A Critique of Tableau's "Visual Analysis Best Practice"
Originally sent to Tableau on 10/9/2014
I want to start this post by complimenting Tableau as being what I believe to be the best data visualization and BI platform on the market today. I am a huge supporter, a long-time Tableau Public user, a corporate user, an academic user, and a shareholder of the company. I teach Data Visualization at the University of Cincinnati, and we use Tableau extensively in class. I have introduced nearly five hundred students to Tableau in the last few years, with some students going on to have careers in data visualization. I also demonstrate Tableau in my corporate training work and various presentations on data visualization, data mining, and data science. I often tell people that Microsoft Excel will do everything it can to make your chart terrible and break rules by default while Tableau will do what it can to suggest things that are best practice. I could write extensively on this topic, but the focus of this post is to examine and discuss Tableau's publication of "Visual Analysis Best Practices".
Unfortunately, many of the things that Tableau has published in this white paper here are NOT best practice. In fact, I would grade some of these visuals harshly because of the use of bad practices. Here are a few major points.
1.) There is not a single mention of color deficiency (aka color blindness) in this white paper. This affects nearly 10% of men and 1% of women and the primary issue for someone who is color blind is distinguishing between red and green. In addition to not mentioning that red and green should be avoided, the white paper uses this color combination over and over again. Moreover, the white paper notes that many cultures perceive green to be good and red to be bad, but yet this combination is used as a categorical color scheme on a number of examples. Is "furniture" bad and "technology" good?
2.) While color is addressed in this white paper, I do not believe it is addressed in the context of "best practice". The white paper states, "Color can make the difference between a boring visualization and an inspiring one." This implies that color should be used to "spice up" your visualization and that visualizations without color are at risk of being boring. This is not best practice and is simply not true. Color should be used purposefully. Further, there are many amazing visualizations that do not use color or use it sparingly, and they are incredibly beautiful.
Here's a list of a few visualizations as examples:
One of my favorite data visualizations of all time is the Wind Map by Fernanda Viégas and Martin Wattenberg.
Tableau Iron Viz Champion and Zen Master Anya A'Hearn's limited use of color in her visualization Twitter TV.
I will counter Tableau's statement regarding color and state, "If your viz is boring without color, then adding color probably isn't going to help."
3.) There are 3 primary ways to use color in visualizations that should be considered best practice: Sequential, Diverging, and Categorical. These are not mentioned at all in the white paper and, in a number of cases, are not used appropriately in the visualizations. For example, in the District of Columbia Crimespotting Viz, there is a diverging color scheme. Besides the red-green problems in this visualization, I would suggest a better use of color overall. A categorical color scheme would work, but the color in this case is encoding the severity of the crime. To accomplish this, a simple sequential color scheme would have been the more appropriate use of color in this case, perhaps all in a sequential red or orange and in context with the severity of the crime. The color scheme seems to suggest a grouping with bad crimes in the red group and the less violent crimes in the green group, with homicide in black. As side note, in using this color scheme I'm not sure I'd classify sex abuse as less violent than robbery. Also related to color, it's generally considered "best practice" to use earth tones unless the intention is to alert the reader. There are a number of visualizations where this is a mixed use, using alerting with earth tones.
4.) Chart types - There are a number of references to the proper chart types, but in many cases, a less than optimal choice has been made. For example, in the section "Trends Over Time", the statement is made that line charts do not have the capability to see the overall funding trend or the funding for all sectors at any given point in time. This is a completely inaccurate statement; in fact, a simple line chart plotting the total funding would be the best way to visualize this data. The stacked area chart and stacked bar are certainly not the best approach for this data.
In the District of Columbia Crime Spotting Viz, there is a tree map with only 9 categories. While this may serve a purpose for interacting with the data, a bar chart would provide a much better visual comparison and would offer the same interactive features. Tree maps are useful for hierarchical data or in the case where there are lots of categories, far more than 9.
The white paper does a good job giving a short explanation of why pie charts are not great visualizations. However, it goes on to "encourage" the use of pie charts on maps. While the use of pie charts on maps is generally accepted by most data visualization professionals (including Stephen Few), I would certainly not categorize them as "best practice". In this case, a better description might be "a necessary evil" as the use of them is not ideal, but there are few alternatives and nothing that is dramatically better.
5.) The white paper discusses the idea of avoiding rotated text, which is certainly a best practice. However, in a few examples, various text labels were not avoided. In fact, one frustrating thing about Tableau's Y-axis format is that it will not allow a user to rotate the text horizontally, only at a 90 degree vertical. To achieve best practice in this case it is necessary to trick Tableau with a calculated field on a worksheet or the use of text boxes on a dashboard.
I sent this critique to Tableau a few weeks ago, and this white paper is now under review. As always, if you have any questions feel free to email me at Jeff@DataPlusScience.com