9 Visualization Toolbox

There are many types of data visualizations and many variations on each 9.1 Exploration Visualizations 179
type. These data visualizations can collectively be thought of as tools in the Scatterplots . . . . . . . . . . 179
data visualizer’s tool box, and good data visualizers will be as familiar with Rug Charts and Histograms 180
them as a master wood worker is with their tools of their trade. Two-Way Tables . . . . . . . 182
9.2 Presentation Visualizations 183
The only way this can come about is through practicing with each type and Text Blocks . . . . . . . . . . 183
variation of visualization, using either teaching datasets or by finding real Tables . . . . . . . . . . . . . 184
world use cases. Line Graphs . . . . . . . . . . 184
Bar Charts . . . . . . . . . . . 185
It also helps to have a good organization to the tool box, which can serve to 9.3 The Rest of The Landscape 188
direct the visualizer to the right general type of visualization. Maps and Heat Maps . . . . 188
Bubble Charts . . . . . . . . 189
As with organizing a tool box, there is no one right way to group data
Small Multiples . . . . . . . 189
visualizations together, and to some extent it is a matter of personal preference. Area Charts and Treemaps . 191
However, some experts in this field have made efforts on this front, and it’s Text Visualizations . . . . . 191
possible to see some commonalities. Parallel Coordinates . . . . . 193
Trees and Networks . . . . . 193
As seen in Figure 9.1, one approach to organizing data visualizations is to
Animated Visualizations . 193
consider which ones best highlight:
9.4 Misc. & Charts to Avoid . . 197
a relationship – show a connection or correlation between two or more Chernoff Faces . . . . . . . . 197
variables,1 such as the impact of an aging population on health care;
Alluvial Diagrams . . . . . . 197
Charts to Avoid . . . . . . . . 198
a comparison – set some variables apart from others, and display how
those two variables interact, such as the number of fans attending
hockey games for different teams in a season;
a composition – collect different types of information that make up 1: Also called dimensions, axes, factors,
etc.
a whole and display them together, such as the various search terms
that visitors used to land on your site, or how many visitors came from
various sources (links, search engines, or direct traffic), and
a distribution – lay out a collection of related or unrelated information
to see how it correlates (if at all), and to understand if there’s any
interaction between the variables, such as the number of bugs reported
during each month after a new software release.

However, this is not the only way to think about data visualizations. Some
practitioners have broken down these categories further, as shown in Figure
9.2. As yet another alternative, it’s possible to consider which combinations
of the 5W questions (who, what, when, where, and how/why) certain data
visualizations are best suited to display, as was shown in Figure 2.14.
 178 9 Visualization Toolbox

Figure 9.1: A Classification of chart types, based on visualization objectives [J. Camoes ].

Figure 9.2: An alternative way to group chart types [D. Hull , A. Abela ].
 9.1 Exploration Visualizations 179

The take home point here is that while it’s possible to group data visualization
types in a number of ways, it is important to hone our own sense of the most
appropriate visualizations for particular situations, which may be informed
by schema developed previously, and so on (see Figures 9.1 and 9.2).
As we have already discussed in Chapter 2, regardless of our preferred
organizing principles, there are some data visualizations that are true
workhorses – they are usable in some way with almost any dataset, will
be familiar to most lay people and are particularly useful for exploring or
presenting data. Other visualization types have more situation-specific uses,
and may be difficult to use or ineffective in other situations – practice and
discernment are required to use them skillfully and appropriately.
Different people may consider different options for a list of workhorse
visualizations. For instance, we could trot out the following “old-faithfuls”:2
2: Another selection has already been re-
viewed in Chapter 2.
Data Exploration
Scatterplots
Rug charts
Histograms
Two-way tables
Data Presentation
Text blocks
Tables
Line graphs
Bar charts
These categorizations – exploration vs. presentation – are not intended to
be hard and fast: at times, we might use a barchart for exploration and a
scatterplot to present a key finding, say. Everyone will develop their own
approach for the use of these visualizations, but we provide some comments
and pointers as guidelines.3 3: As a rule of thumb, these lists provide a
good starting point when deciding which
tools to pull from the toolbox first.

