Introduction to Data Visualization & Storytelling

A Guide For The Data Scientist

Jose Berengueres
with
Marybeth Sandell &
Ali Fenwick
 2nd Edition
Type setting by SNASK
Text Copyright © Jose Berengueres
Where otherwise stated, artworks,
cover & drawings © by Jose Berengueres
Data, value creation & thinking modes
 TOC
TOC
Preface
How to use this book

Chapter 1. Stories, Data & Narratives

What is data?
Data vs. wisdom
Match the words
Information vs. meaning
Narratives & stories
Elements of effective visualization
Quiz: Narratives & stories
Quiz: Storytelling climate change

Chapter 2. Visualizing Information

Visualizing gender
The Chart-narrative fit
Human-centered charts
Gravity & charts
Chart wars: Pies vs. Bars
Quiz: Making useful charts

Chapter 3. Knowledge
Mental frameworks
Visualizing inclusion
Winner takes all
The BRICS framework
Quiz: Visual summaries
Chapter 4. Charts to Think
Use rankings to create situational awareness
The design space
Forecasting with mean-reversion
The design space in business
The Gap Matrix
The Innovation Matrix
Wardley maps
Wheel of Life
Interactive exploration
Quiz: Supporting decisions
Chapter 5. Making your chart pop
Use arrows to unbound your thinking
Decluttering charts
Use personas to win over the audience
Visualizing big differences
Putting the world’s money into perspective
How many Solar Panels are needed to power the USA?
Storytelling age-bias with humor
Use the Golden ratio everywhere
Twists
Quiz: Ice baby
Quiz: Global warming
Quiz: Magic quadrants
Quiz: Visualizing gaps

Chapter 6 Psychology of Visualization

Context
Framing
Gestalt
The Peak–End rule

Chapter 7. Detecting Bias

Bias in narrative: A balanced meal?
Bias in the story: Hard working Germans?
Bias in data: Austerity controversy
Other biases
The Amazon forest fires case
Due diligence checklist
Quiz: Fire Tweets

Acknowledgement
References & Footnotes
 Preface
It was May 2018 when I received an email from Kaggle.com, a data
science community where people all over the world compete in data
modelling challenges. Every year since 2017, Kaggle has surveyed its users
and this year they decided to organize a storytelling competition and offer
cash prizes for the best survey visualization. I was curious to see what
people had submitted so far and sifted through the entries. I was impressed
at how much time and effort had been put into analyzing the data. Some
charts required laborious data wrangling, others crafty SQL inner joins, and
some Python sorcery. And yet, the charts were not doing justice to the
compelling findings and the stories that could be told. No doubt, some were
creative, extensive in length, exhaustive in the exploration. However, there
were also many unimpressive charts. Why was so much IQ not producing
more enticing visualizations?
It’s about the awareness
80% of the data scientists graduating today will do so without having
received any formal education on storytelling[1], and 60% of them place data
visualization at the bottom of their priority list of skills to have. In addition,
the fact that most data science is taught at STEM schools, not Art schools,
does not help either. (Really, STEM schools should aspire to be STEAM
schools, to include Art). In order to improve the quality of data
visualization, there must be a change in how data scientists are trained
along with a mind shift regarding the imperative of good story telling.
However, to make effective visualizations, art sensibility is not all that is
needed.
Death by default settings
A second cause, is that the gallery of default chart styles of Microsoft
Excel is unlikely to match what is required to tell the story[2] and is
laboriously hard to adjust. For example, a single bar chart in Excel has a
whopping +129 configurable options: color fill options, axis options, line
thickness, line types, legend positioning options, scale options, bar type,
spacing options. Each option has between 2 and 10 possible values. This is
a 10 followed by 38 zeroes. Because of the tricky laws of aesthetics[3], very
few of these combinations will produce the Aha! chart that will dazzle the
audience, and even if it could, it is far too laborious for most people. To put
it into perspective, there are more possible combinations than stars in the
Milky Way and if we spent just 3 seconds tweaking each of the 129
combinations it would take 6 minutes and 578 mouse clicks to adjust them
all. It does not help that most visualization software in use today (this
applies to ggplot too) was developed by CS graduates with little training on
the basics of color theory, information design or visual communication. The
exception was Steve Jobs, who took calligraphy lessons and credited that
for the Mac’s beautiful fonts. See Walter Isaacson’s bio on Jobs as a
flashback on how much the field has improved[4].
Start with why
A third cause, is that to produce a meaningful visualization we need to
know why we are visualizing. To get to the why, it’s important to explore
the story that needs to be told and be able to make it relevant.
To succeed at transforming data into a compelling and relevant story, it
helps to connect the data to a context, metaphor or mental framework
(frameworks from Economics, Art, and Sociology are particularly useful).
In order to make these connections, a cross-functional education is
necessary. Unfortunately, this is not the norm resulting in loads of very
interesting data failing to become useful knowledge. To address this, in the
following pages you will find a set of principles by example that I wish I
had learned in grad school.
 Happy visualizations!
Jose Berengueres
Stockholm, Oct 25th 2019
How to use this book
What will you learn
1. Identify the role of a narrative in a chart
2. Transform data into information
3. Synthesize knowledge
4. Apply visual thinking tools to the decision-making process
5. Select visual communication techniques to persuade

Updates to this edition

This edition adds two new chapters, one on bias by Marybeth Sandell
and one on Psychology by Ali Fenwick. Whether you are using this book as
course material for a data visualization course, in journalism, or in a data
science minor; this is a solid foundation before getting hands on with
Tableau, Excel or Python Pandas. We have also upgraded all the images to
600 DPI.
Exercises
There are several exercises in this book so it is helpful to have paper and
black fiber tip pen 0.5-0.7mm such as Pilot EF or a Pilot Sign Pen available.
Slides
This book was born out of workshop full color slides that you can use in
the classroom. To get them send an email to [email protected] with subject
“DATAVIZ101” and link to your verified purchase review on Amazon.
 Chapter 1. Stories, Data & Narratives
– Finding the purpose of the chart –

Fig 1 These three magazines publish some of the most influential charts in
the newsstand. None of them is made with Excel[5].

Before you start

Before visualizing, there is data. However, before working with data it is
important to understand what knowledge is. What is knowledge to you? In
this chapter, you will learn three things; (i) the difference between: data,
information and knowledge; (ii) What wisdom is; and (iii) What the process
to produce knowledge is.
What is data?
Exercise
What is data to you? We always open data visualization workshops with
this exercise.
 Fig 2 An exercise to understand what data is by relating it to knowledge.

Data has many definitions. However, to understand, there’s nothing like

doing an exercise. There are various ways to learn something. By listening,
by talking about it, by reading about it and, by doing it. Of those four, doing
has the highest recall rate. Can you order the four words (data knowledge
information wisdom)? Most students will come up with the following
ordering:

data > information > knowledge > wisdom

Once the order of the words has been agreed, we can discuss the
ordering criteria. Why did we order them the way we did? This is a great
conversation starter. To ground the conversation, it further helps to list the
attributes of the words at the extremes. What are the attributes that
distinguish data vs. wisdom?
Data vs. wisdom
Exercise

Fig 3 An exercise used to understand the arrow of value between data and
wisdom[6].
Solution

Fig 4 The arrow of value.

What is wisdom?
Data is many while wisdom is usually characterized by few. Data is
abundant, wisdom is scarce. Value is closely correlated with scarcity too.
This exercise is great to clarify the pervasive confusion between data,
information, knowledge and its relationship to value, scarcity and wisdom.
How is wisdom made?

Fig 5 This arrow represents a process.

Jackie Chan says in one of his films, “information is not knowledge, and
knowledge is not wisdom”. But what is wisdom? Is wisdom just knowledge
in context? Is wisdom meta knowledge? — knowledge about knowledge?
And more importantly, knowing in which situation to apply a given
knowledge? Even if the definition is not universal, what we are more
interested in here is how to transform knowledge into wisdom. Why?
Because it is a high added value activity and one of the reasons (if not the
only reason) why companies employ data scientists. One way to arrive at
wisdom is the Synthesis process — the dialectic combination of thesis and
antithesis into a higher truth.
Match the words
Exercise
Connect each keyword with its corresponding image. Time 1 minute.

Fig 6 An exercise used to understand what wisdom is. Lego image source:
LinkedIn, anonymous mem.
Solution

Fig 7 To solve this exercise, think outside of the box.

In this exercise, students must match the words to a Lego configuration.

Each Lego piece represents some piece of data. The Lego exercise is a great
way to clarify that wisdom (and to some extent knowledge) is not just data
arranged and visually presented. Let us zoom into this case. Imagine you
are the C.E.O. of Lego and this information has been presented to you. Can
you map the words to the pictures? Where does wisdom map to? In this
exercise wisdom is orphan to drive home the point that wisdom is
something more than presenting and arranging data. In other words,
wisdom is knowing where to apply knowledge. Example of wisdom: “We
do not have any pink Lego pieces in this set. Why is this? Are we blind to
some important customer segment? Are sales not optimal because of that?”
Note how valuable knowledge emerges when we connect our information
(there are no pink bricks) to other existing knowledge (gender studies).
Information vs. meaning

Fig 8 “There is no peace in more”.

Meaning vs. Information

A version of this chart was originally posted by Hugh McLeod, the
author of “Ignore Everybody”. When I first saw this chart, I felt
mesmerized by it. Then I tried to find flaws in the chart. But there are no
flaws. This chart is a masterpiece. I was just flummoxed at the simplicity of
it[7]. The y-axis is the Shannon Information Entropy, measured in number of
bits. The x-axis, is an ordered sequence (not continuous, and has no units).
It is an ordering of cases (visualization examples) that the author ranked in
descending order of meaning along the x-axis; for each element two dots on
the y-axis are plotted. Then the author drinks his own medicine and
simplifies the chart to reduce all non-essential information to deliver the
Aha moment: Less (info) is more (meaning). The info line is called
sometimes hockey stick. It is an exponential function. The meaning line fits
a downward arch. What this chart means is that it is not possible to have
meaning with information overload.

Fig 9 A scientific deconstruction of the McLeod’s chart.

Exercise

Summarize the previous 20 pages in 20 words or less. Time 3 minutes.

_______________________________________

Now, visualize it
Solution

Fig 10 A visual metaphor based on the DIKW model .

Wrapping-up
When delivering a workshop, this exercise is a great way to bring
attention to the point of summary vs. synthesis. A summary is a mere
reduction process whereas synthesis is demonstrating an understanding of
the subject by relating it to other subjects, ultimately adding value through
connective thinking.
In Fig 10, we summarized the relationship between data and wisdom by
way of the pyramid metaphor. Elements at the top of the pyramid are
valuable, scarce and hard to carry to the top because they work against the
force of gravity. This is a great example of a visual summary of the chapter
while also a good example of synthesis.
Now that we’ve learnt the difference between data and knowledge, and
how to transform knowledge into wisdom, let’s look at the role of narratives
in charts.
Narratives & stories
Exercise
Fill in the blank with a verb. Example: “Data _____ stories”. Time 3
minutes.

Fig 11 An exercise used to understand the difference between narratives

and stories.
Solution

Fig 12 The cloud template, use when logic is fuzzy[8].

