charts, 272–273, 272–274,
229; engagement with, 61–62, 114–115, 115, 275; in bubble plots, 256–258, 257–262,
312, 312–313, 325; fonts for, 331–332, 331–332; 260–263; in chord diagrams, 269–272,
graphs for, 150, 150; gridlines for, 76, 76–78, 270–271; comprehension of, 134–135, 135; in
78–79; highlighting for, 78, 78–79, 151; correlation matrix graphs, 275–276, 275–277;
hurdles for, 17–18, 18; integration for, 29, data, 104; hierarchies in, 297, 297, 309; in
33–41, 34–37, 39–40; layout for, 178; legends network diagrams, 277–284, 278–283; in
for, 161, 229–230, 229–231; lines for, 35, parallel coordinates, 263–265, 264–266,
35; maps for, 231, 232, 233; mobile, 61–62, 267; in radar charts, 267–268, 267–269; for
62; numbers of observations for, 227, 227; readers, 144–145, 144–145; in scatterplots,
INDEXփ 4 45
249, 250–254, 251, 253; between series, 20–21, scales, 46–47; area and, 109; colors and, 103;
20–21; in tree diagrams, 284–285, 284–285, icon-scaling approach, 109–110, 110; Likert
286; between variables, 249, 254–256, 287. See Scales, 92–95, 92–97, 97; log, 260–263,
also part-to-whole relationships 261–262; for readers, 234, 235–236, 236
relative change, 262, 262–263 scatterplots, 105; axis settings for, 287; connected,
relative numbers, 263 175–177, 175–177; for maps, 380, 380;
Rendgen, Sandra, 247–248 relationships in, 249, 250–254, 251, 253. See
reports, 4–7, 4–8, 376–378, 377–378 also bubble plots
representation, 17–18, 125, 125, 159–162, 159–162 schedule-tracking devices, 166–167, 167–169
research: analysis and, 350; on curves, 290; schematic plots, 196–198, 196–198
icons for, 313; on images, 13; information schemes, for colors, 358–360, 358–361
visualization, 189; Microsoft Research, 61, 62; school lunch programs, 380–382, 381–382
Pew Research Center, 37, 61–62; on readers, seasonal trends, 155–157, 156
111 Secret Lives of Color, The (St. Clair), 362
responses, 95, 95–97, 97 Sedaka, Neil, 39
responsibility, 47–49, 49–51 segments, 87–91, 88–91
ribbon effect, 154, 155 semantic groupings, 315–316, 316
Richards, Neil, 38–39, 39 senate elections, 378–380, 379–380
ridgeline plots, 201–203, 202–203 sentiment, 92–95, 92–97, 97
right skewed distributions, 182, 182 sequences, 170, 170–173, 172
rings, 299, 299–300 sequential color schemes, 358, 358–359
Robinson projection, 221–222, 222 series: in area charts, 157–158, 157–158;
robustness, 367 associations between, 147, 148, 175; attention
rose diagrams, 300–304. 301–304 to, 91; change in, 162, 162; clutter with, 128–
Rosling, Hans, 249, 251 129, 129; data, 130, 130, 162–163, 163–164, 177,
Rossi, Gabriele, 395 356; dots for, 25, 25; independent data, 145,
Rost, Lisa Charlotte, 262, 262–263 145; in line charts, 134–136, 135–136; plotting,
Rothstein, Richard, 48–49, 49 105; relationships between, 20–21, 20–21;
rounding numbers, 230, 230 specific points in, 141–142, 142; time, 175–177,
rows: cells and, 328, 329; columns and, 333, 175–177; uninterrupted, 142, 142
333–334; total, 339, 339–340 service delivery, 372–373, 372–374
rules, 328, 329 Shaffer, Jeff, 106–107, 107
RVA. See Real Value Added shapes: circles, 118, 118–120; clutter and, 382;
graphs and, 80, 80–81, 192, 192; gray and, 387,
St. Clair, Kassia, 362 387; polygons, 304–305
Sankey, Matthew Henry Phineas Riall, 126 Shneiderman, Ben, 61, 297–298
sankey diagrams, 126–129, 126–129 signals, 142, 142, 189
446փ INDEX