Reading Response

SHOW ME THE NUMBERS

Designing Tables and Graphs to Enlighten

SECOND EDITION

STEPHEN FEW

Analytics Press

BURLINGAME, CALIFORNIA

Analytics Press

PO Box 1545

Burlingame, CA 94011

SAN 253-5602

www.analyticspress.com

Email: [email protected]

© 2012 by Stephen C. Few

All rights reserved. Protected under the Berne Convention.

PUBLISHER: Jonathan G. Koomey

COPY EDITOR: Nan Wishner

COMPOSITION: Bryan Pierce

COVER DESIGN: Stephen Few

PHOTOGRAPHY: john Fernez

PRINTER AND BINDER: C&C Offset Printing Company

Reproduction or translation of any part of this work in any form or

by any means, electronic or mechanical, beyond that permitted by

Section 107 or 108 of the 1976 United States Copyright Act without

the expressed permission of the copyright owner is unlawful.

Requests for permission or further information should be addressed to Analytics Press at the address or URL above.

ISBN-10: 0-9706019-7-2

ISBN-13: 978-0-9706019-7-1

This book was printed on acid-free paper in China.

10 9 8 7 6 5 4 3 2 1

140 SHOW ME. THE. NUMBERS

without changing anything else about the manufacturing process or facilities

may have reslllted in people simply getting in each other's way.

You decide to show the Operations Manager the strength of this relationship

of increased staff to decreased productivity before taking any further steps. You

have daily headcount and productivity statistics for the last year. Both head­

count and productivity remained fairly steady until just after the Operations

Manager's arrival. In what form will you present your information?

Scenario #6

For the first time ever, your organization has built a database that contains

comprehensive and reliable information about donations. Since it became

available, you've been slicing and dicing the information in various ways,

looking for answers to important questions that you've never before been able to

investigate. One of your queries involved a list of every single donation for the

past year, sorted by size in U.S. dollars from the biggest to the smallesit. You took

your list and divided it into 10 equal groups labeled "Over 90 -100%" (i.e., the

top 10%), "Over 80 - 90%", and so on, to the final one labeled "Over O - 10%".

Next, you calculated the running percentage of total income associated with the

donations, beginning with the largest and continuing all the way to the small­

est. You were then able to easily see the amount of income that each group of

donations contributed to overall income.

You were amazed to discover that the top 10% of your donations contributed

87% of your total revenue. After the top 10%, the income contribution of the

remaining 90% of your donations dropped off dramatically, with the last 50%

contributing only 1 % of total income. You have no do..ibt that your organiza­

tion's leadership will find this discovery enlightening. You want to present this

message as concisely and clearly as possible. You realize that if you don't hit

them between the eyes with this important reve.lation in a single page of

information, they won't bother reading it. What form will you give to this

information to ensure that it hits the mark?

Responses to Scenario #6:

Table or graph?

If a table, which kind?

If a graph, what kind of relationship?

If a graph, which graphical objects

for quantitative encoding?

Anything else'

You can find answers to the six

scenarios in Appendix F, Answers to

Practice in Selecting Tables and

Graphs.

7 GENERAL DESIGN FOR

COMMUNICATION

With a basic understanding of visual perception, we can build a set of visual design

principles, beginning with those that apply equally to tables and graphs. Our

primary visual design objectives will be to present content to readers in a manner

that highlights what's important, arranges it for clarity, and leads them through it in

the sequence that tells the story best.

Visual design can serve many purposes, not least of which is to create beauty,

which we can appreciate purely for its own sake. This is the work of the artist.

Without it our Jives would be dismal and our souls malnourished. Artists spend

their lives learning from the masters and their own painstaking experience.

Through each stroke of the brush, angle of t h e chisel, or subtle positioning of

the light, they attempt to move us in some way. As creators of tables and graphs,

our use of visual design serves a different purpose but one that is also funda­

mental to life and deserves no less attention. We use visual design to communi­

cate. There are stories in the numbers that will be perceived and acted upon or

will go unnoticed and be ignored, depending on our knowledge of visual design

and our ability to apply that knowledge to the important task of

communication.

In this chapter we'll examine the aspects of visual design that apply equally

to all visual forms for communicating quantitative information, including

tables, graphs, and text. These general practices of communication-oriented

design support two fundamental objectives:

1. Highlight

2. Organize

We highlight important information to give it a voice that comes through

loudly and clearly, without distraction. We organize information to lead readers

through it in a manner that promotes optimal understanding and use.

Highlight

It is appropriate to begin this section by repeating six incisive words written by

Edward Tufte: "Above all else show the data." 1 These words should be our

mantra. Nothing is more central to our task.

Tufte introduced a useful concept known as the data-ink ratio. Tables and

graphs are composed of ink on the page. Some ink represents actual information

and some does not (e.g., supporting components like grid lines or superfluous

components like ornamentation that play no role whatsoever in presenting the

data). The data-ink ratio is the amount of ink that presents information com­

pared to the total amount of ink. The degree to which we feature data in a table

1. Edward R. Tufte (2001) The Visual

Display of Quantitative Information,

Second Edition. Graphics Press,

page 92.

142 SHOW ME THE NUMBERS

or graph corresponds in large part to the percentage of ink that we used to

represent data rather than non-data.

The object isn't to eliminate all non-data ink. To some degree we always need

supporting visual components to make tables and graphs readable. The object is

to reduce the non-data ink to no more than what's necessary to make the data

ink understandable.

We highlight data through a design process that involves activities of rn•o

types:

I. Reducing the non-data ink

2. Enhancing the data ink

Reduce the Non-Data Ink

The process of reducing the non-data ink involves two steps:

1. Subtract unnecessary non-data ink

2. De-emphasize and regularize the remaining non-data ink

SUBTRACT UNNECESSARY NON-DATA INK

Subtracting unnecessary non-data ink begins by asking the following question

about each visual component: "Would the data suffer any loss of meaning or

impact if this were eliminated?" If the answer is "no," then get rid of it. Resist

the temptation to keep things just because they're cute or because you worked so

hard to create them. If they don't support the message, they don't serve the

purpose of communication. As the author Antoine de Saint-Exupery suggests:

"In anything at all, perfection is finally attained not when there is no longer

anything to add, but when there is no longer anything to take away."2

By subtracting what is not needed to support the message, you bring your

communication one step closer to elegance. The word elegance comes originally

from the Latin term eligcre, which means to choose out or to select carefully. To

achieve elegance in communication, you must carefully select the content that

is essential to the message and trim all else away.