Stories
To understand what a narrative is, we first need to understand what a
story is. A story is an account of events. We humans love stories. Why?
Telling and consuming stories is addictive. For example, listening to an
Aesop fable, reading a book, and watching a movie, all those release
oxytocin in the brain— the feel-good hormone. That is why people get
addicted to Netflix, Venezuelan soap operas and fiction books. Oral
storytelling is thought to be the earliest method for sharing narratives. From
an anthropological perspective, during most childhoods, narratives are used
to shape children on proper behavior and values. This is usually done
through tales.
Narratives
A narrative is a set of beliefs, values or worldview. Therefore, the chosen
narrative interprets the story (and consequently the underlying data /
reality). An example of a narrative popular in European culture is that kids,
especially young girls, should not trust strangers. A story that promotes that
narrative is the story Little Red Riding Hood, a tale from the 10th Century. In
fables, a narrative, is also made explicit at the end of the tale as in the moral
of the story. Another example of a narrative is FUD — Fear, Uncertainty
and Doubt. It is also known as a disinformation strategy[9] used to thwart
change to the status quo. It is said that IBM was one of the first companies
listed on the Dow Jones to use FUD openly. An exemplification was,
“Nobody gets fired for buying an IBM”.
Connection to Aristotle
Because the goal of a story is to persuade; narratives, stories and data are
related to the three modes of persuasion of Aristotle — The Ethos, the
Pathos and the Logos[10]. The narrative is related to the Ethos (to appeal to
the ethical values), the story is related to the Pathos (to appeal to the
emotions), and the data that supports the story is related to the Logos (to
appeal to logic).
Exercise
Identify the narrative, the story, the data and call to action in this photo.
Time 2 minutes.

Fig 13 A Brexit Bus, UK. Original photo by David Beeson.

Solution

Fig 14 Annotations by a student.

At first glance, The Brexit bus story is an example of a narrative where

an out-group steals from an in-group. But it is more than that. In fact, the
bus appeals to so many people because it is connected to four narratives: (i)
appeal to in-group tribal instincts to fight “outsiders”; (ii) doing for
common good is ethical; (iii) fighting unfairness is ethical too; and (iv)
maximizing the utility of a resource is common sense.
These narratives were brilliantly embodied in the Brexit bus. The
statistics and words printed on the bus were later found to be lies, but it did
not matter. This story checked all the boxes of the pro-Brexit voters and
mobilized them.
Is your narrative ethical?
Narratives can be evaluated from an ethical perspective. There are four
ethical frameworks[11] commonly used. Each one maximizes some different
ethical criteria. For each framework, the most ethical choice is the one that
maximizes a different policy.
The four frameworks
In the so-called Utilitarian framework, the most ethical choice is the one
that optimizes global happiness. In the Common good framework, the most
ethical choice is the one that optimizes the Well-being of society. In the
Fairness framework, the most ethical choice is the one that optimizes
equity. In the Virtues framework, the most ethical choice is the one that
aligns best with a set of predefined values.
Equity vs. equality
 Fig 15 Equity is not the same as equality. “Interaction Institute for Social
Change”. Source: Angus Maguire.

Examples of famous virtue frameworks

Liberté, égalité, fraternité[12]
The ten commandments
The Bushido
What are ethical dilemmas?
The so-called ethical dilemmas arise when conflict between the
approaches appears. In addition, dilemmas can also arise when different
regions in the brain (amygdala vs. frontal lobe) evaluate a situation with
opposed outcomes. (See Trolley dilemma[13]).
Choices & ethics
 To be ethical one needs to first consider at least two choices. Many
times, we end up with suboptimal choices simply because alternatives were
not even considered. Before disseminating a visual, always consider two
alternatives. Then apply an appropriate framework to rank the alternatives.
(If something goes wrong at least you can show you followed a process).
Elements of effective visualization
Exercise
Summarize the relation between story, narrative and data in a visual
titled Elements of effective visualization that uses a plane as a metaphor.
Fill in the gaps. Time 3 minutes.

Fig 16 Visual metaphors + annotations, so powerful.

(Solution on the next page)

Elements of effective visualization

Fig 17 Don’t let words like ‘narrative’ get in the way of a great story.

According to this visual, to persuade others you will need:

1. A why or a narrative (message)
2. A story (to help assimilate the message)
3. Data to give credibility to the story
4. A way to visualize the story (metaphors help)
The payload is the narrative (your message), and you are the pilot.
Quiz: Narratives & stories
True or False? Time 5 minutes.
1. A story is an account of events. [ True / False]
2. Story consumption releases oxytocin. [T/F]
3. The purpose of a story is to advocate (transmit, perpetuate) a
narrative (a belief, a culture, an ideology). [T/F]
4. A narrative does not stick to the human brain, it cannot go
viral, it is not easy to remember. [T/F]
5. If your story (chart, visualization) advocates for no narrative,
it will feel like it has no purpose. [T/F]

(Solution in the next page)

Solution

1. A story is an account of events [T/F]. True. To gain

perspective on this question also see Lisa Feldman work on the
Constructed Model of the World as Reality.
2. Story consumption releases oxytocin [T/F]. True. To learn
more about the biochemistry of storytelling and its function in
evolution.
3. The purpose of a story is to advocate (transmit, perpetuate) a
narrative (a belief, a culture, an ideology) [T/F]. True.
4. A narrative does not stick to the human brain, it cannot go
viral, it is not easy to remember [T/F]. True. The message of a
narrative in its succinct form is usually not viral. Stories in form
Ads, movies and books are more suited to go viral.
5. If your story (chart, visualization) advocates for no narrative,
it will feel like it has no purpose [T/F]. True.
Quiz: Storytelling climate change
Time 15 minutes. Online search: allowed. In August 2019 Greta
Thunberg set sail from Europe to a UN climate conference on the other side
of the Atlantic. Identify the data, the story, and the narrative in this story.

Fig 18 Unlike Al Gore in 2006, Greta Thunberg needed no charts to get her
message across [14].
Solution to Storytelling climate change

Narrative (example of answers)

The underdog wins[15]
Role reversal
Dysfunctional family
Climate change is an emergency

Story
(Abridged speech[16]). “You might be grown-ups but you are not
mature enough to understand this emergency. If you did, you would not
jet to the conference like you do. You could Skype, or travel like me to
reduce your carbon footprint. Hence, you (not me) is behaving like the
immature kid.”

Data
Look at the big waves behind me, I am serious, this is
dangerous
A carbon neutral sail-ship = it’s possible to reduce carbon
footprint
The situation is bad enough that I had to skip school classes
Air jet-set travel produces CO2 but there are alternatives look at
me
 Chapter 2. Visualizing Information
– How to transform data into information –

Fig 19 Creating knowledge from data, the secret to winning the Nobel
prize?

In this chapter, you will learn to transform data into information, (a

prerequisite to produce knowledge). In other words, how to make your
charts more useful by displaying information effectively.

“The foundation of a useful chart is seldom graphic design”

The bulk of published books on data visualization focus on how to build
charts and how to make them readable. A few of them spend a considerable
number of pages listing all the types of charts available out there and use
the good chart / bad chart template to teach how a little tweak can make or
break the readability of a chart. Others give color advice and how important
it is not to clutter your chart with colors, (see also the term junkcharts
coined by Tufte). This is helpful for improving readability and aesthetics
but not so helpful in transforming information into knowledge (aka wisdom,
prescriptive analytics). And while poor color choices can kill the readability
of any chart, focusing on chart aestheticism is the equivalent to teaching
about the importance of font types to someone that just wants to become a
writer. After all, the book Harry Potter did not become a bestseller because
of the font type they used. In the same vein, the typical root cause of a poor
chart is failing to transform data into (meaningful) knowledge, not using the
wrong aesthetics. Let’s see an example of transforming information into
knowledge with data about gender distribution from the Kaggle 2018
survey.
Visualizing gender
Exercise
How would you visualize the following gender breakdown of data
scientists? Time 3 minutes.

Survey responses
Female 16.8%
Male 81.4%
NS 1.4%
ND 0.3%
Solution
Solution with Matplotlib

Fig 20 This default Matplotlib chart uses four different font sizes and six
colors (red, green, turquoise, purple, black, grey).

Exercise
 Is the figure above data, information or knowledge? Why? Reason your
answer. Time 2 minutes.
Solution
It is just information. It is not knowledge because it is not significantly
more useful than the original raw data.
Reflection
Unless you are in a preliminary Exploratory Data Analysis (EDA), it is
not a good idea to disseminate a chart unless there is a clear why (narrative)
for the chart. And even if you produce many charts as a part of an EDA,
resist the temptation to show all of them. In this case, we are asked to
visualize the gender distribution of the respondents of the 2018 Kaggle
survey — one of the largest data science communities worldwide. Gender
was one of the 30+ questions on the survey which was answered by about
30k respondents. Fig 20 is the default settings chart produced using the
popular python library Matplotlib. This chart is perfectly fine. It is
informative, but there is no message, there is no why. It lacks a purpose.
Why? One reason is that it is not connected to any narrative. Another
reason is that it does not increase our knowledge. Is it helping us to become
wiser? Is it facilitating the prescriptive analytics function? How would you
make this chart more useful?

Exercise
Draw here at least three alternative charts to Fig 20? Time 3 minutes.
(Solution in the next page).
The Chart-narrative fit
 Fig 21 Three ways to visualize the same data.

Choosing the right chart

 Fig 21 shows the same information as Fig 20 in three different ways, a)
pie chart, b) bar chart, c) stack bar chart horizontal. Let’s consider the
process we followed to produce them.
Simplify
We consider men and women only and ignore the other two outlier
groups.
Avoid colors
The men are represented in white and the women in dashed area. We
avoided color, because we can. Colors are loaded with symbolisms[17],
greyscale less so. Conjugating colors is a very subtle art that is easy to
underestimate. Why use more than six colors without proper graphic design
training? We have also removed the scales, figures and the axis labels. We
traded-off a bit of accuracy for a big gain in clarity.
Consider a narrative
Let’s assume we are advocating for a gender equity narrative. We define
the equity narrative as a world view in which the world will be a better
place if there is less gender ratio imbalance[18].
Check the chart-narrative fit
Once a narrative is set, choosing a compatible chart template is the next
step to transform information into knowledge. In this case, pies and bars are
a poor choice. Why? Because pie charts are connected to the narrative of
how much pie each group gets (finite resources), it is a confrontational
narrative that undermines the narrative of gender equity. The bar plot chart
is also a poor choice because it is connected to the narrative of competing
and to other win-lose narratives such as scarcity and the view that the men
and women compete. Using them risks undermining any growth mindset[19]
or equity (win-win) narrative that you might be advocating for. The stack
bar is a better choice. It is horizontal so it is not subject to the gravity
metaphor. (See gravity & charts). Now that we have found a fit between the
chart and the narrative, let’s design the chart in ways that are easier to
assimilate by humans. We call this charts human centered charts. Human
centered charts leverage the same principles as human centered design
(placing the user at the center of the design process).
Human-centered charts

Fig 22 Four Batmen and a Wonder woman make this chart easier to
remember.

Using superheroes is a one way to visualize the heavy[20] topic of gender

equity in STEM dominated environments. We also use the iso-measure. In
an iso-measure chart, each unit (in this case a superhero) represents the
same quantity (in this case, about 5,000 respondents of the survey). We also
reduced the superhero count to the minimum possible (five persons) which
means we approximate the original percentages to a 1:4 female to male
ratio. This chart can also be read as follows: on average, for every 5 people
in a team (room, meeting), one will be female.
Checking the chart-narrative fit
 Given the same gender equity narrative used earlier, let’s see how the
design choices we made fit such narrative.

A superhero’s job is to make the world a better place, the equity

narrative is about making the world a better place, this playful
theme will help the reader of the chart remember it.
The iso-measure is humans, like the respondents
The number of heroes is less than seven, (we are respecting the
7 chunks rule and not overloading the reader with information)
We use humor[21] to improve the communication effectiveness.

Leveraging humor
Many charts are impersonal because we cannot relate to them. We solved
that with the superheroes. See also user personas in Ch. 6) However, if in
addition we want the audience to remember the chart, we can use humor or
an insider joke as in, data scientists are superheroes because they have to
“wrangle” with data, see the term data wrangling.
Sexism in your chart?

Fig 23 Always check for blind spots.

Chart check list
☑ Type of chart compatible with the narrative (message)
☑ No color overload (3 colors max)
☑ One chart, one message
☑ Metaphors aligned with the narrative
☑ I don’t need to read the caption to understand the chart
☑ The caption is used as a synthesis opportunity
☑ The caption does not explain the chart again
☑ Bias checked by third party

How to bias check? It is important to check for blind spots. Charts are no
different. It is prudent to ask for bias check to a diversity of people, ideally
with different backgrounds. See Chapter 5 for more on bias.
Gravity & charts
 Fig 24 Gravity shapes everything on Earth, including how we interpret
charts.

