Skip to content

NMDS and varimax rotation

6 messages · Bernd Panassiti, stephen sefick, William Revelle +1 more

#
hello,

subsequently to a NMDS analysis (performed with metaMDS or isoMDS) is 
it possible to 
rotate the axis through a varimax-rotation?

Thanks in advance.

Bernd Panassiti
#
have you looked at the vegan viginette- I know there is proscrutes rotation.

On Tue, Sep 9, 2008 at 3:54 PM, Bernd Panassiti
<bernd.panassiti at uni-rostock.de> wrote:

  
    
#
At 9:54 PM +0200 9/9/08, Bernd Panassiti wrote:
Bernd,


Yes.  The output of isoMDS is an object with points and stress.

consider the following example:

  test.data <- Harman74.cor$cov      #twenty four mental measurements
  harm.dist <- sqrt(2*(1- test.data) )  #convert correlations to distances
harm.iso <- isoMDS(harm.dist,k=2)  # find the multidimensional solution
harm.varimax <- varimax(harm.iso$points) #rotate with varimax
op <- par(mfrow=c(1,2))
plot(harm.iso$points, main="unrotated")
plot(harm.varimax$loadings,main="rotated")

Bill

  
    
  
#
hello,
thanks a lot for your help.

@ Stephen: In my opinion the proscrutes rotation is more used to 
compare the 
alignment of different ordination methods. But maybe I'm wrong.

@ Bill: Thanks for the comprehensive description. That was the 
information I was 
looking for. 

Greetings,
Bernd Panassiti
#
On Wed, 2008-09-10 at 14:24 +0200, Bernd Panassiti wrote:
Isn't Varimax rotation what is done in PCA? If so, metaMDS() does this
rotation by default (via postMDS() and it's 'pc' argument). So this
feature may already be implemented in the metaMDS wrapper to
MASS::isoMDS that you mention.

G
#
hello Gavin,

yes, you might be right. I performed a comparison of a normal NMDS 
(with metaMDS) and a 
subsequent rotation with varimax. The rotation didn't seem to improve 
significantly the 
the alignment of the former ordination output.
Thanks for your hint & greetings.

Bernd Panassiti