Above All Else Show the Data

Best Practices in Data Visualization

Jorge Martinez

University of Houston

2023-05-24

What is the point of data visualization?

What pattern do you see?

Example from Anscombe (1973).

Data graphics visually display measured quantities by means of combined use of points, line, a coordinate system, numbers, symbols, words, shading and color (Tufte 2015).

But not all graphics are created equal

Agenda

• Principles of Graphical Excellence
• Data distortions
• Data-ink
• Identify graphic crimes
• The importance of color
• Socially responsible visualizations
• Resources

Principles of Graphical Excellence

Excellence in statistical graphics consist of complex ideas communicated with clarity, precision, and efficiency (Tufte 2015, pg 14)

Graphics should…

• show the data
• induce viewer to think about the substance and not methodology, design, or technology
• avoid distorting the data
• present many numbers in a small space
• make large data sets coherent
• compare different pieces of the data
• reveal the data at several levels of detail
• display a clear purpose

• Focus primarily on the data and the story behind the data

• Excited about how we put it together. Frequent refrain: the Tableau dashboard, the Power BI report

• Spend a good amount of time on distortion and how to avoid it. Distortion=deceit

• Sometime tables or stand alone numbers suffice

• We are always asking compared to what and how other variables interact

• Data are almost always nested. In our context, students are embedded in majors, departments, and colleges. Outside of the institution, students come from neighborhoods, census blocks, counties, cities, and states. These all inform the student experience

• What is the story you want to tell? Who is the audience? What impact will your analyses have on policy and practice?

Data Maps: Death Rates from Cancer, Female

• This map reports age-adjusted death rate from various types of cancer for the 3,056 counties in the United States.

• This map portrays 21,000 numbers. This is a good example of “presenting many numbers in a small space.”

• Shows “many levels of analysis” from overall patterns to detection of fine county-by-county detail.

• Example

  • High death rates from cancer in the northeast part of the country and around the Great Lakes

  • Low rates in the east-west band across the middle of the country

  • Unusual hot-spots including northern Minnesota and a few counties in Iowa and Nebraska along the Missouri River. You can also see some along the Mississippi River

• National Cancer Institute (U.S.). and Thomas J. Mason. 1975. Atlas of Cancer Mortality for U.s. Counties 1950-1969. Bethesda Md: U.S. Dept. of Health Education and Welfare Public Health Service National Institutes of Health.

Data Maps: Death Rates from Cancer, Male

What about cancer rates for men?

• High rates for men than for women in the south, particularly Louisiana which is probably due to occupational exposure from working with asbestos in shipyards
• What other sources of error might there be?
  • The data source itself is flawed - the data come from death certificate reports, which are influenced by doctor diagnosis and coroners that report the data at the time.

Graphical Integrity: Proportionality

• The representation of numbers, as physically measured on the surface of the graphic itself, should be directly proportional to the numerical quantities (pg.56)

• Different people see the same areas differently and perceptions change with experience and context

• Tables sometimes outperform graphics in clarity, but only for small data sets of 20 numbers or less

Fuel Economy Standards

Fuel Economy Standards

\[ \frac{27.5-18.0}{18.0} \times 100=53\% \]

\[ \frac{5.3-0.6}{0.6} \times 100=783\% \]

• On most roads the future is in front us on the horizon. In this case the labels are flipped - the older dates are ahead of us and the future is at our feet. What?

• The dates on the left remain a constant size on the page even though they move along with the road towards the horizon

Graphical Integrity: Labeling

• Clear, detailed, and thorough labeling should be used to defeat distortion and ambiguity

• Write out explanations of the data and label important events

• Includes labeling the coordinate system and providing an informative title

Fuel Economy Clarified

Data-Ink

• A large share of ink on a graphic should present data-information

• Data ink is the non-erasable core of a graphic arranged in response to variation in the numbers represented

• Every bit of ink on a graphic requires a reason - and that reason is to present new information

  • Maximize the data-ink ratio

  • Erase non-data-ink

  • Erase redundant data-ink

  • Revise and edit

Erase non-data-ink

What elements of this chart are not data-ink?