9.1 Exploration Visualizations

Scatterplots

A scatterplot is one of the fundamental tools of data scientists. In a rapid

and accessible manner, it can reveal key relationships between two variables
simply by plotting the data points on a grid.4 4: Which we technically refer to as a Carte-
sian plane, a two dimensional cartesian
When using a scatterplot for data exploration, the emphasis lies not only coordinate system.
on the aesthetics of the plot, but also on what a simple version of a plot
show about the nature of the relationship of the variables being plotted – in
particular, the emergent patterns. Do a circular cloud of points, a straight
line, a diagonal line, a wavy line appear? All of these different patterns denote
different potential relationships between two variables which could be worth
exploring further.
 180 9 Visualization Toolbox

But caution must be exercised when using scatterplots for data presentation.
Because they can represent all of the points in a dataset, there is a risk for
clutter (and overwhelming the consumer). The message can easily get lost.
Consequently, we suggest only using scatterplots for communication when
the pattern is naturally clear and relevant to the broader context of the story
being presented.
Bubble charts, which are a variation on scatterplots, can communicate
relationships between multiple dimensions and provide a powerful tool to
render multivariate relationships. We will discuss them further in a coming
section of this chapter.
Scatterplots (and bubblecharts) are most commonly used for quantitative
data, but it is possible to have one or both of the axes represent a qualitative
variable, using an approach similar to that taken on the horizontal axis
(x-axis) of bar charts. This can lend itself to misinterpretation, however.
Lastly, it is not uncommon for scatterplots to be overlayed with a trend line
or curve, such as in the data storytelling tropes of Section 8.3 (Evolving a
5: Creating trend lines involves calcula- Storytelling Chart).5
tions over the data points represented; this
can be automated by the software used to TL;DR Summary and Comments:
render the chart, as we will see in the next
few chapters. plots show relationship between 2 variables (scatterplot) or 3 variables
(bubble plot)
we can use average lines (or similar curves) to provide context
consider using groupings to add clarity (e.g., colour gradients)
colour and geometry allow us to plot (at least) 2 extra variables on a
2D scatterplot
the data may need to be re-scaled or binned
a movie could be used to visualize an additional ordinal variable
text can also be added to visualize an additional categorical variable
works best when chart is not too encumbered
Examples of scatterplots (and bubble charts) are provided in Figures 9.3 and 2.7.

Rug Charts and Histograms

A rug chart is essentially a single quantitative variable that is plotted on a

horizontal or vertical number line (see Figure 2.4). It is easily overlooked as
a strategy to understand the behaviour of each variable separately from any
of the other variables in the dataset. In some ways it can be thought of as a
“quick and dirty histogram”, since histograms also focus on a single variable,
6: Rug charts are useful building blocks albeit in a slightly more nuanced fashion.6
for more complex visualizations like radar
or spider charts. Among data visualization approaches, histograms are at once one of the
most familiar and one of the least well-understood. It is invaluable for data
exploration, but can be treacherous when used in data presentations, to the
point that to the question of when a histogram should be used in a data
presentation context, our answer is: never, probably.
 9.1 Exploration Visualizations 181

Figure 9.3: Scatterplots and bubble charts: personal collection (top row); Medium (middle left); Towards Data Science (middle right); Ottawa
Senators player usage, 2016-2017 (Hockey Abstract , bottom row).
 182 9 Visualization Toolbox

As with the number line, the histogram focuses on a single quantitative

variable. Once it has been selected, we must then carry out some calculations
on this variable. If {𝑥 𝑖 | 𝑖 = 1 , . . . , 𝑛} are the values taken by the variable in
the dataset, then its histogram should contain the following information:
the range of the histogram is 𝑟 = max{𝑥 𝑖 } − min{𝑥 𝑖 };
√
the number of bins should be ≈ 𝑘 = 𝑛 , where 𝑛 is the sample size;
the bin width should approach 𝑟/𝑘 ,
and the frequency of observations in each bin is then used to determine
the height of the bar.
By abstracting away from the “bumpiness” of the individual bars, it is possible
7: Histograms are related to bar charts, to get a general sense of the abstract variable shape in the dataset.7 Examples
which we will discuss shortly. of histograms and rug charts are shown in Figures 9.4, 2.5, and 2.6.

Figure 9.4: Examples of histograms and

rug charts: frequency of daily number of
road accidents in Sydney, Australia, over
a 40-day period (left); grade distribution
in a 2nd year probability and statistics
class, with mode (black), median (blue),
and mean (blue) overlay, density curve,
and rug chart (right).

Two-Way Tables

Most of our discussion on exploration visualizations has focused on quantita-

tive variables; a straightforward preliminary strategy for exploring qualitative
8: Or 𝑚−way table, more generally. variables is through the use of a two-way table,8 which displays the counts
of one variable level relative to the counts of a second variable level, as is
illustrated in Figure 9.5.

