Sooo I have a data matrix with my sites in the columns and the species across the rows. I have about 200 different species. For each species I measured its proportion (out of 100%) in each of the sites. There are LOTS of zeros.
I have settled on using PCoA and NDMS to see which method best visually represents the data. But I am wondering:
- a) do I need to transform the data beforehand? I.e., log transform or square root? I have read a few papers that have square rooted their % cover plant data.
- b) Can Bray-Curtis be used for proportion cover? Or is there a better dissimilarity measurement to use? I have read papers that use Bray-Curtis for % cover.
Thank you for your help :)
Hello :)
First, I'm not a biostatisticien, just a PhD student in the same situation but maybe it could help you. I have 20 sites (columns) and over 1450 OTU (rows).
Since I have talked with biostatisticiens from my lab and read some papers, I clearly understood that NMDS would be used in carefull conditions (because it's non-parametric). So PCoA could be better.
But, as explained in chapter 6 of this book Book Multivariate Analysis of Ecological Data with ade4
(little bit of advertising because it comes from a friend's lab...) you have other methods like AC (broadly : It organizes sites on their relative composition (NOT ABUNDANCE)) and Centred PCA, Normed PCA, Non-Centred PCA... Taking abundance but not in the same way depending of "Centred" or "Normed" or... So before to do PCoA it could be alternative methods without dissimilarity measurement.Then, to the transformation you'll find some informations in the "Box" of the book mentionned above and in this article : Paliy O, Shankar V. Application of multivariate statistical techniques in microbial ecology. Mol Ecol. mars 2016;25(5):1032‑57. I understood that transformation is not always necessary.
Once again, I'm not a biostatisticien I don't want to confuse you. Here are just some clues...
Bye !
The following article may be useful:
Article The piosphere revisited: Plant species patterns close to wat...
- Badges
- Science topic
- Similar topics
- Papers