Mikel Madina & Miren Berasategi

Visual Analytics

Communicating data-driven insights
through data visualization techniques
and useful dashboards

Mikel Madina & Miren Berasategi

miren.berasategi@deusto.es

0. Introduction

0.1 Key points

Data driven: as seen in previous sessions with Professors Onieva and Lorenzo
Insights: the capacity to gain an accurate and deep understanding of something through
Data visualization techniques: to take the user from data to insight
Dashboards: as situation awareness tools

Tableau Desktop to practice

Section outline

Introduction: the why and the what for of visualization
Graphs: some reminders, idioms to map variables to graphs
Promote insight: by adding meaningful modifications to graphs
Dashboards: situation awareness, dos and don’ts
Epilogue

Practice: build a simple dashboard with online marketing campaign data

0.2 Why use visualization

Sight is our most developed sense
The visual system provides a very high-bandwidth channel to our brains
A significant amount of visual information processing occurs in parallel at the preconscious level
The human brain is trained to identify visual patterns
Summary statistics have the intrinsic limitation of data loss

0.2 Why use visualization

Anscombe’s Quartet

0.3 What to use visualization for

0.3 What to use visualization for

0.3 What to use visualization for

0.3 What to use visualization for

There is a strong relationship between the form of the data (the attribute/variable and dataset types) and what kinds of vis[ualization] idioms are effective at displaying it. (…) Don’t just draw what you are given; decide what the right thing to show is, create it with a series of transformations from the original database, and draw that!

Munzner 2014, p.50

0.3 What to use visualization for

Derived attributes can be directly visually encoded. Munzner 2014, p.52

Practice: meet our sample data

Download and open data.xls: fake data for online marketing goals and tools

from Google Drive or http://mrn.bz/MUMA2018data

Tableau Software

Tableau

Load data
Explore the Data Source tab

See the subtle blue/green colour of the variable type icon? Take notice, it is important:

Understanding the difference between the blue and green items in Tableau is (IMHO) the single most important piece of understanding necessary to make Tableau function well.

Tom Brown, Blue things and Green things

Possible to merge more sources (or Connections), only one for our practice – might require to define New Union to determine how merge takes place
Shows the only sheet in the file, may be more
Connection > Live/Extract, first one allows to modify source file and update work accordingly

Interprets the source field as Abc, rest as # (well done). Only, we would rather improve the quarter interpretation as Date instead of Number

Green & blue: blue fields are discrete, green fields are continuous. We’ll see implications of this later on. For now, it’s all good

Rename, hide… fields

The Excel > Tableau convert is quite straightforward (does a good job), more complex sources may require more manipulation at this point

This screen allows to preview how Tableau is interpreting our dataset: we are not allowed to change values, only variable interpretations

The source file (data.xls) is never modified

1. Graphs

Section outline

Reminder: variable types
Mapping variables to graphs

Marks
Channels, channel types
Using marks and channels

So, which graph?

Practice: explore dimensions, measures and graph types in Tableau

1.1 Reminder: variable types

A question of time

Spatial and time/hour variables are special variable types. Time variables are specially complex:

are there 365 days in every year? 30 days in every month? 24 hours in every day?
timezones make it even more complex to use hours or time of day

Time may be used as a continuous or as a qualitative variable.

as a qualitative variable, it has a hierarchy: year > month > (week >) day > hour > minute
but different hierarchies may be necessary: bimonthly publications, multiple work shifts in a day…

Quantitative
- Continuous
- Discrete
Qualitative
- Categorical
- Ordinal
Special types
- time
- space
- …

1.2 Mapping variables to graphs

Understanding marks and channels provides the building blocks for analyzing visual encodings (Munzner 2014, p.95)

1.2.1 Marks

A mark is a basic graphical element in an image

Marks are geometric primitives (Munzner 2014, p.96)

1.2.2 Channels

A visual channel is a way to control the appearance of marks

Visual channels control the appearance of marks (Munzner 2014, p.96)

1.2.2 Channels

One and only one attribute/variable should be used per channel.

Multiple channels per attribute are possible (redundant encoding), but this approach has limitations.

1.2.2 Channels

The size and shape channels cannot be used on all types of marks, but most combinations are still possible:

lines have two size channels: length + width
points refer to location but can be size and shape coded

1.2.3 Channel types

Two kinds of sensory modalities:

Identity: what, where
Magnitude: how much

It does not make sense to ask magnitude questions for shape, color hue. We can ask about magnitudes with length, area or volume; color luminance or saturation; and angle/tilt/slope.