How gravity affects goals

 Let’s assume your charts have convinced your organization that
achieving gender parity at work is a good goal and after a board meeting a
goal of reaching a 50% female to male ratio has been set. How to visualize
it to persuade and rally the rest of the organization? The arrows point to the
goal. Note how the alignment of the arrows with the direction of gravity
influences how the goal achievability is perceived. In the pie chart, the
downward arrow has a negative connotation. For the vertical stack bar the
upward arrow against gravity makes the goal appear hard to achieve. The
horizontal stack bar is gravity neutral and the freest of connotations that
might distract from the narrative.
Chart wars: Pies vs. Bars

Fig 25 Musk vs. Bezos. Two visions of space exploration. Two ways to
visualize altitude.
When to use pies
There is a fundamental difference between circular charts and bar charts.
The brain is sensitive to angular change and (by comparison) numb to linear
change[22]. This is particularly true when considering motion, and sensitivity
to small changes. If in your narrative, highlighting minute changes in a
variable is important for the story, then circular pie charts (speed gauges)
are the way to go. If on the contrary, too much attention to change is a
distraction, avoid pie charts. Compare for yourself. In the Blue origin cast,
the attitude change is barely noticeable. Whereas in the SpaceX cast, it pops
during all the cast.
Blue Origin cast: http://bit.ly/2NHycmf
SpaceX cast: http://bit.ly/2XwXYxY
Quiz: Making useful charts
True or False? Time 10 minutes.

1. A bar chart is a great way to visualize the odds of the Casino

Roulette. [T/F]
2. Humans are more sensitive to pie charts than bar chars.
However, in animated charts where the ratios change, these
changes are more obvious in a bar chart. [T/F]
3. A drawback of the iso-measure chart is that one cannot use
metaphors or personas with it. [T/F]
4. A way to turn information into knowledge is to use a palette
with the minimum amount of colors possible. [T/F]
5. The purpose of an EDA is to tell a story about the data. [T/F]

(Solution in the next page)

Solution

1. A bar chart is a great way to visualize the odds of the Casino

Roulette [ T/ F ]. False. A bar chart is linked to the compete
narrative. The roulette is a zero-sum game. A Pie chart, an iso-
measure chart or a photo of the roulette itself communicates
odds more clearly.
2. Humans are more sensitive to pie charts than bar charts.
However, in animated charts where the ratios change, those
changes are more obvious in a bar chart. [T/F]. False. They are
more sensitive to pies/ needles in both situations.
3. A drawback of the iso-measure chart is that one cannot use
metaphors or personas with it. [T/F]. False. It is the opposite.
The iso-measure lends itself to metaphors.
4. A way to turn information into knowledge is to use a palette
with the minimum amount of colors possible [T/F]. False.
Avoiding information overload is a pre-condition for
knowledge. However, the fundamental transform is usefulness.
(Usually, by connecting it to other knowledge).
5. The purpose of an EDA is to tell a story about the data [T/F].
False. The main purpose of an Exploratory Data Analysis is to
aggregate and visualize basic statistical information.
 Chapter 3. Knowledge
– How to create knowledge with frameworks –

Fig 26 This chart uses the gravity metaphor.

In Chapter 1, we saw that charts with purpose have a message. In

Chapter 2, we saw an example of how to transform gender data into
information. In this chapter, we will learn how to synthesize information
into knowledge. If we were in an English language class, this skill would be
the equivalent of writing the book synthesis. Your data is the book; the chart
is the synthesis. An effective way to do this is by using reference
frameworks, summarization techniques and visual metaphors. Let’s see an
example that uses age data from a survey.
Mental frameworks
Let’s take a look at the chart in Fig 27. How many chunks of information
can you count? It has information overload. Let’s focus on the color palette
for instance, a rainbow. However, a rainbow does not convey any meaning
here. On the contrary, by using 12 colors, we have increased the
information overload by a whopping 12 memory chunks with a zero gain
in meaning, (see information vs. meaning in Ch. 1).

Fig 27 A ggplot default visualization of the age of the respondents of the

Kaggle 2018 survey. Color scale: Rainbow, 12 colors.

Exercise
How would you make the previous chart more useful? We can start by
reducing the information overload. Draw solutions. Time 2 minutes.
Solution

Fig 28 Same information with a narrative and linked to a reference

framework, closer to knowledge.

What can we learn from senior data scientists?

Like Fig 27, Fig 28 is user distribution by age too. However, here we use
a two-color scheme to highlight which age-group won most competitions
per user. However, just a few too many age bins can overwhelm any reader.
A way to declutter and structure the bins into usable information is to
reduce their numbers and group them in a familiar, relatable form. One way
to do this is to reduce age groups to generation groups. In this case, we used
a reference framework many are familiar with: Generations in the
workforce. It comprises the gen X, Y, Z and the Boomer[23]. Furthermore,
we are interested to see which group is the most productive in terms of
competitions and cash prizes per user. Because everyone belongs to a
generation this chart can become very personable. What can we learn from
the wisdom that each generation offers? Source: 2018 Kaggle Survey Q2
What is your age?
Generation & work-ethic attribute framework
The Baby Boomers, born 1946 – 1964, “often branded
workaholics[24]”
Gen X, born 1967 – 1977, “this generation works to live and
carry with them a level of cynicism”
Gen Y, “Millennials” born 1980 – 2000, “considered the most
educated and self-aware generation in employment”
Gen Z[25], born after 2000
Narrative
Ageism is the narrative that says things like “old people do not have
energy to be entrepreneurial and cannot innovate”[26]. However, a few
papers [27] have disproved this claim. The chart advocates for the opposite
narrative, “(in data science), older generations are as productive as younger
generations”.
Knowledge creation
Note how the key step to create meaning (knowledge) is not only to
summarize and declutter, but to find where the information is most useful
and then by linking it to that context (reference framework). In this case,
Generations in the workplace and productivity. Another way to create
meaning is by way of visual metaphors such as the pyramid. Let’s see an
example based on salary information.
Visualizing inclusion

Fig 29 A default ggplot2 visualization of the salaries of the Kaggle 2018

Survey. Source Q9 What is your current yearly compensation.

Exercise
In the previous section, we saw an example of meaning creation by way
of connecting information to a reference framework. Now let’s do the same
and, in addition, let’s apply a visual metaphor. Let’s look at salary data from
the same 2018 survey. Is Fig 29 data information or knowledge?
Given an inclusion narrative, how would you create a more useful chart?
Time 6 minutes. Hint: If the chart was a building where would highly paid
individuals own apartments?
Intermediate step
(Fig 29 rotated counter clockwise 90 degrees)

Fig 30 Does your chart fit any visual metaphor?

Solution to inclusion

Fig 31 Using text inside charts is a great way to defuse ambiguity.

Earlier we mentioned the importance of the chart-narrative fit. Fig 31’s

narrative is about the digital divide. How inclusive is data science as a
community? As we just saw in the previous section, one way to create
knowledge is to relate our information to existing reference frameworks.
Generations in the workforce was an example of a popular framework to
think about age and workforce. Helping the reader situate the new
information into existing frameworks makes the new information easier to
assimilate, use and recall. Here, we apply the reference framework of
income percentiles – a common analysis framework used by economists
and fit it with the pyramid a metaphor that represent hierarchies. See also
the #onepercent movement. In the US, to belong to the 1% elite, one needs
to earn more than $422k per year[28]. About 23 respondents declared that
they do. In addition, about 6% declared they belong to the 10% percentile, a
very inclusive number because 6% is similar to 10%. The 10% percentile
income is about $166k in US[29], so if the sample reflects the distribution
found in society, it means it is at least somehow inclusive. We add a smiley
emoji to reassure the reader that yes, this is good. However, those numbers
are for US household incomes. When we look globally, the 1% percentile
threshold is $32k per year. This puts 60% of the respondents in the top 1%.
60% is very different from 1%, so globally this data point does not support
inclusiveness because it does not reflect the global distribution. Aha
moment. One way to create such moments in the story is to A/Bify the story
by switching between two points of view. We just saw how powerful visual
metaphors can be. Let’s see three more examples.
Winner takes all

Fig 32 The top three ML libraries.

Here, we are visualizing the data science libraries the respondents use.
Using survey Q20: Of the choices that you selected in the previous
question, which ML library have you used the most? Given a winner takes
all narrative (so common in the software world), what visual metaphor can
we apply? This chart is an example of less is more. In this case, Sci-Kit (a
famous scientific Python library) has a 48% share, Google’s TensorFlow
has a 16%, followed by Keras 14%. Let's see how this visual is connected
to the narrative.
Chart-narrative fit
In scenarios with strong network externalities at play such as social
network, a phone OS or an Olympic race; being on the podium (being first)
has a disproportionate effect on the reward. In such cases, the winner-takes-
all narrative is in place. Anthropomorphizing the ranking with a podium
conveys a memorable narrative and affordance — the glory the winner
deserves for the great utility this library provided to the community. This
narrative is also connected to other memes famous in the software world
such as the developer’s glory. (See S. Balmer in “Developers, developers,
developers, developers”).
 All or nothing
In the previous section, we visualized data about the most popular ML
libraries with a winner takes all narrative, here we do the same with a
different narrative.

Fig 33 The house of Shiva. When colored areas occupy large areas, use
50% grey and pastel colors instead of 100% solid bold colors.

House of Shiva
Fig 33 is a combination of, (i) a chart template called Marimekko, with
(ii) a symbolic chart called House of Shiva. The House of Shiva is used to
emphasize all or nothing relations. The metaphor is that the roof falls if just
one column collapses.
Symbolism
The columns support the visualization efforts of the community (roof
load, common good). The width of the “columns” expresses how much
work/load each column supports. Grey columns on the right represent other
less mainstream libraries such as: D3, Shiny, Bokeh, Leaflet, Lattice.
Source: Survey Q22 Which specific data visualization library or tool have
you used the most?
Metaphor
The goal is the roof. As with a house, the integrity of it becomes clearly
impaired if one column is weak.
Narrative
The narrative here is that non-mainstream visualization libraries are
important but with different degree. Note here that if we had used a pie-
chart we would have conveyed a win-lose scarcity narrative, not faithfully
representing the win-win ethos of the open source movement.
The BRICS framework

Fig 34 An example of meaning loss by information overload. Source:

http://bit.ly/2K7ZLBk

A common way to summarize international data is to count grouping by

country. However, this often leads to clutter because there are more than
200 countries in the world. Another common way is to use a geographical
world map, and to modulate the color of the country with the count.
However, some countries are very large in extension while other countries
become a pixel on the screen. This is not great for usability.
Exercise
Summarize the previous chart. There are more than 200 countries. Only
24 are shown in this view[30]. How would you go about this? Draw
solutions. Time 5 minutes.
Solution to BRICS
If we look at the 2018 survey and compare it to the 2017, there is an
increase of 1145 new respondents that identify as “data scientist”. So,
where do new data scientist users come from? Fig 34 displays 2018 and
2017 data. However, there are too many countries for a human to make
sense of it! Remember that a human brain’s working memory is limited to
5-7 chunks [31]. This means we cannot juggle more than seven countries at
the same time and neither should your chart. What would Marie Kondo do?

Fig 35 This chart was made with PowerPoint because it was faster than
tweaking the parameters of ggplot. Notice how the golden ratio is used
across the chart.

