About 35 years ago, when I started my career as the managing editor of an economics journal, producing figure and tables was expensive. My memory is that at Princeton University, where I was based at the time, there was still a skilled draftsman who hand-drew beautiful figures, plotting the points and and then putting in a best-fit curve for the data using French curves.
For those born after 1980 (or 1970?), a French curve was a clear piece of plastic with a smooth edge that combined many different types of curves. The design is commonly attributed to the German mathematician Ludwig Bermester (1840-1927). There were three primary French curves: one for hyperbolas, one for ellipses, and one for parabolas. You rotated the French curve over your data until one of the curves seemed to fit the points, and then used the edge of the curve to draw a smooth curved line. Drawing publishable figure was once a specialized skill. In 8th-grade shop class in my Minnesota junior high school, the boys were required to take a one-quarter class in mechanical drawing, where we sat at sloped drafting tables and learned how to make blueprint-style detailed drawings of a screw, top and side views, with the head, the threaded shank, and the point precisely delineated. French curves were far too high-end for us.
Figures have gone from time-consuming and expensive to dirt cheap. As an economist, I’m generally in favor of improvements in quality accompanied by a sharp drop in price. But the related economic lessons are that when something gets much cheaper, it may be used much more often. When something is used much more often, diminishing returns may arise: while the first few figures may be illuminating and valuable, the last few are likely to range from overkill to actively confusing.
In addition, researchers have an incentive to generate lots of figures and tables as the basis for seminar presentations, so that listeners have something to look at. By the time that the researcher writes up the paper for publication, long propinquity to the figures has made them part of how the researcher conceives of the ideas.
Put it all together, and a few decades ago it was fairly common for me to receive a first draft with zero figures, or just a few. Now, it’s not unusual for me to receive a first draft with 10, 15, even 20 figures and tables—some of those with four or six or 12 separate panels. The paper can ends up feeling like a string of figures, each accompanied by bite-sized chunks of text.
Thus, I would promulgate some guidelines both for reducing the clutter of too many figures and for improving the quality of the remaining figures in written presentations.
- Written and spoken communication differ, just as reading and listening differ. In spoken exposition, summarizing a simple pattern with a simple bar graph or a line chart can help a listener focus. But for a reader, sometimes it’s just better to give people the numbers. Just because software will generate a figure doesn’t mean the figure is a good way of explaining to readers.
- Remember the basics. Figures need a title, and labels on the axes. Multiple lines or bars need labels, too. Use the note under the figure to list sources of data and to explain any abbreviations.
- A set of five or seven or eleven lines on a graph, each with its own separate key—one solid, one dashed, one dotted, one dash-dot-dash-dot, one dot-dot-dash, dot-dot-dash, and so on—requires some rethinking.
- It’s generally better—although admittedly not always practical— to label lines or bars directly, rather than using a separate key under the paper, which requires the eyes of the reader to jump back and forth.
- A greater ability to use color is one of the great and useful breakthroughs of modern graphics. Take advantage. But remember that some readers are red/green or blue/green or yellow/red colorblind. When using shades of color, the human eye is best at perceiving multiple shades of green, and worst at perceiving multiple shades of yellow.
- Sometimes it’s useful to label points: perhaps they represent countries or US states. But be cautious about labelling every country or state, which can lead to a blur of overlapping and unreadable labels. It’s often better to pick a few points worthy of labelling—perhaps specific points discussed directly in the text.
- A figure takes up the space of several hundred words. If it takes one or two sentences to convey the message of a figure, then the figure becomes just a big weirdly-shaped exclamation point for a message already fully conveyed in the text and dropping the figure probably makes sense.
- If you can’t make a good Figure 1 for an empirical paper with raw data, you ultimately aren’t likely to convince a lot of people (a saying I have heard attributed to the economist Steven Levitt).
- In the same way that you will automatically try to keep current with the evolution of terminology in your field of research, keep expanding your vocabulary for presentation of data. William Playfair, who invented the basic pie graph, bar graph, and line graph, died almost 200 years ago in 1823. More options are now available: bubble charts, waterfall charts, the bullet-graph variation on a bar chart, Mekko charts, heat maps of the cluster or spatial variety, and others.
- If you’ve been a victim of figures ranging from purposeless to indecipherable, don’t be a perpetrator. If you can’t remember having been a victim, take a closer look at the figures you are producing.
An advertising executive named Frank Barnard popularized the phrase, “A picture is worth a thousand words.” But Barnard was writing a century ago, and his topic was how to design advertising for the sides of moving streetcars. Now that modern statistical software can spit out a suite of figures on demand to be cut-and-pasted over to Powerpoint slides, Barnard’s rule of thumb needs rethinking.
The best figures are worth much more than a thousand words; indeed, a few well-chosen and well-constructed figures can sometimes convey the main arguments of an article (at least to a reader already somewhat knowledgeable in the subject). But figures can also be extraneous and unclear—imposing higher costs on readers than any plausible benefit. Expository wisdom involves learning the difference.