25 VISUALIZATIONS

E D U A R D O A R I Ñ O D E L A R U B I A C H I E F D A T A S C I E N T I S T

E D U A R D O @ D O M I N O D A T A L A B . C O M

A N “ O U T O F M Y L E A G U E ” P R O D U C T I O N

A N D W H E N T O U S E T H E M

PICTURE SLIDEDATA SCIENTIST

A BIT ABOUT ME

{Robotics, Vision Systems

Job Shop Scheduling, Optimization/Ops,

Neural Networks, NLP

GOD

THE GUY GOD ASKS DATAVIZ ADVICE

???

A DISCLAIMERThere are many kinds of dataI am only talking about tabular data.

That is, arranged in a table or systema7c arrangement by columns,

rows, etc…

There is non-tabular data out there, like networks and trees and

whatnot. I ain’t messin’ with that. (Except maps)

C O W A R D L Y S T A T E M E N T

STANDING ON THE SHOULDERS OF GIANTS IS NICE…

This presentation is based on the work of Dr. Andrew Abela’s “Extreme Presentation” method, as well as the Financial Times fantastic Chart Doctor feature. There is a lot of amazing work out there

to help you pick the right way to present your data. None of what I’m saying is my own personal research. It’s reading other smart peoples stuff and then telling you.

CITATION

Product: Open/Flexible + Full-Lifecycle Support

3. Opera&onalize / Deploy

2. Experiment & HardenFaster Experimenta&on

More Collabora&on

Reproducibility &Audi&ng

Integrate models into the business

More Time for Research

AutomaVc Version Control

Environment Management

Sharing and Discussion

Publishing & DeploymentTools

Data

Code

Compute automaVon

https://app.dominodatalab.com/u/earino/plotcon2016

DEVIATION

Emphasize varia7ons (+/-) from a fixed reference point. Typically

the reference point is zero but it can also be a target or a long-

term average. Can also be used to show sen7ment (posi7ve/

neutral/nega7ve).

OUR CATEGORIES

CORRELATION

Show the rela7onship between two or more variables. Be mindful

that, unless you tell them otherwise, many readers will assume the rela7onships you

show them to be causal (i.e. one causes the other).

RANKING

Use where an item’s posi7on in an ordered list is more important than its absolute or rela7ve value.

Don’t be afraid to highlight the points of interest.

DISTRIBUTION

Show values in a dataset and how oSen they occur. The shape (or ‘skew’) of a distribu7on can be a memorable way of highligh7ng

the lack of uniformity or equality in the data.

CHANGE

Give emphasis to changing trends. These can be short (intra-

day) movements or extended series traversing decades or

centuries: Choosing the correct 7me period is important to

provide suitable context for the reader.

OUR CATEGORIES

COMPOSITION

Show how a single en7ty can be broken down into its component elements. If the reader’s interest

is solely in the size of the components, consider a

magnitude-type chart instead.

SPATIAL

Used only when precise loca7ons or geographical paXerns in data

are more important to the reader than anything else.

1DEVIATIONEmphasize varia7ons (+/-) from a fixed reference point.

Da

DEVIATIONA simple standard bar chart that can handle

both negative and positive magnitude

values.

DIVERGING BAR

Da

DEVIATIONSplits a single value into

2 contrasting components (eg Male/

Female).

SPINE CHART

DDEVIATION

The shaded area of these charts allows a balance to be shown – either against

a baseline or between two series.

AREA CHART

2CORRELATION

Show the rela7onship between two or more variables.

SCATTERPLOT

The standard way to show the rela7onship between two

con7nuous variables, each of which has its own axis.

C O R R E L A T I O N

BUBBLE

Like a scaXerplot, but adds addi7onal detail by sizing the

circles according to a third variable and color to a fourth

C O R R E L A T I O N

ANIMATED BUBBLE

Like a scaXerplot, but adds addi7onal detail by sizing the

circles according to a third variable and color to a fourth

and anima7on for a fiSh!

