For some reason nMDS always separates the samples nicer than other ordination methods. Therefore I use it a lot. However it is hard to see the variance explained by each nMDS vectors. Therefore one should always consider a principle component method first (such as PCA, PCoA or CA).

My reference is mainly from this book: Numerical Ecology with R-2nd edition.

Today we will talk about CA(correspondence analysis). Some key facts:

1. It handles: presence-absence or abundance data (frequencies or frequencies like, dimensionally homogeneous, non-negative)
2. It is not influenced by double zeros.
3. Asymmetrical
4. No pre-transformation is needed.
5. The variation explained by each orthogonal axes is measured by total inertia (sum of squares of all values in matrix Q bar)
6. Two scalings: 1. rows(sites) are at the centroids of the columns (species). Default.