COMP 7990

Principles and Practices of

Data Analytics

Chapter4: Data Visualization

Eric Zhang Lu

Acknowledgements:
Lan Liang, School of Communication, HKBU
Katy Börner, School of Informatics and Computing, Indiana University, USA
Robert Putnam, Research Computing, IS&T, Boston University, USA
Outline
 Concepts of Data Visualization
– Value, goals
– History
 Elements of Visualization
– Choosing the Right Chart
– Color, Size, Text, Titles, Labels
 Design of Visualization
– A Visualization Workflow
 Case study: COVID-19
– Mask wearing trends
– People concern about impact of COVID-19
Outline
 Concepts of Data Visualization
– Value, goals
– History
 Elements of Visualization
– Choosing the Right Chart
– Color, Size, Text, Titles, Labels
 Design of Visualization
– A Visualization Workflow
 Case study: COVID-19
– Mask wearing trends
– People concern about impact of COVID-19
Value of Information Visualization

 Exploring information collections becomes increasingly difficult as the

volume grows

 With minimal effort, the human visual system can process a large amount
of information in a parallel manner

 The occurrence of advanced graphical software and hardware enables

the large-scale visualization and the direct manipulation of interfaces

4
Goals for Information Visualization

 Provide insight
– Explain data to solve specific problems
– Support the analytical task, showing the comparison or causality
– Explore large data sets for better understanding

 Relieve the cognitive overload

– Conveying abstract information in intuitive ways

In today’s world, successful decision-making has everything to do with

turning data insights into action. And because the goal of data
visualization is impact, not numbers.
5
What is Visualization?

 Visualization is any technique for creating images, diagrams, or animations to

communicate a message.
 Visualization through visual imagery has been an effective way to
communicate both abstract and concrete ideas since the dawn of man.
– Cave paintings
– Hieroglyphs
– Maps
– …

6
Brief History

 First known map: 12th century (Tegarden,1999)

 First presentation graphics: William Playfair (1786)
 Multidimensional representations appeared in 19th century (Tufte, 1983)
 Examples:
– William Playfair (1821) – Chart showing at one view the price of the quarter of wheat, &
wages of labor by the week, from the year 1565 to 1821
– John Snow (1854) – Cholera Map in London
– Charles Joseph Minard (1861) – Napoleon and The Russian Campaign of 1812

7
Brief History

• A graph by Playfair (1821) shows the price of wheat, weekly wages, and
reigning monarch over a two hundred fifty year span from 1565 to 1820.
Integration of bar charts and line graph 8
Brief History

John Snow mapped the

cholera cases on the map.
The cases were clustered
around the pump in Broad
street.

• Each death case is a bar

• “Spot map” – Geo-
spatial based mapping

John Snow (1854)

9
Brief History

In this visualization, Minard visualized: Charles Joseph Minard (1861)

• Napoleon’s marching and retreat routes
• Army Size
• Temperature during retreat 10
Outline
 Concepts of Data Visualization
– Value, goals
– History
 Elements of Visualization
– Choosing the Right Chart
– Color, Size, Text, Titles, Labels
 Design of Visualization
– A Visualization Workflow
 Case study: COVID-19
– Mask wearing trends
– People concern about impact of COVID-19
Choosing the Right Chart

 There are many different charts, how

to choose the effective chart for your
own purpose?
Choosing the Right Chart

 When a single number is important

 How two or more numbers are alike or different

 How we are better or worse than a benchmark

 When there are parts of a whole

 How things change over time

When a single number is important

 Single number
– Showing count, frequency, mean, median etc

 Stories can be told with a single number

– On average, our mathematics student achievement is 64%
– We served 123,000 clients last year
– 9 out of 10 consumers prefer our pizza over our competitors
– Only 27% of children in this key neighborhood had a dental visit in the past year
– Chances of dying from a snake bite are just 1 in 50 million
– …
Displaying a Single Number

 Simple text with a single large number

 Icon Array

 Donut or Pie Graph

 Bar
Simple text with a single large number

 Simple text can be a great way

to communicate if you have just
a number or two to share.

