Perhaps the main problem is that a principal components analysis is not
intended to achieve good discrimination between groups. It is used just to
get a low-dimensional overall view of any patterns of diversity. To see how
well groups can be distinguished you need to compute a discriminant function
(if you have just two groups) or a canonical variates analysis (for more
than two groups). 

As Philipp suggested, it is valid to add log centroid size as an additional
variable. Adding additional information such as size might improve
discrimination. However, there is also a risk. If you happen to have
collected young specimens of one group and older ones from another then size
might discriminate perfectly but would not work in future samples where you
had specimens of similar ages. Thus size provides important additional
information but it may be less useful than shape variables.

--------------------
F. James Rohlf, Distinguished Professor, Graduate Program Director
Dept. Ecology & Evolution, Stony Brook University, Stony Brook, NY
11794-5245
Web: http://life.bio.sunysb.edu/ee/rohlf
Morphometrics: http://life.bio.sunysb.edu/morph

 

> -----Original Message-----
> From: morphmet [mailto:[EMAIL PROTECTED] 
> Sent: Wednesday, November 16, 2005 6:06 AM
> To: morphmet
> Subject: DA with PCA+CS
> 
> One quick question:
> 
> is it a mistake to use PCA of the shape + Centroid Size to 
> obtain a better discrimination in a Discriminant Analysis?
> 
> If I use only PCA of the shape the discrimination of the 
> groups is not very good...
> 
> Thanks,
> Diego
> --
> Replies will be sent to the list.
> For more information visit http://www.morphometrics.org
> 
> 
-- 
Replies will be sent to the list.
For more information visit http://www.morphometrics.org


Reply via email to