Sharing how we can communicate our science effectively, whether in posters, papers or on social media!
Share
SciCom - Better Figures: Choosing the Right Graph Type
Published about 2 months ago • 6 min read
Graph Types & Design
Hi Reader, ever heard of chord diagrams?
Today, we’ll take a look at which graph types exist.
Moreover, we will see when to apply them and the fundamental principles for clear design.
All in all, this will enable you to prepare figures that are easy to understand, scientifically robust, and good-looking.
Establishing Purpose
It might seem counterintuitive, but before we talk about design, we must zoom out.
First, you want to decide which part of your overall data you want to share – i.e., what is your main message? Second, you choose the data that should be included and form the storyline - that is, what data your reader must see to be convinced. Finally, you tell the story - that is, you choose the wording, design the graphs, and do the fine-tuning. Although this comes from a blog focused on sales data, I think it is a well-written piece that is helpful to scientists as well.
To choose the right kinds of graphs, we must first know what broader story our paper is meant to tell.
That means we have to decide what our main message is and in which order to present each piece of information to make our paper as understandable as possible.
Always guide your reader from what they already know to what is new - and from the simplest to the more complex. For example, Nature papers typically progress from simple observations - for example, protein presence in a specific tissue - to more complex analyses, such as the effect of blocking an associated signaling pathway in an organ-on-a-chip model. Therefore, read Nature papers and pay particular attention to the setup of the introduction and results sections.
Therefore, it can be helpful to summarize the key message of your paper in one sentence. Then, write one sentence for each figure.
This can help you clarify what each figure needs to convey and in which order to present them. From there, you can determine sections and eventually paragraphs for your results.
Human mental processing is not path-independent - the order in which we receive things matters a lot for our understanding.
Once we know our message, we can decide which data we actually want to include and how to present it.
Knowing Your Graphs
First, it helps to know what kinds of graphs exist.Covering all of them would exceed the scope of this piece, so here are some useful resources:
Ardigen provides a nice walkthrough of the graph types available and also offers a clear overview that you can keep or turn into a poster.
Luzmo provides a solid overview of the basic chart types you should know.
Atlassian compiled an almost exhaustive overview, with more in-depth explanations of each chart type.
And Datawrapper offers a clear overview along with practical design tips.
The Data Visualization Catalogue also has a helpful search page where you can look up suitable graph types based on what you want to display.
The Data Visualisation Catalogue is a really useful blog - and under the Resources tab you will find many other websites and projects that might be helpful.
Finally, many bioinformatics blogs and websites provide excellent overviews of graph types, design principles, and the code needed to create them.
In contrast to Bioinformatics sources, in standard overviews graph types such as chord diagrams are typically not included. P.S.: Even if you don’t have any coding experience, you can search for dedicated websites that are built for that single purpose (although editing options are often more limited).
For ggplot2/R, check out Corytophanes or the R-Graph-Gallery. Especially the latter offers a quick and well-structured overview.
Choosing The Right Graph
To find an appropriate figure type, you want to ask two questions:
A) Will it display all relevant information? Think
Distribution
Summary statistics such as means (i.e., effects & quantities)
Inferential statistics (SD, CI, or significance)
Inter-sample patterns or comparisons
B) What is visually the easiest to interpret?
An amazing example, taken from “Design Strategies for Scientific Figures” by the College of Natural Science at the University of Texas at Austin. Moreover, we could denote the mutation type on the x-axis and use MCF7 or V1 as the heading if we had only one figure or wanted to eliminate the legend. Note that the color code and apparently some aspects of the data composition have changed.
That means: what is easy to process, what your reader is used to (and therefore finds intuitive), and what shows important data clearly?
For example, use a bar chart when you have a small dataset with little scatter and want to highlight clear differences in means, including non-overlapping confidence intervals. However, use a box plot when you have more than 30 data points and the distribution itself is important to show.
This is also why you rarely see pie charts in scientific publications - they are a poor choice for comparing values and are clearly inferior to bar charts for assessing relationships.
Sometimes, transforming the data helps.
While helpful for visualization in some cases, please keep in mind that transformations can influence statistical analyses. In some instances, they are used to adjust the data distribution in order to achieve a more normally distributed dataset.