Figure 9.5: Example of a two-way table for

an artificial dataset with 89 observations
and 2 variables: window type and window
size; for exemple, the are 11 medium-sized
door windows among the 80 observations.

We could use 𝑛2 pair-wise 2−way tables to display information for a dataset

with 𝑚 = 3 categorical variables (speed, size, season), there would thus be
three 2-way tables: speed × size, speed × season, and season × size, say.
large medium small autumn spring summer winter large medium small

high 13 56 73 high 32 34 38 38 autumn 19 33 28

low 32 24 2 low 16 13 12 17 spring 21 34 29
medium 38 56 46 medium 32 37 36 35 summer 19 36 31
winter 24 33 33
 9.2 Presentation Visualizations 183

But this only presents a part of the picture. The 1−way tables provide a
univariate summary:

season autumn spring summer winter

80 84 86 90

size large medium small

83 136 121

speed medium high low

140 142 58

while the 3−way table speed × season × size provides the full picture, but it
is not the only way to represent it:9 9: What would the other combinations
(size × speed × season, etc.) look like?

speed = high speed = low speed = medium

large medium small large medium small large medium small
autumn 3 13 16 autumn 9 6 1 autumn 7 14 11
spring 3 14 17 spring 7 6 0 spring 11 14 12
summer 3 16 19 summer 6 6 0 summer 10 14 12
winter 4 13 21 winter 10 6 1 winter 10 14 11

9.2 Presentation Visualizations

Text Blocks

The simples of the presentation visualizations, the text block, may not even
seem like it belongs in a list of data visualizations, given its lack of visual
elements outside of the written word. However, when treated as a graphical
element, where the focus is on a fact containing one or two numbersat most,
they are excellent at “setting the scene”.
This is particularly useful in a dashboard or in a report context, where text
blocks can be used to draw the focus to an area of the report which contains
a more detailed breakdown or analysis of the data in question.

Figure 9.6: Bar chart (left) vs. text block

(right): in this case, the chart is overkill –
the insight is much more easily conveyed
with text.

Don’t neglect this simple approach to conveying insights.

 184 9 Visualization Toolbox

Tables

Tables are another text-heavy visualization which interact with our ver-
bal system: we read them. They are useful for comparing values across
variables.
One complicated aspect of tables is that the audience has considerably leeway
in regard to how they elect to read them: they may focus on the relationship
between numbers across each row, or down each column. Furthermore, if
the data relates to specific individuals (or units) audiences are expected to
be most interested in, they will certainly look for and focus on their rows,
10: Designers may need to use the Gestalt potentially to the exclusion of everything else.10
guidelines of Section 4.2 to draw the audi-
ence’s eye to another location in the table. Importantly, table design should blend into the background.11 It is the data
11: Although it seems to go against easily- that should stand out, not the borders – if you must display large, dense
accessible MS PowerPoint templates... tables,12 consider alternating the table row colour from white to a very lightly
12: Must you, really? noticeable shade to help the eye scan across the rows and seperate one row
from another.

Figure 9.7: Fanciful table (left, not recom-

mended) vs. simple table (right, recom-
mended).

The table heat map provides a variant on the table, where cells contain
a colour as well as a number, and in which the colour is leveraged to
convey magnitude, by mapping the colour hue and saturation to the cell
13: A single colour saturation with a leg- value.13 Eventually, the numerical values or cell text labels may be removed
end (white = low, blue = high) is preferable without altering the message, leading to a more holistic reading of the data
to colour differentiation (rainbow scale).
visualization.

Figure 9.8: From table (left) to heat map table (middle) to holistic heat map table (right).

Line Graphs

Line graphs have some similarities with scatterplots. As a consequence, it

can be difficult for people new to data visualization to appreciate how they
differ from the latter and to get a good sense of when to choose one over the
other. Both involve plotting data points on a grid, and both can accommodate
curve overlays, for instance – the critical distinction is that in the case of a
14: The dataset must thus contain an or- line graph, the line (curve) connects all the points in sequence,14 and passes
dinal variable that is used to create the exactly once through each of the points.
sequence of points, and typically at least
one quantitative value used to place the While some line graphs only display one line (curve), they can also be used
point on the chart.
when the dataset contains also at least one categorical variable whose levels
 9.2 Presentation Visualizations 185

Figure 9.9: Various line graphs and sparklines, from personal files.

can be used to separate the observations in different groups. Each of these

categories can then be given its own curve on the line graph, allowing us to
compare the behaviour of the qualitative values across the categories.
Although describing the mechanics of a line chart can be complicated, they
are typically familiar to a broad audience and so are likely to be more readily
interpretable than some other data visualizations (as in Figure 9.9).15 15: Sparklines are a line graph variant
which are meant to play the role of a
TL;DR Summary and Comments: “word” in a larger chart or paragraph [3].

line charts can show a single series or multiple series of data;

they particularly useful to display time series;
axis scale should be clear and relevant;
the 𝑦−axis should be “anchored” when using dynamic filters so that
the graph does not “jump around” as users interact with it.

