-------- Original Message --------
Subject: Re: statistical tests in Morpheus et al
Date: Mon, 17 Aug 2009 08:53:10 -0700 (PDT)
From: Dennis E. Slice <[email protected]>
To: [email protected]
References: <[email protected]>
The statistical testing referred to are part of a set of routines in the
old Morpheus et al. that never made it into documentation because I
never considered them complete and fully tested. When it works, though,
it does produce good randomization p-values in line with those of
parametric stats for mv data I have tested. So, I have used it myself
and told colleagues and workshop participants about it on an as
needed/useful basis.
This is a wholly unsupported feature in the old Morpheus et al., but I
plan similar routines for the new version.
If you'd like to try it, from the command line:
list ptest options
- lists the two options for the ptest procedure: nSamples (# of
randomized samples to compute) and alpha (nominal alpha value used for
Bonferroni adjustments to multiple tests
set ptest optionname value
- change a particular option, I seldom have need to do this
ptest manova
- assuming you have grouped data with no missing values, this generates
a design matrix with a column of ones and g-1 contrast vectors and
regresses the grouped data onto this matrix. In keeping with the
morpheus spirit, no superimposition is assumed or required and data of
any dimensionality should work. So if you want the data GPAed, you had
better have done that yourself.
The routine then randomizes the group labels to generate the prescribed
number of samples and test several stats to compute the proportion
(p-value) of randomized samples that exceed that observed in the
original data. The stats are based on the explained SSCP matrix from the
regression and are that matrix's Determiniant, Trace, and Largest
eigenvalue. These are the basis of the parametric tests: Wilks' lambda,
Pillai's and the Hotelling-Lawley traces, and Roy's greatest root.
P-values are reported underneath.
Of particular interest are:
Det - which is always zero for Procrustes coordinates since the
covariance matrix is singular due to the estimated parameters.
Trace - computable even with singular matrices and has the nice additive
property that Trace_TotalSS = Trace_ExplainedSS + Trace_unexplainedSS
Max Root - included for completeness
For GM, I get my p-values from the trace.
Of course, if you get a significant manova result for more than two
groups, your next probable question is which groups are different/same.
To wit,
ptest pairs
This will test all pairs of groups and produce a final table the trace
p-values and satisfying asterisks based on a Bonferroni correction of
the default critical alpha value. The correction is simply the division
of the original alpha by the number of pairs tested.
If it works in the old Morpheus, the results can be informative. If it
doesn't (I have had stability problems on newer updates of Windows),
then it doesn't.
I hope to do a better and more thorough job in the new Morpheus.
Best, ds
morphmet wrote:
-------- Original Message --------
Subject: statistical tests in Morpheus et al
Date: Fri, 14 Aug 2009 04:56:39 -0700 (PDT)
From: Fabio Murtas <[email protected]>
To: morphmet <[email protected]>
Hi all,
I'm trying to figure out how to run a MANOVA on groups in Morpheus et
al. (rev 1-30-98 beta 2002).
which are the commands for statistical tests available in Morpheus?
thanks in advance
Fabio Murtas
[email protected]
Nothing in biology make sense if not in Evolution light
--
Dennis E. Slice
Associate Professor
Dept. of Scientific Computing
Florida State University
Dirac Science Library
Tallahassee, FL 32306-4120
-
Guest Professor
Department of Anthropology
University of Vienna
-
Software worth having/learning/using...
Linux (Operating System: Ubuntu, CentOS, openSUSE, etc.)
OpenOffice (Office Suite: http://www.openoffice.org/)
R package (Stats/Graphics environment: http://www.r-project.org/)
Eclipse (Java/C++/etc IDE: http://www.eclipse.org/)
========================================================
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