1.2.4 Using marks and channels

All channels are not equal.

The selection of marks and channels should be guided by the principles of expressivenes and effectiveness.

Once the most important attributes/variables for the desired insight have been identified, the selection of marks and channels should ensure that they are encoded with the highest ranked.

the same data attribute encoded with two different visual channels will result in different information content in our heads after it has passed through the perceptual and cognitive processing pathways of the human visual system (Munzner 2014, p.100)

expresiveness: the visual encoding should express all of, and only, the information in the dataset attributes. Ordered data should be shown in a way that we perceive as ordered; unordered data SHOULD NOT be shown in a way that implies an ordering that does not exist.
effectiveness: the importance of the channel (task abstraction, targets and actions) should match its salience, how noticeable it is. The most important attributes/variables should be encoded with the most effective channels, and less important attributes can be matched with less effective channels.

These can be combined to create a ranking of channels according to the type of data that is being visually encoded.

1.2.4 Using marks and channels

Channels ranked by effectiveness according to data and channel type. Ordered data should be shown with the magnitude channels, and categorical data with the identity channels (Munzner 2014, p.102)

1.2.4 Using marks and channels

The choice of which attributes/variables to encode with position is the most central choice in visual encoding.

1.2.4 Using marks and channels

Error rates accross visual channels (Munzner 2014, p.105)

1.3 So, which graph?

A. Abela (2006), Choosing the right chart. Interactive version: Chart chooser

Tableau: let’s explore

Dimensions and measures (remember also blue vs. green)
Encode = drag
Show me tab

Left column, Data tab:

dimensions contain qualitative values
measures contain numeric, quantitative values (you can apply calculations to them and aggregate them)

Reminder (blue vs. green, learn more):

green = continuous. Its values are treated as an infinite range. Generally, continuous fields add axes to the view.
blue = discrete. Its values are treated as finite. Generally, discrete fields add headers to the view.

Mapping variables to graphs is done, in Tableau, by dragging items from the left column to either: Columns/Rows, or Marks.

Show me tab displays the most common graphs and the minimum requirement of data for each graph. If graphs = recipes, lets us know which ingredients are required to cook a certain recipe.

Also works the other way round: if I select my ingredients from the left bar, the Show me tab highlights the recipes available for those ingredients.

(test some graphs) - Source only > table - add spent > table, pie chart etc

Fields can be used more than once (redundant coding) for easier identification, by dragging to more than one place. barchart income (rows) + source (columns), source to colour

Options in Show me are different from the Marks box. Show me only allows graphs that make sense, very basic and tested graphs. Marks allows to force graphs in case we need more complex displays. USE WITH CAUTION, possible to create graphs that make no sense or are misleading (change to “line” in Marks box)

2. Promote insight

Section outline

How can we enable easier insight through data visualization?

Change default settings
Make simpler graphs
Highlight observations
Add attributes as context
Add statistical information

Practice: build (not so) basic graphs

2.1 Change default settings

2.1 Change default settings

2.2 Make simpler graphs

Data-ink is the non-erasable core of the graphic, the non-redundant ink arranged in response to variation in the numbers represented.

Tufte 1983

2.2 Make simpler graphs

A step-by-step example: Data looks better naked

2.2 Make simpler graphs

2.3 Highlight observations

Through preattentive attributes:

they are processed in spatial memory without our conscious action
make it easier to understand what is represented through a design: saves from consciously processing data

2.3 Highlight observations

2.3 Highlight observations

2.3 Highlight observations

2.3 Highlight observations

Nussbaumber, Do you see it? The importance of contrast when communicating with data [video]

2.4 Add variables (as context)

Adding preexisting variables (in moderation)
Creating conditional variables from preexisting variables
- binaries or with few levels are best
- example of calculated field or variable: weekend date

2.5 Add statistical information

statistical summaries
(mean, variance)
models

Tableau: (not so) basic graphs

Tableau: (not so) basic graphs

Tableau: (not so) basic graphs

Source to rows, income to columns > draws bars.

Where is goal? Drag to columns > two different graphs (not what we want).

Select source + income + goal > bullet graph from Show me.

