[R] vegan metaMDS

[Michael Denslow]

> Hi,

Hi Radu,

> 
> I'm trying to use metaMDS with a dissimilarity matrix
> of angles, not
> Bray-Curtis, and I wanted to know if there is an in-built
> function to
> produce a plot of stress values against dimensions, that
> could be used to
> determine the 'true' dimension of the solution. 

I am unaware of any formal test for determining the proper number of dimensions in an NMDS ordination. It probably makes sense to think about your data and interpretation of your solution. For example it can be hard to present a 4D solution. Also, as I understand it the first dimension of a NMDS ordination with 2 dimensions will not be the same as the first dimension of an NMDS ordination with 3 dimensions. The axes are arbitrary in this method. 

One idea is to simply plot the stress by dimensions and look to see where the stress levels off or drops to an acceptable level.

here is a rough example!

plot(c('2','3','4','5'),
	c(23.26272,16.37268,12.9641,10.50830), 
	xlab = 'dimensions', ylab = 'stress', type = 'b')

hope this helps,



Michael Denslow

I.W. Carpenter Jr. Herbarium [BOON]
Appalachian State University
Boone, North Carolina U.S.A.

-- AND --

Communications Manager
Southeastern Regional Network of Expertise and Collections
sernec.org