DE-EMPHASIZE AND REGULARIZE THE REMAINING NON-DATA INK

Once you've subtracted all the unnecessary non-data ink, you should push the

non-data ink that remains far enough into the background to enable the data to

stand out clearly in the foreground. This can be achieved by reducing the visual

prominence of non-data ink components.

Tab.les and graphs consist of three visual layers: 1) data as the top or promi­

nent layer, 2) 11011-dntn items as the middle layer, and 3) the bnckgrormd (the

surface on which the data and supporting components reside). Kon-data items,

consistent with their supporting role, should stand out just enough from the

background to serve their purpose but not so much that they draw attention to

themselves. This can be achieved through the use of thin lines and soft, neutral

colors (e.g., light gray). To clo otherwise, for example to give the same visual

weight to data and non-data items, would provide no visual cues to lead the

reader's eyes to what's important. When everything stands out, nothing stands

out.

2. This quotation ol Antoine de

Saint-Exupery and the explanation of

the term elegance were taken from

Kevin Mullet and Darrel Sano (1995)

Designing Visual lnrerfaces. Sun

Microsystems, Inc., page 17.

GENERAL DESIGN FOR COMMUNICATION 143

BecaU5e a reader's eyes are drawn to contrast, you can go one step further to

reduce the visibility of non-data inJ< by making it as consistent as possible, so

that none of it stands out. Multiple instances of the same supporting compo­

nents throughout a report should look precisely the same everywhere they

appear. Any differences work against your purpose by inviting your readers' eyes

to notice and their brains to assign m�aning to those differences.

Take a few minutes now to examine two or three of your own reports to

identify opportunities to reduce the non-data ink. You may be surprised to find

how much there is that could be subtracted, muted, and regularized for greater

effect.

Enhance the Data Ink

You can enhance the data ink through a process that consists of two steps:

1. Subtract unnecessary data ink

2. Emphasize the most important data ink

SUBTRACT UNNECESSARY DATA INK

You must carefully avoid the common mistake of saying too much. Not all

information is equally important. This is especially true when your readers dm1't

have the time or the patience necessary to savor a message in all its subtlety.

Don't remove anything that's important, but be sure to remove all that is

peripheral to the interests and purposes of your readers. Every step taken to

reduce data causes what remains to stand out even more. The more you earn

your readers' trust by giving them only what they need, the more they'll pay

attention to everything you give them.

The intention here is to summarize when detail Isn't necessary and to trim

away what's not important, not to arbitrarily reduce the content of your mes­

sage. It's appropriate for a single table or graph to deliver a great deal of informa­

tion or to a rti.culate a ·comp.lex (but not overly complicated) message. Strive to

give your readers what they need, and a!J that they need, but nothing more.

EMPHASIZE THE MOST IMPORTANT DATA INK

Data values are encoded differently in tables than in graphs. In tables, they are

encoded entirely in verbal language (i.e., words and numbers), but in graphs

they are encoded primarily in visual language (e.g., points, lines, bars, and

boxes) although words and numbers are used as well. Regardless of the encoding

method, certain visual attributes of objects, words, and numbers stand out more

than others.

In the earlier chapter on visual perception, you learned that preattentive

visual attributes differ in the degree to which they stand out. Size is a good

example. You can make something stand out by making it bigger. Objects,

words, and numbers that are bigger stand out more than those that are smaller,

all else being equal. You can take advantage of this to emphasize the most

important data ink relative to the rest. Here is a list of the preattentive visual

144 S H OW M E T H E N U M B ER S

attributes that are especially useful for emphasizing data ink in tables and

graphs:

Attribute

Width

Oriental ion

Size

Enclosure

Hue

Color intensity

Values Useful for Emphasis

Thicker lines (including words and numbers that are

boldfaced) stand out more than thinner lines.

Slanted words and numbers (i.e., italic�) stand out more than

those that are oriented normally (i.e., not slantec.l), assuming

that vertically oriented fonts are the norm.

Bigger objects, words, and numbers stand out more than

smaller objects.

Objects, words, and num bers that arc enclosed by lines or

background fill color stand out more than those that are not

enclosed.

Objects, words, and numbers that have a hue that is different

from the norm stand out.

Objects, words, and numbers that are dark or bright stand out

more than those that are light or pale.

Each step in the process of highlighting data results in greater simplicity. In

the communication of quantitative information, simplicity of design is the

essence of elegance. Your message might be complex, but its design-the form i n

which you present it-should b e s o simple that t o your readers it i s nearly

invisible.

Organize

When your readers look at a page or screen of information, they immediately

begin to organize what they see in an effort to make sense of it. As a designer of

communication, it is your job to organize the information for them in a manner

that tells the story as clearly as possible. If you fail to do this effectively, the

result is not information that is unorganized, but information that is organized

in a manner that does not support its essential story, resulting in ineffective

communication. In fact, your readers may get a different message entirely-per­

haps one that is wrong. Communication involves much more than knowing

what to say; it also involves knowing how to say it.

The page or screen that serves as your medium of communication will often

contain more than a single table or graph. Your message may require multiple

tables, multiple graphs, or a combination of both, along w ith additional text in

the form of annotations, sentences, or even whole paragraphs. When you

arrange information on the page or screen, you must consciously do so to tell a

story. What should I say first? What should I save for last? What should I

emphasize more than the rest? The answers to these questions take on the form

of v isual attributes designed to accomplish the following:

1. Group (i.e., segment information into meaningful sections)

2. Prioritize (i.e., rank information by importance)

3. Sequence (i.e., provide direction for the order in which information should

be read)

GENERAL DESIGN FOR COMMUNICATION 145

Group

You must always begin with a clear sense of what belongs together, that is, what

your readers should perceive as belonging to tJ1e same group because those units

of information have something in common. Once this is clear, you can select

from various visual design techniques that organize the information into

groups.

Grouping takes place on several levels. You begin with your overall message

and then break down its content into different topics. The various topics are

then grouped into the appropriate modes of expression: tables, graphs, and text.

Within tables and graphs, information is naturally grouped into categorica I and

quantitative data. Finally, categorical information is broken down into its

various items, and quantitative values are associated with each of those categori­

cal items.

It's youI job to make this grouping obvious to your readers. It shouldn't be up

to your readers to do the work of arranging the content into meaningful groups

when you can do this in advance for them.