For example, displaying log-transformed values may be more informative than sticking to a linear axis.
Powerful Design Principles
Choosing the right graph type is essential, but design matters just as much.
A good design guides the eye and minimizes how much information the reader has to hold in working memory. Legends are a classic example where smart design can make a big difference.
This example is taken from an excellent overview titled “Design Strategies for Scientific Figures” by the College of Natural Science at the University of Texas at Austin. As a rule of thumb, the faster you can understand a figure, the less mental effort is required.
In some cases, including absolute numbers is helpful.
In others, it's about applying principles of visual composition that help readers grasp information quickly:
As outlined by Aiora Zabala in “Designing More Effective Scientific Figures” (VTP Graphic Design, Cancer Research UK), grouping, ordering, and containing can help readers analyze data at a more hierarchical level and therefore process it faster.
Grouping, ordering, and containment can make your main message much clearer.
If you choose proper labels, axis titles, and graph layout, your readers should be able to understand the gist of your figure even without a description. Again, this is important, as we often just skim papers due to limited time.
Therefore, always pay close attention to what feels intuitive to your reader.
Another example from Rougier et al.: while differences in disc radius may be easier to distinguish visually, viewers typically interpret displayed differences as differences in area.
As mentioned before, avoid overloading figures with too many samples or visual elements. It’s about delivering a message with your figure, not telling a story.
Don’t overcrowd your figures. Readers should be able to instantly recognize the graph type, understand the axes, identify samples via the legend, assess differences, evaluate spread and deviations, and interpret individual data points. This is only possible when the data are clearly discernible, colors are unambiguous, and axis labels are readable and clear. What happened in the figure above is an example of chartjunk, as outlined by Rougier et al.
When assembling multi-panel figures, use grids and aim for geometric harmony. This often takes time and may require fine-tuning axis widths and spacing.
Another tip: read the journal guidelines.
Checking the journal guidelines is not only important for your submission. As in this case with Nature, they can also provide generally applicable best practices.
In some cases, design work or assembly will be taken over by the journal - saving you a lot of time.
In Essence
This lesson should convince you to spend a little time deciding which graph type fits your data best, instead of defaulting to what you are used to.
Have you ever been impressed by a figure? I certainly have been by such as the one on the right or the circus plot on the left. However, these figures often take several minutes to decode due to missing expertise in the field, the display of higher-dimensional or meta-level data that is difficult to interpret, variation in the types of data depicted, and the use of abbreviations that differ from paper to paper (compare circus plots from here, here or here). In contrast to, for example, bar graphs, these figures serve a different purpose. They exist to (A) demonstrate which factors have been analyzed, (B) highlight general patterns, and (C) be visually impressive.
While we normally try to finish our figures as quickly as possible in order to publish, choosing a fitting design will:
Make more people read, understand, and therefore cite your paper.
Save you time by reducing work during revisions and when reusing your figures for presentations.
Enhance your perceived professionalism and trustworthiness.
Graphical Abstract - Contents Hi Reader, what should a graphical abstract include? While there is no single right answer, what you choose to show will influence how easy (or difficult) the design process will be. Just like with posters or figures for papers, it's a lot about prioritization. Therefore, let's see how to decide what to include: What To Show Essentially, you want to show the key points that make the implications of your work clear. To my mind, this is a very well-designed...
Discussing Graphical Abstracts Hi Reader, want a short and concise option to communicate your science? Graphical abstracts are a fantastic way to do so. However, don’t make the mistake of thinking they are the same as a written abstract, just with pictures. To design a good graphical abstract, we must properly understand its nuances: What's Their Purpose? Fundamentally, a graphical abstract is a visual way to express the main idea of a scientific paper. There is an almost endless variety of...
Developing a Sense for Design Hi Reader, do you pay attention to advertising on the sides of roads or in the city? I believe you should, as it will make you a better science communicator. Whether we design graphical abstracts, bar charts, or overviews we cannot do that without graphical design. So, here’s how to develop a sense for it: Why It Matters For Science Designing good figures and diagrams is not always a scientific process. We have to convey scientific data, but finding the right...
