# Do I need to standardize environmental data before canonical correspondence analysis?

I have a dataset of 15 environmental variables (soil physical and chemical properties) and about 25 "species" variables for about 40 sampling sites. I want to do a CCA to analyse the effects of the environmental data on the species pattern. However, these variables are measured on quite different scales and units (eg. pH, clay content and salt concentration in %, total N in mg/g etc.). The species variables are all measured in the same unit.

For PCA, I would definitely standardize these environmental data to zero mean and unit variance, otherwise the ones with higher absolute values would shift the results inappropriately. But I couldn't find any information about if standardization is necessary or even appropriate for CCA.

• I have the same question. Did you find the answer? Here they write that yes, but without references: researchgate.net/post/… Nov 28, 2019 at 11:39
• Yes I did found the answer. It is it doesn't matter. CCA uses chi2 distances, which are not affected at all by standarization (I tested it by R just to be sure and found no difference in the results). Also, I've found out since then, that CCA is much more limited in usage, than I initially thought. It has a bunch of assuptions regarding the datasets, and is seriously biased towards rare species. So the only case I would use it is if I was interested in some environmental effects on rare species (e.g. indicator species). In almost any other cases, there are better solutions, e.g. RDA. Feb 5, 2020 at 15:45

• It sounds like you are claiming correlation coefficients can exceed $1$ in size--but of course they can't. What did you intend to write?