-------- Original Message --------
Subject: Re: Principal coordinates analyses
Date: Fri, 24 Jul 2009 11:08:55 +0200
From: [email protected]
To: morphmet <[email protected]>
References: <[email protected]>


  Dear Jessica!

    You should start by studying the classical paper by John C. Gower
in the English statistical journal Biometrika, 1966. In this paper he
formalizes the concept of Q-mode analysis of a data-matrix with the
end in view of producing a graphical representation of points based on
the statistical distances between them. Please note that the
components of the latent vectors are not possible to reify, although
some try this out of ignorance of what is implied by the geometry.
The project was given to Gower by the renowned English
biomathematician Maurice Bartlett in order to tidy up what was known
as Q-mode factor analysis.

  Also to be noted is that the residual variability - ie variability
remaining after the extraction of 2 latent roots should not be too
high (too high is often considered to be > 50%).

  Practical details, and references to literature, can be obtained
for example in Richard Reyment and Enrico Savazzi (1999). Aspects of
Multivariate statistical analysis in geology. Elsevier.  Another
reference in which interrelationships between multivariate techniques
are explored is the book by Reyment and Karl G. Joereskog (1993)
entitled Applied Factor Analysis in the Natural Sciences. Karl was,
until his retirement, Professor of mathematical statistics at the
University of Uppsala.

   Richard A. Reyment



Citerar morphmet <[email protected]>:



-------- Original Message --------
Subject: Principal coordinates analyses
Date: Wed, 22 Jul 2009 11:30:21 -0700 (PDT)
From: Camp, Jessica A <[email protected]>
To: [email protected] <[email protected]>

Hi all,
I'm working with a data set that has a lot of missing information. I
want to do PCAs to assess variability and disparity between my species,
but making subsets to run them decreases my sample size until it's
ridiculously small with both the number of specimens and variables. I
was advised that doing PCOs instead might be a way to include more data
while ignoring the gaps to some degree. I haven't had much luck finding
literature that describes the pros and cons of PCOs and exactly what
they do, though. Can anyone guide me to something that would? Or to
another test I could try?
Thanks,
Jessica Camp


--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org



----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.



--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org

Reply via email to