Can we tell how a community was constructed? A comparison of five evenness indices for their ability to identify theoretical models of community construction.

Evenness indices provide a simple measure of community structure. It might therefore be possible to use them to identify the process by which a community has been assembled. To test whether this is practicable, we constructed simulated community samples using five stochastic models of community construction, proposed by Tokeshi. We then examined the ability of five evenness indices to identify the model under which the community samples were produced. Each model produced samples with a range of evenness values, but mean evenness differed between the models. The dominance decay (DD) model produced community samples with the greatest evenness, followed by MacArthur fraction (MF). Evenness was lowest under the dominance pre-emption (DP) model. The differences between models were quite consistent across indices and consistent between 5-species and 15-species communities. Samples produced under the DD and MF models varied least in evenness between 5-species and 15-species samples, and those produced under the random assortment (RA) model varied most, irrespective of the evenness index used. Evenness varied considerably between replicate samples, as expected with stochastic processes. In 5-species communities, the greatest robustness in evenness across replicates was seen using indices O or E'. In 15-species communities, O and E(Q) were the most robust. The index best able to identify the model which had generated the sample differed between models and with species richness. If the model of interest is not known in advance, the best index for identifying the generating model is E(var) for communities of more than 10 species for RF, RA and DP models and O for other communities. The number of samples required in a data set before it could be effectively identified was, for example, more than 30 for 5-species samples produced under the random fraction (RF) model, and 3 for 15-species RF samples. In contrast, a 5-species DD data set could be effectively identified when it contained 15 or more samples. We conclude that evenness can be used to identify the process by which the community has been constructed, out of the five models considered here. The best evenness index for doing this varies with the species richness and the evenness of the community but we can suggest the use of O without knowing the model because its the more stable one against species richness and the more robust and unbiased against simulated samples and it encompass a good discriminating power between the models in most of the cases studied.