 Solely using the number and a

few support words to clearly
make your point.
Simply Text

Original Graph Simple Text

 You have some numbers does not mean that you need a graph!
 A simple sentence would suffice in this particular example.
 Concern: Overload. Use simple text if you have a single large number you want to convey.
Icon Array

 Icon arrays
– One icon (e.g. square, circle) is repeatedly 10, 100 or 1,000 times and then some of the
icons are colored to represent a percentage or proportion.
– Icon arrays have been shown to be especially effective for people with low numeracy
skills.
Combing icon array with a single large number
Donut or Pie Chart

 Donut chart is very similar to pie chart.

 One major difference: the middle of pie is gone for donut chart.

Pie Chart Donut Chart

The weakness of donut or pie chart

 People are not good at interpreting angles and curvature.

The best way to use a
pie/donut chart is to
highlight just one chunk to
support a single number.
Bar Chart

 Bar chart can be more effective than pie chart

 In the bar chat, the baseline are aligned, we can easily compare the end
points.

Bar Chart Pie Chart

Bar Chart

 A simple bar can be focused to

help audience to remember a
single number.
Choosing the Right Chart

 When a single number is important

 How two or more numbers are alike or different

 How we are better or worse than a benchmark

 When there are parts of a whole

 How things change over time

How two or more numbers are alike or different: visualizing
comparisons
 Stories can be told about comparison of two numbers
– Option A is better than Option B
– Group F and Group G disagreed on five of the eight measures
– There were no differences between the treatment and control groups
– Youth rates are significantly lower than adult rates
– We have equal participation from boys and girls with the exception of girls aged 10 to 13
– All sites increased performance, but Site L did so at a faster rate than other sites
– …
Visualizing Comparisons 考试可能会要求选择哪种图表最合适

 Side by Side Column or Bar Charts

 Slopegraph
 Back-To-Back Bars
 Dot Plot
 Dumbbell Dot
 Small Multiples
Side by Side Column Charts (or Bar Charts)

 Column charts and bar charts are similar. Their difference lies in orientation.
– A bar chart is horizontally orientated whereas the column chart is vertically orientated.
 Side by side column charts (or bar charts) are possibly the most ubiquitous
way to showing how two or more numbers compare.
– Side by side column charts (or bar charts) are also referred to as clustered columns or
bars
Limitation of side by side column charts (or bar charts)
 Effectiveness really stops with two comparisons
– It is difficult for people to compare between nonadjacent bars.
 Once we start slicing in a third, fourth, or fifth column within each group, we
are asking brains to do too much

This side by side column has too many columns

Slopegraphs

 Slopegraphs are perfect for highlight the story of how just one category
decreased when other categories increased
 Or to show that one change at a rate much faster than others.

Side-by-Side Bar Chart Slopegraph

 Slopegraphs

 It is a bit hard to see that sales of cheese are actually

down at the new site from side-by-side bar charts (only
one new site bar is shorter than its corresponding old
site bar).
 Slopegraph will make the story clearer.
Old site New site
Back-To-Back Bars

 Two sets of bar graphs that share a spine

 The objective here is NOT to ask people to

compare specific values from each bar graph.
Such comparisons are difficult to make when
the bars are back-to-back.

 These visualizations are best for showing the

shape of the overall dataset. When you are
trying to show natural distributions or bell
curves in the two datasets, this is the graph
type for you.
Dot plot

 Dot plots are flexible and easy to read.

– Humans are accurately interpret location on line, when those lines share a common axis
Dumbbell dot plot

 Compared with dot plot:

the line that connects
each pair of dots
emphasizes their
distance.

 Good at telling a story

about the gap or growth
between two things.
Small Multiples

 Trying to compare too many numbers at once can quickly become

overwhelming and meaningless.
 When we are trying to compare multiple categories, but the graph gets
complicated, one good solution is to break it apart into small multiples.
 Small multiples are several graphs, all on the same scale, arranged together.
Choosing the Right Chart

 When a single number is important

 How two or more numbers are alike or different

 How we are better or worse than a benchmark

 When there are parts of a whole

 How things change over time

How we are better or worse than a benchmark: displaying
relative performance
 Help a reader to determine whether performance was good, bad or close to
the benchmark.