Use Mega Regions

One way to summarize in a more humane way is to group countries in
economic mega regions that have a certain degree of homogeneity (cultural
or economic). In this case, we apply the BRICS framework. BRICS stands
for Brazil, Russia, India and China. It is a grouping coined by Goldman
Sachs to classify countries with similar economic indicators under one
label. (The term “BRIC” was coined in 2001 by then-chairman of Goldman
Sachs Asset Management, Jim O’Neill). This framework provides a
meaningful way to group countries by bagging them by economic and
social development affinity. In this case, we chose four groups: the US,
Europe, BRICS and the rest of World. When we do that, we see that not
only is BRICS the top contributor to growth with 42% of total growth for
2018, but it was also the fastest growing among the big three. In 2018, in
the category “users that define themselves as data scientist”, Europe added
302 users, US 131, the rest of the world: 231, and BRICS: 481. By 2020 the
growth of BRICS will outnumber Europe and US combined. Data source:
we forked and modified a snippet of the code from Kaggle user ash316 and
took the top 20 countries whose respondents identified as “data scientist”.
EU-6 means the top 6 European countries.
Aesthetics
Note how different font sizes are used in the chart. How the horizontal
layout neutralizes the gravity metaphor. And how flags and labels are used.
The individual countries growth added together should stack up to 100%,
What narrative would we fall into if the author had used a pie or stack bar?
Narrative
The narrative here is that to be wise, one must look where the ball is
going to be not at where it is now. Don’t look at absolute numbers of 2018
only, look at growth. Of course, an alternate narrative is that emerging
countries are catching up in strategic fields such as data science.
Quiz: Visual summaries
True or False? Time 5 minutes.
1. Visualizing 200 country names in an 800x1200 display is
ridiculous because there are not enough pixels. [ True / False ]
2. Visualizing 200 countries in a chart violates the 7 chunks limit
rule. [T/F]
3. Using PowerPoint or Adobe to build a chart one needs is
professional. [T/F]
4. To visualize ‘a winner takes all situation’ we can use a
Marimekko chart. [T/F]
5. House of Shiva is best to visualize all or nothing relations.
[T/F]

(Solution in the next page)

Solution

1. Visualizing a chart with 200 country names in a 800x1200

display is ridiculous because there are not enough pixels [ True /
False ]. False. It is ridiculous because a chart with 200 text
labels is information overload unusual.
2. Visualizing 200 countries in a chart violates the 7 chunks limit
rule [T/F]. True.
3. Using PowerPoint or Adobe to build a chart one needs is
professional [T/F]. True. That is how Bloomberg Businessweek
charts are built.
4. To visualize a ‘winner takes all’ situation we can use a
Marimekko chart [T/F]. False. Podium 10x is clearer in this
case.
5. House of Shiva is best to visualize ‘all or nothing relations’
[T/F]. True. If one column is missing the roof falls.
 Chapter 4. Charts to Think
– How to use charts to make wiser decisions –

So far, we have seen examples of how to transform data into information

and information into knowledge. Now let’s consider prescriptive analytics
— the charts the board room uses to decide what to do in policy and
decision making. Unfortunately, policy challenges are complex to deal with.
Particularly, when the variables involved have unexpected dependencies
that are not properly understood, or even known!
A way to deal with these complex problems is to simplify them by
describing them in a more humane way. One popular way, is to transform
the problem into a map — a process called mapping. Mapping can be done
if we just use two dimensions to describe the problem. Once in 2D, the
canvas becomes a design space where we can search for solutions. Because
we are in a map setting, we can leverage the extraordinary spatial cognitive
abilities that have been gifted to us[32]. Glorious examples of “mapping” are
war rooms, the BCG growth-share matrix, Gantt charts, The Business
Model Canvas, Kanban boards, Gartner’s Magic quadrant and Wardley
maps — to date, the most advanced thinking tool to think about strategic
innovation.
Let’s start with 2D mapping. Imagine you are a high official in charge of
innovation in Singapore. You have only been given Fig 34’s data (number
of data scientists per country). How would you use this data to inform the
next innovation policy looking forward? Hint: A first step is to create
situational awareness. One way is to rank the countries, so we can see
where we stand. The second step is to use a meaningful indicator in the x-
axis. (Solution in the next section).
Use rankings to create situational awareness

Fig 36 Two is the maximum amount of colors you should use in a chart.
Grey does not count.

It’s per capita or die

As the economist and comedian Harald Eia politely implies in his Oslo
TEDx Talk, when comparing countries we are hardly ever interested in
absolute numbers. In fact, any non-per-capita measure is rather useless. Yet,
how often we forget this! Charts not normalized per country population are
setting themselves up for stereotypes and unhealthy narratives. From the
2018 Kaggle data science survey, it is possible to count the number of data
scientists in each country, and most participants rendered such charts. But
how useful is that? We already know that China and the US are large
countries, so in absolute numbers they will also have a larger quantity of
data scientists. That is expected. However, if it is expected it means we
already knew it. The more expected, the less information (100% expected
means zero information). What is useful, is to look at per capita measures.
How do countries compare on data scientist density? In this chart, we
highlight Singapore in red and the US (home to the largest survey
community, like a center of mass) in black so the reader has a reference
point. This chart has a lot going on:
USA mean: 0.14 per 10,000
EU6 mean*: 0.09 per 10,000
BRICS mean*: 0.03 (5x less than US)

Mind the gap

Mind the gap is a common strategy to think about differences between
categories in the data, in this case, countries. Thinking about why the gap
exists can help explain the reality that the chart is representing. For
example, a linguist might think about the gap in terms of English
proficiency and its correlation to the prevalence of data scientists. Is the
language barrier an explanatory factor for the gap? What are the policy
implications? Note: The BRICS, and EU6 mean is mean of country means,
not weighted by respondents. Source: World Bank Population Data 2016,
Survey Q11 Current country of residence.
Aesthetics
This color scheme is called the red on grey, it is my favorite scheme for
charts. Unlike, other schemes such as purple on grey, it is gender neutral.
However, for it to work, the red surface must be kept to a minimum,
otherwise it comes across as strident.
Narrative
Michael Porter’s Competitive Advantage of Nations.
Exercise
Let’s take Fig 36 chart a step further. Earlier we saw that frameworks can
help the reader make sense of new information. Using the global innovation
index, how would you relate it to the prevalence of data scientists? Use 2D
mapping. Time 5 minutes. (Solution in the next section).
The design space

Fig 37 A scatter plot between two correlated variables always yields a

similar cloud.

As Mr. Wardley would say — When you need to understand the territory
it helps to have a map. Here we use 2D mapping by scattering the countries
along two dimensions[33]. The technique of projecting into two dimensions
has been successfully used in famous charts such as Wardley Maps, the
BCG growth share matrix, The Urgent-Important matrix and Gartner’s
magic quadrant. This map can be used to cluster countries by policy to help
elucidate success factors that influence the position in the map (See also
Gapminder).
Narrative
Porter’s Competitive Advantage of Nations.
About the Innovation Index
Every year, INSEAD MBA, Cornel University and the WIPO publish
the Global Innovation Index. In 2018, the most innovative country was
Switzerland. A Spearman rank correlation between GII and user prevalence
yields 79%.
Exercise
Let’s take this chart a step further. One of the most valuable skills is
prediction. Given Fig 37, can you predict where Japan will be 10 years from
now? Use a linear regression. Time 5 minutes. (Solution in the next
section).
Forecasting with mean-reversion

Fig 38 Mean reversion, always right in the long run?

Here, we just added a regression line and removed the outlier Singapore.
The 95% standard error margin is shown in grey. Some countries are below
and some above. Highlighted in red is Japan, as an outlier with high in
Innovation Index (y-axis) but low in x-axis relative to peers. Let’s assume
that the principle of mean reversion applies here as a baseline predictor and
a hidden hand continually pushes countries towards the mean (dotted line).
The principle of mean reversion is based on the idea that there are no
permanent competitive advantages to either companies (See introduction
chapter in Blue Ocean Strategy) or nations. It has shown its worth,
particularly in finance. For example, in betting on the composition of the
DOW JONES, very few companies have what it takes to last long in the
Dow Jones. Of the original members of the index formed in 1896, only GE
remains.
Reflection
What can we forecast about the 2019 GII rank position of Japan?
Applying the principle of mean reversion, it is unlikely that Japan will
increase its rank because it is already high. Even if Japan catches up in data
scientist prevalence, it is likely that it will still go down (towards the mean).
Indexes are just weights. Assuming the Data Science weight in the
innovation economy will only increase in the coming decades and that the
GII index calculation method will be updated accordingly, what countries
are more likely to improve their “nominal" ranking in 2019? When the GII
index weights are rebalanced, is it likely that countries Canada, Australia
will jump a few places? Source: Global Innovation Index 2018, World Bank
Population Data 2016, Q11 - Current country of residence.
A note on the origin of linear regression
The name linear regression as in the line that minimizes the sum of the
square of the errors, was popularized in a paper where the principle of
“regression to the mean” was verified in how offspring height is related to
the parent’s height. Spoiler alert! Only 60% of the offspring height is
explained by the parents’ heights. The rest is explained by the mean of the
race. Which means that the Mean reversion principle applies in height with
a 40% influence approximately. However, the mathematical method is
completely unrelated to any concept of regression. The paper got famous
and the regression word stuck to the method. A great trick question is to ask
students to explain why linear regression is called linear regression. I am
always amazed at the inventiveness of some students[34].
The design space in business

Fig 39 Can you visualize a problem as a map?

Reframe it
One of the most important roles of a data scientist, is to realize when the
customer cannot articulate his own needs (see Jobs-to-be-done theory). This
skill is what distinguishes the A+ data scientist from the rest. The chart here
is adapted from the book The Accidental Investment Banker. The author, a
banker, came up with it during a business engagement. He used it to map
out the M&A strategy for a client. Once he made this chart, everybody in
the room could visualize where value was. In his book, he credits this chart
as an important moment in his career.
The Gap Matrix
Fig 40a Business Innovation is sometimes as easy as finding a white space.
Source: McKinsey Global Institute[35]

Finding gaps in the market

Another use of 2D design space is to find overlooked or underserved
segments. We can apply this method to matrixes too. Examples are, a
company-employee skills matrix, technology roadmaps, and the innovation
matrix. In Fig 40, McKinsey took a sectorial view. However, other
viewpoints can yield discoveries too. For example: instead of viewing by
sector we could view by software vendor (Salesforce, JIRA, Autodesk…).
The Periodic table

Fig 40b The most useful visualization in the history of Science? Source:
Bloomberg BusinessWeek.

A famous application of the Gap Matrix is Mendeleev’s 1869 Periodic

Table of Elements. In 1869, 150 years ago, Dmitri Mendeleev published a
periodic table of the chemical elements in based on properties that appeared
with some regularity as he laid out the elements from lightest to heaviest.
When Mendeleev proposed his periodic table, he noted gaps. At the time,
only 63 of the 118 known elements identified today were known. Then he
predicted the properties of five undiscovered elements -- a genius coup for
any young scientist[36]. Mendeleev visual predictions spurred a discovery
race. Shockingly, he never received the Nobel prize – a testament to
scientific rivalries, but also of how visual thinking has been historically
despised by an academia dominated by word oriented thinking.
The Innovation Matrix
The Innovation Matrix is a knowledge discovery tool – Leo Tschirsky

Fig 41 The Innovation Matrix (IM)[37].

Leo Tschirsky, Professor Emeritus of Business Management at the Swiss

Federal Institute of Technology (ETH), facilitated this microwave oven
workshop in Tokyo Institute of Technology around 2006. The IM helps to
formalize and organize functional relations between:
Market needs
Product functions
Technology
Science fields
How to use
1. Draw a matrix format by rows
 Write the market needs (Why people buy ovens)
Product functions (heat, boil...)
Technologies behind those functions (Magnetron, LCD...)
Basic science field supporting those technologies. (RF, IC)

2. Link concepts with arrows

Clarify
Seek deep truths
Use 5-Why root cause analysis, jobs-to-be-done theory
Once your product is clear and mapped out...

3. Innovate
Now that you have a clear picture of relationships between value,
customer needs, costs and technology. You are in a better position to
innovate using a variety of techniques such as:
Brainstorming
Planning an ideo style shopping cart workshop
Using Edward deBono creativity tools
Finding gaps
Serving new needs with exiting functions
Exercise
Groups of four. Time 20 minutes. Think about this microwave and its
components…

Fig 42 The cyclotron is the element that generates the microwaves.

