Dear Macarena,

You could use a SIMPER analysis, which is, for instance, provided in the free software PAST (see manual attached, p. 118). It is specifically designed to assess which taxa are primarily responsible for an observed difference between groups of samples. Like nMDS, SIMPER uses similarity measures such as Bray-Curtis or Euclidean distance. The results are given in percent (SIMPER means "similarity percentage").

Diversity indices might give you an idea of differences in diversity among the samples but won't help to identify the species responsible for it.

Best regards,

Thomas

You can compare Shannon or Simpson diversity indices.

See: Researches on Population Ecology

August 1999, Volume 41, Issue 2, pp 203–215

A comparison of species diversity estimators

Or you can perform anova, which assigns percentages that account for differences. I hope this helps :-)

Dear Macarena,

You could use a SIMPER analysis, which is, for instance, provided in the free software PAST (see manual attached, p. 118). It is specifically designed to assess which taxa are primarily responsible for an observed difference between groups of samples. Like nMDS, SIMPER uses similarity measures such as Bray-Curtis or Euclidean distance. The results are given in percent (SIMPER means "similarity percentage").

Diversity indices might give you an idea of differences in diversity among the samples but won't help to identify the species responsible for it.

Best regards,

Thomas

Macarena, you may find this useful:

http://www2.ib.unicamp.br/profs/thomas/NE002_2011/maio10/Magurran%202004%20c2-4.pdf

Tomas

Thank you very much everyone!

Dear Macarena,

When mentioning biodiversity, is it beta-diversity?...in which case a simple similarity index could be used. Twinspan and or Decorana could also be useful.

Best,

Guy

Hello Macarena, I perform SIMPER on PRIMER-e software. Just in case this program is more familiar to you.

Best regards

Juan

i also consider that SIMPER is a good option. However, is also strongly recommended that y ou check for the turnover of species between your communities. For that you could use a Jaccard index, in this form you could also consider the identity and composition of species for the change observed between communities. And later this could be useful to compare if you have some environmental gradient.

Call me old school, but no amount of this kind of analysis can establish cause-effect in ecology. Time for an experiment.

I would also recommend SIMPER as one of a suite of tests to carry out on your data. I would also throw Indicator Species Analysis (e.g. as in PCOrd, see Legendre and Legendre, Numerical Ecology). You do need to pursue beta-diversity mesures. If you have environmental variables, I would also look at CCA as an option. It always will come down to the interpretation of the results of the analyses, the data and your knowledge about the ecosystems you are studying.

Another option compatible with R is to run a Multinomial analysis (nnet package) with species as your response. It requires rearranging your dataset a little but is not difficult to run and when combined with allEffects plots shows the relationship for all species with your variables. I am happy to send you example code if this approach interests you.

I agree with Christopher Beachy. These descriptive and associative measures discussed here can suggest patterns, but the only way to know if particular species are drivers of differences in biodiversity is to conduct manipulative experiments. In order to understand the patterns better, you might explore network analysis, which will tell you the relative strengths of (direct and indirect) interactions among the species (although you still might need experimental data).