Bar Charts

One of us16 thinks that the bar chart is the workhorse of all workhorse among 16: *cough* Jen *cough*
data visualizations. It is almost immediately familiar to most people and, if
used in an expected fashion, readily interpretable as well. Although some
people may hesitate to use a bar chart simply because it is so familiar and
frequently used, we believe that this is a strength, a weakness.17 17: If novelty is desired, there are many
variations on the bar chart, both with
The basic bar chart represents a single numeric variable broken down by the respect to what types of data are incor-
values of a categorical variable.18 porated and aesthetic and presentation
choices, that can add interest and nuance
These charts are quite versatile and useful. Apart from very rare instances, to the basic bar chart.
they should always have a zero baseline. When constructing them, we 18: See Section 2.2 for more information.
 186 9 Visualization Toolbox

recommend using either the graph axis or the data labels: axis for broad
19: Horizontal charts are apparently easier statements, data labels for detail information.19
to read, as we have discussed in Section 4
[38]. From a design point of view, the basic bar chart can be transformed in a
number of ways which impact how viewers interpret and prioritize different
20: Variations include: pie charts, gauge aspects of the visualization.20
charts, funnel charts, lollipop charts, wa-
terfalls, stacked bar charts, cluster bar Funnel charts are typically used to represent decreasing proportions amount-
charts, 100% bar charts, percentage bar ing to a 100% total.21 These can be very useful to help audience quickly
charts, etc.
prioritize items without having to actively filter the data.
21: Although that is not always the case.
Gauge charts are often used as a dashboard component (with or without
needle) – they typically display single value measures on the way to some
goal or key performance indicator (KPI), in a manner that can quickly be
scanned and understood. While gauge charts are particularly useful to show
22: Not that there is anything wrong with progress, they may ultimately prove to be a management fad.22
that, of course.

Figure 9.10: A funnel chart representing

the % of total sales for each salesperson
× product (fictional example, left); gauge
chart, with target (right).

Stacked bar charts are designed for comparing totals, but can quickly become
overwhelming. They are hard to sort and order. Filtering is complicated in
dashboard applications like Power BI because it is unclear how the chart
should respond when filter is applied.
100% bar charts work well for visualizing portions of a whole on scale from
negative to positive. They have a consistent baseline at each of the extremities
(either left/right, or top/bottom), making it easy to compare the bars. The
issue, however is that there is no relative measure of the magnitude of data.
As with other bar charts, research shows that horizontal is easier to process
than vertical.
Waterfall charts shows how the initial value increases or decreases using a
series of intermediate values; different colours should be used to represent
increases and decreases. One drawback is that it is difficult to remove charts
23: In other words, it is difficult to declut- elements without removing context.23 Note that large increases or decreases
ter waterfall charts. may look odd (as in Figure 9.11).
 9.2 Presentation Visualizations 187

Figure 9.11: Bar charts and variants: basic (top row); stacked bar charts (2nd row); 100% bar charts (3rd row); waterfall charts (bottom row).
 188 9 Visualization Toolbox

9.3 The Rest of the Visualization Landscape

As we have seen throughout (especially in Chapter 3), modern data visualiza-

tion endeavours often go above-and-beyond the workshorse visualizations.
In this section, we will present some of the more sophisticated approaches
used in data presentations.

Maps, Heat Maps and Choropleths

A more thorough treatment of geographical maps and map-based visualiza-

tions would probably require a chapter in its own right. Most of us are quite
familiar with geographical maps, so they tend to be easier to interpret; we
can play with this to produce a striking effect when the data visualization
shows unexpected results, and thus change the viewer’s perception in the
process (see Figure 9.12).

Figure 9.12: Maps, maps, maps: a sprinkle of maps and distortions – Canadian airports (top left, personal file); population cartogram (top right,
Paul Breding); global warming culprits, by population and by size (bottom row, New Scientist).

