[R] metaMDS in vegan: zero stress

[Jari Oksanen]

Mia Bengtsson <miamynta <at> gmail.com> writes:

> 
> Dear R and vegan package users,
> 
> I have been experiencing problems with the metaMDS function when working
on a dataset (euk) consisting of 9
> "sites" (RNA extracts of 9 biofilms samples) and 340 "species" (microbial
taxa based on rRNA sequences).
> The problem is that I get nMDS ordinations with overlapping points, so
that it looks like 6 samples are
> identical, while the remaining 3 are well separated. The dataset is rather
large, and quite complex so I do
> not think that this is a correct representation of dissimilarities. Every
time this happens, I get the
> warning message:
> 
> "In metaMDS(t(euk_red), distance = "bray") :
>   Stress is (nearly) zero - you may have insufficient data"
> 
> Removing relatively rare species using: 
> 
> euk_red<-euk[rowSums(euk)>100,]


Nine (9) points is not a "rather large" data set, but it is very small.
You have 9*8/2 = 36 pairwise dissimilarities among your nine points, and
you try to estimate 2*9 = 18 coordinates in mononotonic regression. No
wonder you get a warning of insufficient data. 

You may use vegan::stressplot() function to inspect the result and
see how it fails. My guess is that your three points are well separated
from others, and probably have some disjunction with microbial taxa
that do not occur in the rest.

I suggest you do not use nonmetric MDS with nine (9) points. It may
be that metric scaling (cmdscale in R or wcmdscale in vegan) works
better.

Cheers, Jari Oksanen