Ten Simple Rules for Better Figures: The 2014 Paper Every Chart-Maker Should Read

In short: Ten Simple Rules for Better Figures is a 2014 editorial in PLOS Computational Biology by Nicolas P. Rougier, Michael Droettboom and Philip E. Bourne.¹ Its main argument is that a figure is not a picture of your data. In the authors' words, it is "a graphical interface between people and data." Their ten rules come in a deliberate order. You decide who the figure is for and what it should say before you touch colors, defaults or tools.

The paper is short. It is open access and was released under a CC0 public-domain dedication. The authors also published the matplotlib code for every example figure on GitHub.² This post covers each rule in plain language, explains what the paper's example figures show, and closes with a checklist you can apply to the next chart you make.

What is the paper actually arguing?

The paper's main claim is that one dataset can be drawn in many ways, and different viewers read the same drawing differently. Plotting is not a neutral, automatic step. The authors start by noting how many options there are: "scatter plots, linear plots, bar plots, and pie charts, to name just a few". They add that "the same data, using the same type of plot, may be perceived very differently depending on who is looking at the figure."¹

So they reject the textbook definition of visualization as simply "graphically displaying scientific data." They call a figure an interface, something designed for a user. The ten rules follow from that idea, and they fall into three groups:

Rules 1–3: decide the job before you draw

Rule 1: Know your audience. The authors say problems start "when how a visual is perceived differs significantly from the intent of the conveyer." A figure made for you and your collaborators can skip steps. A journal figure has to be correct and complete for a wider audience. A figure for students has to explain a concept. The paper says the general public "may be the most difficult audience of all," because the figure has to be simple and may need to approximate, showing only the most important part of the result.¹

The paper's Figure 1 remakes a 2007 New York Times cancer chart. Its layout works well for general readers, because the central labels put "cancer" front and center. But the authors admit that the same layout makes it hard to estimate the number of kidney cancer deaths. That is fine for a newspaper. It would not be acceptable in a scientific publication unless the actual numbers appeared somewhere else in the article.

Rule 2: Identify your message. A figure should state one idea that would take too long to explain in words. The authors compare this to writing: you develop the figure only after you know its point, "just as you would take the time to craft your words and sentences when writing an article only after deciding on the main points of the text." Figure 2 makes the point with an artificial checkerboard pattern that never appears in real experiments. They use it because it shows the key result clearly: how strongly the retina-to-brain mapping magnifies the foveal region.¹

Rule 3: Adapt the figure to the support medium. A slide, a poster and a printed page are different viewing situations. In a talk, viewers see the figure briefly, from a distance, while listening to you. So you need thicker lines, bigger text, strong contrast and no vertical text. On a page, readers can look as long as they like, so you can add detail and put explanation in the caption. The paper's advice is blunt: "you should abandon the practice of extracting a figure from your article to be put, as is, in your oral presentation."¹ Figure 3 shows one simulation drawn twice. The talk version has fewer trajectories, fewer ticks, larger labels and no title.

Rules 4–8: say it clearly, and don't distort it

Rule 4: Captions are not optional. A caption "explains how to read the figure and provides additional precision for what cannot be graphically represented." The authors suggest writing the caption as the explanation you would give in front of a poster, while anticipating the questions people would ask. If exact values matter, don't expect readers to measure bar heights. Put the numbers in the text or label them clearly on the figure.¹

Rule 5: Do not trust the defaults. Every plotting tool comes with default settings, and they are tuned for no plot in particular. The paper puts it this way: "they are good enough for any plot but they are best for none." Figure 4 shows a matplotlib sine/cosine plot with default settings next to a hand-tuned version. The data is the same, but the tuned one is easier to read.¹

Rule 6: Use color effectively. Citing Edward Tufte, the authors say color "can be either your greatest ally or your worst enemy." Their simplest test is a question: "Is there any reason this plot is blue and not black?" If you don't know the answer, just keep it black. They warn against defaulting to jet or rainbow colormaps. They also give a three-way classification worth memorizing:¹

• Sequential: one color varying in intensity, for values running from low to high.
• Diverging: one color shading into another, to highlight deviation from a median value.
• Qualitative: rapidly changing colors, for discrete or categorical data.

In Figure 5, a rainbow map and a diverging map both hide detail in the high-frequency part of a signal. A sequential purple map shows it. The paper also tells you to avoid too many similar colors, because readers with color vision deficiency may not be able to tell them apart.

Rule 7: Do not mislead the reader. This is the most practical rule in the paper, and Figure 6 shows two common ways charts mislead:

1. Area vs. radius. The values 30, 20, 15 and 10 are drawn as discs. When the value sets the disc's radius instead of its area, the 10 looks tiny next to the 30, even though the real ratio is only 3:1. The authors note this mistake is "actually very frequent in the literature," because many plotting commands set marker size by radius by default.
2. Truncated axes. Ten bars on a 0–100 axis look nearly equal. The same bars on an 80–100 axis look very different. The authors say that even with a labeled axis, "the effect would persist because the bars are the most salient information."

