Radar obscura
Misrepresenting data
yet such lovely shapes
Recently I’ve been looking at how to visualise some data in a way that’s engaging for the reader. Radar charts (also known as spider charts, web charts and star plots) seemed to fit the bill. My data with its 6 variables per subject can create a variety of irregular hexagonal shapes that are interesting and informative. Curiosity even had me wondering whether I could look at the surface area of the shapes to compare between different subjects.
Problem solved?
Perhaps not.
In reading up on radar charts I found articles by Chandoo (2008), Odds (2011) and Morin-Chassé (2020) outlining why my plans might not be ideal, and why despite their good looks, comparing between subjects can be less intuitive than simple bar charts. The issue lies with what is added to the data when visualising it. Let’s take my radar chart with 6 variables as an example.
The radar chart consists of a centre point with the 6 variables coming out from it like the spokes of a wheel, with the length of each spoke being the value for that subject in that category. The end of each spoke is then connected to its immediate neighbours with straight lines (although some radar charts are circular and the connecting lines are curved).
The main thing the reader sees and focuses on is those connecting lines and the hexagonal shape they create. But that shape actually means very little – it’s not the actual data, instead it’s a circular sequence of the relationships between pairs of neighbouring variables. Ultimately the reader can easily be distracted from the data by how its visualised.
It gets worse – the shape created depends on a couple of factors:
1. The scale of each spoke.
2. The order of the variables arranged around the graph.
A different scale for one or more of the variables or a different order of the variables around the graph and the resulting shape can look very different. These decisions can also make it harder for readers to interpret the graph – imagine trying to read 6 different axes with their own units and trying to understand what they mean.
What’s more, the areas of the resulting shapes change as the shapes themselves change – simply swapping the location of two variables can result in a different shape and so a different area. And even if the shapes were always regular hexagons, the area doesn’t increase proportionally to the spoke length – a radar graph with longer spokes would have a disproportionally large area compared to one of the same shape with shorter spokes.
All this means that in many situations radar charts can actually cloud interpretation of the data rather than make it clearer. Doesn’t stop them being a good looking graph though!
Curious about what to use instead of radar charts? Check out the articles below for alternatives (including stellar charts and petal charts) and to get a more detailed (and far better written and explained!) understanding of some of the flaws of radar charts.
A note about the sciku: I’ve used the word obscura in the first line. In this case I mean to suggest how the radar chart obscures or obfuscates the data. I could have written obscurer but I wanted to reference the camera obscura that were used from the second half of the 16h century onwards as drawing aids to produce highly accurate representations and were later integral to the development of the camera. I liked the comparison between something that made things clearer and something that purported to make things clearer but often doesn’t.
Further reading:
Chandoo (2008) You are NOT spider man, so why do you use radar charts? https://chandoo.org/wp/better-radar-charts-excel/
Odds (2011) A critique of radar charts https://blog.scottlogic.com/2011/09/23/a-critique-of-radar-charts.html
Morin-Chassé (2020) Off the “radar”? Here are some alternatives. https://www.significancemagazine.com/science/684-off-the-radar-here-are-some-alternatives