Data Graphics

We owe a collective debt to Edward Tufte for his brilliant book The Visual Display of Quantitative Information. The book is a work of art in itself and contains a handful of critical principles for effective information design. His impact can be seen in every issue of the New York Times and the Wall Street Journal, among other publications. Regrettably, Tufte's work is not readily available on the web. Furthermore, while quite elegant, his ideas are sometimes hard to translate into the practical world of Excel in which most of us live.

Stephen Few is a consultant and educator who has put into practice many of the Tufte principles. Few parrots Tufte's approach in some ways. In other ways, however, Few's approach is more applied and thorough, and he has made some of his work available on line. His book Show Me The Numbers is practical, concise, and filled with examples using Excel. This white paper is a pretty comprehensive summary of the ideas in the book.

I would like you to review at least five of the examples on Few's website. He does a nice job of transforming real business graphics into clear and informative alternatives. You may also find Few's quick quiz on data graphics useful in illustrating a summary of some key ideas. (It takes just a minute or two.)

A lot of subtleties are involved in doing the transformations in the examples on Few's website. My goal here is to get you to start to view data graphics critically and to avoid the most common pitfalls. There is a lot of material to explore here. Some of you will find this very interesting; others will not. At a minimum, you are responsible for knowing the following ideas, and demonstrating that you know them in the data graphics assignment. If you can master these ideas, you'll be able to create above-average graphics in professional life. If you devote a bit more attention to studying Tufte, Few, and the other resources here, you have the potential to be perceived as a data genius in your professional lives. This would be a very high return for any ambitious management professional.

1. Data Ink, Non-Data Ink, and the Data-Ink Ratio
The best summary of this idea is probably this short white paper by Stephen Few.

2. Direct labeling of data and the use of legends.

Simply put, you should almost never use a legend to identify data points or bars in a graph. Whenever possible, just directly label the points, lines, or bars with text. You can do this by inserting a text box on your chart using the drawing tools in Excel. Legends require the user to first encode the meaning of the legend and then apply that meaning to the chart. You help the viewer avoid all that cognitive effort by just directly labeling the chart. You may not be able to avoid a legend when, for instance, you are showing the same four bars (e.g., N, S, E, W) in several groups (e.g., 2009, 2010, 2011, 2012). However, most of the time, look for a way to delete the legend altogether. Following is a typically excellent graphic from the New York Times (editorial page, Oct. 9, 2011) showing direct labeling, and using Few's recommended approach for showing components of a total. Note also that there is almost no non-data ink in this graphic.

3. Making dimensions of a data graphic meaningful

A common defect in data graphics is to waste a dimension. For instance, consider this graphic produced by one of your classmates in the past. The data-ink ratio is pretty good, the information is interesting, but the horizontal axis is meaningless. Many alternatives are possible. For instance, why not order the regions in decreasing order of the metric? One could even order them by location west-to-east...but never order the categories arbitrarily.

4. Using scatter plots for relevant comparisons

This bar chart is what Excel will give you if you "insert chart" and select "bar chart" for a set of variables with two values each. This chart is intended to enable you to compare the perceptions of two groups of people (in this case US and Non-US MBA students) on a set of issues. (The values shown are the fraction agreeing/strongly agreeing.)Yet, comparing two values for a set of different categories or issues is exactly what a scatter plot is intended to do. A simple scatter plot with direct labeling of the data points (see rule 3) makes the comparisons much easier. The outliers are color coded for emphasis.

5. Avoiding chart junk.
Stephen Few whitepaper on chart junk.

6. Don't connect points that are not in a logical series (like a time series)

A common error in data graphics is to use a line graph to connect points on a graph which are not in fact related to each other by a series. For instance, consider below this student work as an example. It's a fairly nicely implemented graphic, in terms of data-ink and so forth. However, it's really a bar chart masquerading as a line chart. The vertices of these lines are the heights of the bars for each of the five categories. They really should be just shown as bars, so that no one is confused into thinking that these points are in some kind of series, as in a time series.

You may find these instructional modules by your classmates useful in reinforcing these ideas.

Simon Lu's Video on Data Graphics
(you may want to pause this on the final graphic to study it more carefully)

Deb Crandall's video on Data Graphics
Crandall_DataGraphics.m4v

More on Data Graphics (optional):

Stephen Few's full library of white papers (lots)
Stephen Few's book Show Me the Numbers is a practical, Excel-oriented text. The Few book mimic's Tufte and is more sensible and concrete, even if less elegant.
Blog on data graphics junkcharts.typepad.com
Stanford alumni magazine article on Tufte: http://www.stanfordalumni.org/news/magazine/2007/marapr/features/tufte.html