I would like to be able to display our data in a Radar Chart i VA.
Proc GRADAR would be a good place to start until they add it into VA.
What is the argument for the benefits of a radar chart rather than, say a bar chart? Particularly in this example where the there is no logical ordering to the radar axis (anitsocial attitiudes, antisocial peers, etc), reordering them can have a huge impact on the "shapes" that you see in the radar plot.
I have yet to see a radar plot that makes sense to me. I suppose I could imagine maybe one where the axis was cyclical (hours or seasons or something like that).
Anything that relates to compass directions is a start. Weather data uses a display called a windrose that show the the accumulated windrun or percent of time wind comes from a compass direction. Also depending on display space a largish number of bars, say 15 or more with labels may not fit but the radar chart is more compact.
I would say the order argument is not a good one when comparing to bar charts as the order, if changed on a bar chart, does the exact same change in impact for quick visual comparisons.
If you have directional data then you could definitely benefit from a visualization that encodes that direction. But Radar Chart (at least as SAS defines it) isn't intended for that usage since it tends to arrange the dimensions in a somewhat arbitrary fashion. You could set the order so that it works out for cardinal directions of course, but having an axis graph that operated on polar coordinates would be the more powerful answer for that use case and would make the windrose easy to achieve.
For general visualization of multidimensional patterns it is likely that a Parallel Coordinates plot will make its way into VA before a Radar Chart since it is more scalable and removes a bit of distortion Radar introduces by not having parallel axes. This makes meaningful comparison between the slopes of lines easier.
Where Radar arguably has some merit is in comparing multidimensional patterns across a spatial dimension (like a geographic map). If instead of overlaid shapes each Radar Chart is shown as a separate glyph then you have the advantage that the radar glyph has a clear center that can be placed spatially. Hard to find a good example of this, but it would be something like this, but with the shapes distributed across a map or other coordinate plane.
@Quentinthe benefit cames out when one can read the chart as if it was a "profiling chart" (and yes, reordering can have a huge impact).
In the @Allan_dk 's example we can see that the use of substance abuse seems to "pull" the profile's shape toward antisocial attitudes and antisocial's peers while the antisocial personality is reduced.
Or one may say that the antisocial personality is pulling the profile in the opposite direction.
I don't know if A and B concern the same indiviual at differrent times, or two distinct personnalities.
There is obviously no "good" answer but this reading can help in the understanding of some kind of tradeoffs, or patterns.
And when it comes to comparing a lot of different profiles the human brain is able to remember and compare the shapes very quickly.
The high level of education is a remarcable peak in these shapes that we can remember easely, thanks to its South-West position.
If we used a parallel coordinates as @RileyB suggests, one could hardly remember if the peak was in the 3th or 4th coordinates. Parallel coordinates are very good illustrations to figure out "scenarios" or "trends", or kind of "roads".
For example if I ask someone to quickly compare these products which graph would you prefer ?
Can one easely remember which cameras are better on the color scale using the parallel ordinates ?
I support this idea, as I also have a great need of this kind of chart.
Thanks for the example @Julien_Marandet, but I'm still not convinced. : ) I'm not sure what you mean by "profiling chart."
When I look at your example, I would want the series plot rather than the radar chart. And really I wouldn't want the series plot, since there is no logical ordering to the axis and it is not really a series. I would want a bar chart or dot chart.
I can accept that the human mind is good at remembering shapes, but in this case remembering a shape won't help me unless I also remeber all of the radar plots' axes. Tomorrow, even if I remember that Sony won, I'm not going to remember if the color scale axis was southeast or northwest. And I'm not sure if the mind is good at measuring distances from a centroid.
The mind is good at measuring (and comparing) the length of vertical or horizontal distances. Thus the appeal of the bar chart. Or even with the series plot you showed, I can quickly scan the five plots to see which camera looks best overall, and also which is best on color.
My main concern with the radar plot is the fact that it gives you a shape, and leads you to think that the area of that shape is somehow meaningful. But of course the shape and the area of the shape can vary dramatically, depending on the ordering of the axes. I've seen a radar plot with overlaid semi-transparent shapes (comparing pre and post intervention results on several domains), and I was flummoxed by it, and wondered why it was attractive, other than being pretty shapes. Wanting instead to enter the data myself and make a bar chart with two groups.
Just read a long anti-radar plot blog post (which I agree with) here: http://blog.scottlogic.com/2011/09/23/a-critique-of-radar-charts.html Also liked the first example in this post: http://fansided.com/2017/05/17/nylon-calculus-possessions-overestimated/
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.