 Stories
– Our key indicators met the pre-established targets in three out of seven areas
– Regions A and B did not meet quarterly benchmarks
– Here is how our groups compared to the national norm
– Students in the Chemistry Department are above average on final exams this year
– We did not meet our fundraising goal, but we got very close
– Our click rate was 25.6% while the industry standard is 4.3%
– …
Displaying Relative Performance

 Benchmark Line
 Combo chart
 Bullet Chart
 Indicator Dots
Benchmark Line

 Adding a benchmark line to a graph gives loads of context for the viewer.
 This simple line packs so much power.
Combo Chart

 Benchmark line works fine when the

benchmark is the same for every
group. But, what about when relative
performance is different for each
group?

 Combo Chart
Bullet Graph

 A bullet graph is a variation of a bar graph.

 It has several key design elements
– The dark center line represents actual value
– The white vertical line represents target value
– The colored bands represents ranges (e.g., unacceptable and acceptable)
Bullet Graph

• The dark center line represents the actual

value.
• The dark vertical line represents a target
value.
• The colored bands represent ranges, such
as poor, average, and good.
Indicator Dots

 Indicator dots are little colored markers that show up next to a target value
listed in a table.
 They indicate whether or not the target has been met. Show students
who are failing
(score under 50)

Show students
who are on the
borderline (score
between 50 and
60)
Choosing the Right Chart

 When a single number is important

 How two or more numbers are alike or different

 How we are better or worse than a benchmark

 When there are parts of a whole

 How things change over time

When There Are Parts of a Whole: Visualizing Beyond the
Pie Chart
 Stories:
– Most students persisted to graduation, but some dropped out and some had incomplete
attendance
– Our employees are mainly female
– Our newest phone already has 5% of the market share
– The student body composition is 75% Free and Reduced Lunch and 25% paid lunch
– This map shows the percentage of our users who come from each state in the United
States
– The majority of our visitors are ages 10 and younger and 56 and older
– Market data show we have three main profiles of customers
– …
Visualize The Parts of a Whole

 Pie/Donut

 Stacked Bar

 Histogram

 Tree Map

 Map
Pie Chart

 Typically, we show parts of a whole as pie

chart.

 Pie charts work best when

– Roughly four or fewer wedges
– Wedges are really different (or really similar) to
another
 Where is my Pie?

1. Don’t use pie charts if your data

involves time.
2. Don’t use pie charts if you have
too many slices.
3. Don’t use too many labels.
4. Keep the color-scheme flat.
5. Two-dimensional is better.
6. Use six categories or fewer.
7. Data distribution matters.
https://www.idashboards.com/blog/2017/03/14/have-your-pie-eat-it-too-when-to-use-pie-charts/
https://www.linkedin.com/pulse/eat-pies-dont-share-them-avinash-kaushik
https://www.perceptualedge.com/articles/visual_business_intelligence/save_the_pies_for_dessert.pdf
100% Stacked Bar

 Swap the pie for bar chart if there are more than four categories in your
dataset

Bar Chart 100% Stacked Chart

Histogram

 A histogram is nothing
more than a standard bar
or column chart, with
absolutely no space
between the bars or
columns.

 It gives a good picture of

overall shape of the
dataset.
Treemap

 Treemaps are square or rectangle shapes that represent parts, all positioned
inside a larger square or rectangle that represents a whole.
 Treemaps can visualize hierarchy

• Peter oversees Kori and Mary, and

that Kori and Mary both oversee
two other employees.

• The sizes of the boxes represent

sales. So that Mary’s group has
higher sales than Kori’s group

• Color represents the growth in

sales over the previous year
Maps

 Increasingly, our social sciences understand the importance of geolocation

when studying outcomes. How respondents or constituents are spread over
location can reveal patterns and insights that can’t be seen any other way.
Choosing the Right Chart

 When a single number is important

 How two or more numbers are alike or different

 How we are better or worse than a benchmark

 When there are parts of a whole

 How things change over time

How things changed over time: depicting trends

 Stories:
– Things changed/didn’t change
– After is so much better/worse/the same as before
– We started this intervention and the outcome improved as a result
– When the legislation took effect, we saw peaks in services for several years
– Sales increased 0.5% over the last quarter
– Maternal services have steadily increased their proportion of hospital use
– Here’s how much change occurred on this measure in the last decade
Visualize How Things Changed Over Time

