Monday, July 9, 2018

Data Science 101: Radar Charts

I actually wanted to write down what I don't like about radar charts, but then I found out that this already has been done. So I won't. But, adding to "Misreading 1: Area" in Ghosts on the Radar , I was thinking of the following puzzles:

Suppose you have $n$ real-valued, non-negative measurements $x_1$ through $x_n$. You are a super-smart data scientist and you want to use a radar chart to trick your customer to believe that the measurement result is "good" or "bad" -- i.e., the area under the polygon obtained by connecting points corresponding to these measurements should be large/small. Since the angle between the axes is fixed to $360/(n-1)$, you can influence the size of the polygonal area only by selecting the order in which you plot the measurements to the axes. What is the optimal assignment of measurement values to axes such that the area under the polygon is maximized/minimized?

Edit on July 12th, 2018: Here is the solution to the problem of maxizing the area.