Wow. you didn't say how you did CCA but if it used stepwise selection watch out!!! see the attached. Again you didn't say what was plotted. Best wishes, David Booth

Dear David,

Thank you for your answer. I performed CCA on a matrix of relative abundances of different taxa over sample sites, using a matrix of 8 environmental gradients (related to climate, land use and human impact) to explain taxonomic variance. I did not use stepwise selection.

I think I found my answer, but encountered the next question right away:

Calculating Euclidean distance of a specific environmental gradient using the CCA scores of all canonical axes of the CCA result should give the exact length of that vector.

So I did. However, it turned out that the length of those vectors representing the environmental gradients in my triplot does not compare to the Euclidean distance calculated using the CCA scores over all canonical axes, but just the CCA scores of each environmental gradient over the first 2 canonical axes (the ones also displayed in the CCA triplot). As a result, the vector length of one on my less strong environmental gradients, precipitation, is half as long as it would be when calculated using all its CCA scores instead of only the two that belong to CCA axis 1 and 2.

It seems to me that the choice to calculate length (= impact) of environmental gradients on taxonomic composition only using the CCA scores of axes 1 and 2 is a simplification of complete calculation that is compromising the reliability of the results. Or did I misinterpret something here?

Sincerely, Serge

I am not familiar with the software but I would guess that calculation was done on the basis of the most important components not all of them because the other components are usually noise. However, you really need to consult someone more familiar with this than I am. Sorry, David Booth

No worries, thank you David!