Shows totals, may not be what we want. Anyway, what did this do? Now manually:

source to rows, income to columns
now we want goal as black bar (reference line), goal valorations as background (distribution band)
Double click on ref. line, adds a line of average income

scope: default per pane: avg per each combination. per table, whole table (same as per pane when there is only one dimension). per cell avg. per each mark.
line: only shows income for now. How to add goal? Drag to Marks > detail, not going to display but is available to use. Now line: sum(goal), label (none).
formatting: bolder, black

Double click on distribution band.

scope: per cell
computation: value: 60%/80%, percent of sum(goal) instead of income. label: none
formatting: check fill below (used normally in bullet graphs because below is usually bad)

Style:

Marks > size, make thinner bars to make background more visible
Format > lines:
- rows: grid lines, solid white
- columns: none

Rename sheet to bullet graph, save.

Tableau: (not so) basic graphs

3. Dashboards

Section outline

What is a dashboard?
Common design mistakes
Key goals in the visual design process
Example

Practice: layout and format graphs into a dashboard

3.1 What is a dashboard?

Visual display of the most information needed to achieve one or more objectives which fits entirely on a single computer screen so it can be monitored at a glance.

Few 2013

3.1 What is a dashboard?

Visual display: I see = I understand → insight to achieve specific objectives: may require gathering information that is otherwise unrelated or disperse
fits in a single computer screen: it must all be seen at once (short-term memory effect)
monitored at a glance: doesn’t need to provide all the details, but if it doesn’t, it should make it as easy and seamless as possible to get to that information

combine text and graphs, with emphasis on graphics. There is such an intimate connection between what we see and how we think, we actually use the expression “I see” as a substitute for “I understand”. when we make sense of something, we refer to what we’ve learned as “insight”. the more you can rely on images to tell the story, the faster it can be perceived. We perceive images in a parallel manner, which is much faster than the serial/sequential perception of language. We have very limited short-term memory: dashboards address this issue in two ways:
- making the content in each slot more “information rich” (only 4-5 slots; numbers/graphs?)
- reducing the need to rely on it, everything relevant is available simultaneously. External form of memory.

Must the information be constantly refreshed in real time? Only if the objectives that it serves require real‐time information. If you are monitoring air traffic using a dashboard, you must immediately be informed when something is wrong. On the other hand, if you are making strategic decisions about how to boost sales, a snapshot of information as of last night, or perhaps even the end of last month, should work fine.

3.2 Common design mistakes

Exceeding the boundaries of a single screen
Supplying inadequate context for the data
Displaying excessive detail or precision
Choosing a deficient measure
Choosing inappropriate display media
Introducing meaningless variety
Using poorly designed display media

continues…

3.2 Common design mistakes

…continued

Encoding quantitative data inaccurately
Arranging the data poorly
Highlighting important data ineffectively or not at all
Cluttering the display with useless decoration
Misusing or overusing color
Designing an unattractive visual display

3.3 Key goals in the visual design process

From previous section:

make simpler graphs (declutter)
highlight observations
add attributes/variables as context or statistical information

3.3 Key goals in the visual design process

In other words:

Reduce non-data pixels

eliminate all unnecessary non-data pixels
de-emphasize and regularize the non-data pixels that remain

Enhance data pixels

eliminate all unnecessary data pixels
highlight the most important data-pixels that remain

3.4 Example

Dashboards in Tableau

Dashboards in Tableau are containers of sheets of graphs.

Allow for quite basic but functional formatting.

Tableau: Actions

Some degree of interactivity with Actions: highlight and filter

Tableau: What else?

Calculated fields

…

Epilogue

What? I want more!

Tableau, Free Training Videos
- Tableau, Análisis visual: como aprovechar las características cognitivas humanas para comprender tus datos
Coursera, Visual Analytics with Tableau
Udemy, (Ranked) Data Visualization Courses

References

Abela, Andrew (2006). Choosing a good chart.

Few, Stephen (2009). Dashboard Design for Real-Time Situation Awareness [White Paper]

— (2013). Information Dashboard Design. Analytics Press: 316.763 F 44 s

Kirk, Andy (2016). Data Visualisation: A Handbook for Data Driven Design. SAGE: London 316.763 K 63 a

Munzner, Tamara (2015). Visualization Analysis and Design. CRC Press: Boca Raton, Florida 316.763 M 92 t

Tufte, Edward R. (1983). The Visual Display of Quantitative Information. Graphics Press: California 316.763 T 87 e

Thank you!

This presentation is available at
http://mrn.bz/MUMA2018

Miren Berasategi
miren.berasategi@deusto.es

License

Copyright © 2018 University of Deusto
This work (except for the quoted images, whose rights are reserved to their owners) is licensed under the Creative Commons “Attribution-ShareAlike” License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/3.0/