 Line Graph
 Area Graph
 Stacked Column
 Deviation Bar
 Slopegraph
 Dot Plot

Area
Line Graph

 Line Graphs are the most common way to depict trends over time
Area Graph

 Objective: to show trends that make up 100%

Pie chart can show parts of a whole but

Line graph can show the trends but can can not show the trends over time.
not represent parts of a whole.
Area Graph

 Objective: to show trends that make

up 100%

 Area Graph
– A line graph with each segment stacked on
top of one another.
– Can show both parts of whole and trends.
Stacked Columns

Categories are parts of a whole AND one

categories is really tiny
Deviation Graph

 Situation: focus on the how

much change has happened
since the beginning
Slope Graphs

 Slopegraphs can also be used to show change over time.

Dot Plots

 Dot Plots can be used to

show change over time as
well
Outline
 Concepts of Data Visualization
– Value, goals
– History
 Elements of Visualization
– Choosing the Right Chart
– Color, Size, Text, Titles, Labels
 Design of Visualization
– A Visualization Framework
– A Visualization Workflow
 Case study: COVID-19
– Mask wearing trends
– People concern about impact of COVID-19
Color

Color is one of the most powerful

aesthetic features because it’s an
attention-grabber. It’s the first thing we
notice, and it can immediately highlight
specific insights or identify outliers. The
data, not personal favorites or brand
colors, should drive the use of color
to make a point.
63
Measurable

• Does the color scale match my data?

• Does the color move from light to dark, or is it stepped to
best represent what you’re measuring?
Relatable

• Semantically-resonant colors help people process

information faster.
• So use yellow to depict bananas, red to represent heat.
 Data from 100 observations of global surface temperatures (°C) around the world from 1961- 1990

While these visualizations are accurate, the color Swapping in semantically- resonant colors to the
red represents cooler temperatures, and doesn’t same dashboard really heats things up. With a
resonate with the information the data is trying to little extra thought into color choices and pallets,
portray. the data points now tell the story they were meant
to tell—and faster.

Good Great
 Color Blind Friendly - Accessibility
1. Avoiding problematic color combinations, e.g., red
& green / green & brown / green & blue…

2. Selecting color blind friendly palettes

3. Using different textures and patterns to highlight

important information–not just color

4. Using symbols and icons to supplement color-

Left: Normal vision, Right: Deuteranopia coded messages, warnings and alerts

5. Using highly contrasted color combinations

6. Adopting minimalistic design to help avoid

unnecessary confusion.

https://venngage.com/blog/color-blind-friendly-palette/
Size

The bigger the object, the bolder it looks.

Bold shapes and colors might work well with
bar charts and area charts, but they may
also look gaudy and garish when used in a
different chart, like a treemap. Use size to
draw emphasis to your key message, not
obscure it.

68
Text

• Readability is essential.
• Make the most important information stand out.

69
Titles

• Keep them short, but powerful.

• Convey the point, message or story in the fewest words
possible.
Labels

Find the sweet spot. Too many mark labels can be very
distracting.
Try labeling the most recent mark, or min/max. Save
additional and more detailed information for tooltips.
Outline
 Concepts of Data Visualization
– Value, goals
– History
 Elements of Visualization
– Choosing the Right Chart
– Color, Size, Text, Titles, Labels
 Design of Visualization
– A Visualization Workflow
 Case study: COVID-19
– Mask wearing trends
– People concern about impact of COVID-19
A Visualization Workflow

A needs-driven workflow
• How to select visualization types?
• How to visually encode data?
• How to visualize dynamically? – Animation and Interaction

73
Types and Levels of Analysis

Levels
Types
Needs‐Driven Workflow Börner (2014) – Visual Insights

75
Line graph Map visualization
Needs‐Driven Workflow – Select Visualization Type

Börner (2014) 76
Visualization Types

 Börner (2014) emphasized five major types:

– Charts
 No reference system—e.g., Wordle.com, pie charts
– Tables
 Categorical axes that can be selected, reordered; cells can be color coded and might contain
proportional symbols. Special kind of graph.
– Graphs
 Quantitative or qualitative (categorical) axes. Timelines, bar graphs, scatter plots.
– Geospatial maps
 Use latitude and longitude reference system. World or city maps.
– Network graphs
 Node position might depends on node attributes or node similarity.
 Tree graphs: hierarchies, taxonomies, genealogies.
 Networks: social networks, migration flows.

