Week 05

VISUALIZATION
Structure & Design

ETW3483 Applied Analytics. ULO2 #Viz I.

Dr. Ewilly Liew
 Visualize and interpret trends and
ULO#2 patterns based on visualization
Viz design and validation concepts
and methods. (visualize)

ETW3483 ULO2
The art of exploring your data. Rapidly visualizing trends and patterns on charts, generating
hypotheses, quickly testing them, then repeating again and again and again... generate many
promising leads that you can later explore in more depth.
“Far better an approximate answer to
the right question, which is often vague,
than an exact answer to the wrong
question, which can always be made
precise”
-- John Tukey
Strength of Evidence
versus
Quality of Question

It takes time to get to

the "right" question.
“You have to shuck
a lot of oysters to
find a pearl.”
Lecture 1 Viz design choices (chart types)
Outline 2 Viz design structure and system

3 What - Data Abstraction

4 Why - Task Abstraction

5 Matching the {action, target} pair

Read
1
Viz Design Choices
Different Chart Types for Satisfy
Different Purposes of Use
“Computer-based visualization systems
provide visual representations of
datasets designed to help people carry
out tasks more effectively. ”

-- Tamara Munzner
Visualization is about external cognition
How external resources outside the human brain can boost the
cognitive capabilities of the mind to process information visually.
 1 2
⭐Data Visualization Catalogue ⭐SWD Chart Guide
“Visualization is crucial to understanding your
data and communicating your message to
others. It’s about more than just making it look
good. There’s a whole system and structure for
the way data is visualized.”

-- Dianne Cook,
Professor of Business Analytics, Monash University Australia
 Effectively use
Derive insights preattentive attributes
to reduce time
to insights

Comprehended by
Correct representation of
human perception but
data but poor match to
poor match to the
human perception.
intended task/message.
How do we choose?
Satisfy Optimize

What is the chance of randomly making the viz design looks effective?
2
Viz Design
Structure & System
Visual analytics cycle.
Nested model of visualization design
and validation.
Visual Analysis Cycle
Start with questions (it takes time)
What: Find the right data source
Why: Pick the right type of visual

How: View and iterate

So what: Discover answer (insight) to your question

Share and act
Scaffolding Structure
of Visualization Design

Source: Visualization Analysis and Design. Tamara

Munzner, with illustrations by Eamonn Maguire. A K Peters
Visualization Series, CRC Press, 2014.
System of Visualization
Design and Validation

Source: Visualization Analysis and Design. Tamara

Munzner, with illustrations by Eamonn Maguire. A K Peters
Visualization Series, CRC Press, 2014.
Example: how we use the nested model
to validate a visualization design
Munzner's nested model of what - why - how
visual analysis framework
3

What - Data Abstraction

What data do we have to show?
Data abstraction means we consider data in its
abstract form, i.e. domain-general meaning,
without requiring any domain-specific
knowledge to understand what the data means.

-- Tamara Munzner
What?
Data Abstraction
Import: Where do we get the data? Is the
dataset available to us?

Cardinality: What types of dataset,

variable/attribute types?

Tidy/ Transform: Is data wrangling or

transformation needed?

Risk/ limitation: Any other data concerns?

 Semantic meaning
of the data
Real-world meaning.
e.g. A number is day of a month,
age, height, postal code, position in
a space, or unique ID for a person?

Domain-specific terms Domain-general terms

Technical but generic meaning that a
Real-world meaning that requires general person or computer can
specific knowledge to understand it. understand it.
e.g. Sales trend for the past three years e.g. Present ordered continuous (numerical)
sales data in a monthly view from Jan 2019 to
Dec 2022.
 Technical meaning
of the data Items Attributes Links Positions Grids
Individual Property that is Express Spatial 2D or 3D Sampling
Types = Cardinality entity measured/ relationship data in a space. strategy for
recorded on observed. between two continuous data.
Structural/ mathematic items (usually
interpretation of the data a dataset.
(not the same data types in R). Variables in within a
terms of network). Geo coordinate Geometric &
How are the data represented at the measures (longitude, topological
"data" level, "dataset" level, (numerical) and latitude). relationships
"attribute" (i.e. variable) level. dimensions Photo pixels. between cells.
(categorical)
 Dataset Types
Flat table and Multidimensional table
Flat table
One item per row.
Each column represents each
attribute (i.e., variable).
Each cell holds a value for each
item-attribute pair.
Can also have a unique key
(primary/foreign)

E.g. CSV file

Mutldimensional
table
Indexing based on (explicit)
multiple unique keys.
E.g. Order ID in “Orders”;
Region in “Returns”
E.g. Multiple sheets in an Excel file
 Quantitative Qualitative Qualitative
Continuous/ (Ordinal) (Categorical)
Numerical Factor levels Nominal no levels

Weight (10kg, 20 lbs) Low, middle, high school Malaysia, Singapore

Height (100 cm, 2 m) Gold, silver, bronze Sarah, John, Maria
Discount (5%, 10%) Poor, good, excellent Sprite. Pepsi
health
Sales ($2000, $50)
Quiz: Variable/Attribute Types
Which types/ classes are these values or measurement?

