What Makes a Good Data Visualization?

Robert Norberg
August 19, 2015

5 Commandments

  1. Know your audience and your medium
  2. Tell a story
  3. Do not mislead
  4. Less is more
  5. Maximize your data/ink ratio

links

You can find these slides

Here: http://rpubs.com/rnorberg/good_data_viz

Know your audience

Team Memebers

  • Quick and dirty
  • Lots of shorthand (e.g. cst_cnty_fips)
  • Default colors, styles, etc
  • More complicated is OK

General Public

  • Make sure axis labels, title, legend, caption are perfect
  • Use a colorblind friendly palette
  • Simpler is better

Know your audience

Team Memebers

plot of chunk unnamed-chunk-1

General Public

plot of chunk unnamed-chunk-2

Know your medium

Journal

  • Only static images of static data are possible
  • Achromatic usually required
  • Standard page size
  • Comes with a particular audience
  • Is there a precedent for displaying similar data?

Web

  • Interactive, animated, dynamic data all possible
  • Colors encouraged
  • Screen sizes vary widely
  • The world is your audience
  • Eschew precedents; creativity gets clicks/likes/re-tweets

Tell a story

  • What is it you are trying to convey to viewers?
  • Choose one feature to highlight
  • Highlight that feature while providing appropriate context

Tell a story

plot of chunk unnamed-chunk-3

Do not mislead

plot of chunk unnamed-chunk-4

plot of chunk unnamed-chunk-5

Do not mislead

The classic example: Pie charts are misleading.

Do not mislead

  • NEVER USE 2 Y AXES
  • SERIOUSLY DON'T
  • Instead, make 2 plots and juxtapose them

Avoid truncating the Y axis

Density plots >> Histograms

plot of chunk unnamed-chunk-6

plot of chunk unnamed-chunk-7

Things to keep in mind

  • We perceive height/area of rectangles better than angles/area of pie charts
  • We perceive area in 2D much better than area in 3D
  • Lines imply progress/movement through time
  • Adjacency implies similarity
  • Differences in color imply differences in the data
  • With great color comes great responsibility

Some color theory

  • Think carefully about the connotations that accompany colors (e.g. don't shade profit in red)
  • Use color only when needed to serve a particular communication goal
  • Use soft, natural colors to display most information and bright and/or dark colors to highlight information that requires greater attention
  • When using color to encode a continuous range of values, use a single hue and vary intensity from pale (for low values) to dark/bright colors (for high values)
  • 4.5% of all people are colorblind - use colorblind friendly palettes

Less is more

I would have written a shorter letter, but I did not have the time. - Blaise Pascal

I would have made a simpler figure, but I'm under some tight deadlines. - Robert Norberg

Less is more

  • 5 colors or fewer
  • Don't over plot
  • No chart junk
  • Use as little text as possible on the plot (detailed captions are encouraged)
  • If the legend can fit inside the plot area, put it there
  • Consider making an “Other” category

Maximize your data/ink ratio

If I erase this, will my visualization convey less information?

  • Avoid chart junk
    • Making bars 3D
    • Giving objects shadows
    • Unnecessary animation
  • Avoid excessive text
  • Grid lines should be understated

Recommended reading

Inspiration

You can find these slides

Here: http://rpubs.com/rnorberg/good_data_viz