You have been hired by a microwave oven brand. Recently, due to

Chinese competition, the oven margins are paper thin so the survival of the
company might very well depend on you coming up with a new oven
design for which customers want to pay more. Required. Use the IM to
innovate the microwave.
Solution

Fig 43-44 Functional Innovation example, can we replace light and

window by a cheaper component with same function?

Light and oven window seem two unrelated technologies in an oven.

However, they serve a common need: The need to check for mistakes! Cost
of (Window +Light) > Cost of (X)? Consider the cost of a window + light.
Their sole purpose is now clear. Can X do their function better, cheaper?
Let’s find X!
Wardley maps

Fig 45 A Wardley map, Simon Wardley CC BY-SA 4.0 .

A Wardley map can be thought of as a 2D mapping of commoditization

vs value chain, or as an evolution of Leo Tschirsky’s Innovation Matrix. In
any case, when there are dilemmas about whether to outsource or not, this is
the go to tool in 2019. So far we have seen a few visualization tools used in
business and policy making. However, visualization is also used in
coaching. Let’s see an example called the wheel of life.
Wheel of Life
“A chart to think about your life goals”
Amy is a working, single mom who just resigned from a Fortune 500
job. A successful career woman in her own right, Amy just left her
ascending executive role because she wanted to travel less and spend more
quality time with her 14-year-old teenage daughter. I am sitting with Amy.
We are meeting to talk about life, career and charts, yes charts! Amy is
about to show me a chart I have never seen before. She calls it the Wheel of
Life. This chart, she tells me, was instrumental in helping her realize that
she needed to change her life.
How to use

Fig 46 Start with a dot in the middle of a paper.

First, you draw spokes. Eight spokes of a wheel. Each spoke represents a
different category of your life and will help you measure your satisfaction
in each area of your life. The first one is Money (How satisfied are you with
the money you have saved/make?) Second, Career (How satisfied are you
with your path, progress and current career?). Third, Wellness (both
spiritual/mental and physical). Then, Friends & Family, Love, Fun, Physical
Environment (Do you like the country, city/ house/ neighborhood you are
in?), and finally, spiritual and personal growth. We put a grade on each
category marking a dot on the spoke on a scale of 1 to 10, 1 being at the
center and 10 being away from the center, and then we connect the dots.
Connecting the dots

Fig 47 Citizen Kane-type wheel of Life.

As I connect the dots of my chart, a chill goes up my spine. My wheel of

life is not round! It looks like an asteroid and has a big dip where the spoke
of love is. Is that bad? “Oh, well” I think, mortified, I guess I need to
reprioritize my life.
An x-ray of your life
 Then Amy leans over and says, “This is the snapshot of your life. I learnt
this exercise at a drop-in class at Stanford. It turned out to be one of the
most useful things I learnt there because it helped me visualize my blind
spots. Many of my classmates did this chart too, and they were just as
shocked as you are now...” Then she adds, “The typical comment of my
classmate was: Here I am, focusing on my career and personal
development, as I have done all my life, while equally or more important
parts of my life are being neglected.” The wheel of life together with other
visualization tools such as the good time journal are part of the design your
life trend where design principles of iteration, visualization and design
thinking in general are applied to coaching. The results are spectacular. To
learn more, the book by Atari co-founder Designing your life is a great start.
Exercise
Make your Wheel of Life. How would you make this chart more
compelling than Fig 47? Time 7 minutes. (Solution in the next page)
Solution

Fig 48 Icons, use to reduce the time to understand a chart.

Icons and emoji are an underused resource in chart making. On the other
hand, emoji use is correlated with employee engagement
Interactive exploration

Fig 49 A screenshot of a real time, visual SQL inner join operation between
three tables; Source: Square 2001.

Square’s Crossfilter
Crossfilter is a JavaScript library for exploring large multivariate
datasets in the browser. Extremely fast (<30ms), it allows “Doherty
threshold” interaction with coordinated views, even with datasets
containing a million or more records; Square built it in 2001 to power
analytics for Square Register.

Exercise
Groups of two. Time 12 minutes. Using Square Crossfilter demo, find
three insights about air travel. Example: to avoid delays fly in the morning.
Visualize your findings.
Quiz: Supporting decisions
True or False? Time 10 minutes.

1. Working with per capita measures is the only rational way to

compare countries. [ True / False ]
2. The principle of mean reversion states that, in the long-run, a
‘hidden hand’ pushes outliers towards the mean. [T/F]
3. The gap matrix is used primarily to discover niche or gaps in
the market. [T/F]
4. Wardley maps have rendered Innovation Matrix obsolete.
[T/F]
5. The Crossfilter library is an appropriate choice to visualize
linear relationships between two variables such as delays and
departure time. [T/F]
Solution

1. Working with per capita measures is the only rational way to

compare countries. [ True / False ]. True.
2. The principle of mean reversion states that, in the long-run, a
‘hidden hand’ pushes outliers towards the mean [T/F]. True.
There are no long lasting competitive advantages [98].
3. The gap matrix is used primarily to discover niche or gaps in
the market [T/F]. True. However, it can also be used in other
areas.
4. Wardley maps have rendered Innovation Matrix obsolete
[T/F]. False. Innovation Matrix is still better than Wardley to do
product innovation and to map a product to its market.
5. The Crossfilter library is an appropriate choice to visualize
linear relationships between two variables such as delays and
departure time [T/F]. False. A scatter plot is clearer. Crossfilter
is great for EDA.
 Chapter 5. Making your chart pop
– Tips to make it viral –

Fig 50 Jazz hands, use to make a visual compelling.

In this chapter, we will see hacking strategies to make your chart

memorable. Now that you know the principles of meaningful chart making
(by reducing information overload, linking to frameworks, fitting narratives
and leveraging visual metaphors), let’s look at simple tips on how to jazz it
up, or in design thinking jargon, make it “pop”. Pop means it pops into the
eyes. My favorite resources on this are: Dan Roam’s books, Information is
Beautiful by McCandless, and the Instagram account “chartr”. Be careful!
Some charts are so beautiful on their own that they are like a piece of art in
their own right. In any case, remember, the chart should never be the main
character of your story, what matters is the narrative (message) and how
well it is communicated to your audience. Unfortunately, aesthetically
pleasing visuals and a visual that gets the job done do not always coincide.
Incidentally, this is one of the reasons that in an Advertising agency you
will see the art director and account manager always at odds, the art director
wants to win ad awards to go to Cannes Lions Creativity Festival, the
manager wants effective advertisement. Let’s see of the main techniques
use to make a chart “pop”.
Use arrows to unbound your thinking
Fig 51 The arrows give you permission to think beyond.
Decluttering charts

Fig 52 Marie Kondo applies the principle of throwing away things in one’s
life that do not spark joy.
Fig 53 Animation on decluttering a pie chart bit.ly/2OgCLUO
Use personas to win over the audience
 Fig 54 Humans are adept at recognizing faces.

Did you know that we can recognize a face faster than many other
objects in the world? Use it! In 2007 Honda used this principle when they
designed a motorbike that, from the back, looked like a human face
(anthropomorphic).
Visualizing big differences

Fig 55 This chart has an astonishing range of four orders of magnitude.

 How to calculate the dynamic range in a planet chart. In a planet chart,
the dynamic range is n3, where n is the times small planet radius fits in big
one. In Fig 56, n=12. Therefore, n3 = 123 = 1720. Therefore, the share of the
small planet is approximately 1 in (1720+1), 0.05%.

Fig 56 The small planet fits 1720 times in the volume of the big one.

Other successful attempts to visualize enormous differences are the

famous post “What does a PhD mean to you”, and the 1977 film Powers of
Ten; both used the 2D zoom technique to illustrate gargantuan size
contrasts. However, of all chart metaphors, the planet metaphor is excellent
to visualize vast differences. See the next page to find out why.
Putting the world’s money into perspective

Fig 57 Humans are evolutionary adapted to estimate mass from height.

The dynamic rage of this chart is an astonishing four orders of

magnitude! It can visualize 41bn and something 20,000 times larger. This
chart was combined with the Fear Of Missing Out (FOMO) narrative before
the Bitcoin crash of January 2018. Original source: HowMuch.net.
Fig 58 Comparison between a) Log of value, b) Cubic root of value (Radius
of equivalent weight sphere), and c) Linear scale.

In Fig 58, note how (a) has the most dynamic range but is not-intuitive;
(b) is the radius of the sphere equivalent. It would be intuitive if expressed
as volume, like in Fig 57; (c) is a linear scale; is intuitive, but lacks dynamic
range.
Why 3D spheres work so well
We humans have evolved to estimate weight of an animal from sight. Of
course, this was a very useful skill for our ancestors in the Savannah. Notice
how much easier it is to understand relative sizes when we use volume,
versus any other option. For primates, estimating the weight of a fellow
primate visually was a crucial survival skill useful to determine how
dangerous the threat of an opponent was before contact. Given most living
forms have a similar weight density, a way to do this was by estimating
volume. At the same time, we humans are struggle to understand bar charts
when the bars differ more than 2 orders of magnitude. Luckily, if shown 2D
projections of 3D objects, most humans can estimate the weight well. This
comes in handy to compare magnitudes as different as 3 or 4 orders of
magnitude on a flat surface such as this book. Using the cubic relation, a 1
to 10 change in height becomes a 1 to 1,000 change in weight — a great
dynamic range.
Log charts
Note that the log plot solves the issue of dynamic range but we humans
are not born with logarithmic intuition built-in (Fig 58). In other words, a
kid will understand the balls, but it takes hours for undergrads to become
familiar with semi-log plots.
How many Solar Panels are needed to power the
USA?

Fig 59 In 2017, Elon Musk used such a chart template to advocate for Solar
Energy. — It was a flop.

Fig 59 narrative is scarcity. In 2017, Elon Musk used a chart like this
one. He was advocating for Solar Energy. He said, “We just need one pixel
of the map covered in panels to power the whole USA, remember just one
pixel.” It was a flop. Why? Because it connected to a win-lose narrative. It
is also hard to trust what we cannot see (one pixel is not a great
visualization). Unfortunately, 2D charts do not have enough dynamic range
to visualize differences larger than 2 orders of magnitude. He was trying to
visualize 4 orders.
 Fig 60 A chart that uses the growth mindset narrative. Adapted from Q-
Cells.

Fig 60 narrative is the growth mindset. It visualizes more than 5 orders

effortlessly. Circa the year 2005, German solar maker Q-cells used a similar
chart in their PR. This chart is more effective and trusted than Fig 59
because it connects us to the growth mindset narrative by visualizing the
astonishing abundance of renewable energy available.
Storytelling age-bias with humor
Fig 61 is based on an arrest dataset provided by Minneapolis Police and
other sources. It was published by Kaggle in 2018. It shows the correlation
between arrests with no charges and the age difference between policeman
and subject.

Fig 61 Of all the biases, age is one of the most pervasive and less talked
about.

Full kernel: http://bit.ly/2JXWP9d. Regression shown in blue. If one

considers not being charged as a positive outcome, then one would prefer
interacting with an older cop, but if one considers “a no charge arrest” as a
negative outcome, then one would prefer interacting with a cop closer to
one’s age. The lower the age gap, the lower the chance of not being
charged.
Exercise
Make this chart more compelling. Time 3 minutes.
Solution

Fig 62 Use personas to make a chart memorable.

Use the Golden ratio everywhere

Fig 63 Which one feels more harmonious?

Fig 64 Never miss a chance to use the golden ratio.
 Did you know that if you show two different cards to a human, she will
choose the one whose sides ratio is the closest to the 1.6.
Twists

Fig 65 Use twists to make your chart stick.