The Gestalt principles of visual perception reveal a number of techniques that

can be used to group information meaningfully. The simplest approach­

proximity-is often the best. This is especially true for arranging content into

various topics. If you were communicati11g quarter-to-date sales performance to

a sales manager, your overall message might consist of regional sales perfor­

mance compared to forecast, the top orders, and the top customers. You would

have a single story consisting of three related topics. By placing the information

related to each topic close together and by separating the topic groups by white

space, you create a simple and clear arrangement with nothing to hinder your

readers' eyes as they move from one group to another.

Sometimes your message consists of several separnte but related topics that

need to be appropriately grouped and arranged on a page or screen, and among

those individual topics reside relationships that should be identified. Let's

continue the previous example. The story of quarter-to-date sales consists of

three primary topics; but each includes sales expressed both as bookings and

billings. Let's assume that you have appropriately arranged the three topics into

a graph that displays regional sales performance compared to the forecast, a

table listing the top orders, and a table listing the top customers, each separated

by enough white space to render it distinct. Even though bookings and billings

both appear in the graph as well as the tables, it would be helpful to tie the

separate instances of each together visually. This would make it easier for your

readers to quickly scan for all bookings information separately from billings

information, and vice versa. You could do so simply by selecting one of the

remaining Gestalt principles or one of the preattentive attributes that you

learned about in the chapter on visual perception. Given this scenario, what

attribute or principle might you choose to visually group bookings as distinct

from billings? Take a moment to run through the list of available methods,

weighing the potential advantages of each.

(

146 SHOW ME THE NUMBERS

There is no one right answer. You might have realized while assessing the

alternatives, however, that many of the available methods suggest that either

bookings or billings are more important than the other. For instance, if you

chose color intensity to distinguish bookings from billings, rendering bookings

as black and billings as gray, bookings would stand out more. This would be

appropriate if the purpose of your message were to emphasize bookings over

billings, but if you wished to treat them equally, variation in color intensity

would not be the best method. You could, however, select different hues for

each.

Tables and graphs both use conventional means to organize information into

categories. Tables primarily use the Gestalt principles of proximity and continu­

ity to organize different categories into columns and rows. Graphs use many

techniques, such as the principles of similarity (e.g., common hues or shapes)

and connection (e.g., the use of a line to connect points) to group data. We'll

examine these techniques further in the coming chapters on table and graph

design.

Prioritize

Whenever you communicate quantitative information, it is important to step

back and ask yourself, "What are the important numbers here?" Once you've

established a clear sen se of what's important, you should make that information

stand out clearly from the rest. This is a vital part of your job. Don't just high­

light i mportant numbers when you happen to think about it or when their

importance hits you over the head. Consider it every time.

Not all numbers are equally important. In fact, some numbers are so much

more important than others that a few seconds spent examining and under­

standing them produces benefits that could never be equaled by years of

concentration on all the others. Help your readers develop a productive

approach to numbers by pointing out those that most deserve attention.

We learned in the chapter on visual perception that some preattentive visual

attributes are perceived quantitatively. Their values can be arranged along a

continuum ranging from less to more, small to big. Such attributes have the

built-in ability to make some information stand out as more prominent than the

rest. Here's a reminder of the attributes that are perceived quantitatively, along

with examples of how each can be used to highlight important sections of text,

tables, and graphs:

Attribute

Width

Size

Color intensity

2-0 position

Tables and Text

• Boldfaced text

• Bigger tables

• Larger fonts

• Darker or brighter colors

• Positioned at the top

• Positioned at the left

• Positioned in the center

Graphs and Objects

• Thicker graph lines

• Wider bars

• Bigger graphs

• Bigger symbol shapes

• Darker or brighter colors

• Positioned at the top

• Positioned at the left

• Positioned in the center

GENERAL DESIGN FOR COMMUNICATION 147

As the final item in the list suggests, certain positions in a 2-D space stand

out as more prominent than the rest. When language is read from left to right

and top to bottom, you can make text, tables, or graphs appear more prominent

by locating them at the top left of a page or screen.

In addition to these quantitatively percejved visual attributes, you can also

take advantage of the fact that visual contrast of any kind can make particular

information stand out from the rest. Here are additional attributes that you can

use to emphasize particular information by means of contrast:

Attribute Tables and Text Graphs and Objects

Orientation • Italics • Data points with an

orientation that is different

from tlle norm

Shape • Any font that is different • Any symbol shape that is

from the norm different from the norm

Enclosure • Border around tables, rows, • Border around graphs or

columns, or particular particular values

values • Fill color in a graphs behind • Fill color behind tables, particular values

rows, columns, or particular

values

Hue • Almost any hue that is • Almost any hue that is

different from the norm different from the norm

2-D position . Any position that is out of • Any position that is out of

vertical or horizontal vertical or horizontal

alignment with the norm alignment with the norm

The final attribute of 2-D position highlights the significance of alignment in

visual design. We are more sen sitive to the vertical and horizontal alignment of

text and objects than you might imagine. The slightest misalignment jumps out

at us, and we react by trying to impose meaning on that difference. Unless you

intend to make something stand out, be careful to keep the edges of text and

objects aligned so tbat your readers' eyes can scan down and across without

disruption.

You may use differences in horizontal alignment quite consciously to estab­

lish a hierarchical relationship between different sections of content, with

subordinate content indented to the right of higher-level content. When you use

indentation in this manner, be sure to indent far enough to make your intention

clear. Using alignment in this manner makes it easy for your readers to sepa­

rately scan higher-level content without distraction from subordinate content

when they wish to take in the main points quickly.

There is actually one more method that we haven't considered yet because it

doesn't involve a visual attribute of objects but is instead a special type of object

that's used for a particular purpose. I'm referring to a collection of objects called

pointers. This includes objects like arrows, asterisks, and check marks. Put one of

these next to or pointing to any content, and your reader's attention will

definitely be drawn to it. Pointers are not subtle, especially arrows, so you should

use them with ruscretion to avoid visual clutter.

148 SHOW ME THE NUMBERS

Sequence

The final objective of visual design is to provide clear direction to your readers

regarding the best sequence in which to read a report's contents. The strongest

sequencing method is the location of content in 2-D space. Because we read

from left to right and top to bottom, this is generally the order in which your

readers will scan the page or screen. If you clearly divide the contents into

columns, such as those in newspapers, readers will first scan the left-most

column from top-to-bottom, then move to the top of the next column, unless

you've done something to draw their attention elsewhere.

