Guiding principles for effective data visualization

Slide Deck available at https://bit.ly/2MFU9kn

William R. Buchanan, Ph.D.
Director, Office of Grants, Research, Accountability, & Data
Fayette County Public Schools

Background

What is the purpose of data visualization?

  • Explore
  • Expose
  • Explain

What is communication?

Visual adaptation of Shannon's (1948) communications model.

How does the visualization process get us there?

Adaptation of Jeff Heer's (2015) framework for data visualization.

Pop Quiz

Go to: PollEv.com/billybuchana011

OR

Text BILLYBUCHANA011 to 22333 to connect

Q1. Which slice in this pie chart is the SECOND largest?

Q2. Which slice in this pie chart is the SECOND largest?

Q3. Which bar is the SECOND largest?

What advice do I have for visualizing data?

  • Mapping between visual and measurement scales should be one and the same.
  • Do not alienate potential end-users.
  • Avoid obfuscating the message.
  • Bind helpful information in the UI without blocking the user's view.
  • Establish clear and consistent standards for visual encodings.

Mapping between visual and measurement scales should be one and the same

Nominal Scale

  • Used for categories that have no natural order.
  • Can only test equality and inequality.

Ordinal Scales

  • Used for categories that have a natural order.
  • Can test for equality/inequality.
  • Can test order of magnitude.
  • No other mathematical operators are defined on this scale.

Intervallic Scales

  • Used in cases where there is no true zero or where the location of zero is arbitrary.
  • Can test for equality/inequality.
  • Can test order of magnitude.
  • Mathematical operations are defined on this scale.

Ratio Scales

  • Used when there is a true zero or when the location of zero is fixed.
  • Has all of the same prop

Visual Scales

Visual Scale type 1 2 3 4 5
Qualitative 5 1 4 2 3
Sequential 1 2 3 4 5
Divergent -2 -1 0 1 2

Quiz Time

Q4. What type of measurement scale is used for the letter grades below?

F D C B A

Q5. What type of visual scale is used for the colors for the letter grades below?

F D C B A

Q6. The visual scale mapping from the previous example is aligned with the measurement scale of the values.

Q7. What type of visual scale is used for the risk groups below?

Low Risk Moderate Risk High Risk

Q8. The visual scale mapping from the previous example is aligned with the measurement scale of the values.

Q9. What type of visual scale is used for the colors for the risk groups below?

Low Risk Moderate Risk High Risk

Q10. The visual scale mapping from the previous example is aligned with the measurement scale of the values.

Takeaway

  • Visual scales should reinforce the underlying measurement scale.
  • Stop light colors are nominal.
  • Divergent scales are useful/helpful when you want to highlight distances in multiple directions.

Do not alienate potential end-users

Example of Stata graph created with ggplot2 style aesthetics.
Example of Stata graph created with ggplot2 style aesthetics with colors simulated to show tritanopia.
Example of Stata graph created with ggplot2 style aesthetics with colors simulated to show protanopia
Example of Stata graph created with ggplot2 style aesthetics with colors simulated to show deuteranopia.
Example of Stata graph created with ggplot2 style aesthetics with colors simulated to show achromatopsia.
Example of same Stata graph using Kelly's contrasting colors.
Example of same Stata graph using Kelly's contrasting colors with colors simulated to show tritanopia.
Example of same Stata graph using Kelly's contrasting colors with colors simulated to show protanopia.
Example of same Stata graph using Kelly's contrasting colors with colors simulated to show deuteranopia.
Example of same Stata graph using Kelly's contrasting colors with colors simulated to show achromatopsia.

Color contrast issues can also alienate end users.

Color contrast issues can also alienate end users.

Takeaway

  • Be careful about the selection of colors to denote groups/categories/values.
  • Be mindful of color contrast issues for any text in your visualizations.
  • Use available tools/technologies to proof your visualizations prior to publication.

Avoid obfuscating the message

  • What types of comparisons are easy for humans to perceive?
  • What element do you want/need end users to focus on?
  • How does your choice of graph/chart support/hinder that comparison?

Takeaways

  • Humans + Visual Comparison of Areas = Incorrect Interpretation
  • To see change over time, you need to display the slopes that represent the change.
  • Stacked bar/area charts can be challenging to interpret due to their construction.

Bind helpful information in the UI without blocking the user's view

  • If an end user isn't familiar with your graph/chart type, how do you know if they will interpret the visualization correctly/consistently?
  • How can end users actively engage with your visualization if they need additional contextual information?
  • What can you do to support multiple stakeholder groups interacting with the same visualizations?
  • How can you help end users avoid unsupported inferences/conclusions from your visualizations?
Example of Fayette County Public Schools' data dashboard including the click for additional info illustrating binding info to visualization.
Example of Fayette County Public Schools' data dashboard after user clicks on the click for additional info element in the user interface.
Example of Fayette County Public Schools' data dashboard after the user clicks on the click for additional info element and expands the what story is this chart telling section.
Example of Fayette County Public Schools' data dashboard after the user clicks on the click for additional info element and expands the how to use these data section.
Example of Fayette County Public Schools' data dashboard showing how external resources can also be bound to the visualization.

Takeaways

  • Guides to help users interpret the visualizations can drive consistent use and interpretation.
  • Make the UI elements obvious so end users can find the information easily.
  • Make the information as proximal to the visualization w/o obscuring the view.
  • Bind external resources that are regularly sought to create additional value.

Wrap Up

Wrap Up

  • Make your visualizations accessible to all potential end users whenever possible.
  • Reinforce the underlying meaning of the data with the visual mapping.
  • Humans and visual comparison of areas don't mix.
  • Empower the end users to make appropriate inference from your visualization with supporting materials.
  • Create/implement visualization standards to provide end users with consistency.

Additional Resources

People
People (continued)
People (continued)
Books
Books (continued)
Other
Papers/Other References