In improvisation[38] theatre, we always say that a good story should have

a twist. An effective twist it what makes a TikTok video viral, or in business
cases, the so called Aha! moment. We can use negative space to create
twists in charts. In this case, the reader expects to find the label #5 in a
certain place but it is not there! This sends the brain into search mode[39].
When a few milliseconds later the brain finds the missing label, some
endorphins are released. A bonus of this twist is that it fail-proofs (Poka-
Yoke[40]) the chart — changing the label location makes sure that the reader
will not miss the title.
Now that we know the main hacks to make a chart pop, let’s see a few
examples.
Quiz: Ice baby
Evaluate the following chart. Then prose a better way to visualize it.
Required: 300 words essay + two mockup sketches of proposed improved
visualization. Time 25 minutes.

Fig 66 Sales of ice cream in Sweden, Iceland and Finland. Source:

Meniga/Wrapp.

(Solution in the next page)

Solution
Main observations
This chart contains useful information such as: weekend sales
are higher than Wednesday sales, and Summer sales are higher
than Winter. However, the chart time line is not optimal to
showcase this.
To give this chart more utility, connect it to other knowledge.
For example, comparing to sales in Spain, (where the Winter:
Summer ratio of sales is lower, 4x vs 10x in Sweden).
This chart lacks narrative. An example of a narrative could be
how culture influences consumer habits. Example: “It’s funny
actually what happens in Sweden during summer months, we
sort of lose our minds. We eat, drink and swim like we never
saw the sun - that’s what happens when you live close to the
North Pole and July comes.”. Thérèse Lundquist -- Head of
Marketing & PR at Wrapp.

Color choices
Orange: cute at first glance, annoying after 20 seconds.

Time line
The data is weekly and yearly periodical, therefore a yearly or weekly
periodical time scales would reduce clutter. For example, a weekly radar
chart.
Alternative charts
Radar weekly
Bar plot monthly, weekly with iso-measures of ice cream
The grey line is constant therefore carries no meaning
(remove?)

Other suggestions
Normalize per capita & per app user
Annotate Christmas weekend and other peak days
Explode into a scatter between temperature (x-axis), sales (y-
axis), color of dot is day of week
Calculate % of sales due to temperature and % sales due to
seasonality from a linear regression analysis using as factors
summer and weekends.
Use Crossfilter to let users explore and discover hidden
relationships for the variables weekday, month, temperature,
sales of ice-cream and individual variance in consumption
Quiz: Global warming

Fig 67 Al Gore's CO2 Emissions Chart was not viral.

In 2006 Al Gore's presented this CO2 Emissions Chart. X-axis span is

+100 years; In Y-axis, red line is concentration of CO2; the blue line is
temperature. Note the high correlation between both. To the right is Al Gore
on top of a scissor lift. It was a flop. How would you help Al Gore visualize
global warming? Time 25 minutes.
Solution

Fig 68 Ed Hawkins made this spiral chart in May 2016. It went viral in
minutes.
 From Pies vs. Bars we know that humans are more sensitive to circular
than linear change. If we want the chart to align with the narrative that
“climate change is an emergency”, then let’s leverage that!.
Quiz: Magic quadrants

Fig 69 Most wanted Data Science skills in 2019. Source KDnuggets. Image
at original resolution (might appear blurry in some devices).

In 2019, Gregory Piatetsky, editor of kdnuggets, published this chart. It

used three visualization principles:
1. Color labelling to identify the cold/hot skills,
2. Dot size to indicate a magnitude, and
3. Mapping to a 2D space (but did not utilize its design
potential)
The chart was based on a poll. The poll had just two questions:
 1. Which skills/knowledge areas do you currently have? and
2. Which skills do you want to add or improve?
KDnuggets received 1,500 answers, and the aggregates by skill look as
follows.
Table 1. Aggregates of the survey.
Skill Have it Want it ratio
Python 71.2% 37.1% 0.52
Data Visualization 69.0% 25.3% 0.37
Critical Thinking 66.7% 15.5% 0.23
Excel 66.5% 4.6% 0.07
…

How might we visualize this data into a more meaningful way? What
design space is most appropriate given the data? Time 4 hours. Hint: See
Magic Quadrants.
 Fig 69b Fig from Most wanted Data Science skills in 2019. Source
KDnuggets. Image at original resolution (might appear blurry in some
devices).

Solution
 Let’s apply what we have learned so far. (Find a why, transform data into
information, synthesise knowledge by linking to frameworks, make it useful
for decision making). Before finding a why, let’s first explore the data.
The first instinct is to do a scatter plot to identify interesting clusters.
The x-axis can be the percentage of respondents that have a given skill, and
the y-axis, percentage of respondents that would like to have that skill
(want). However, there are too many data points for a human to make sense
of it. It is a textbook case of death by information overload and in Fig 70 we
used the Jackie Chan meme to convey it.
 Fig 70 A victim of information overload?

How to transform this information into knowledge? Let’s take a cue

from Gartner magic quadrant, a framework used in IT business
intelligence. It reduces complexity to human levels by using a quadrant
hierarchy (Fig 71).
Fig 71 Gartner uses the quadrants framework to cluster, reduce complexity
and make meaning.

However, Fig 71 is far from ready. The y-axis is aligned with the gravity
metaphor (highly wanted, high y). However, the x-axis is not aligned with
another unspoken rule, (this one by Guy Kawasaki): “you want (desired
goals) to be high and to the right”. In this case, the most desired skill (Deep
Learning) is on the wrong side – we need to flip the x-axis, (Fig 72).
 Fig 72 Goals should be “high and to the right” – Guy Kawasaki.

If you make a chart and no one remembers it. Did it still happen? In Fig
72, we grouped the skills in four categories but what good are they if no one
remembers them? One way to help your audience to remember is personas
(memes, in Gen-Z speak). Let’s apply user personas. In Fig 73 each
quadrant means:
Unwanted skills (Have but, don't want = Excel)
No-thank-you skills (don’t want and don’t have = JAVA)
Hot skills (want but don’t have = Tensor Flow)
Loved skills (want and have = Python)
 Fig 73 Pop culture, use to make your chart stick.

Red pen
Don’t be afraid to red pen your canvas! Captions are an opportunity to
clarify meaning and add punch to your story (not everyone is visual). Note
how in Fig 74 we broke the symmetry by tilting the “loved” label, that is
Feng-Shui for charts. We also added a twist in the Java quadrant by not
having a label for it. This ensures that the reader will go to this quadrant
after visiting all the others. The label for this quadrant is inside the meme
(No thank you).
Layering
Note how we have layered information in hierarchies (meme, quadrant
labels, quadrant representative). We have respected the seven-chunk limit in
each layer to avoid overload. Meaning was achieved by linking to an
existing framework and organizing the data into quadrants and creating a
labelling them.
Narrative
Finally, charts should have a purpose. It is reasonable to go through the
process of knowledge creation without knowing why. Once knowledge has
been found and visualized, the why will be easy to find. My personal why
for this chart is: “I like to see more Python and less Java in my classroom”.

Fig 74 Can Magic Quadrants make a scatter plot memorable?

Thinking about strategy

Fig 75 Salesforce acquired Tableau. Why?

Now that we created knowledge, can we use this chart as a thinking tool?
One way is to imagine contexts where this chart might be useful. Where
could this chart be used to create situational awareness? The figurative
“cloud wars” between Microsoft and Google are fought via proxies such
PowerBI, kaggle and other cloud software lock-in levers. A similar play
book developed in the 90’s in the database market. Fig 75 visualizes who
sponsors which language to see where allegiances stand.
Quiz: Visualizing gaps

Fig 76 Data science skills - How to conjugate absolute magnitude and

relative differences. Source: KDnuggets. Image at original resolution
(might appear blurry in some devices)
 In the same post, KDnuggets posted this chart. The purpose of this chart
was to display the gap between have and wanted skills while not losing
perspective of the absolute magnitudes involved in each skill. Why does
this chart fail? How would you declutter this chart and make it more
meaningful without loss of information? (no summarization). Time 1 hour.
Hint: See waterfall stack bars with negative components by McKinsey.
Solution

Fig 77 If people could unlearn things, would they? (Only 3 items are shown
due to resolution limits of the device)

Explanation of the decluttering process

The author wants to display for each skill three attributes:
Have skills
Wanted skills
Wanted minus have (the gap)
Of these three only two are independent variables. Applying the
principle of eliminating superfluous information we display only two. Have
(as the underlying) and the gap (want minus the underlying).
Zoom in
 Fig 78 What if you could learn anything you wanted? (Only 6 items are
shown due to resolution limits of the device)

This chart has now become a predictor of what would happen if people
learnt what they say they want. Have is the current (prevalence) level. Have
+ gap is the future level. In the case of negative gaps, the gap bars are
plotted on the other side of the y-axis, (a glitch of the stacking function of
ggplot2 or a feature - one cannot unlearn). Note how, from a compete
narrative of Fig 71 (competing bars), we have switched to a growth
mindset narrative with the “what if you could learn anything you wanted”.
How might we use this chart to prioritize what skills to teach.
 Chapter 6 Psychology of Visualization
with Ali Fenwick

Psychology plays a central role in designing effective visuals, especially

when it comes to persuasion and influence. This chapter will take a further
look at how behavioral principles can be used to influence visual design and
data interpretation. Specifically, we will focus on three psychological
concepts that apply to visualization: Context, Framing, and Gestalt theory.
Familiarizing yourself with these principles and techniques and learning
how to apply them effectively will greatly improve how you design
persuasive visualizations.
Behavioral science
The field of behavioral science is a broad set of specializations focusing
on theory and frameworks ranging from anthropology to psychology.
Insights from cognitive psychology and behavioral economic theory can be
useful in understanding how visualization techniques can affect human
judgment, especially when deciding how to visualize data and how to tell a
story. In real-life settings, important decisions are not just based on facts,
but also on subjective partialities. As we saw in the previous chapter, the
way information is presented, rather than what is presented, can be equally
if not more influential on how people interpret and act upon the data. As
Tversky & Kahneman explained in their seminal paper, written over 40
years ago, judgment and decision-making are heavily affected by
unconscious factors such as psychological, emotional, social, and
contextual factors [41]. Being aware of how these factors affect cognition is a
first step toward developing effective and ethical visualizations. In the next
section, we will start by exploring the power of context and how it plays a
crucial part role in data visualization.
Context
According to Thomas & Millar, context is one of the strongest factors of
influence on judgement and decision-making[42]. This is because our brain
uses all the senses (e.g. sight, sound, smell) to collect information about the
environment to create a coherent story of what is going on. In presentations,
for example, visual design, speaker voice, dress, body language, and the
physical environment all together communicate a ‘single narrative’ to help
the brain give meaning to the experience. The brain also tries to make sense
of all these inputs by relating them to our existing narratives (e.g. our
values, beliefs, and mental frameworks). In other words, individual
information is not (and cannot be) processed in isolation or independently
from its environment.
When speaking to an audience for the first time, the brain uses all of
these inputs to create an opinion about someone or something. The science
shows that this happens within the first 11 seconds of your encounter and
influences how people will further evaluate you or your presentation. That
is why, for example, when I give a presentation to an audience I have never
met before, I make sure to rehearse the first minute of my presentation
meticulously so that I can recite it word by word during the opening of
presentation, I suit up to look the part, make sure to face my audience
confidently, and if possible have someone else introduce me. Techniques I
know will significantly improve that first impression and consequently my
success in delivering an impactful presentation. Sir David Ogilvy, the
advertising pioneer, was a master at using context to influence his audience
and built his career upon it! In his 1963 biography ‘Confessions of an Ad
man’ he recounts several techniques he used. Next, let’s look at an example
of how context, in this case relativity, affects visual perception.
Principle of relativity
Take a look at the Fig 88. Which blue circle is bigger?

Fig 79 Context influence on perceived size.