The strength of this left-to -right and top-to-bottom navigational sequence is

greatest with textual content because text can only be perceived through the

sequential process of reading. This same sequence works for graphs as well but

not as strongly. For instance, if your page contains a collection of graphs without

sections of text to introduce them, readers will still g.ive attention to each graph

in the normal left-to-right, top-to-bottom sequence, all else being equal.

However, if any one of the graphs has been highlighted as important using any

of the prioritizing methods noted in the previous section, readers' eyes will

likely be drawn immediately to that graph. If your message requires that your

readers work their way through a collection of tables and/or graphs in a particu­

lar order, you can further reinforce the navigational route by using numbers (1,

2, 3, etc.), alphabetical letters (A, B, C, etc.), or some other form of sequential

labellng.

Integrate Tables, Graphs, and Text

Tables, graphs, and text form a powerful team, working together intimately to

communicate quantitative information. Each brings a different set of strengths

to the effort. We've already examined the separate strengths of tables and

graphs. In this section, we'll focus on the contribution of text and the way it can

be integrated with tables and graphs to create clear and powerful messages.

The Role of Text

To complement or enhance tables and graphs, text can:

• Label

• Introduce

• Explain

• Reinforce

• Highlight

• Sequence

• Recommend

• Inquire

Let's take a quick look at each role.

LABEL

GENERAL DESIGN FOR COMMUNICATION 149

We've already examined the role of labeling. Tables and graphs both use text to

label information. Tables use text (i.e., words and numbers) not only to express

quantitative and categorical data but also to label what the columns and rows

contain. Graphs incorporate text in the form of titles, labels for categorical items

and quantitative values along the scale lines, and legends to interpret the visual

encoding of categorical items (e.g., the orange bars represent the eastern sales

region). Text in the form of labels supplies critical information to enable readers

to interpret tables and graphs.

Reports containing tables and graphs also use text in the form of titles. Clear

titles are vital data in themselves. How many times have you seen a report with

a title that revealed nothing definitive about its contents? When people scan

lists of available reports in an effort to find one that contains the information

they need, they often do this with no information other than the titles. Good

titles are invaluable.

INTRODUCE

Quantitative displays often require an introduction to set the reader on a clear

path to understanding. Text is the ideal medium for providing introductions.

Introductions are especially useful in new reports and for new readers of old

reports, po tentially saving readers a great deal of time and frustration. Among

other things, an introduction should preview what readers will find in the

report, what they should especially notice, and what they should do with the

information. Because you can't always hand a report directly to all its eventual

readers, the introduction is your chance to set the stage for the report using text

that states what you would tell them in person if you could.

EXPLAIN

An introduction to a report is not the best place to put every bit of text that

might be needed to explain the data the report contains. Explanations work

most effectively when they are provided right where they're needed to clarif y

something about the message. I f you provide a time-series graph that displays a n

unusual brief up-tick i n donations during the month o f May, you may want to

mention right there, in or just underneath the graph, that a successful promo­

tional campaign beginning in late April was responsible for the anomaly. If a

few words are what it takes to make the message clear, then they belong there.

Whenever a table or graph doesn't speak clearly enough on its own, its design

should be improved to solve the problem, if possible, or a little text should be

added.

REINFORCE

Some information is so important that you should say it more than once and in

more than one way to increase its likelihood of getting through to your readers.

If you encode that information visually in a graph or verbally in the columns

and rows of a table, and then present it again in a few well-chosen words, you

will increase the odds that the message will be heard. You don't want to overdo

150 SHOW ME THE NUMBERS

it though. Don't say everything in multiple ways or you'll waste your readers'

time and lose their confidence. The important stuff, however, deserves a little

extra.

HIGHLIGHT

We've examined several methods for visuaily highlighting important data.

Sometimes it's also useful to highlight particular information by referencing it

with words as well. This is different from reinforcement because in this case

you're not repeating the information in a different form; you're simply calling

the reader's attention to it. For instance, if the sales ran king of a particular

product warrants special notice, you may say so in words right in or underneath

the graph. Perhaps it isn't appropriate to make that product stand out above the

rest visually in the graph itself because that would distract from the other

products that are also important, but a short note following the graph could do

the job without creating a visual distraction in the graph.

SEQUENCE

Sometimes it's challenging to use visual methods alone to clearly reveal the

order in which your readers should examine the contents of a report. Perhaps

information in a report cannot be positioned from left to right and top to

bottom in the order it should be read because of a greater need to use 2-D

location to highlight the importance of some data or to group data in a particu­

lar way. In circumstances such as these, you can use text to instruct your readers

to navigate through the contents in a particular way.

RECOMMEND

As a communicator of important quantitative information, your job often

involves more than simply informing. Sometimes it's your responsibility to

recommend what could or should be done. Recommendations for action are best

communicated in words. Whether or not making recommendations "is your

explicit role, your organization might appreciate it if you take the ini.tiative to

offer recommendations that you deem warranted.

INQUIRE

Inquiry is vastly underrated and too often ignored. Quantitative information

frequently invites questions that ought to be asked. You can sometimes add

more value to your organization by asking a single important question than by

prov.iding hundreds of answers. We so often get caught up in business as usual

that we fail to question why things are as they are or whether things might be

better if they were different. As a communicator of quantitative information,

you're in a great position to recognize opportunities for further exploration,

important speculation, and valuable questions that somebody ought to be

considering. Why not ask such questions by placing a few words in your reports

near the information that prompted the questions? I realize that your readers

might respond by assigning you the task of exploring those questions, but who

better? lf you're like me, discovering the right questions to ask and then doing

the research and analysis to find the right answers is the real fun of working

with nwnbers.

Text Placement

GENERAL DESIGN FOR COMMUNICATION 151

Tables, graphs, and text are complementary. There is no need to arbitrarily p.lace

them in separate areas in a report. Blend them together, placing each unit of

content precisely where it is most needed. Just be careful, when you place text in

the plot area of a graph, to do so in a way that does not obscure or distract from

patterns in the data.

The importance of minimizing distraction was suggested indirectly in the

chapter about visual perception. Our eyes have a limited area on which they can

focus at any one time. lf you place the legend for a graph too far from the data it

labels, you force your readers to jump back and forth over and over to read the

graph because they can't keep all the encodings (e.g., the blue line represents

widgets) in working memory. If you place the explanation for a table that

appears on page l at the end of the report on page 10, you'll cause unnecessary

effort and frustration. Perhaps your message involves a great deal of text spread

across several pages, which refers to a single table or graph. In this case, you

might actually want to reproduce the table or graph in multiple locations so that

it's always available where it's needed.

