3-variable modeling?

Subject: 3-variable modeling?
From: Geoff Hart <ghart -at- videotron -dot- ca>
To: techwr-l -at- lists -dot- raycomm -dot- com, Paul Strasser <paul -dot- strasser -at- ennovationinc -dot- com>
Date: Thu, 22 Jan 2004 11:10:04 -0500

Paul Strasser wondered: <<I'd like some ideas on displaying three variables [for baseball players] on a piece of paper in such a way that the result is as easy to comprehend as two variables are in your basic x-y axis graph.>>

Can't be done. <g> I'm actually saying that semi-seriously, since very few people are good at comparing three simultaneously varying variables. It's just not something that comes naturally, and even with training, it's a bit of a mental stretch until you've practised it for a while. (I say this as a former scientist who watched many colleagues--not to mention me--grapple with the problem of interpreting 3D relationships.)

If you actually want to show the numbers rather than their behavior, a table is by far the simplest means of presentation. Any 3D presentation creates a distortion of the values because it's inordinately (pun not intended) difficult to extrapolate along multiple graphical axes in three dimensions. Try it some time with a group of friends and calculate the standard deviation of the group's estimates. Scary.

That being said, there are two good possibilities. The first is the "response surface"--basically a topographical map of the three variables, displayed in perspective view. Unlike 3D bar charts, which are evil, this graph lets you see the dips and peaks in the values for any given combination of the three variables; the angle of the lines around each dip and peak provides a qualitative indication of the sharpness and direction of the change. You can't easily discern the magnitude of the change or precisely compare numbers between peaks and valleys, but you can at least see the direction of the change.

Spider (radar) graphs are probably the best way to show this: they're both simpler to interpret and easier to use for objective (quantitative) comparisons. These graphs are built by creating one axis per variable, then distributing these axes at equal angular intervals around a circle, with each axis starting at the center of the circle. So for example, with three axes, you'd have something that looks like the hood ornament on a Mercedes. Choose a different line color or pattern for each player you'll display on these axes, then connect the points along each axis for that player, and you get a visually simple, elegant way of comparing the different players. Of course, this gets nasty if you have more than a few players because the lines interweave like copulating octopuses. <g>

<<Sure, I can just make three vertical (or horizontal) bar graphs, one for each variable, but there is a certain lack of elegance in such a prosaic graph.>>

Depends on what you mean by "elegance". I consider successful communication "elegant", and a bar chart (with the bar's for each player's set of skills grouped side by side above that player's name) is an extremely effective way to present the information. If you want it visually elegant, screen in a faint graphic of a ball player behind the bars, where it creates visual interest without interfering with visual comparison of the data.

--Geoff Hart ghart -at- videotron -dot- ca
(try geoffhart -at- mac -dot- com if you don't get a reply)





References:
RE: white paper: From: whitedh
white paper - let's try again.: From: Andrea Brundt
Re: white paper - let's try again.: From: Dick Margulis
3-variable modeling: From: Paul Strasser

Previous by Author: Users writing their own procedures?
Next by Author: Dreamweaver upgrades?
Previous by Thread: 3-variable modeling
Next by Thread: Re: 3-variable modeling


What this post helpful? Share it with friends and colleagues:


Sponsored Ads