By Helena Jambor
Article DOI: doi.org/10.6084/m9.figshare.7001156.v2
Scientists have used visualizations since antiquity but in recent years their number and complexity has increased. Since visual communication is not part of university curricula figures in publications are sometimes hard to read or inadvertently misleading. But good and clear visualizations are essential for reproducibility.
After three years teaching and working with students, postdocs, and faculty to design better figures, I’ve come up with the following tips:
1. Know your message.
Incomprehensible figures are very often those with an unclear message, with more than one message, or those that unselectively present all available information.
How. Start drafting a figure only after you know what you want it to explain*. When you are able to formulate a succinct and clear title for your figure, you are on a good path. If you need several sentences or subtitles, you likely still have too many messages. During drafting, you may decide that a figure is not appropriate. Tables are generally better to present complete datasets. A hybrid formed of tables and charts is also an option for visualizing one aspect of the dataset next to the tabular information.
2. Get feedback on drafts.
You will not have time to re-make complete figures for posters, talks, or publications. So, do get feedback before you even start a first electronic draft.
How. Sketch on paper, iPad, chalk board – it is much easier and faster to trash a drawing instead of an illustrator file or a printed poster. When you draw, try out different approaches (see #3). Ask several people who are not familiar with your work for feedback on the draft. Specifically, ask if your sketch conveys your main message.
3. Select a truthful and functional format**.
You need a figure that informs and doesn’t mislead. Not all information is visible in a microphotograph, and not all data is suited for a bar chart. Knowing which presentation format is required for your data is key to conveying your message and to avoid making misleading visualizations.
How. Familiarize yourself with good practices in data visualization. Images need scale bars, summarizing statistics need the sample size; trends are presented with line charts, category sizes with bar charts; bar charts need a zero-baseline. You find more details in books by the statistician William S. Cleveland or the data journalists Alberto Cairo, or in blogs on these subjects (e.g. thefunctionalart.com, flowingdata.com).
4. Add enough text.
Figures often lack enough explanation and contain an overabundance of technical abbreviations. The goal is to create self-explanatory figures, and this requires useful text.
How. Titles: Shorten as rigorously as possible. What helped me learn how to condense messages into few words was tweeting and using active language.
Abbreviations: Try googling it. If your abbreviation isn’t among the top results, reconsider using it. Avoid it even if it is a common abbreviation in your small subfield – your audience almost always is wider that you think and includes journalist, grant agencies, or patient groups.
Axis/image labels: Try formulating the take-home message of your chart, and make sure the words you use in the title are also found in the axes. For example, if your title is “Increase in temperatures over the past century across Europe” the y-axis should include “temperature”, the x-axis “year” and the lines should be labelled with country names.
Remove redundancies: If words are repeated in every axis and data label, add them to the title/figure legend instead.
5. Good layout helps.
The visual system perceives information quickly; and does so faster when this information is well organized on the page, poster, or slide. Research on how we perceive visualizations was initiated in the 1920s and is summarized in the “gestalt principles”. These investigated how factors such as the form (gestalt), the symmetry, proximity, connectedness, and enclosure of objects influence our perception of them in visualizations.
How. We read figures similarly to how we read text: from left to right, from top to bottom. We look for titles as guides to a visualization, therefore they should be placed above a figure where we see them first. Details, such as the scale bar and the sample size, can be placed at the bottom right where they will be seen last.
Proximity suggests relatedness, therefore related information should be close together. Instead of placing a legend below the chart for example, try labeling important data directly. Overall, strive for symmetry and use enough white space to group and separate information.
When preparing a composite figure or poster, always think of the page with an underlying grid structure, similar to the front page of a good newspaper: information is ordered in columns and images are placed within these columns, while headers extend across the entire page and so on. You can help yourself during the design process by actually drawing out the boxes (and then removing them for the final figure!).
Dreyfuss, a famous designer, once said “Color produces an immediate reaction and is the explanation mark of visual communication”. This hints to the fact that colors are immediately perceived by our visual system, whether we want to notice them or not. Colors should therefore be used strategically. A good time to apply them is during the last steps of the design process.
How. Strategic use: For example, a red hue stands out in an otherwise black/white graph and should encode the key data, not an irrelevant control. Colors must also be used consistently across the figure, and ideally become a color code that guides the audience through the entire work. Every color must be explained in the figure or the legend. Also, keep in mind that color is by far not the only method for emphasizing: alternatives are black among grey, or in scatterplots varying symbols might be used for data subcategories.
Colors should be accessible: Certain colors look the same to color-blind people. Try using software which simulates what a color-blind person would see in your figures, for example ColorOracle. Colors that are too similar may appear indistinguishable to some of your audience, very light tones fade on some devices – be careful to avoid both. Good color schemes are suggested by colorbrewer2.org, which is designed to select colors that are visible and distinguishable in maps! Here you can select also color-blind and printer-save color schemes for your data types.
Colors should be functional: Quantitative data along a common scale, for example weight of specimen, are best encoded in one hue with varying saturation. Quantitative data along a diverging scale, such as up- or down-regulation of gene expression, are encoded by diverging color scales. Categorical data may be shown with varying hues but is not always necessary.
7. Improve, remove, include…
1-second test. Show the final figure for just one second (or 10 or 20, but short!) to a colleague and ask what they see first. If the reply is the axis ticks or gridlines you will need to de-emphasize them and work harder to make your main message stand out.
Is your figure self-explanatory? Does the figure include all the information that is necessary to de-code it? Ask someone not yet familiar with your work to explain your figure back to you – and don’t interrupt as they do so! Based on their explanation you will immediately realize what you need to improve: more text, missing units, legends or labels, remove acronyms or gridlines, increase line weights etc.
Finding a clear visual language is key for communicating your research. It is also important for allowing future researchers to reproduce and re-use your discoveries. For reusing your own figures, especially conceptual diagrams, think about copyright. One strategy is to deposit them to figshare and citing/licensing the associated DOI when you use them in a publication. This allows you and others to continually use and reuse figures without legal problems.
Lastly, for innovative ideas, innovative visualizations are often indispensable: after all, Darwin did not only write the theory of evolution, but also invented the phylogenetic tree visualizing it.
Dr. Helena Jambor is an RNA expert and now works in science communication and management at the TU Dresden. She teaches visual communication to postgraduate scientists. Find Helena on Twitter @helenajambor.
* this applies for published figures in slides, publications, and posters. For exploratory figures, where you want to yourself learn about your data, making drafts if of course not necessary.
** paraphrasing Alberto Cairo.