You might have noticed that in this book I don't follow the traditional

practice of placing notes and references at tJ1e bottom of the page, the end of the

chapter, or worst of all, the end of the book. I also don't force you to turn to a

middle section of illustrations but instead have integrated all illustrations right

where they're needed. This was a conscious design choice to support your

reading experience. You face similar design choices regarding the integration of

tables, graphs, and text whenever you construct a quantitative message. The

tighter the integration, the better.

Required Text

Text should be included on every page of every report to answer the following

questions:

• What?

• When?

• Who?

• Where?

Excerpts from multi-page reports are often copied and distributed. If the infor­

mation that identifies the report only appears at the beginning, readers who

have only a portion of the report will have no way of knowing where it came

from. It takes only a minute to include this identifying information in the page

header or footer of your reports.

WHAT

As l said before, a good title is invaluable. A simple glance at the title should

clearly tell your readers what the report contains. The title should describe,

without being long winded, the type of quantitative and categorical information

that the report presents. The title "Sales" isn't enough. How is this report

152 SHOW ME THE NUMBERS

different from all the other reports that deal with sales? A title like "2011

Bookings by Month and Region" says a great deal more.

WHEN

Two facts should be provided with every report to inform your readers about its

relation to time:

• The range of dates the information represents

• The point in time when the information was collected

Does the informati.on represent a single hour, day, week, month, quarter, year?

Does it represent some range of hours, days, etc.? Perhaps it represents an odd

range of time, such as from April 23'd of 2010 through January 14th of 201 l .

Whatever the range, .if it isn't clearly labeled in the table or graph, then make

sure it appears in the title or subtitle.

The point in time when the information was collected is often called the "as

of" date. "This represents expenses for February as of March 4th ." This informa­

tion is important because more expenses could be recorded later, or corrections

could stiJI be made to expenses after March 4th . Multiple reports ,covering the

same period of time often differ simply because they were produced at different

times, and the data changed in between them. A simple "as of" followed by the

date when the information was collected, noted in the header or footer of the

report, conveniently satisfies this need.

WHO

The reason to include your name or the name of the group you represent on

yom reports is not self-promotion; it is to let people know whom to contact if

they have questions. I've spent many frustrating hours during the course of my

career trying to track down the creator of a report because I needed to ask a

simple question about it. Save your readers this annoyance. Provide your name,

along with some means to contact you, such as an email address or phone

number.

WHERE

By "where'' I am referring to page numbers, which tell your readers where they

are in a multi-page report. Try describing to someone where he or she can find a

particular piece of information in a multi-page report that doesn't include page

numbers. I find that the format "Page# of##" (e.g., "Page 13 of 197") is best

because it informs your readers from the very first page how many pages they're

facing in total. This is especially helpful when reports are distributed and read

electronically because there is no physical stack of pages to alert readers to the

size of the report. Have you ever started to print an electronic report only to

realize later when you saw the line of angry coworkers at the printer that it was

more than a thousand pages long?

Summary at a Glance

GENERAL DESIGN FOR COMMUNICATION 153

General Design Objectives of Quantitative Communication

HIGHLIGHT

• Reduce the non-data ink

• Subtract unnecessary non-data ink

• De-emphasize and regularize the

remaining non-data ink

• Enhance the data ink

• Subtract unnecessary data ink

• Emphasize the remaining data ink

Highlight What's Important

ORGANIZE

• Group

• Prioritize

• Sequence

USING QUANTITATIVELY PERCEIVED VISUAL ATTRIBUTES

Attribute

Width

Size

Color intensity

2-D position

Tables and Text

• Bold faced text

• Bigger tables

• Larger fonts

• Brighter, more vivid colors

• Positioned at the top, l.eft, or

center

Graphs and Objects

• Thicker graph lines

• Wider bars

• Bigger graphs

• Bigger symbol shapes

• Brighter, more vivid colors

• Positioned at the top, left, or

center

USING VISUAL ATTRIBUTES IN CONTRAST TO THE NORM

Attribute Tables and Text Graphs and Objects

Orientation • italics • Data points with an

orientation that is different

from the norm

Shape • Any font that is different • Any symbol shape that is.

from the norm different from the norm

Enclosure . Border around or shading . Border around or shading

behind table, rows, columns, behind graph or particular

or particular values values

Hue • Almost any hue that is • Almost any hue that is

different from the norm different from the norm

2-D position . Any position that is out of . Any position that is out of

vertical or horizontal vertical or horizontal

alignment with the norm alignment with the norm

154 SHOW ME. THE NUMBERS

Sequence Information

• Using left-to-right, top-to-bottom positioning

• Using visual highlighting

• Using sequential labels (e.g., 1, 2, 3 . .. )

Include on Every Page

What it is In the form of a good title

When it is In the form of the range of dates and an "as of" date

Who produced it So readers know whom to contact

Where readers are In the form of page numbers

9 GENERAL GRAPH DESIGN

The visual nature of graphs requires a number of unique design practices. The

volume and complexity of quantitative information that you can communicate with a

single graph are astounding but only if you recognize and avoid poor design

practices that would undermine your story.

Because of their visual nature, graphs tap into the incredible power of visual

perception to communicate quantitative information. When the story that you

wish to tell is contained in the data's patterns, trends, and exceptions; or when

it depends on your audience's ability to compare entire series of values to one

another (e.g., monthly domestic sales for the entire year compared to interna­

tional sales), a graph will do the job best, but only if you avoid far-too-common

design pitfalls.

We've already covered the aspects of quantitative communication that apply

to both tables and graphs. None is more important to the design of graphs than

the fundamental principle that was stated so eloquently by Edward Tufte:

"Above all else show the data."1 Quantitative stories reside in the facts, not in the

containers that we use to present them. The general practice of highlighting the

data and subduing all else is even more important in the design of graphs than

in the design of tables. Tables are a bit more forgiving of visual design flaws

because tables encode data through the use of verbal language (i.e., text),

visually displayed. Graphs, in contrast, encode data as visual objects. These

objects must be prominent, accurate, and clear.

Two fundamental principles of quantitative communication apply exclusively

to graphs:

• Maintain visual correspondence to quantity.

• Avoid 3D.

Both principles are firmly rooted in practical concerns; you can wreak havoc on

communication if you ignore these principles.

