Two Ways to Easily Improve Your Graphics

Having been thoroughly impressed by Edward Tufte’s first book, The Visual Display of Quantitative Information, I decided to pick up another, Visual Explanations. Tufte envisions his books fitting together like so: The Visual Display of Quantitative Information is about pictures of numbers. Envisioning Information is about pictures of nouns, and Visual Explanations is about pictures of verbs. I was not drawn in by Visual Explanations (VE) as I was by The Visual Display of Quantitative Information (VQ). VE felt disjointed, and the lessons learned are less applicable for graduate students. However, there are still two very important lessons I gleaned: be subtle and avoid legends.

In a nutshell, VE states that images should be honest and scientific. Images should lend themselves to easy comparison through similar composition and repetition. Lastly, images alone can make an argument or tell a story through juxtaposition and symbolism.

I want to highlight a couple of these lessons, starting with one that I initially thought was wrong.

Subtlety is better than garishness.

Tufte states, “Make all visual distinctions as subtle as possible, but still clear and effective.” I would think you would want to eschew subtly to make sure the point comes across. However, this can be far too overwhelming, as shown above.

I’ve seen this lack of subtlety in lab presentations where the presenter shows a chart with ten or more curves. Every curve is a different color and has different markers for the points. Either a different color or a different set of markers would suffice, but both is excessive. Adding insult to injury, the default Excel colors are garish and unsubtle. To fix my own graphics, I’ve taken to using shades of black and gray to differentiate between each curve.

That same chart had a legend. Imagine trying to look at the legend, then the curves, then the legend, then the curves, then the . . . I simply gave up. You might think I give up too easily or that I am nit picking, so I’ll let you experience the difference between a legend and direct labeling for yourself.

Which image is quickest to understand? Which do you finish reading?

In light of this striking difference, I always label my curves directly. To do so in Excel, do not use a text box. Use a data label. You can overwrite the label and move it anywhere in the graph, so it is just like a text box. However, unlike a text box the label will stay in the same relative place on the graph and rescale along with the graph.

Gerrymandering.

The most important message for me is the same message Tufte stressed in his previous book. Graphics should be honest and scientific. To make a graphic honest, there must be a sense of scale and orientation for the viewer. If time-averaging or area-averaging has been applied, it should be done carefully as it can easily obscure important trends. The graphic to the right illustrates how area-averaging may doomed the residents of Broad Street. To make a graphic scientific entails quantification, comparison, and investigation of cause-and-effect. Quantification means applying a scale and going further to assign numbers to seemingly qualitative data. Comparison means plotting similar graphs on the same scale so that when they are put side-by-side, a line falling an inch in one means the same thing in the other. Investigation of cause-and-effect is the most difficult, but it means always plotting the suspected cause on the X-axis and the effect on the Y-axis, rather than plotting both against time.

Tufte does a fantastic job illustrating his message through the Challenger explosion. This is definitely a favorite case-study of his as it shows up in at least three of his books. Here he shows the actual set of slides the engineers sent to NASA the night before trying to persuade them not to launch. Tufte dissects the slides to show why they failed to persuade. The slides omitted many critical data points, failed to quantify the extent of damage to the O-rings, and presented important comparisons with many slides in between. At the end of the dissection, he shows the graphic below.

The slides only discussed the two labeled data points. This shows the full trend and makes a case for cold being dangerous.

By creating a damage-index, Tufte quantifies the data that was previously only qualitative. By plotting this data against temperature, Tufte makes a case for cold temperatures causing damage. I think that if this graphic had been shown to NASA, the launch would have been postponed.

While not all design decisions involve life or death, design can make or break the viewer’s comprehension of your argument. Taking the time to make your graphs easy to read will lead to better questions from the audience. Better questions will push your research along more quickly. Or maybe good design will help a grant-reviewer understand your argument and see why it is significant. In any case, this is not a small matter. While content is definitely king, it must be presented in an intelligible manner for it to take over the kingdom of the viewer’s mind.