Quantitative

Ordinal

Categorical
Quiz: Global Superstore.xlsx
Can you identify the data types and dataset types?
the entire dataset is Real-time/ live data Such as ordered Deconstruct into Values wrap around
available on a local trickles in over the continuous/ two sequences back to a starting
computer all at course of the viz numerical data. pointing in opposite point (rather than
once. session. Real-time directions that meet continue to increase
changes take place Where there is a at a common zero indefinitely).
on the dataset. homogenous range point.
of values from a
finite minimum to a E.g. Elevator going up
In Tableau: In Tableau: or down. Mountain E.g. Clock. Hours of
when you use finite maximum.
when you use height (bathymetric) the day, day of the
dataset in “Extract” dataset in “Live” from sea level or week, month of the
format format above (below). year
4

Why - Task Abstraction

Why do we show the things we show in a viz?
 The Photographer's Perspective

Visualization design involves choices of abstraction

(aspect to focus) and emphasis (frame to include)
depending on the problem we intend to solve or
answer. We call it, task abstraction.

-- Tamara Munzner
Why?
Task Abstraction
People need details but no Reframe domain-specific user's
trusted automatic solution task to a domain-general
exists (yet) abstract form (free of jargons)
01 02

People does not know exactly what Without requiring specialized

questions to ask for the details they need. knowledge about a domain (i.e. field of
People want to speed up exploratory data study), people can easily compare
analysis to know what's inside their dataset. and contrast the details inside a chart
People want to present known results to against the information they really
stakeholders ASAP. need for insights.
Visualization is the first step towards
automation (e.g. dashboard).
Why?
Task Abstraction
Use {action, target} pair to

frame the viz design choice

why we want to create this viz?
{Action: analyze, search, query}
that matches the user’s intended task
what we want to show the audience?
{Target: all data (trends, outliers, features),
attributes (one/many), network data, spatial
data}
Action: Analyze
High-level choices whether to consume or produce information
Action: Analyze: Derive
Transform variable/ attribute using calculated field

Don’t just show data as is. Determine the right thing to show!
In Tableau, use calculated field (compute new
variable, simple change of data type, acquire more
data, create a new formula for complex transformation)
 Action: Analyze: Derive (Example)
Transform GDP to % change of GDP

Data: Greece's GDP data

Task: Compare and derive

% change of GDP same
month this year vs. the
same month a year earlier.

Mark: Deviation bar

Channel: Use two-tone

color to emphasize the
positive change (dark
blue) and the negative
change (light blue).
Action: Search
Search for a known or unknown item/ observation
Action: Query
Retrieve characteristics of item/ observation

One Target Some Targets All Targets

Identify a data point. Compare between Summarize all data.
data points.
E.g. Bar, Line E.g. Bar, Line, Deviation E.g. Table, Heatmap
 5
Matching the {action, target} pair
Action - task - why are we showing the viz?
Target - data - what are the right thing we are showing?
Target
Focusing on the data level
 Target
Focusing on the attributes (i.e. variable) level

Single variable Multiple variables Network/ Spatial

{present, distribution}. Identify dependency paths. variable
{identify, extremes}. Derive correlation. General topology or
Compare similarity. specific path.
Spatial shape.
Quiz: Task abstraction
Identify all the possible {action, target} pairs in the two charts below.
Solution: Task abstraction
Identify all the possible {action, target} pairs in the two charts below.

Task A: present (sort) attributes

Task B: compare pair of attributes (Direct vs Distributor)
Task C: compare pair of attributes (Distributor vs OEM)
Task D: present trends across all attributes
Task E: spot outlier attributes
Task F: enjoy / engage
Rules of thumb Experimental visualization design
The old design should inform the new.

for abstraction The new design should improve the old.

1. Translate from domain-specific tasks to domain-general tasks

that do not require specialized knowledge to understand the
data or the task.
2. Distinguish the cardinality of data and dataset types.
3. Think about the purpose of plotting a viz.
4. Frame your viz design choice by focusing your audience's
attention on the aspects you want to emphasize.
5. Deliberately match the {action, target} pair when you design
a viz. Target (data) first, then action (task). View the first draft of
viz, then iterate between data and task until you finetune and
get it right.
Sketching
Low-cost, low-risk, quick-and-
easy way to try different
possible layouts and
solutions.

Useful to illustrate spatial

relationships that are
otherwise hard to describe
in words.
 Reflection
Consider reflecting on how you might
abstract data for the tasks you intend to show in
visualizations. Think about how you might frame
your viz design choices that closely match the
{action, target} pair of viz design. Iterate between
data and tasks until you get the right viz. But first,
start with a good question. It takes time to get there.
 Q&A
#AskUsAnything on Padlet General Forum.

Look out for assessment briefing (F2F/video)

2 weeks before the due date.