Hello all,
I have been struggling the past few days with the results of
elliptical fourier analysis on fossil clams.  EFA went well enough; I
used the HAngle and HMatch programs.  I have been using PCA for
visualizing the data (by treating the Fourier coefficients as
variables; I have been getting 11 pairs = 22 coefficients = 22
variables), but I would very much like to run a statistical test on
the "sameness" of the groups that I have, and also be able to place
ungrouped specimens within the framework I develop.

I have read this entire thread [http://www.mail-archive.com/
[EMAIL PROTECTED]/msg00328.html] about PCA and multivariate
normality, but I would like to know more about how 'real'
discriminant analyses are actually run.

Every textbook source I have claims that you need to be able to
assume multivariate normality and equal variance-covariance matrices
to be able to run discriminant analysis and Hotelling's T^2 test.
Unfortunately, both skewness and kurtosis of the data are not normal,
and Box's M tells me that the variance-covariance matrices are unqual
(although I hear that test is rather stingent).

However, I have found a number of papers where the authors either use
discriminant analysis or do something similar, but I have not found
any mention of meeting these assumptions.  So my question is this:
is it okay to run the test without multivariate normality just
because everyone else seems to have done so, or is there something I
am missing?  If the latter, is there a way to normalize this
multivariate data?

(As an aside, it would be best for me to rely on the data from the
original specimens rather than an aggregation of the group, simply
because the question I am trying to ask is 'what group(s) do these
specimens fit into?)

Thanks for any light you can shed on these problems,
Matt BK




-----------------------------
Matthew Burton-Kelly
Graduate Student
Department of Geology and Geological Engineering
University of North Dakota
(802) 922-3696
[EMAIL PROTECTED]
Web site:  http://uweb.und.nodak.edu/~matthew.burton.kelly/
--------------------------------------------
"About thirty years ago there was much talk that geologists ought
only to observe and not theorize; and I well remember someone saying
that at this rate a man might as well go into a gravelpit and count
the pebbles and describe the colors.  How odd it is that anyone
should not see that all observation must be for or against some view
if it is to be of any service!"
-Charles Darwin, in an 1861 letter to Henry Fawcett.




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

Reply via email to