[R-sig-eco] Defining Different Groups with Nonparametric Data (vegan package)

[Jari Oksanen]

On 20/11/2012, at 21:01 PM, Charles Jones wrote:

> Hey Folks:
> 
> I am trying to define groups within a dataset of water-quality analyses.
> The data need to be analyzed with non-parametric statistics.  However, I
> am unsure of my current approach  (See below).
> 
> Current Procedure:
> Step 1:  Non-metric Multidimensional Scaling  (Euclidean Distance on Scaled
> Data)
> Step 2:  Hierarchical Cluster Analysis  (Wards Method, Euclidean Distance
> of NMDS coordinates)
> Step 3: Multi-response Permutation Procedure (Euclidean distance of the
> rank-transformed waterquality data, groups defined by Step 2)
> 
> Question:
> Since NMDS coordinates are non-normal, I am violating the assumptions
> behind the Ward's clustering algorithm. However, according to the MRPP
> test, the groups are significantly different.  Since MRPP is a
> non-parametric test, can I justify using the groups defined through
> clustering?
> 
Nate Jones,

There are a couple of things here that I do not understand:

1) in which sense NMDS coordinates are non-normal?
2) in which sense they should be normal?
3) how should this influence Ward's clustering?


Cheers, Jari Oksanen
--
Jari Oksanen, Dept Biology, Univ Oulu, 90014 Finland