A Radar Chart, also known as a Spider Chart, visually compares several entities (products, organizations, investment opportunities, or even people) on multiple dimensions. For example, a manager of a diagnostic imaging (radiology) center might want to compare her facility with the competition on dimensions related to patients such as Time To Appointment, Report Turnaround Time, No Show Rate, and Wait Time.
Or, one might want to compare 6 products A, B, C, D, E, and F on 6 different attributes: Aesthetic Appeal, Compatibility, Strength, Market Size, Durability, and Reliability. The Custom Radar Chart lets one easily compare all the products along each of the dimensions on interest. This is the example used in the documentation. With the axes normalized, the resulting chart would be
For more on this shareware product, please visit www.tushar-mehta.com/excel/software/custom_radar/index.htm
AKA one of the ugliest and misleading chart types =)
While radar/spider charts may seem like a good idea, in practice they are very difficult to interpret. The misaligned axes (radii) make judgments difficult, and the resolution at lower values is degraded because the distances between these axes is severely compressed.
Parallel coordinates charts are a much more effective technique for displaying this kind of data. The factor axes are parallel, which makes comparisons between factors much easier, and avoids the problem of poor resolution at lower values in the radar chart.
Jon, as the intro to the add-in indicates the add-in creates “independent* axes, and even those can be normalized. The weaknesses that you identify essentially disappear. :)
I like the idea of this type of chart, and have run across situations where they might be a good choice of display, except that they’re too hard for most people to interpret. Maybe if you had a small, set number of people who needed some kind of dashboard metric for which this might be right — a sort of quick look of “are these factors roughly in balance” — you could explain up front and the users would be familiar with the display. But there’s no way I’d put one of these in, say, a report being produced and sent out to a client.
As Jon points out, parallel coordinate charts would almost always be a better choice.
Though there are some god-awful parallel coordinate charts out there.
The “axes” in a parallel coordinates chart are usually independent, like those in a typical radar (spider) chart, scaled from low to high or from 0 to 100%. It is not the scales of the radar chart axes which hinder comprehension. It is their radial (not parallel) orientation, and the emphasis of larger values at outer edges of the radar chart at the expense of smaller values at the center. The shapes of the radar chart series change drastically with order of the independent axes, further impeding understanding.
I’ve written a number of articles describing the deficiencies of radar charts and suggesting alternatives, including Spider Chart Alternatives, Radar Charts are Ineffective, Spiraling Down the Drain, and Composite Baseball Player Evaluation.
But that’s just me, what do I know?
Data visualization authority Stephen Few begins Keep Radar Graphs Below the Radar – Far Below (pdf) with
“The test of a graph’s usefulness is its ability to communicate efficiently and effectively. If it expresses the right information clearly and accurately in a way that speaks to your audience, then it is effective. If not, regardless of how pretty it is, it’s not only useless, it might even be harmful. If you are using radar graphs to communicate typical business information, you could be making a costly mistake.”
In Communicating Numbers (pdf) Few recommends a simple table as a more effective means of presenting radar-charted values.
Graham Odds discusses the effectiveness of radar charts in A Critique of Radar Charts, and presents a number of graphical alternatives.
Even Wikipedia’s article on Radar Charts has a good discussion of the limitations of these charts.
I first encountered a “parallel coordinate chart” during my MBA, and the chart has a much simpler name, though I do not recall it. {grin}
To state the obvious, a parallel coordinate chart has a *single* valued axis. Every attribute *has* to be scaled between 0 and 1. And, if you are willing to do that, you can use the default Excel spider chart.
I looked at the pdf you referenced. Anyone who prefers the table over the chart to evaluate the candidates should definitely stick to the table. I got an overall impression of the candidates in a fraction of a second using the chart whereas I found it impossible to do the same from the table. But, obviously, the table works for the author and, I presume, for you. :)
To me that is the benefit of a chart, and for certain types of analysis, a Radar chart. It conveys overview information allowing one to form a general impression without getting stuck in the nitty-gritty.
Similarly, to claim, as Graham Odds does, that a “small multiples” chart built from bars charts is “a more effective alternative” to a radar chart is, at the very least confusing. I could make no comparison across the nations from the presented chart. The *only* things I noticed were (1) the Mongolia and Hong Kong seem to have the same size and (2) the population of the former was much larger than that of the latter! {grin} On the other hand, if there was a pattern in the data it is — and will not be — evident from that chart panel.
Clearly, there are people who don’t like Radar charts. That’s their prerogative. I’ve worked with clients who like their compact representation, who have *no* problem interpreting the results and prefer working with *actual* numbers rather than the artificial 0..1 values imposed by a parallel coordinate chart.
Every chart has its pros and cons. The use by the Economist of whatever that particular chart type was probably not the optimum decision. Of course, if it had converted that chart into a 3D effect representing a drain, it might have been effective. {vbg} In any case, one should not extrapolate from a single example to an overall conclusion about the chart type.
The bottom line is that I just don’t get what people complain about. The normalized radar chart shown in my post is really a parallel coordinates chart “folded into a circle.” If nothing else, it uses less space while conveying the same amount of information.
I find that it is better to apply the correct tool to the task at hand than to state “I will use only these tools come what may.” A Radar chart is suited for certain tasks, just as a Bubble chart is suited for some other tasks, and a Line chart is suited for yet other tasks.
Tushar,
I welcome a practical case study that shows its usability and its impact on decision making. Until then I follows Jon’s recommendations.
Kind regards,
Dennis
The parallel coordinates axes need not be scaled from 0 to 1, they can be scaled however is appropriate, using the same kind of scaling as in your radar charts. This scaling is no less artificial than a scaling of 0 to 1.
Folding the parallel coordinates chart into a circle is what makes the radar chart more difficult to interpret. This is one of the main points in most of the articles cited. To easily compare values, they must be encoded such that measurement begins at a common baseline and proceeds in a common direction.