Maintain Visual Correspondence to Quantity

You can only use two attributes of visual perceptjon to encode quantitative

information in a way that can be easily and accurately interpreted: length and

2-D position. Quantitative values in graphs are either encoded visually as length

in the form of bars or boxes or as 2-D position in the form of points and lines.

Other visual attributes are either not perceived quantitatively at all (e.g., hue) or

not well enough (e.g., 2-D area and color intensity) to justify their use for

quantitative encoding when length and 2-D position are available.

A bar that is twice as long as another is perceived as having twice the quanti­

tative value. Visual objects that encode quantitative values in graphs are inter­

preted by means of a scale line along the vertical or horizontal axis. When a bar

1. Edward R. Tufte (2001) The Visual Display of Quantitative Information,

Second Edition. Graphics Press, page 92.

192 SHOW ME THE NUMBERS

that is twice as long as another corresponds to a value of two on the scale line,

vi5ual perception alone tells us that the value of the shorter bar is one, or very

close to it. If the shorter bar actually corresponds to a value of 1.75 or 0.5,

something is amiss.

People sometimes intentionally manipulate graphs to mask the truth con­

tained in numbers. Darrell Huff, in his 1954 classic How to Lie with Statistics/

was one of the earliest to express this concern. Advertisements are notorious

sources of deliberately misleading graphs, but deception is not confined to

advertising. You'll be faced many times with the temptation to manipulate

graphs to give your case more strength than it deserves based on the actual

numbers. Given the understanding of visual design that you are developing by

reading this book, you will be even better equipped to manipulate visual design

to exaggerate or hide the truth. It's easy to rationalize little design manipula­

tions here and there to shade the truth slightly for a just cause. Be aware,

though, that this manipulation does not qualify as design for communication.

The goal of design for communication is always to promote an accurate under­

standing of the truth.

Here's a simple illustration of the potential for deliberate misinformation:

Sales are Skyrocketing!

Jul Aug Sep Oct Nov Dec

2011

$20,000,000

$15,000,000

$10,000.000

S5,000,000

$0

Jul Aug

Sales are Flat

Sep Oct

2011

Nov Dec

The graph on the left has been deliberately manipulated to make an increase in

sales from $19,500,000 in July to $19,560,000 in December, which is an increase

of less than one-third of 1%, look like an increase of more than 200%. The

graph on the right more accurately presents the data. Do you see the specific

aspects of the graph on the left that were used to exaggerate the increase in

sales? Take a moment to see how many you can find, and list them in the

margin to the right.

Five design characteristics of the graph on the left give the false impression that

sales have risen dramatically from July to December:

1. The scale on the Y axis does not start at zero. Rather, it starts at

$19,475,000 and extends only to $19,560,000, thus making minor changes

in sales appear extreme.

2. Darrell Huff (19S4) How to Lie w,th

5tatistic5. W. W. Norton & Company.

FIGURE 9.1 These two graphs

display the same information in

dramatically different ways,

producing radically different

messages.

2. The plot area of the graph is taller than it is wide. This dramatically

increases the slope of the line.

::3. The line is green. The color green carries the meaning of growth and

health in English-speaking cultures and dollars in the United States, so it

reinforces the positive spin of the message. Also, placing a bright green

line on a black background makes it pop with visual impact.

4. The highest value-the final value of $19,560,001-is set as the top of the

scale. This gives the green line the appearance of extending right off the

top of the graph.

5. Placi.ng the boldfaced axis label Millions in the prominent upper left

position near the title "Sales are Skyrocketing" suggests that they are

increasing by millions.

This design certainly exaggerates the good news about sales, but I've seen

much worse. Can you think of any additional design changes that could be

made to further hide the truth?

Here's one that I've seen:

Sales are Skyrocketing! $19,560,001

Jul Aug Sep Oct Nov Dec

2011

Notice the changes? Values along the Y axis have been removed, and only the

final data point has been labeled. Without at least one more value on the scale,

there is absolutely no way to know the extent of the increase. The single vaJue of

$19,560,001, combined with the characteristics we've already discussed, together

suggest a huge rate of increase. By making the graph 3D and manipulating the

angle, r could exaggerate the increase even more, which is done all the time.

Now, back to the principle that prompted our journey through the dark alleys

of visual obfuscation. A quantity that is visually encoded in a graph should

match the actual quantity that it represents. Two specific design practices will

help you honor this correspondence:

• Make the distance between tick marks on a scale line correspond to the

differences in the values that they represent.

• Generally include the value zero in your quantitative scale, and alert your

readers when you don't unless you're confident that they won't be misled.

GENERAL GRAPH DESIGN 193

FIGURE 9.2 This is an extreme

example of intentional deceit

through graph design.

l 94 S H OW M E T H E N U M B E R S

Correspondence to the Tick Marks

You should always keep the distance between tick marks on a scale line consis­

tent with the difference in the quantitative values that they represent. Software

that generates graphs for you based on speci fiect sets of values automatically

enforces this practice. If the tick marks represent the values 1, 2, 3, 4, and S, they

will be positioned an equal distance from one another. If you ever produce

graphs without the aid of graphing software, you should be sure to honor this

practice. Approaching this from the opposite perspective, if you have a set of

tick marks that are positioned at equal distances from one another, the values

that you use to label them should also represent equal numeric intervals. Never

place a gap in the values, such as in consecutive tick marks labeled as 1, 2, 7, 8,

and 9, even if there are no values in the graph that fall within the missing range.

To do so would undermine the graph's visual integrity.

Even if you indicate a break in the quantitative scale where a section of values

has been eliminated, your readers could still be easily misled.

Influenza Cases

1 0.000

j 8,000

400 1

,oo 1 200

100

0

2007 2008 2009 2010 2011

Despite the fact that the scale starts at zero, the increase in influenza cases from

2010 to 201 1 is underrepresented to a huge degree because of the scale break

between 400 and 8,000 along the Y axis. Here's how the same values appear

with a proper scale.

Influenza Cases

1 0,000

8.000

6,000

4,000

2,000

0

2007 2008 2009 2010 2011

You may recognize that these lick

marks would not be equidistant if

you were using something other

than a standard scale, such as a

logarithmic scale. We'll look at the

special qualities and uses of

logarithmic scales a little lc1ter.

FIGURE 9.3 Scale breaks can be

misleading.

FIGURE 9.4 With a proper scale, the

dramatic rise from 2010 to 2011 in

influenza cases is striking.

