Message-ID: <A4E5A0B016B8CB41A485FC629B633CED39C1330D1E@GOLD.corp.lgc-group.com>
Date: 2012-05-08T10:35:37Z
From: S Ellison
Subject: How to plot PCA output?
In-Reply-To: <F2C2F721-4BE4-4424-9FCD-4FE152C015B4@depauw.edu>
> -----Original Message-----
> I avoid the biplot at all costs, because IMHO it violates one
> of the tenets of good graphic design: It has two entirely
> different scales on axes. These are maximally confusing to
> the end-user. So I never use it.
I think you're being unnecessarily restrictive there. The confusion that arises when using multiple scales in the same graphical dimension arises from a tendency to read distances and locations on the wrong scale. In a biplot, the PC's have essentially no intuitive physical interpretation (by which I mean a 1:1 mapping onto an identifiable variable) so this doesn't matter much even if it happens (in fact you cold probably lose the scales entirely in a biplot without compromising its interpretation much). And the alternative - sticking rigidly to the 'one axis per dimension' rule and to plot them with the _same_ scales - often leads to unreadable plots: invisibly tiny arrows or an invisibly tiny cloud of data points.
But having indicated that I don't see a biplot's multiple scales as particularly likely to confuse or mislead, I'm always interested in alternatives. The interesting question is 'given the same objective - a qualitative indication of which variables have most influenced the location of particular data points (or vice versa) and in which general direction - what do you suggest instead?'
Steve Ellison
*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}