Heat maps are ideal when we want to look at the relationship between 3 or
4 variables. If one of these represents a percentage or a value within a set
range, it can be used to fix the colour scale, for comparison purposes. The
other variables are then used to locate and size markers on the display.
If the axes variables are continuous, it could still be preferably to bin them:
this decreases the number of required observations for usefulness. It is
typically easier to read such charts if colours are selected along natural
24: More sophisticated gradients can be colour gradients, such as White → Blue or Red → Black.24 When the
used (Red → Yellow → Green, say), but background canvas is a non-distorted geographical map, heat maps are
those are less than ideal from a Gestalt
perspective, or if some viewers are colour
known as choropleths (see Figure 9.13).
blind – this is another clear case where
“less is more”.
 9.3 The Rest of The Landscape 189

Figure 9.13: Heat maps and choropleths: The Horizon or Pedestrian Risk (J. Nelson, IDV Solutions, top left); basketball shooting charts
(NBAsavant.com , top right); Election choropleth (A.E. McCann, bottom left); Canadian population choropleth (Statistics Canada, bottom
middle); US elevation choropleth (author unknown, bottom right).

Bubble Charts

Unlike scatterplots, which have already been discussed both in Chapter 2 and
earlier in this chapter, bubble charts can serve to illustrate the interactions
between multiple variables (when used correctly). Importantly, however, they
are usually most useful when there relationships between the variables in
questions are strong, resulting in clear patterns in the chart. We must also
be careful when choosing how to represent the many variables involved –
there are likely more bad options than good choices on this front, and so
experimentation is likely to be required; examples can be seen in Figures 1.4
and 9.3.

Small Multiples

Combining multiple visualizations of the same type and presenting them

as tiles in a larger composite visualization can have a powerful effect – we
call such combined charts small multiples, after Tufte [3]. Depending on the
choice of visualization, the small multiple can depict change over time, as
in a flip book, or encourage high-level comparisons across categories (see
Figure 9.14).
 190 9 Visualization Toolbox

Figure 9.14: Small multiples: US electoral results choropleths, by year (author unknown, top); debt line graphs, by G7 country (Pew Research
Center, bottom).
 9.3 The Rest of The Landscape 191

Area Charts and Treemaps

Area charts use physical areas to represent various quantities, such as in

Figure 9.15.

Figure 9.15: Area chart with three cate-

gories.

We suggest, however, to try to avoid area charts, except in situations where

the plotted quantities have vastly different magnitudes as human brains
have a hard time attributing a value to a 2D area (see Section 2.4).25 25: Area maps (such as a pie chart) are
perhaps most useful when viewer need to
An exception to this warning might be the treemap, if used with care. A see that a quantity is much greater than
treemap can simultaneously show the big picture and compare categories another, but the numerical factor by which
it is not relevant.
(or sub-categories) easily. They are useful for prioritizing “big ticket items”
in dynamic dashboards.26 26: Although labeling and colouring can
be tricky... but that is potentially a problem
with just about any data visualization.

Figure 9.16: Treemap with five categories

and three sub-categories.

Text Visualizations

Not to be confused with text blocks, text visualizations use text attributes
(such as size and colour) to represent some other variable associated with the
words. For maximal impact, font size may be a function of frequency. These
visualizations are typically used for univariate categorical data, but small
multiples, cloud shape, word placement, colour, and hue could be used to
integrate more variates.
In many implementations, the word placement and colour choice algorithms
are “hidden” from the users. As an example use case, text visualizations can
be used to answer authorship questions.
 192 9 Visualization Toolbox

Figure 9.17: Text visualizations: word cloud (most pirated artists, 2007-2010 , top row); Ottawa Senators most frequently named players in AP
articles, 2016-2017 (middle row, first two charts); comparison of word usage in Shakespeare and Marlowe plays (middle row, last two charts);
various text visualizations for Shakespearean plays (bottom row).
 9.3 The Rest of The Landscape 193

Radar Charts and Parallel Coordinates

Although parallel coordinate charts, which stack and connect multiple rug
charts to show relationships between potentially large numbers of variables,
are a relatively obscure type of visualization, a variation has been increasing
in popularity in recent years.
Radar charts, which arrange the axes radially as spokes coming out of a
central point, are often seen in social science or business contexts, where they
are used to show survey results. When used in this manner, the overall shape
of the connected line on the radar chart gives a gestalt sense of the response
profiles.27 27: E.g., are they mostly low or mostly
high? and so on.

