Data Visualization, Neuroscience, and Why It Matters to the Analyst

​

​

Tim Wilson

Data visualization misconceptions abound!

@tgwilson

Data visualization is NOT about making the data pretty.

@tgwilson

Data visualization is about �being understood.

@tgwilson

It’s about�neuroscience.

@tgwilson

Iconic Memory

@tgwilson

Iconic Memory

Short-Term Memory

Long-Term Memory

@tgwilson

Miller's Law (1956)

The number of objects an average human can hold in short-term memory is 7 ± 2.

@tgwilson

Miller's Law (1956)

The number of objects an average human can hold in short-term memory is 7 ± 2.

4

@tgwilson

Miller's Law (1956)

The number of objects an average human can hold in short-term memory is 7 ± 2.

4

1

@tgwilson

It’s ALL about reducing the�cognitive load

for your audience.

@tgwilson

How many 5s are in this list?

49790275647902894728624092406037070570279072�08029007302501270237008374082078720272007083�02602703793775709707377970667462097094702780�79709723097230979592750927279798734972608027

Source: Information Dashboard Design by Stephen Few

@tgwilson

How many 5s are in this list?

49790275647902894728624092406037070570279072�08029007302501270237008374082078720272007083�02602703793775709707377970667462097094702780�79709723097230979592750927279798734972608027

Source: Information Dashboard Design by Stephen Few

@tgwilson

How many 5s are in this list?

49790275647902894728624092406037070570279072�08029007302501270237008374082078720272007083�02602703793775709707377970667462097094702780�79709723097230979592750927279798734972608027

Source: Information Dashboard Design by Stephen Few

6!

@tgwilson

How much larger is circle B than circle A?

A

B

Source: Information Dashboard Design by Stephen Few

@tgwilson

How much larger is circle B than circle A?

A

B

Source: Information Dashboard Design by Stephen Few

16X

@tgwilson

How much longer is bar B than bar A?

A

B

16X

@tgwilson

@tgwilson

@tgwilson

Gestalt psychologists were smart folk.

Humans perceive entire patterns or configurations, not merely individual components.

The perception of those patterns are controlled by distinct Gestalt principles: proximity, similarity, closure, connectedness, and continuity.

This is… one group?

@tgwilson

Two groups?

@tgwilson

DEFINITELY two groups!

@tgwilson

Columns!

@tgwilson

Or...rows?

@tgwilson

Four groups?

@tgwilson

Definitely four groups!

@tgwilson

On to the Tips!

@tgwilson

“Maximize the data-pixel ratio!”

Data-Information Pixels

All Non-Background Pixels

– Stephen Few

“Pretend your printer is running out of toner.”

- Michele Kiss

Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away.

– Antoine de Saint-Exupery

@tgwilson

We already looked at an example of this idea.

@tgwilson

Maximizing the data-pixel ratio works for tables, too!

@tgwilson

Maximizing the data-pixel ratio works for tables, too!

@tgwilson

Maximizing the data-pixel ratio works for tables, too!

@tgwilson

Pie charts are generally evil!

vs.

See http://bit.ly/timin2009 for details.

@tgwilson

Pie charts are generally evil!

vs.

See http://bit.ly/timin2009 for details.

@tgwilson

And...adding a hole to the pie isn’t an improvement.

vs.

See http://bit.ly/timin2009 for details.

@tgwilson

Pie charts with 2-3 sections can be okay.

@tgwilson

But there is a BIG problem with this one.

@tgwilson

9-10% of U.S. males have deuteranopia.

@tgwilson

And plenty of people still use black-and-white printers.

@tgwilson

When in doubt...check!

Sim Daltonism (Mac)

Color Oracle (Windows or Mac)

@tgwilson

Horizontal bar charts are generally awesome!

SESSIONS

@tgwilson

Put them side-by-side for metric comparisons.

SESSIONS

BOUNCE RATE

@tgwilson

Use color for emphasis!

SESSIONS

BOUNCE RATE

@tgwilson

They work in compact form, too!

@tgwilson

They work in compact form, too!

@tgwilson

Use plain English.

@tgwilson

Use plain English.

@tgwilson

Don’t forget to include the data source!

SESSIONS

BOUNCE RATE

Data Source: Google Analytics, July 2019

@tgwilson

What’s wrong with this chart?

3.00%

2.00%

1.00%

0.00%

2.13%

2.51%

Control

Test

@tgwilson

Kill extraneous significant digits!

3%

2%

1%

0%

Control

Test

2.1%

2.5%

@tgwilson

Possibly...even kill the y-axis!

Control

Test

2.1%

2.5%

@tgwilson

Or (subtly) represent uncertainty!

Control

Test

3%

2%

1%

0%

@tgwilson

Have we had enough bar (charts) yet?

@tgwilson

There are other visualizations!

(But ALWAYS consider the basic ones first!)

The Extreme Presentation^TM Method

@tgwilson

The Juice Analytics Chart Chooser

@tgwilson

Sparklines FTW!

@tgwilson

Text size variation FTW!

@tgwilson

Boxplots show variability.

@tgwilson

Boxplots show variability.

@tgwilson

Bar charts are good...but not always the best.

@tgwilson

Slopegraphs can show change really well.

Direct

Organic Search

Paid Search

Referral

Display

@tgwilson

Heatmaps can be handy!

@tgwilson

Pseudo-heatmaps can be handy!

@tgwilson

Pseudo-heatmaps can be handy!

@tgwilson

Pseudo-heatmaps can be handy!

@tgwilson

Keep it monochromatic!

@tgwilson

Scatterplots

@tgwilson

Word clouds can be nice (but have limited utility).

Take the time to clean up the data first:

Force everything to lowercase

Remove stopwords

Stem the words (if possible)

Remove obviously dominant words (if warranted)

@tgwilson

Geographic overlays can be revealing (but aren't always).

@tgwilson

Interactive geographic overlays can be good for exploration.

This was built with R, but most BI tools support this sort of visualization.

Excel now supports map charts, too (interactivity through slicers and pivot tables, perhaps?)

@tgwilson

Network diagrams are neat (if they reveal something)!

Product Content

Support Content

@tgwilson

To summarize...

@tgwilson

It’s ALL about reducing the�cognitive load

for your audience.

@tgwilson

Recommended Reading: Data Visualization

@tgwilson

Recommended Reading: Data Storytelling

@tgwilson

Any questions?

@tgwilson

tim.wilson@searchdiscovery.com