The paper also warns about choices that are "wrong by design," such as pie charts or 3-D charts used to compare quantities. Its rule of thumb: use the simplest plot type that conveys your message, show the full range of values when it matters, and ask colleagues how they read your figures.¹

Rule 8: Avoid "chartjunk." Tufte coined the term for visual elements that add nothing or add confusion: extra colors and labels, colored backgrounds, grid lines that don't help. The paper quotes his test: "Regardless of the cause, it is all non-data-ink or redundant data-ink, and it is often chartjunk." It also quotes Stephen Few, who says graphs should "represent all the data that is needed to see and understand what's meaningful."¹

The paper does not treat this as a ban on all decoration. Figure 7 takes seven overlapping, auto-colored series with a legend that covers the data and redraws them as seven small panels. Each panel shows one series, with the others drawn faintly behind it. The labels sit directly beside each panel, so no legend is needed. That redesign keeps a gray background band on purpose, because it marks the [−1, +1] range. The same element can be junk in one figure and useful in another.

Rules 9–10: judgment over convention, and the right tool

Rule 9: Message trumps beauty. Each field has its own figure conventions. They are worth knowing because they make results comparable and help you spot errors. When no convention fits, the authors suggest adapting (not copying) good figures from the literature. They also warn about the web, where "the frontiers between data visualization, infographics, design, and art are becoming thinner and thinner." Their conclusion: "in science, message and readability of the figure is the most important aspect while beauty is only an option."¹

Figure 8 is a fun example. It plots the "uncanny valley" curve in a hand-sketched xkcd style. The sketchy look is the point: it tells viewers the data is hypothetical and that only the overall shape matters.

Rule 10: Get the right tool. The tool you use to analyze data doesn't have to be the tool you use to draw it, because "you can always export data." The paper lists open-source options for different jobs: matplotlib and R for plotting, Inkscape for vector editing and touch-ups, TikZ/PGF for programmatic graphics in TeX, GIMP and ImageMagick for raster editing and conversion, D3.js for interactive web graphics, Cytoscape for networks, and Circos for relational and genomic data.¹

Why does a 2014 paper still matter?

The rules are about how people read figures, not about software. Tools have changed a lot since 2014, but the problems in Rules 5–7 still come up whenever someone accepts a tool's default output: generic default styling, color maps chosen by habit, disc sizes set by radius, and axes cropped by auto-scaling. The paper's most reusable idea is its order of work. Decide the audience and the message first, then the medium, and only after that the styling. Most bad charts come from doing it the other way around.

The paper is also unusually easy to learn from. Its code is public, and every rule has a before-and-after figure you can rebuild.²

A checklist you can paste into your notes

Keep this next to wherever you draft figures. It's a plain Markdown list, so it works in any notes app that supports checklists (MNMNOTE included).

## Figure check: Rougier, Droettboom & Bourne (2014)
- [ ] Audience named (self / journal / students / public)
- [ ] One-sentence message written before plotting
- [ ] Version made for this medium (slide ≠ paper ≠ poster)
- [ ] Caption says how to read it + where exact values live
- [ ] Defaults changed on purpose (fonts, ticks, sizes, colors)
- [ ] Every color has a reason; colormap matches data
      (sequential / diverging / qualitative); colorblind-safe
- [ ] No radius-for-area; axis range honest; no 3-D/pie for comparison
- [ ] Removed every element that doesn't carry information
- [ ] Readable > pretty
- [ ] Right tool for the job (export data if needed)

FAQ

Who wrote Ten Simple Rules for Better Figures? Nicolas P. Rougier (INRIA Bordeaux Sud-Ouest / LaBRI / Institute of Neurodegenerative Diseases), Michael Droettboom (Space Telescope Science Institute), and Philip E. Bourne (Office of the Director, U.S. National Institutes of Health). It was published September 11, 2014, as an editorial in PLOS Computational Biology, volume 10, issue 9, e1003833.¹

Can I reuse the paper's figures? The article is released under a Creative Commons CC0 public-domain dedication. The scripts for all of its figures are on GitHub at rougier/ten-rules.¹²

What colormap does the paper recommend? It doesn't name one "best" colormap. It tells you to match the colormap type to your data: sequential for low-to-high values, diverging for deviation from a midpoint, qualitative for categories. It also says to avoid jet or rainbow "unless there is an explicit reason," and to stay away from easily confused similar colors.¹

Why are disc and bubble charts often misleading? Readers judge a disc by its area, but many tools size discs by radius. Because area grows with the square of the radius, a 3:1 ratio in the data looks like a 9:1 ratio on the page. The paper's Figure 6 demonstrates this.¹

Source

Rougier NP, Droettboom M, Bourne PE (2014). Ten Simple Rules for Better Figures. PLOS Computational Biology 10(9): e1003833. https://doi.org/10.1371/journal.pcbi.1003833