You can already expect what the correct answer is: they are both the same
size. But perception-wise, they are not. The blue circle looks bigger when it is
accompanied by smaller circles around it and looks smaller when we put
bigger circles around it. Cover the outer balls with your fingers and you see
that both blue circles are exactly the same size.
This visual illusion mirrors how our mind interprets the world. It is
therefore unwise to ignore the broader impact of contextual factors on your
presentations. This effect not only holds true for charts, but also for ephemeral
things such as emotions, attitudes, and points of view. This means that our
brain uses both the external context (e.g. visuals, sounds, smells) as well as
the internal context (e.g. emotions, past experiences, desires) to interpret the
narrative as it happens.
The mental processing of contextual information, be it visual or not,
happens predominantly at the unconscious level. Less than 20% of our
awareness is conscious awareness, which means that factors such as context
can easily influence judgment and decision-making without people being
aware of it, having both intentional and unintentional effects. Let’s see with a
food example how context can influence eating behavior unconsciously.

Fig 80 Professor Wansink studied the effect that the plate’s size has on your
intake of calories. Source: Pelle Guldborg Hansen, inudgeyou.com

The same amount of food is presented on two different size dishes; one
being bigger that the other. Visually, the smaller dish makes the food look
bigger and studies have shown that people who eat from the smaller dish feel
full quicker and therefore eat less. People eating from the bigger dish tend to
eat more than the people eating from the smaller plate. Now think how this
applies to charts.
Cognitive Overload
According to Sweller, cognitive overload is a phenomenon during which
the brain is unable to process information effectively due to the sheer amount
of information presented[43]. Cognitive overload causes mental fatigue and
reliance on mental shortcuts (which are subjected to heuristics and biases). It
is therefore important that design incorporates visual elements which help
overcome cognitive resource limitations and prevent faulty decision-making.
As we saw in the Magic Quadrants in Ch. 5 the main mechanisms to avoid
overload are:
1. Limit the amount of data presented on a table or slide
2. Structure your data into bite-size pieces of relevant
information, and
3. Highlight key words and phrases in sentences which help the
reader pick up the most important parts of the text
Framing
Triggering mental shortcuts in visual design can also be beneficial to
improve visual effectiveness. One way mental shortcuts can be triggered in
visualization is through framing. Framing is a cognitive bias which affects
judgment and decision-making using positive (gain) or negative (loss)
messaging (see again Kahneman’s work[44]). People tend to be more
motivated to take action when messages are framed as either positive (a
gain) or negative (a loss) depending on the situation. For example, studies
show that people are more likely to seek risks when a message is framed as
negative or a loss. Here is an example.

Fig 81 Avoiding deaths (loss thinking frame) leads to more persuasion than
phrasing a policy proposal as “saving lives” (a gain thinking frame).
Source: bi.team

According to the above diagram, government officials were three times

more likely to take riskier options to combat a disease outbreak when policy
design was framed as preventing lives lost versus lives saved. In medical
situations, studies show that it wiser to use positive framing, instead of
negative framing, to help patients make better health-decisions for
themselves, especially in high-uncertainty situations. Interestingly, recent
findings in framing effects have found that certain factors might moderate
the impact of framing. For example, a study by Thomas and Millar in 2011
showed that framing effects change with age, affecting decision-making and
memory of younger and older people differently. Framing effects have also
been found to disappear when information is presented in a second
language or a language which does not come natural to the reader. This can
be explained by the fact that more deliberate thought is needed to process
information in a foreign language which reduces the brain’s tendency to
automatic (biased) responses.
Gestalt

Fig 82 Four of the eight Gestalt principles. Source: Annemarie Bufe

In the early 20th century, a group of German and Austrian psychologists

provided an understanding of how the human brain interprets and organizes
visual elements. Their investigation lead to the creation of the Gestalt
Principles for better visual design to reflect the brain’s natural way of
processing information. Gestalt is a German word which means pattern or a
unified whole. Let’s take a look at how some of these principles can be
applied to data visualization.
The Law of Proximity
The law of proximity reflects the need to group similar items together or
in close proximity to each other. Grouping similar items together makes it
easier for the reader to understand what is being communicated. The closer
parts are placed together, the higher the perceived relatedness. The mean
reversion chart, for example, in Chapter 4 (Fig 38), is a great example of the
law of proximity. The closer the label “Japan” is to its corresponding dot,
the less the confusion and cognitive load. Tip: Watch out not to put
unrelated objects too close to each other, as the eye will aim to create
relatedness based on the proximity of items presented.
Enclosure law
Enclosing elements together creates relatedness. Apply enclosure to
highlight certain elements in a visual. Enclosure could reflect the usage of
(dotted) circles on a scatter plot to indicate belonginess amongst specific
pieces of information (e.g. clusters, outliers). In the Fig 83-84 below we
have added an example of enclosure. The circular enclosures of Fig 84 help
reduce cognitive load of Fig 83 and increase its meaning.
Fig 83 A scatter plot with two classes of data points.
Fig 84 Enclosures, use to make meaning.
The Peak–End rule

Fig 85 Peak-end rule, use to maximize impact.

Finally, let’s see a fascinating psychological principle called the peak–

end rule, a psychological heuristic in which people judge (or remember) an
experience largely based on two specific moments: a peak moment
somewhere in the middle of the experience and the experience at the end,
rather than evaluating the experience in its totality. This surprising effect
occurs regardless of whether the experience is pleasant or unpleasant. In
other words, when storytelling your visual, the audience will remember the
experience more favorably if it ends on a high note (save the loudest bang
for last!). Fig 85 shows the intensity of two presentations by Mr. A and Mr.
B. According to this rule, the presentation of Mr. B will be rated nearly as
favorably as Mr. A’s presentation even though Mr. A presentation was
longer and had more Aha moments.
Steve Jobs used this rule to maximize the impact of his own
presentations by always ending in a high-note. His “One more thing…” is
now part of the pop culture. Throughout 1998 to 2010, Jobs used it in no
less than 25 occasions. Greg Wyatt of @appleexplained made an hour-long
compilation of such moments. The peak-end rule not only can be applied to
presentations but to charts as well. An example is Fig 65 in “twists”. There,
(at the expense of readability), the author maximized the chart’s punch by
making sure that the Aha moment is not reached until the audience went
through all the other elements of the chart first.
 Chapter 7. Detecting Bias
with Marybeth Sandell

Fig 86 Bias is unethical when it is unfair, usually against minorities.

In this chapter, you will learn what bias is and how it can affect a data-
driven visual. Bias not only can be sorted by their point of entry (data, story,
narrative) but also by the area they exploit in the cognition system (optical
illusions, cultural biases). It is easy to assume that bias is intentional.
However, bias can emerge for many reasons.
 First, bias can be embedded in the data itself, intentionally in the way it
is gathered but also accidentally by not realizing what is missing.
Second, bias can appear as the story is crafted. Again, this can be
intentional by cherry-picking from existing data, or accidental from cases
where not enough time is spent exploring all data available (usually due to
time pressure).
Third, it can be embedded in the narrative itself. Often this is intentional,
as in propaganda. But it can also be unintentional as in cultural bias.
Types of bias
In broad terms, bias is any systematic error. In other words, a systematic
difference between a model and the “truth” it supposedly represents. In
social sciences bias is judged to be unethical when it is unfair (usually
towards a minority). See also ethical frameworks in Ch. 1.
Bias can affect the producer of a visual (as in selection bias in data), but
also the consumer of the visual (as in Groupthink, and the hot hand fallacy,
and so on). Psychologists and behavioral economists have identified more
than 200 types of cognitive biases. Those can be classified in three groups:
Belief bias
Social bias
Memory bias
In addition, a part of the mentioned cognitive biases, when dealing with
data visualization, visual perception biases also apply. Let’s see some
examples.
Bias in narrative
The broadest forms of unconscious bias are due to unawareness and are
so rooted in society they usually are cultural (moral) norms too. Note that
not all cultural norms are biased but that most norms evolve slower than
society does and thus are usually lagging behind reality. Examples of
conscious narrative bias are: Propaganda and disinformation. Typical
techniques used are FUD (Fear, Doubt and Uncertainty) as seen with the
tobacco industry and FLICC (Fake lies, Logical fallacies, Impossible
expectation, Cherry-picking and Conspiracy theories) as seen with the
climate change denial[45]. Let’s see an example.
Bias in narrative: A balanced meal?
In the second half of the 20th century, a balanced diet was assumed to be
optimal for health. In school, many kids (myself too) were shown charts
with relatively balanced food groups. Yet, other cultures and a few
independent research papers show that perhaps that balanced food narrative
isn’t the healthiest one. For example, Okinawan diet contains less than 5%
of animal protein and no milk derivatives. Their diet would be considered
dangerously unbalanced by any Western standards. However, Okinawans
report one of the healthiest and longest lifespans in the World.
Fig 87 Logical fallacy? How could this “balanced” diet not be healthy if it
is balanced? Source: US Department of Agriculture choosemyplate.org

Fig 88 Okinawa diet in 1950 is correlated to the world’s longest and

healthiest life spans. Does it disprove the popular belief that an unbalanced
diet cannot be healthy?
Bias in the story: Hard working Germans?
When telling a visual story, how the data is used and presented can
induce conscious and unconscious biases (well known in advertising and
neuropsychology). This can be done through the selection and the
presentation of data, colors used and so on. Let’s see an example.
Exercise: Hard-working Germans
Fig 89 is a graph created by the German Economic Development
Agency. They used EU data to show that Germans work longer (harder)
than the average. But how much harder? Can you identify where is the
bias? Time 4 minutes.

Fig 89 A partial view?

Solution

Fig 90 A fuller view.

In the chart, Germany optically doubles France. In reality, the difference

is less than 10% because the axis starts at 36. A website redrew the graph to
start at zero and suddenly, German doesn’t stand out much at all. This chart
might also suffer from data biases that for a trained statistician are obvious.
For example, (due to historical reasons), Germany has more industrial jobs
than any other EU country. Germany has more double the amount of
industrial jobs per capita than Spain for instance. If industrial jobs have
different regulation or a different mix of full time / part time workforce than
other sectors (service jobs) then the chart is not comparing apples to apples.
The chart just reflects a different job mix in each country. This type of bias
is called data skewness and it is one of the common errors of amateur
statisticians: to confuse correlation with skewness. Let’s learn more about
data bias in the next example.
Data bias
In journalism, bias in data comes under three broad umbrellas: selection
flaws, skewness and omission. Some examples include: how questions are
asked (as in leading questions); bias in hypothesis or assumptions;
mathematical errors and survey design flaws. Let’s see an example of a
famous selection bias[46] discovered by the student Thomas Herndon[47].
Bias in data: Austerity controversy
Growth in a Time of Debt (2010) was a non-peer reviewed paper
published by Carmen Reinhart and Ken Rogoff - a former chief economist
of the International Monetary Fund. It said that a country’s economy slows
significantly when debt rises above 90% of GDP. This is not true. However,
the paper was used to justify austerity for the sake of the “common good”,
(see ethical frameworks in Ch. 1). A student trying to replicate the results
discovered that from the 20+ countries considered to make the statistic, key
countries had been arbitrarily omitted in the calculations. Later, Prof.
Michael Ash noted another unbecoming behavior in the weighting average
method used. For example: “New Zealand's (a country of population less
than 5 million people) that had one single year, (1951), at -8% growth was
held up with the same weight as Britain's 20 years at 2.5% growth
(population 60 million)”. On a dollar basis, that is a bias ratio of 1 to 240.
By now we have seen three examples of bias in the story, data and narrative.
Now let’s see a few more biases often at play in charts.
Other biases
Confirmation bias – we tend to believe statements and data that
confirm our beliefs and ignore evidence we dislike, (see previous section).
Picture superiority effect – presentations with charts have more
credibility.
Self-generation bias – we tend to remember what we say better than
what others say. So, if you ask a question during your visual presentation,
and let the audience come up with the conclusion they will remember it
more.
Illusion-of-truth bias – we are more likely to identify statements we
hear in repetition as true. Just because it’s been retweeted by multiple
sources doesn’t mean it is correct. For example, the previous austerity paper
was cited more than 2000 times before the bias was uncovered.
Optical deception – Another way in which visuals can be misleading
are optical deceptions. In 2015, in the paper “How Deceptive are Deceptive
Visualizations?”, Dr. Pandey studied how such manipulations are used in
charts [48]. As we have seen in the German working hours’ example, most of
the manipulations in charts consist in playing with axes zoom levels and
praying that no one will notice it. Note it is not an excuse to say that we
didn’t hide the scale of the axis, the way Panday measures deception is by
measuring what people understood after watching a visual, not what its
written in the visual.
The Amazon forest fires case
Bias can be found at multiple levels. For example, look back at the fires
of the amazon forest in Brazil in the summer of 2019. This story dominated
the major news outlets for a few weeks as some readers expressed outrage
the impact the fires would have on the earth’s climate. Apply what we
learnt so far to find potential biases in the following chart. Time 5 minutes.
Exercise

Fig 91 Where is the bias?

