RE: PCA of landmark data on wings

[morphmet_moderator]

----- Forwarded message from "Sheets, H David" -----

Date: Wed, 26 Sep 2012 17:00:30 -0400
From: "Sheets, H David"
Reply-To: "Sheets, H David"
Subject: RE: PCA of landmark data on wings
To: "morphmet@morphometrics.org"

The sign, and thus the orientation, of a PCA axis is arbitrary, it is determined by the way in which the algorithm generating the PCA works and the nature of the data.  PCAGen creates a more or less arbitrary orientation of the axes, there is a button on the program that will reverse the orientation (and scores) along any given PCA axes reverse the axis orientation, which is meant to allow you to manually flip the axes to compare results from one program to another.

Drop me a note if you need more explanation, or details on where to find the button

Best,

-Dave

H. David Sheets, PhD
Professor
Dept. of Physics
Canisius College
2001 Main St
Buffalo, NY 14208

---

From: morphmet_modera...@morphometrics.org [morphmet_modera...@morphometrics.org]
Sent: Wednesday, September 26, 2012 4:48 PM
To: morphmet@morphometrics.org
Subject: PCA of landmark data on wings

----- Forwarded message from Jason Mottern -----

Date: Wed, 26 Sep 2012 12:50:47 -0400
From: Jason Mottern
Reply-To: Jason Mottern
Subject: PCA of landmark data on wings
To: morphmet@morphometrics.org

Hello,

I am a new list member, as well as a novice with respect to geometric morphometric analysis. I am doing a morphometric analysis of landmark data on wasp wings, and I'm doing both PCA and CVA. I am analyzing males and females separately, and there are three species. When I run all four analyses (CVA and PCA for each sex), the directions along the principal component axes are reversed for the female PCA analysis only. In other words, the signs are all reversed on the PC scores relative to the other three analyses, so its graph is "mirrored" compared to the other three. I am most intrigued as to why this reversal occurs between the female PCA and female CVA. These two analysis are based on the exact same set of partial warp scores, though, of course, subsequent calculations are different. I don't understand how the directions of the vectors are determined, and why they might differ between a PCA and CVA analysis of the same data. The programs I'm using for the analyses are PCAGen and CVAGen (Sheets, 2002). I apologize if I'm not articulating the phenomenon very well, but, like I said, I'm very new to this stuff. Any thoughts would be greatly appreciated.

Cheers,

Jason Mottern

jmott...@ucr.edu 

----- End forwarded message -----

----- End forwarded message -----