The 10 Commandments for Figures

If you need to satisfy me because I’m your prof or you think I might be a referee, then just follow the rules. If you want more information about the rationale behind the rules, they are mainly based on the books by Edward Tufte which are really worth reading for the examples and interesting discussion.

1. Avoid pie diagrams. Strongly avoid multiple pies. They have a very high ink-to-information ratio and the eye finds it difficult to rank similar sized areas. People find it difficult to perceive relative angles which is a key task involved in reading a pie chart.

2. Use scatterplots. They show relationships in the data. Ideally, you will find a way to illustrate the key relationship in your paper in the form of a scatterplot. Then all the rest of your regression tables can show the relationship is robust to more sophisticated technique.

3. Embed lots of useful information in your scatterplot. For instance use different symbols for points drawn from different samples (e.g. males with squares and females with circles). Include 45 degree lines or regression lines where it assists in interpretation. Label outliers.

4. In multiple line figures, label the lines rather than providing a legend . The labels should be placed where the different curves are furthest from each other to avoid confusion. It is very annoying to the reader to have to look back and forth, trying to match patterns with remote legends.

5. Avoid 3d surfaces. Some people think these are cool but most viewers find them very hard to understand. Paper is two-dimensional. Try to find two-dimensional ways to illustrate the third dimension in data. Examples include contour lines (like a topographic map) and multiple small side-by-side figures that are drawn for different values of a third variable. NEVER use three-dimensional bar diagrams. The information is uni-dimensional. Standard (2d) bars are OK but the 3d bar distorts perception, particularly at low levels.

6. Use tables or just the text of the paper for reporting small bits of information . Not all pictures are worth a thousand words. Some can be replaced entirely with two or three words. Do so.

7. Think hard about whether it is important to include zero in your axes.

8. Use log scales for strictly positive variables with no natural units, especially when you care about multiplicative relationships. For example stock prices over long periods should probably be shown in a log scale so you can interpret the slope as the rate of change. Almost never show time or shares in log scale since these variables have natural units.

9. Do your figures in a command language like R (very powerful and free!) or Stata. These programs take some time to master but it is worthwhile. Most figures done in Excel are deficient in one way or another.

10. Insert figures inside the body of the paper. Journals insist upon figures at the end for the final submitted version of the paper. This does not mean you should do it for working papers or first submissions. There is a reason why the printed version of your article puts the figures back into the text: it is easier to read a paper that way without having to constantly flip to the end to find a figure and then flip back to the text for interpretation. By putting the figures in the text you will also be more aware of whether your paper has the right mix of text and visual information.