Solution to fires
Data was central to this story. Just how bad were the fires? There was
one main data source — the National Institute for Space Research
(INPE[49]). Initially, the news coverage presented data that encouraged the
reader to be outraged and suspect a crisis was at hand. The chart displays
data on the number of fires from 2013 to 2019. The graph leads the reader
to think that this was the highest level of fires ever. In fact, CNN on Aug.
22 wrote a story that the forest was “burning at a record rate”. The first
sentence of that CNN story said the rate of burning was a record “since
INPE began recording tracking fires in 2013”. Between the use of words
like ‘double’ and ‘record’ and the use of a visual with the 2019 bar looming
over all the other years shown, the narrative was set. To be sure, agencies
like CNN added to their written story lead: “...and scientists warn it could
strike a devastating blow to the fight against climate change.”
 Fig 92 Good journalism?

However, a mark of due diligence is to always ask if there is more

data. Often it is there but not available online, or not free online. It may be
in a drawer or behind a paywall. It may exist but the time-starved reporter
didn’t have time to look, or even possibly intentionally ignored it.
 Fig 93 Better journalism.

In this case, INSE did indeed have more data regarding number of fires
(it was just not online!). The BBC updated its chart to include more years.
Here is how it looked when digging further into the non-online archives.
Still, is this the best data we can get? Is there more? Consider what is
being measured: number of fires. Does this mean that if we light five small
fires today and one big one tomorrow, the total number of fires is declining?
Then, The New York Times published the chart below. Here, we have the
same source, INSE, but the measurement is square miles burned. The
numbers used initially weren’t wrong, but rather they were not complete or
fully reflective of the situation, a case of unconscious selection bias.
Due diligence checklist
1. Pay attention to words
Any good set of data will offer transparency into the methodology of
how the data was gathered. This means paying particular attention to what
and how questions asked in surveys or statements made. A red flag is any
use of adverbs and adjectives. They are usually loaded with bias.
2. Follow the money. Who paid for the research?
Big tobacco showed us that the organization that pays for the research
can control its results. For example, the egg industry lobbyists are paying
for research at accredited universities to promote research that says eggs
won’t boost bad cholesterol in humans. In the 1960s, the sugar industry paid
researchers to produce data that made consumers believe fat was a bigger
health hazard than sugar. The list goes on. Next time you see research about
health or the environment, try to discover the identity of the ultimate
financial backer. Follow the money.
3. Pay attention to the statistical methods used
As we saw, sometimes the data is being selected to intentionally support
a position. After performing some statistical analysis, a good rule of thumb
is to always ask a more proficient data scientist to find flaws. It works
wonders and one can learn a lot.
4. Consider the availability of data
Just because the data isn’t publicly online it doesn’t mean it is non-
existent. Post millennial journalists who were never taught how to do
research before the internet existed are particularly vulnerable to this bias.
Quiz: Fire Tweets

Fig 94 Where is the problem in this tweet?

 Answer: The photo is fake. Use Search google for photo to find out that
this is Aerial view of the Taim Ecological Station on fire, in Rio Grande do
Sul state, southern Brazil, on March 27, 2013. It is not from the Amazon
forest fires of 2019. NYTIMES wrote on Aug. 23 about how misleading
visuals went viral during the Amazon forest fires.
Fig 95 Direction pole metaphor, more human-centric than a list.
 Acknowledgement
This book received help from many friends. Special thanks to Cheryl
Shuttle, Ricardo Bragança, Liz Ortiguera, Agatha Braja, Kumiko Sasaki,
Louise Taylor, Natalia Rodriguez, Eric Nylen, Ferran Pujol, Eva Mitchell,
Cliff Hazel, James Piecoway, Amal, the Kaggler user Head or Tails and
Paul. Warm thanks to Anthony Goldblom, Carolina Ampuero, Francois
Cadiou, Nicolas Paris, Alex Rios, Steven Anthony, DiDi library, the 2019
digital government conference organizing committee, SNASK, Susanna
Vento, Kai Bruns, 710stories, the team at Ivory & Gold, and the Kaggle
team. The ebook style is inspired on the 3rd edition of “Don’t make me
think’.
 References & Footnotes

[1]
Own estimates [go back]

[2]
Rose, T. (2016). The end of average: How to succeed in a world that values sameness. Penguin
UK. [go back]
[3]
Rand, P., 1985. Paul Rand: A designer's art. New Haven: Yale University Press. [go back]
[4]
Isaacson, Walter. Steve Jobs. 2011. [go back]
[5]
“It takes too long to style a chart like you want by using R or Excel. At the end, it is faster to use
Adobe Illustrator” – Marybeth Sandell, former bureau chief London, Bloomberg. [go back]
[6]
Berengueres, J., 2019, June. Visualization & Storytelling Workshop. In 20th Annual International
Conference on Digital Government Research (pp. 532-533). ACM. [go back]
[7]
See also “how to detect genius ideas” in Ogilvy & Advertising. [go back]
[8]
See also Tanaka, K., 1997. An introduction to fuzzy logic for practical applications. [go back]
[9]
See also the film Merchants of Doubt and Cambridge Analytica scandal. [go back]
[10]
Gallo, C. 2019. The Art of Persuasion Hasn’t Changed in 2,000 Years. [go back]
[11]
Reynolds, George. Ethics in information technology. Nelson Education, 2011. [go back]
[12]
Jefferys, S., 2003. Liberté, Egalité and Fraternité at work: changing French employment
relations and management. Springer. [go back]
[13]
Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., & Cohen, J. D. (2001). An fMRI
investigation of emotional engagement in moral judgment. Science, 293(5537), 2105-2108. [go back]
[14]
In 2006, Al Gore hired Nancy Duarte to make his famous CO2 chart presentation. A scissor lift
was used on stage to show that the CO2 is “off the charts”. Al Gore was later criticized by emitting a
lot of CO2 due to his jet travelling. [go back]
[15]
See also David and Goliath in the book of Samuel. [go back]
[16]
Thunberg, G., 2018. Speech at UN Climate Change COP 24 Conference, Poland 2018. Published
online on YouTube by Connect4Climate. [go back]
[17]
Hyman, J., 2006. The objective eye: color, form, and reality in the theory of art. University of
Chicago Press. [go back]
[18]
Schippers, M. C., Scheepers, A. W., & Peterson, J. B. (2015). A scalable goal-setting intervention
closes both the gender and ethnic minority achievement gap. Palgrave Communications, 1, 15014.
[go back]
[19]
Dweck, C., 2015. Carol Dweck revisits the growth mindset. Education Week, 35(5), pp.20-24. [go
back]
[20]
Corbett, Christianne, and Catherine Hill. Solving the Equation: The Variables for Women's
Success in Engineering and Computing. American Association of University Women. 1111 Sixteenth
Street NW, Washington, DC 20036, 2015 [go back]
[21]
Norman, D., 2013. The design of everyday things: Revised and expanded edition. Basic books.
[go back]
[22]
Mourant, R.R. and Rockwell, T.H., 1972. Strategies of visual search by novice and experienced
drivers. Human factors, 14(4), pp.325-335. [go back]
[23]
See also “Generations in the workforce” [go back]
[24]
Sinton, E (2011). ‘Baby boomers are very privileged human beings’
https://www.telegraph.co.uk/finance/personalfinance/pensions/88409
63/Baby-boomers-are-very-privileged-human-beings.html retrieved October 23, 2013 from
www.telegraph.co.uk [go back]
[25]
] Lukianoff, G. and Haidt, J., 2018. The coddling of the American mind: How good intentions
and bad ideas are setting up a generation for failure. Penguin. [go back]
[26]
Palmore, Erdman. Ageism: Negative and positive. Springer Publishing Company, 1999. [go
back]
[27]
Backes‐Gellner, U. and Veen, S., 2013. Positive effects of ageing and age diversity in innovative
companies–large‐scale empirical evidence on company productivity. Human Resource Management
Journal, 23(3), pp.279-295 [go back]
[28]
https://en.wikipedia.org/wiki/Affluence_in_the_United_States [go back]
[29]
https://www.epi.org/blog/top-1-0-percent-reaches-highest-wages-ever-up-157-percent-since-
1979/ [go back]
[30]
https://www.kaggle.com/harriken/kaggle-journey-2017-2018 [go back]
[31]
Cowan, N., 2010. The magical mystery four: How is working memory capacity limited, and
why?. Current directions in psychological science, 19(1), pp.51-57. [go back]
[32]
Wynn, T., 2002. Archaeology and cognitive evolution. Behavioral and brain sciences, 25(3),
pp.389-402 [go back]
[33]
. [go back]
[34]
Galton, F., 1886. Regression towards mediocrity in hereditary stature. The Journal of the
Anthropological Institute of Great Britain and Ireland, 15, pp.246-263. [go back]
[35]
Gandhi, P., Khanna, S. and Ramaswamy, S., 2016. Which industries are the most digital (and
why). Harvard Business Review, 1. [go back]
[36]
Stewart, P.J., 2019. Mendeleev’s predictions: success and failure. Foundations of Chemistry,
21(1), pp.3-9. [go back]
[37]
Source: Berengueres, J., 2015. The Brown Book of Design Thinking: A workshop based
approach. UAE University College. [go back]

[38]
Johnstone, Keith. Impro for storytellers. Routledge, 2014. [go back]
[39]
Norman, D., 2013. The design of everyday things: Revised and expanded edition. Basic books.
[40]
Berengueres, J., 2007. The Toyota production system re-contextualized. Lulu. com. [go back]
[41]
Tversky, A. and Kahneman, D., 1974. Judgment under uncertainty: Heuristics and biases.
science, 185(4157), pp.1124-1131. [go back]
[42]
Thomas, A.K. and Millar, P.R., 2011. Reducing the framing effect in older and younger adults by
encouraging analytic processing. Journals of Gerontology Series B: Psychological Sciences and
Social Sciences, 67(2), pp.139-149 [go back]
[43]
Sweller, J., 1988. Cognitive load during problem solving: Effects on learning. Cognitive science,
12(2), pp.257-285. [go back]
[44]
Kahneman, D. and Tversky, A., 1981. On the study of statistical intuitions (No. TR-6).
STANFORD UNIV CA, DEPT OF PSYCHOLOGY. [go back]
[45]
Cook, J., Supran, G., Lewandowsky, S., Oreskes, N. and Maibach, E., 2019. America Misled:
How the Fossil Fuel Industry Deliberately Misled Americans about Climate Change. [go back]
[46]
Because the authors of the paper refused to retract their beliefs after the data showed otherwise,
this is also evidence of confirmation bias. [go back]
[47]
Herndon, T., Ash, M. and Pollin, R., 2014. Does high public debt consistently stifle economic
growth? A critique of Reinhart and Rogoff. Cambridge journal of economics, 38(2), pp.257-279. [go
back]
[48]
Pandey, A.V., Rall, K., Satterthwaite, M.L., Nov, O. and Bertini, E., 2015, April. How deceptive
are deceptive visualizations?: An empirical analysis of common distortion techniques. In
Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (pp.
1469-1478). ACM. [go back]
[49]
The National Institute for Space Research (Portuguese: Instituto Nacional de Pesquisas
Espaciais, INPE). It is a research unit of the Brazilian Ministry of Science, Technology and
Innovation. [go back]