Figure 9.18: Radar chart and parallel co-

ordinate chart of NFL players Cruz vs.
Fitzgerald performance (A.E. McCann).

Trees and Networks

Using networks to both model and visualize systems can give us insights into
the system. Having a solid conceptual understanding of the system through
the use of these visualization types can help us draw legitimate and sound
conclusions.28 Examples are provided in Figure 9.19. 28: Here are some clues that suggest that
using a tree or network visualization could
be useful:

Animated and Interactive Visualizations are we dealing with flow (of some-
thing) along pathways?
are we dealing with a collection
Animation and interactivity do not always improve a visualization. What of objects that input and output
insights can they provide? That depends on the data, and on the visualization things?
method. are the inputing/outputing objects
homogeneous?
Even when done well, 85% of users don’t bother with interactive viz, according are we dealing with relationships
to a NY Times analysis of their own viusalizations at The Upshot. This and connections between objects?
Are we dealing with a situation
very strongly supports the notion that the default visualization (i.e., the one
where one object influences an-
that greets viewers when they first load the website on which it is found) other object?
should be coherent and self-consistent as is (see Figures 9.20 and 9.21 for
examples).
 194 9 Visualization Toolbox

Figure 9.19: Trees and network diagrams: disease progression [32] (top left); US airport hubs (top right, author unknown); classification [1]
(bottom left); tree of life (P.Z. Meyers, bottom right).
 9.3 The Rest of The Landscape 195

Figure 9.20: Animated and interactive charts: The Clubs That Connect the World Cup , NY Times, 2014 (top left); Who Marries Whom
, Bloomberg, 2016 (top right); Hipparcos Star Mapper , European Space Agency, 2016 (middle left); The Internet of Things – a Primer ,
Information is Beautiful, 2016 (middle right); Visualizing the Riemann 𝜁 Function and Analytic Continuation , 3Blue1Brown, 2016 (bottom row).
 196 9 Visualization Toolbox

Figure 9.21: Animated and interactive charts II: The Genealogy and History of Popular Music Genres , Musicmap, 2016 (top left); Sequences
Sunburst , Kerry Rodden, 2015 (top right); Health and Wealth of Nations , Gapminder Foundation (middle left); Small Arms and Ammunition
– Imports and Exports , Google, 2012 (middle right); Mobius Transformations Revealed , D.N. Arnold, J. Rogness, 2007 (bottom).
 9.4 Misc. & Charts to Avoid 197

9.4 Miscellanea and Charts to Avoid

Some data visualizations are sufficiently unique that they cannot easily be
grouped or categorized. 29: The idea is perhaps intriguing and
might even work well in some instances,
but in most cases it fails to provide a useful
Chernoff Faces rendering; among other issues, most facial
features are not ordinal, faces are more
than the sum of their parts, and not all
Consider, as a singular example, Chernoff faces, which were designed on the facial features carry emotions.
premise that people can easily understand facial expressions. The Chernoff
visualization can accommodate up to 18 or 36 facial feature variables.29

Figure 9.22: Chernoff faces of MLB managers characteristics during the 2007 season (SC. Wang, NY Times).

Alluvial and Sankey Diagrams

Alluvial and Sankey diagrams (see here and here for examples,
respectively) are similar in appearance to one another, and both allow for the
visualization of proportions; however, in the case of the alluvial diagram,
the focus is on datasets with multiple categorical variables, and the chart
displays the percentages of each variable relative to other variables.
Sankey diagrams, conversely, focus on quantity breakdowns relative to
particular categories and how those quantities change when considering
other categories.
 198 9 Visualization Toolbox

Charts to Avoid

One the one hand, we are agnostic when it comes to tools and methods:
anything that helps convey the data story is on the table. On the other
hand, some of the commonly-used approaches really put a damper on
comprehension.
We strongly suggest avoiding:
ANYTHING with an arc (except for gauge charts) such as pie and
30: Sometimes we need to be pragmatic... doughnut charts.30 Human brains cannot easily compare angles and
but there are limits. arcs, so these can become misleading: without labels, how easy is it to
compare Steve & Bob below?

3D visualizations, which we suspect are flat-out EVIL! As with arcs,

we cannot easily visually compare data series in a 3D context; such
charts are usually way too cluttered.

Note that there is always a danger that if certain types of visualization

techniques take over, the kinds of questions that are particularly well-suited
to providing data for these techniques will come to dominate the landscape,
which will then affect data collection techniques, data availability, future
interest, and so forth.