-------- Original Message --------
Subject: Re: Problem with relative warp
Date: Thu, 8 Dec 2011 19:55:35 -0500
From: K. James Soda <[email protected]>
To: [email protected]
Dear Kalpana,
Have you considered using a cannonical variates analysis instead of a
relative warp analysis? Relative warp analysis is essentially the same
thing as a principle components analysis. As such, it finds the axes in
sample space along which there is the most variation in your data set.
However, these axes are not necessarily the best axes for
differentiating pre-set groups. A cannonical variates analysis finds
the axes in sample space that maximizes the ratio of between group
variance to within group variance, which can be thought of as the
subspace that best differentiates the groups. In addition, it also
represents Euclidean distances between points on a scatter plot in a
manner that reflects the statistical distance between them. On the
downside, it does make more assumptions than a PCA (particularly that
each group has identical covariance matrices).
It may not be a final solution for you, but at least CVA might be
something worth looking into. There was recently some buzz here on
morphmet about CVA's. That might be a good place to start reading, if
you're interested. In addition, you can get a pretty good description
of how CVA is performed from:
Campbell, NA and WR Atchley. 1981. The geometry of cannonical variate
analysis. Systematic Zoology 30: 268-280.
From a software standpoint, CVA's are very easy to run in R using the
lda() function in the MASS package.
Hope this helps,
James Soda
On Thu, Dec 8, 2011 at 3:52 PM, morphmet
<[email protected]
<mailto:[email protected]>> wrote:
-------- Original Message --------
Subject: Problem with relative warp
Date: Sat, 3 Dec 2011 16:42:23 -0500
From: kalpana das <[email protected]
<mailto:[email protected]>>
To: [email protected] <mailto:[email protected]>
Dear All,
I am facing problem with relative warp analysis.As i have mentioned
earlier,I have been working on frogs.I have chosen different number of
landmarks to capture snout shape of a frog but still i am not getting
different clusters of species in Relative warp plot.
I am going with the assumption that snout shape of same species should
be similar and it should be cluster together but its not happening.
I wanna know the techniques specifically the landmarks i have selected
is correct or not.Because i struggled a lot to get results since it is
the new technique.
It would be a great help if you can tell me where i am going wrong?
Waiting for your response.
--
Kalpana
Research Intern
Center for Ecological Sciences
Indian Institute of Sciences,Bangalore
[email protected] <mailto:[email protected]>
<mailto:Email-kkalpanaa1988@__gmail.com
<mailto:[email protected]>>
Contact no-9620313751
