#whoami

Resources (free)

Healy (http://socviz.co)

Wilke (https://serialmentor.com/dataviz/)

Other Resources

• My classes!
• Sequence
  • EDLD 651: Introductory Educational Data Science (EDS)
  • EDLD 652: Data Visualization for EDS
  • EDLD 653: Functional Programming for EDS
  • EDLD 654: Machine Learning for EDS
  • Capstone

Where to start?

• I really recommend moving to R as quickly as possible

http://r4ds.had.co.nz

ggplot2!

Last note before we really start

• These slides were produced with R

• See the source code here

• The focus of this particular talk is not on the code itself

Other elements to consider

• Text

  • How is the text displayed (e.g., font, face, location)?

  • What is the purpose of the text?

• Transparency

  • Are there overlapping pieces?

  • Can transparency help?

• Type of data
  • Continuous/categorical
  • Which can be mapped to each aesthetic?
  • e.g., shape and line type can only be mapped to categorical data, whereas color and size can be mapped to either.

Talk with a neighbor

How would you encode these data into a display?

Comparison

• Both represent three scales

  • Two position scales (x/y axis)
  • One color scale (categorical for the first, continuous for the second)

More scales are possible

Additional scales can become lost without high structure in the data

Thinking more about color

Three fundamental uses

1. Distinguish groups from each other

2. Represent data values

3. Highlight

Discrete items

• Often no intrinsic order

Qualitative color scale

• Finite number of colors
  • Chosen to maximize distinctness, while also be equivalent
  • Equivalent
  • No color should stand out
  • No impression of order

Some examples

See more about the Okabe Ito palette origins here: http://jfly.iam.u-tokyo.ac.jp/color/

Sequential scale examples

Colors to represent continuous values

Diverging palettes

Earth palette

Common problems with color

Too many

More than 5-ish categories generally becomes too difficult to track

Use labels

still too many...

Better

Get a subset

Best

(but could still be improved)

Problem with default ggplot2 palette

Alternative: viridis

Revised version

Last few note on palettes

• Do some research, find what you like and what tends to work well

• Check for colorblindness

• Look into http://colorbrewer2.org/

What is it?

Above all else, show the data

-Edward Tufte

• Data-Ink Ratio = Ink devoted to the data / total ink used to produce the figure

• Common goal: Maximize the data-ink ratio

Example

• First thought might be... Cool!

Minimize cognitive load

• Empirically, Tufte's plot was the most difficult for viewers to interpret.

• Visual cues (labels, gridlines) reduce the data-ink ratio, but can also reduce cognitive load.

An example

Which do you prefer?

Advice from Wilke

Whenever possible, visualize your data with solid, colored shapes rather than with lines that outline those shapes. Solid shapes are more easily perceived, are less likely to create visual artifacts or optical illusions, and do more immediately convey amounts than do outlines.

emphasis added

Another example

Labels in place of legends

Prior slide is a great example of when annotations can be used in place of a legend to

• reduce cognitive load
• increase clarity
• increase beauty
• maximize the figure size

Grouped data

Distributions

How do we display more than one distribution at a time?

Boxplots

Violin plots

Jittered points

Sina plots

Stacked histograms

Overlapping densities

Ridgeline densities

Titanic data

## # A tibble: 1,313 x 5
##   name                                          class   age sex    survived
##   <chr>                                         <chr> <dbl> <chr>     <int>
## 1 Allen, Miss Elisabeth Walton                  1st   29    female        1
## 2 Allison, Miss Helen Loraine                   1st    2    female        0
## 3 Allison, Mr Hudson Joshua Creighton           1st   30    male          0
## 4 Allison, Mrs Hudson JC (Bessie Waldo Daniels) 1st   25    female        0
## 5 Allison, Master Hudson Trevor                 1st    0.92 male          1
## 6 Anderson, Mr Harry                            1st   47    male          1
## # … with 1,307 more rows

Boxplots

Violin plots

Jittered point plots

Sina plot

Stacked histogram

🤨

Dodged

Better

Overlapping densities

Note the default colors really don't work well in most of these

Ridgeline densities

Bar plots

Flipped bars

Dotplot

Heatmap

Tuition data

## # A tibble: 6 x 13
##   State `2004-05` `2005-06` `2006-07` `2007-08` `2008-09` `2009-10` `2010-11`
##   <chr>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>
## 1 Alab…  5682.838  5840.550  5753.496  6008.169  6475.092  7188.954  8071.134
## 2 Alas…  4328.281  4632.623  4918.501  5069.822  5075.482  5454.607  5759.153
## 3 Ariz…  5138.495  5415.516  5481.419  5681.638  6058.464  7263.204  8839.605
## 4 Arka…  5772.302  6082.379  6231.977  6414.900  6416.503  6627.092  6900.912
## 5 Cali…  5285.921  5527.881  5334.826  5672.472  5897.888  7258.771  8193.739
## 6 Colo…  4703.777  5406.967  5596.348  6227.002  6284.137  6948.473  7748.201
## # … with 5 more variables: `2011-12` <dbl>, `2012-13` <dbl>, `2013-14` <dbl>,
## #   `2014-15` <dbl>, `2015-16` <dbl>

By state: 2015-16

🤮🤮🤮

Two puke emoji version

One puke emoji version

Kinda smiley version

Highlight Oregon

Not always good to sort

Much better

Heatmap

Better heatmap

Even better heatmap

Quick aside

• Think about the data you have
• Given that these are state-level data, they have a geographic component

Line drawings

As discussed earlier

Much worse

Unnecessary 3D

Much worse

Unnecessary 3D

Horrid example

Used relatively regularly

Pie charts

Especially w/lots of categories

Alternative representation

A case for pie charts

• $n$ categories low,
• differences are relatively large
• familiar for some audiences

The anatomy of a pie chart

Pie charts are just stacked bar charts with a radial coordinate system

Horizontal

My preference

Dual axes

• One exception - if second axis is a direct transformation of the first
  • e.g., Miles/Kilometers, Fahrenheit/Celsius

See many examples here: http://www.tylervigen.com/spurious-correlations

Truncated axes

Not always a bad thing

It is tempting to lay down inflexible rules about what to do in terms of producing your graphs, and to dismiss people who don’t follow them as producing junk charts or lying with statistics. But being honest with your data is a bigger problem than can be solved by rules of thumb about making graphs. In this case there is a moderate level of agreement that bar charts should generally include a zero baseline (or equivalent) given that bars encode their variables as lengths. But it would be a mistake to think that a dot plot was by the same token deliberately misleading, just because it kept itself to the range of the data instead.

Scaling issues

Conclusions

Practical takeaways to make better visualizations

1. Avoid line drawings

2. Sort bar charts in ascending/descending order as long as the other axis does not have implicit meaning

3. Consider dropping legends and using annotations, when possible

4. Use color to your advantage, but be sensitive to color-blindness, and use the right kind of palette

5. Consider double-encoding data (shapes and color)

6. Make your labels bigger! Didn't talk about this one much but it's super common and really important

Some things to avoid

• Essentially never

  • Use dual axes (produce separate plots instead)

  • Use 3D unnecessarily

• Be wary of

  • Truncated axes

  • Pie charts (particularly with lots of categories)