Something is missing in this graph, however, that you might want your readers

to see: the pattern of changes that occurred from 2007 through 2010. To

accommodate the high number of cases in 2011, tile graph's scale now forces all

other values into a tiny space near the bottom, which makes the line appear

almost flat during that period. How can we show the earlier pattern of change

and yet still tell the more important story that the number of cases dramatically

increased from 2010 to 20ll? Can you think of a solution?

To tell this entire story, two graphs are needed, such as the following.

Influenza Cases 10,000

8,000

6,000

4,000

2,000

0

2007 2008 2009 2010 2011

From 2007-2010 Only in Greater Detail 450

400

350

300

250

200

2007 2008 2009 2010

It's important to know that quantitative stories can often only be told with more

than one graph. Nothing is gained by attempting to squeeze into a si.ngle graph

what can be more effectively presented in several.

Zero-Based Scales

When you set the bottom of your quantitative scale to a value greater than zero,

differences in values will be exaggerated visually in the graph. Usually, you

should avoid starting your graph with a value greater than zero, but when you

need to provide a close look at small differences between large values, it's

GENERAL GRAPH DESIGN 195

FIGURE 9.5 Two graphs are needed

to tell all aspects of this story dearly.

196 SHOW ME THE NU M 8 ER S

appropriate to do so. When you do so, alert your readers to the fact if you have

any doubt that they'll notice. Perhaps you observed that the scale in the lower

graph in Figure 9.5 doesn't start at zero. Because the same information was

already shown using a zero-based scale in the upper graph, the fact that the

scale was adjusted in the lower graph wouid be hard to miss. If the sales manager

of a company with the subtly ri.sing sales that we examined in Figures 9.1 and 9.2

wanted to examine that increase in great detail, however insignificant it might

be as a percentage increase, the following graph would make this possible, but

textual alerts similar to those shown in red might be needed.

U.S.$

19, 580,000 ]

19,560,000

19,540,000

19,520, 000

19,500,000 �

19.480,000

Subtle Rise in Sales in Recent Months

Jul Aug Sep Oct Nov Dec 2011

Atte ntion: The dollar scale along the vertical axis has been narrowed to reveal the small but steady rise in sale s since July

Never eliminate zero from the quantitative scale when bars are used to encode

the values, however. Why? Because a bar encodes quantitative value primarily

through its length, and, without zero as the base, the length will not correspond

to its value. In the following software ad, whlch I clipped from a magazine, how

much greater is customer loyalty to MicroStrategy than Cognos Powerplay?

1 MicroStrategy

7 Appl,x TMi

, SAP BW

,1 M crosol1 AS

� MISAl<.><l

6 o, "' Iii' OLAP Servers

7 811,,1 ss Ob1eCT�

8 Hype11011 Essbase

q Or ade Discoverer

• 0 Cog nos PowerP:ay

60¾ 70% 80% 90%

The MicroStrategy bar appears to be more than six times greater in Length than

the Cognos PowerPlay bar. The difference between the values, howe ver, is about

83% versus 63%-quite a different story. Here's the same information, properly

displayed:

FIGURE 9.6 This is an example of an exception to the zero-based scale, illustrating how such an exception can be clearly n oted to prevent misunderstanding.

FIGURE 9. 7 This graph misrepre­ sents the values by starting the scale at 60 %.

0% 10% 20% 30% 40% 50% 60%. 70% 80% 90%

MicroStrategy

Applix TM1

SAPBW

Microsoft AS

MIS Alea

Oracle OLAP Servers

Business Objects

Hyperion Essbase

Oracle Discoverer

Cognos PowerPlay

0% 10% 20% 30% 40% 50% 60% 70% 80% 90%

When a graph represents both positive and negative numbers, zero will not

mark the bottom of the scale, but it will still represent the base from which all

values extend. The following two graphs contain the same set of positive and

negative values. The graph on the right correctly displays zero as the base of its

scale from which bars extend upwards for positive values and downwards for

negative values, but the one on the left mistakenly sets the base to slightly below

the lowest value, resulting in a confusing and misleading representation of the

values.

$10,000

$8.000

$6.000

$4.000

$2,000

so

(S2.000)

(S4.000) Carl Dan John Nancy Patty Terri

Avoid 3D

$10.000

$8.000

$6.000

J _lJ

$4,000

$2,000

(S2,0::l -1- --

(S4.000) Carl Dan John Nancy PaHy Tern

When 3D is used in graphs, it takes one of two possible forms:

• The addition of a third dimension of depth to objects (e.g., bars) that are

used to encode quantitative values, without the addition of a third

quantitative scale.

• The addition of a third dimension of depth to the overall graph with an

associated quantitative scale (the Z axis).

Neither form is effective, but the reasons are entirely different.

GENERAL GRAPH DESIGN 197

FIGURE 9.8 This graph displays the

values in Figure 9. 7 properly.

FIGURE 9.9 Both of these graphs

display both positive and negative

numbers. The graph on the right

correctly sets zero as the base of its

scale at the point w here the X axis

intersects the Y axis. The graph on

the left incorrectly sets the base to

slightly below the lowest value.

198 SHOW ME THE NUMBERS

Data Objects with 3-D Depth

We're usi11g four objects to encode quantitative values in graphs: points, bars,

lines, and boxes. The addition of depth to a value-encoding object does not

affect the object's value. Add depth to a series of bars, and what do you have?

Nothing more than bars that now occupy more space and are harder to tie to

values along the scale. If you add depth to value-encoding points, li:ke dots and

squares, you get spheres and cubes that represent the same values as before, but

now their depth makes it harder to align them accurately with the scale. 3-D

versions of lines look like thick ribbons and suffer from the same problems.

Here are four variations of the same graph, three of which have 3-D effects

added to the bars:

$1.000.000 $1,000.000

$800,000 $800.000

$600, 000 5600. 000 I

$400, 000 S400.000

S200.000 j $200,000

$0 $0 01 02 03 Q.4 01 02 03 04

$1,000,000 1 $1,000,000

$800,000 $800,000

$600,000 S600.000

S400, 000 $400,000

$200.000 $200.000

so $0

04 01 02

Which graph is easiest to read? When shown all of these at once, the answer is

obvious, isn't it? I was careful in the graphs above to keep the 3-D effects simple.