Erase non-data-ink

The grid lines!

Erase non-data-ink

Erase non-data-ink

Erase non-data-ink

Pie Charts

• Pie charts are all about proportions and this one is outta control.

• It’s not necessarily about equal proportions, it’s just that we are not built to see angles so well.

• The best measure of proportion are straight lines.

• There are 20 slices - is this easy to read despite it being in German?

The Pie Proof¹

What is the pattern here?

1. What’s Wrong with Pie Charts? (Holtz 2023)

Can you see the pattern now?

Pies are for eating, not charting

What’s wrong with this visualization?

What’s wrong with this visualization?

What’s wrong with this visualization?

COLORS!

Colors are an important aesthetic

• All data visualizations use color, depending on how you define black and white

• Three ways to use colors

  • Qualitative colors show category

  • Sequential colors show order, rank, or numeric values

  • Diverging colors show values that move around a zero mark

Qualitative¹

1. RColorBrewer Package

Sequential

Diverging

Color Brewer

https://colorbrewer2.org/

Improving graphics with/out color¹

Move color mapping to bar height

1. Examples from “What to consider when choosing colors” by Lisa Muth (2018)

Improving graphics with/out color

Move color mapping to bar height and position

Color consistency

Color to highlight relationships in context

Color ramps must match psychology

Accessible colors

Equity Awareness

• Published by the Urban Institute, the report talks about how to be socially responsible with data

• It addresses many aspects of data distortion and deceit

Equity awareness in data visualization¹

• Data analysts should think intentionally about how we can learn from and speak to audiences that reflect the diversity of the people and communities we focus

• Systemic discrimination is and can be generated by how we use and misuse data

  If I were one of the data points on this visualization, would I feel offended? - Kim Bui

• A neighborhood earned a red color if African Americans lived in it, even if it was a solid middle-class neighborhood of single-family homes

1. Urban Institute (2021)

Demonstrating empathy

• Put people first: data shown reflect the lives and experiences of real people

• Use personal connections to help readers and users better connect with the material: pair charts with personal stories

• Create a platform for engagement: interactive graphics allow users to find themselves in the data

• Consider how framing an issue can create a biased emotional response

• If the underlying data is biased, graphics can amplify bias and the harm that bias generates

Put people first

Put people first

• Titles, text, and labels are among the first things readers scan when encountering a chart (Borkin et al. 2015)

• Show Tableau eye-tracking study

Language and racial equity awareness

• Word choice: “inmate” has been argued by some as dehumanizing and references people by their crimes and punishment. There is stigma involved and is antithetical to a rehabilitation model.

• Explicitly stating “Racism” focuses on the main root of the problem

• Saying “People of color” instead of singling out a specific race shows that it is not just a black/white divide - other groups are just as marginalized

• Anything else? What about other? How does that make you feel if you fall into that group?

  • The authors of this report suggest using different labeling which is people-first such as

    • Another race

    • Additional groups

    • All other self-descriptions

Ordering data purposefully

• Many of the largest demographic surveys in the US start race categories with white and black as the first two options

Colors and equity awareness

Colors and equity awareness

Conclusion

• Remember the Principles of Graphical Excellence: clarity, precision, and efficiency

• Avoid data distortion through Graphical Integrity

• Maximize data-ink ratio

• Have a clear purpose for your graphics: what do you want to communicate?

• Do no harm: be aware how your data graphics perpetuate bias or systemic discrimination

Above All Else Show the Data!

Additional Resources

Data-to-Viz: Find the Right Viz

https://www.data-to-viz.com

WTF Viz: Examples of what not to do

https://viz.wtf

Atkinson Hyperlegible Font

Free downloadable font for low-vision readers by the Braille Institute used throughout this presentation.

Contact

Jorge Martinez

University of Houston

Director of Research and Reporting

Enrollment Services

jxmartinez.com