Well-designed graphs and diagrams can effectively communicate complicated science. However, poorly designed visuals can lead to misunderstandings and potentially mislead readers. In a recent article in Knowable magazine, Betsy Mason explains that as the volume and complexity of scientific data increase and data are shared with wider audiences, the importance of data visualisation is growing.
The topic of a keynote presentation at the recent 2020 ISMPP European Meeting, data visualisation is sparking much interest within the medical publications field. A good visual should clearly communicate the underlying message of a dataset, revealing trends and patterns that may be lost if the data were presented in, for example, a table. Research carried out in the 1980s to determine how well different types of graphic are perceived revealed that people are best at reading charts based on the lengths of bars and lines, such as bar charts, but struggle to judge differences in area, shading, and colour saturation. For this reason, the use of the ever-popular pie chart is rarely the best choice if the aim is to compare the sizes of each slice. Similar research conducted over two decades later confirmed that these findings still apply in the modern digital environment.
Mason explains that while bar charts are easy to read and understand, they are not always the best option for presenting certain types of data. For example, bar charts are not effective for the visualisation of continuous data and should not be used for small sample sizes where outliers can have a big impact on the mean represented by the height of the bar. Scatterplots, box plots, and histograms are all better ways of visualising the distribution of the data. Mason also discusses the impact of colour on the interpretation of visual displays of data, noting that the thoughtful application of colour can enhance the message of a graphic.
One of the problems contributing to poor data visuals is that scientists rarely have training on how to best present their data and typically just follow convention, thereby perpetuating bad practices.
Mason highlights the potential role of scientific publishers in improving standards in data presentation, noting that several journals prohibit or discourage the use of bar charts for continuous data. Mason concludes that increased training, awareness and better software tools are needed if scientists are to adopt better approaches to data visualisation, which is particularly important as scientific data become increasingly complex.