77
Visualization Type Selection: Temporal

 Temporal data analysis and visualization answer “WHEN” question and can
help to
– Understand the temporal distribution of datasets
– Identify growth rates, latency to peak times, or decay rates
– See patterns in time-series data, such as trends, seasonality, or bursts.

78
Visualization Type Selection: Temporal

 Temporal trends
– Increasing
– Decreasing
– Stable
– Cyclic

79
 Visualization Type Selection: Temporal

 Line graph shows the trends

over time.

80
 Visualization Type Selection: Temporal

 Stacked graph illustrates

individual and total trends.

81
Visualization Type Selection: Geospatial

 Geospatial data analysis and visualization originated in geography and

cartography, but are increasingly common in statistics, information
visualization, and many other areas of science.

 The analyses aim to answer “WHERE” questions that use location

information to identify their position or movement over geographic space.

82
 Visualization Type Selection: Geospatial

– Choropleth map represents data variables such as densities, ratios, or

rates by proportionally colored or patterned areas.

83
 Visualization Type Selection: Geospatial

– Proportional symbol map

represents data variables by
symbols that are sized,
colored, etc. according to their
amount. Data is (or can be)
aggregated at points within
areas.

84
Visualization Type Selection: Topical

 To answer “WHAT” question, we will be using texts to identify major topics,

their interrelations, and their evolution over time at different levels of analysis
– micro to macro.

 To generate visualization from text, text processing or natural language

processing is needed to generate qualitative or quantitative features of the
text.

85
 Visualization Type Selection: Topical

 Word cloud visualize the frequency of the words.

– IMDB movie titles word cloud created with Wordle.

86
 Visualization Type Selection: Topical

 Circular Visualization shows the relationships between entities.

87
 Visualization Type Selection: Topical
 Concept maps are
network graphs that
show the relationships
among concepts.

88
Visualization Type Selection: Network

 The “WITH WHOM” question can be answered by tree and network

visualization.
– Tree visualization utilizes structural, hierarchical data.
– Network visualization can deal with more complex relationships between nodes.

89
 Visualization Type Selection: Network

 Radial Tree Map is a

method of displaying a tree
structure in a way that
expands outwards, radially.

90
 Visualization Type Selection: Network

 Subway map layout

– Aim for evenly distributed
nodes, uniform edge lengths,
and orthogonal drawings.
– Metro Maps - E-mail History

91
 Visualization Type Selection: Network
 Network overlays on
geospatial maps
– Use a geospatial reference
system to place nodes
– E.g., airport traffic

92
Needs‐Driven Workflow – Visually Encode data

Börner (2014) 93
 Data Scale Types
 Categorical (nominal): A categorical scale, also called nominal or
category scale, is qualitative. Categories are assume to be non- More
overlapping. Qualitative

 Ordinal: An ordinal scale, also called sequence or ordered, is

qualitative. It rank-orders values representing categories based on
some intrinsic ranking but not at measureable intervals.

 Interval: An interval scale, also called value scale, is a quantitative

numerical scale of measurement where the distance between any two
adjacent values (or intervals) is equal but the zero point is arbitrary.

 Ratio: A ratio scale, also called proportional scale, is a quantitative

numerical scale. It represents values organized as an ordered
sequence, with meaningful uniform spacing, and has a true zero point. More
Quantitative
94
Data Scale Types – Example

 Categorical (nominal): Words or numbers constituting the names and

descriptions of people, places, things, or events.

 Ordinal: Days of the week, degree of satisfaction and preference rating

scores (e.g., Likert scale), or rankings such as low, medium, high.

 Interval: Temperature in degrees or time in hours. Spatial variables such as

latitude and longitude are interval.

 Ratio: Physical measures such as weight, height, (reaction) time, or intensity

of light; number of published papers, co-authors, citations.

95
 Graphic Variable Types

 Quantitative  Qualitative
– Position – Form
 x, y; possibly z  Shape
– Form  Orientation (Rotation)
 Size – Color
– Color  Hue (tint)
 Value (Lightness, Brightness)

 Saturation (Intensity) – Optics

 Crispness
– Texture  Transparency
 Pattern  Shading
 Rotation
 Coarseness
 Size
 Density gradient

96
Dynamic Visualization

 Previous sections demonstrates the workflow to create a static

visualization.

