Choose Language Hide Translation Bar
One less radar chart (or be square for Pi Day!)

In recent years, I’ve used Pi Day to draw attention to inappropriate pie charts by remaking one under the banner of #onelesspie. This year, I focus on another round chart, the radar chart, also known as a spider chart or star chart. As with the pie chart, the roundness of the radar chart has a certain aesthetic appeal. But the roundness also makes it harder to see the data.


The vice president of sales and marketing for JMP, Jon Weisz, assembles his own bicycle wheels and ran across this radar chart showing the tension measurements for the spokes.



The distances of the orange and blue lines from the center correspond to the tension values. Half the spokes connect to the left edge of the hub and half to the right edge. In case you’re wondering, the left and right tensions are different because this is a rear wheel and the gears on the right side affect the spoke tension. The spokes are numbered 1 – 16, going clockwise from the top.


Here’s the same data on a “normal” Cartesian coordinates graph.



My first reaction was that this doesn’t even look like the same data as in the radar chart. All the tension variations are easy to see in the Cartesian line plot but not on the radial version. I made the axes have the same scales (1 – 16 spoke numbers and 0 – 150 tension values), even though the Cartesian coordinates provide the flexibility to zoom in on the lines if we wanted an even clearer picture of the variation.


What do I mean about the data being harder to see? Take a look, for instance, at spokes 12, 13 and 14, highlighted in pink.   


tensionradarCrop.png                        tensionxyCrop.png


The orange tension values are uniformly decreasing and appear as a straight line in the Cartesian view, but not in the radial view. Conversely, the blue values form a straight line in the radial view, but not in the Cartesian view. To read the radar chart effectively, we have to remap our basic visual encodings for data. I like graphs where I don’t have to consciously think about the visual encoding.


It’s natural to think that the disadvantage of remapping our perception is offset by the advantage of representing circular data, like bicycle wheels, with a circular graph, but a 2012 IEEE journal paper, "Graphical Tests for Power Comparison of Competing Designs," found the perceptual disadvantages outweighed the advantage. Heike Hofmann and her collaborators tested radial versus Cartesian graphs of data related to wind direction, and the Cartesian view was still more effective for seeing interesting features.


So be square for Pi Day!

Article Labels

    There are no labels assigned to this post.

Level V

How do you draw a radar chart like that in JMP?





Radar/Star/Spider Plots can be easily created through JMP Scripting, such as in this available script by Clay Barker in JMP Community -> File Exchange -> Scripts: Star Plot Script  However strongly agree with Xan.  While Radar/Star/Spider Plots make for pretty graphs, these round graphs make it harder to see true differences in your measurements.  Graphing the real story in your data often is much better served by using a clearer chart type such as a Line, Bar or Area Chart.

Level I

This is a case of comparing the utility of tools intended for completely different applications. Cartesian plots are meant to visualize data with an abstract functional dependence f(x), where the independent variable is (or has properties that are meaningful if thought of as) a rankable scalar of potentially infinite range (count, instance, time, money, distance etc.). "Radar" plots are indended for data (R) whose dependence on the independent variable (θ) is cyclical in nature. Rather than higlight the correlation or evolution of trends between factors, these plots show asymmetries within the cycle and are of immense value for analyzing/predicting behaviors of cyclical systems or any system in repeating conditions. When used as intended, they are obviously incomperable.