If we take advantage of the many options that most soft waxe provides, we can

bury truth in visual effects. In the following example, I've manipulated perspec­

tive anq angles to make a steady increase in expenses from $100,000 to $121,000

look like a flat series of consistent values.

FIGURE 9.10 These four examples

display the same values using bars in

four different ways, three of which

incorporate 3D.

S 125.000

SI00,000

S7MOO

S�.000

S25 000

so

Jan

2011 Expenses Are Holding Steady

Most software makes it far too easy and tempting to add a third dimension to

objects in graphs. This functionality is thrown in because people expect it, not

because it's useful. It is far better to impress your readers with graphs they can

easily understand and use than graphs that look like video games and are

difficult to interpret.

Remember the data-ink ratio. The addition of 3D to value-encoding objects

adds ink but not data. That is, it adds meaningless visual content that your

readers must take in and process, resulting in nothing but wasted time and

effort.

Graphs with 3-D Depth

A third dimension of depth may be added to an entire graph through the use of

a third axis, conventionally called the Z axis. The Z axis may be used for eHher a

categorical or a quantitative scale. A categorical scale along the Z axis allows you

to add another set of categorical items that extend back along the axis, accompa­

nied by related rows of quantitative values. A quantitative scale along the third

axis can display a third quantitative variable in a scatter plot. In theory, this is a

valid way to include more information in a graph. In practice, with rare excep­

tions, it is simply too hard to read. Simulating 3-D space on a 2-0 surface works

nicely for paintings or technical illustrations but almost never for graphs.

A few examples will vividly illustrate this point. Let's start with the same data

that we examined earlier as the dark gray bars in Figure 9. 10.

GENERAL GRAPH DESIGN 199

FIGURE 9.11 3-D effects are

sometirries used to tell lies.

200 SHOW ME THE NUMBERS

$1,000.000

S800.000

S600.000

$400,000

$200,000

$0 Q1 Q2 Q3 Q4

So far we have a very simple 2-D graph. Now let's say that we want to display

these quarterly bookings by the four sales regions of North, East, South, and

West. To do so, we could encode each region as a different hue and keep the

graph 2D, as follows:

$350,000

$300,000

$250,000

$200,000

$150,000

$100.000

$50.000

so 01

North East South West

02 03 04

This is still fairly easy to read. Rather than using hue to encode the four sales

channels, we could instead add a Z axis to the graph, making it 3D, a11d display

the sales channels along that axis .

$350,000

$300,000

S250.000

$200,000

$150,000

$100.000

$50,000

$0 '

.-

02 03

04

� West South

East

North

FIGURE 9.12 This is an example of a simple 2-D graph.

FIGURE 9.13 This 2-D graph has two sets of categorical items: quarters along the X axis and sales regions encoded as different hues.

FIGURE 9.14 This is a 3-D graph, with sales in dollars along one axis, quarters along another, and sales regions along a third.

This is a very simple example of a 3-D graph with two categorical scales (quar­

ters and regions) on one quantitative scale (dollars). \Vhat do you think? Does it

work? Examine it for a moment, attempting to read and compare its values. Try

to describe what makes th is graph dHficult to read.

When a third dimension is added to a graph, adjustments are usually made to

the graph automatically by software-tilting, rotating, and adding perspective­

to make its data more visible. A 3-D display like this is called an nxo110111etric

projection. The previous example was tilted down 15 degrees, rotated clockwise

20 degrees, and given 30 degrees of perspective. These variables can be altered in

an effort to make the graph easier to read. Even though the graph has been tilted

and rotated in an attempt to make the rows of bars more visible, some bars will

always remain partially or entirely hidden. Also, it's nearly impossible to line the

bars up with values along the quantitative scale.

Software that generates 3-D graphs often includes grid lines on the walls in

an effort to make the quantitative values easier to align with the scales lines.

Here's the same graph as before with the addition of these features, along with

black borders around the bars to more clearly delineate them.

$350,000

$300,000

$250,000

S200,000

$150,000

$100,000

$50,000

$Ci

Even though this is a fairly simrle graph, these enhancements still don't solve

the problems. Software vendors some times argue that this problem can b e

solved b y rotating the graph to see bars that are hidden. The problem with this

approach is that any new perspective will reveal some bars and hide others,

never allowing us to see and compare all the values at once, which is a key

benefit of graphs. Changing from the use of bars to lines to encode the data

doesn't fix the problem either, as you can see in the following example:

GENERAL GRAPH DESIGN 201

FIGURE 9.15 This is a 3-D graph that

has been enhanced in an effort to

make the values easier to read

through the use of grid lines on the

walls and borders around the bars.

202 S H OW M E TH E NU M B ER S

$350,000

$300,000

$250,000

$200,000

$150,000

$100,000

$50,000

$0

03 04

West South

East

North

Which of the lines in this graph represents the south region? When [ ask this

question in classes, fewer than 50% of my students answer correctly. The lowest

of the four lines represents the south region, but this isn't at an obvious, is if!

Support components called drop lines were invented to help us locate data objects

in relation to scales along axes, especially in 3-D graphs, but they clutter the

graph and reduce its interpretation to a slow series of look-ups.

$350,000

$300,000

$250,000

$200,000

$150,000

$100,000

$50,000

$0

03 04

East

North

So far we've only examined the association of a categorical scale with the

third axis. The problems don't get any better when the third axis is used for a

quantitative scale. Imagine a scatter plot that correlates employee salaries in

dollars along one axis, tenure on the job in years along another axis, and level of

education in years along the final axis. It's too difficult to tell where the points

are positioned along the third axis.

3-D renderings of quantitative information rarely work. Don't sacrifice

effective communication through the use of 3-D fluff. Even when you are driven

by a sincere desire to give your readers more information by using a third

dimension, there are better ways to realize these good intentions. One effective

technique is to use multiple related graphs in a series, which we'll explore in

Chapter 11, Displaying Many Variables at Once.

FIGURE 9.16 This graph displays the

same data as Figure 9. 15 but this time

using lines to encode the values.

FIGURE 9.17 This graph is the same

as the one in Figure 9.16 with the

addition of drop lines.

Summary at a Glance

• Encode quantities to correspond accurately to the visual scale.

Keep the distance between tick marks on a scale line consistent with the

difference in the quantitative values that they represent.

• In most cases include the value zero in your quantitative scale, and alert

your readers when you don't. Always start the quantitative scale at zero

when you use bars to encode the values.

• Avoid 3-D displays of quantitative data.

G EN ER AL G RAP H D ES I G N 203