 In order to present more information and make visualization

dynamic, we can show multiple static images side by side or as an
animation.

 Another way to make dynamic visualization is to introduce

interaction into the design process.

97
Shneiderman's Mantra

 User-Interface Interaction
– Immediate interaction not only allows direct manipulation of the visual objects
displayed but also allows users to select what to be displayed (Card et al.,
1999)
– Shneiderman (1996) summarizes six types of interface functionality
 Overview
 Zoom
 Filtering
 Details on demand
 Relate
 History
– “Overview first, zoom and filter, then details-on-demand.”

98
Two Interaction Approaches

 User-Interface Interaction
– Overview + detail
 First overview provides overall patterns to users; then details about the part of interest to the
use can be displayed. (Card et al., 1999)
 Spatial zooming & semantic zooming are usually used
– Focus + context
 Details (focus) and overview (context) dynamically on the same view. Users could change the
region of focus dynamically.
 Information Landscape( Andrews, 1995)
 Cone Tree (Robertson et al., 1991)
 Fish-eye (Furnas, 1986)

99
Interactive Vis. Example: Gapminder World

http://www.gapminder.org/world/ 100
https://www.youtube.com/watch?v=BPt8ElTQMIg (Hans Rosling telling story)
Outline
 Concepts of Data Visualization
– Value, goals
– History
 Elements of Visualization
– Choosing the Right Chart
– Color, Size, Text, Titles, Labels
 Design of Visualization
– A Visualization Workflow
 Case study: COVID-19
– Mask wearing trends
– People concern about impact of COVID-19
Case study - COVID 19

 What are the mask wearing trends?

– This data viz shows the prevalence of mask wearing in different scenarios: while grocery
shopping, while exercising outside, while visiting with family and friends, and while
working at the workplace. See what behaviors have changed over time as vaccinations
have rolled out and restrictions lifted.

 How worried are people about the impact of COVID-19?

– As we enter into a "new normal", how worried are people about the impact of COVID-19
on personal finances and the economy? How does this break down by party affiliation,
race, education, gender and age?
https://www.tableau.com/about/blog/2021/6/new-momentive-tableau-
data-shows-americans-split-over-mask-wearing
https://www.tableau.com/data-insights/us-election-2020/covid-worries
Dashboard layout

Your dashboard’s purpose is to help guide the reader’s eye through more than
one visualization, tell the story of each insight, and reveal how they’re
connected.

The more you employ better dashboard design, your users will discover
what’s happening, why and what’s most important. Take into account how
you’re guiding their eyes across the dashboard.

107
Guide the user
Don’t leave people high and dry
without guidance on how to use a
visualization.
Try swapping a filter title with
explicit language directions
about how to navigate.
Rule of three

Don’t make a lot of important

information compete for attention.

Most of the time, more than three

visualizations on one dashboard is
too many.
Tell a story

If you need more than one

visualization, or are preparing
for a presentation, connect
the different visualizations
with story points.

Tell the narrative of your

data with visuals that build
on each other, highlight
specific insights, and provide
additional context, all in one
seamless presentation.
Great visualizations will not only help you understand more
about your data, they’ll offer faster, more meaningful
answers, and even inspire others to ask and answer new
questions.
Resources
 Tufte, E. R., & Graves-Morris, P. R. (1983). The visual display of quantitative
information (Vol. 2, No. 9). Cheshire, CT: Graphics press.
 Tufte, E. R. (1991). Envisioning information.Optometry & Vision Science, 68(4), 322-
324.
 Tufte, E. R. (2006). Beautiful evidence. New York.
 Tufte, E. R., & Weise Moeller, E. (1997). Visual explanations: images and quantities,
evidence and narrative (Vol. 36). Cheshire, CT: Graphics Press.
 Börner, K. (2014). Atlas of knowledge.
 Steele, J., & Iliinsky, N. (2010). Beautiful visualization: looking at data through the
eyes of experts. " O'Reilly Media, Inc.".
 Fry, B. (2007). Visualizing data: exploring and explaining data with the Processing
environment. " O'Reilly Media, Inc.".
 Good enough to great - A quick guide for better data visualizations,
https://www.tableau.com/good-to-great

112