C O R R E L A T I O N

HEAT MAPA good way of showing the

paXerns between 2 categories of data, less good at showing fine differences in amounts.

Ordering the entries can be quite powerful!

C O R R E L A T I O N

3Use where an item’s posi7on in an ordered list is more important than its absolute or rela7ve value.

RANKING

Ra

RANKINGStandard bar charts display the ranks of

values much more easily when sorted into order.

ORDERED BAR

Ra

RANKINGSEE PREVIOUS SLIDE

ORDERED COLUMN

Ra

RANKING

Perfect for showing how ranks have changed over

time or vary between categories.

There are many ggplot2 implementations :)

SLOPE GRAPH

With great power, comes great responsibility. These can quickly become an unmanageable mess…

Ra

RANKINGLollipops draw more

attention to the data value than standard bar/

column and can also show rank and value

effectively.

LOLLIPOP CHART

4Show values in a dataset and how oSen they occur.

DISTRIBUTION

HISTOGRAM

The standard way to show a sta7s7cal distribu7on - keep the gaps between columns

small to highlight the ‘shape’ of the data.

D I S T R I B U T I O N

BOX PLOT

Summarize mul7ple distribu7ons by showing the median (centre) and range of

the data

D I S T R I B U T I O N

VIOLIN PLOTSimilar to a box plot but more

effec7ve with complex distribu7ons (data that cannot

be summarized with simple average).

Also, only nerds understand it

D I S T R I B U T I O N

POPULATION PYRAMID

A standard way for showing the age and sex breakdown of

a popula7on distribu7on; effec7vely, back to back

histograms.

D I S T R I B U T I O N

5 Give emphasis to changing trends. These can be short (intra-day) movements or extended series

CHANGE

Ca

CHANGEThe standard way to

show a changing time series. If data are

irregular, consider markers to represent data

points.

LINE CHART

Ca

CHANGEUse to show the

uncertainty in future projections - usually this

grows the further forward to projection.

FAN CHART

Ca

CHANGEUse with care – these are good at showing changes

to total, but seeing change in components

can be very difficult.

AREA CHART

Ca

CHANGEA great way of showing

temporal patterns (daily, weekly, monthly) – at the

expense of showing precision in quantity.

CALENDAR HEAT MAP

6 Show how a single en7ty can be broken down into its component elements.

COMPOSITION

STACKED COLUMN

A simple way of showing part-to-whole rela7onships but can be difficult to read with more

than a few components.

C O M P O S I T I O N

PIE CHART

A common way of showing part-to-whole data – but be

aware that it’s difficult to accurately compare the size of

the segments.

C O M P O S I T I O N

WAFFLE

Good for showing % informa7on, they work best

when used on whole numbers and work well in mul7ple

layout form.

C O M P O S I T I O N

7 Used only when precise loca7ons or geographical paXerns in data are more important to the reader than anything else.

SPATIAL

Sa

SPATIAL

A great way of showing how areas have different

population sizes and different behaviors, not distorted by geographic

size.

(tilegramsR is amazing)

POPULATION TILES

Sa

SPATIALKeeps the overall shape

and layout of the geography so that it’s

identifiable, yet let’s you focus on the state or

province level analysis

REGION HEX

Sa

SPATIALGrid-based data values

mapped with an intensity color scale. As choropleth

map – but not snapped to an admin/political unit.

HEAT MAP

AND FINALLY…

Gosh there are a lot of choices. You mean you can’t just pick whichever one is prettiest? Well, you can, it just may not communicate anything to anyone, that’s up to you. Understanding what you’re

trying to communicate, and what the key components of that communication are, makes the difference between effective and ineffective data visualization.

CONCLUSION

THANK YOUE D U A R D O A R I Ñ O D E L A R U B I A

C H I E F D A T A S C I E N T I S T D O M I N O D A T A L A B

P L O T L Y A N D P L O T C O N A N D A N N A !

H T T P S : // A P P . D O M I N O D A T A L A B . C O M / U / E A R I N O / P L O T C O N 2 0 1 6