What are best practices for designing publication-quality figures?

Question

I am a postgraduate student in mathematics. I am writing my thesis, for which I need to prepare some Matlab plots and figures. I had sent some plots to my supervisor, and he suggested to use black for the line color because most journals want to make sure that plots look good when printed black-and-white.

I am curious to know some other advice for figures like line width, styles, colormap for surface and contour plots, the figures’ dimension, etc.

Answer 1

Someone general advice:

• do a Google search. There are already plenty of articles talking about all different aspects of creating good figures, and they probably do a better job than my answer here.
• experiment. Just create a figure and look at the font size, linewidth, color selection and arrangement, etc. If you have an average person's understanding of aesthetics, you should be able to spot issues in your figures. Print the paper out, and see if everything is legible.
• imitate. Find papers in top journals and important dissertations, and see how other people create figures. I am not in mathematics, but if you look at Science, Nature or top journals in my field like Physics Review Letter, you should get a good sense of the elements in a graph, how to convey data and ideas effectively, and how to make a graph look great.
• look up guidelines. Journals in my field often have a very specific style guide (like this), which can be a very useful reference. They can include things like "minimum linewidth is 0.5 pt" and "font size must be greater than 18 pt". They are generally good guidelines to follow.

Some specific things you can look into:

• Don't be afraid to use large font size. They are probably not large enough.
• using colors, and especially color maps, is tricky. There are many considerations that you may not be aware. You can easily find advice of how to use colors, and there are websites that help you select colors, so I will not repeat them here. Here are some discussions on colormap:
  • A Better Default Colormap for Matplotlib
  • Rainbow Color Map Critiques: An Overview and Annotated Bibliography
  • Color Map Advice for Scientific Visualization
  • A survey and task-based quality assessment of static 2D colormaps
• make things simple. Don't overcrowd your figure.
• use annotations where necessary. A figure with adequate annotations will often explain itself, even without captions.

Additional resources:

• Look up a book named "Trees, maps, and theorems" in which there is a chapter about figures. (I have been to a great talk given by Jean-luc Doumont about how to give a good presentation.)
• Figure Design (MIT EECS Communication Lab)
• Choosing color palettes (Seaborn)

Answer 2

At its heart, every figure tells a story. Before you worry about the "polishing" details, you first need to make sure that you're organizing the information to tell the story as clearly as you possibly can.

A classic book on the subject, which I have found very helpful, is Edward Tufte's "The Visual Display of Quantitative Information." This has a lot of great first-principles concepts for communicating information with figures, as well as a lot of great examples of figures that do well in doing so.

When you are working on your own figures, try playing with different arrangements and densities of information within a figure. Our first instincts are often wrong, so try splitting dense figures into separate charts and try combining figures together. Likewise, our eyes like to move in lines, so try shifting locations and orientations of objects in diagrams into different organization (e.g., vertical, horizontal, circular) to minimize overlapping lines and to bring the most important conceptual relationships into simple and salient geometric relations with one another. Color is another valuable dimension that you can use to organize relationships, whether to make certain things pop or to turn a bunch of individual lines into an organized gradient whose gist can be grasped with a glance.

Once you've got the core narrative elements of presentation sorted, then you can think about how you might want to polish the figure by adjusting fonts, line weights, colorblind compatibility, etc. Most modern software, however, starts you with pretty reasonable defaults for fonts and colormaps, so you may not need to worry much about those. Matlab, for example, has switched to parula, which is designed to work well for both colorblind viewers and greyscale printing.