[R-sig-eco] MetaNmds and ordinal data

[Sarah Goslee]

Hi Frank,

On Thu, Mar 26, 2015 at 3:37 AM, Frank Berninger
<frankberninger at gmail.com> wrote:
> Hello,
>
> I intend to use MetaNMDS on ordinal data and was wondering if there are
> any guidelines and examples what to do (and especially what not to do).
>
> At the moment my design decisions are to use an Euclidian distance
> matrix (since I use already ordinal data anything more complecated seems
> to be inappropiate). I want to test dependencies on environmental
> factors post-hoc. Are there any thoughts, articles or anything else to
> help a newby in ordinations on that subject?

While it's common to treat ordinal data like interval data, that can
be a problem for distance-based tests if the difference between 0 and
1, for example, is substantially different than the difference between
10 and 11: both will have the same Euclidean distance, even though
they aren't equally dissimilar. There are various alternatives,
including ranked distances, Gower dissimilarity, etc.

Without knowing anything else about your data or hypotheses, it's
impossible to give you advice beyond spending some time with the many
journal articles in the Journal of Vegetation Science discussing
multivariate analyses of ordinal data.

Sarah

-- 
Sarah Goslee
http://www.functionaldiversity.org