0

I have a dataset composed by presence of different bacterial families in function at different pesticide treatment. I need to find a good representation of my data but I don't know which method (Principal Component analysis or Principal Coordinate analysis or Nonmetric Multidimensional Scaling) is better for species' abundance data.

ttnphns
  • 51,648
  • 40
  • 253
  • 462
Giorgia
  • 53
  • 1
  • 8
  • 1
    In answers to [this](http://stats.stackexchange.com/q/14002/3277) question it has been shown that PCoA _is_ fundamentally PCA. When we use PCA approach in order to map in low dimension a matrix of distances between observations (rather than observations X variables data) we call it PCoA. Iterative forms of MDS (not necessarily nonmetric, NMDS) are more advanced tools to do the task: they can map the distances with less error (loss) and they have more options, such as selecting the loss function, for example. – ttnphns Jul 20 '15 at 12:36
  • It has also been shown that NMDS is better for zero-rich species abundance matrices because it has fewer assumptions. It is not clear, though, where the pesticide treatment comes into your question as currently worded. – katya Jul 20 '15 at 23:52

0 Answers0