Methods for detecting non-randomness in species co-occurrences: a contribution.

Comparison of co-occurrences between species on a group of islands with those expected from a randombased null model could provide evidence on community structure. However, it is difficult to decide on the appropriate null model. Gilpin and Diamond proposed a model and a test for departure from it, but this test is shown to indicate significant structure even when applied to a matrix of random numbers. An alternative method is suggested, using the distribution of Gilpin and Diamond's deviation as test statistic, but determining the expected distribution by Monto Carlo simulation, and using many such simulations as a randomisation test of significance. The null model used accepts the observed totals of occurrences for islands and species; it therefore offers a somewhat conservative test. Applied to the Vanuatu bird data that Gilpin and Diamond used, significant departure from a null model is seen, but with an excess of extreme negative associations, the opposite result from that given by Gilpin and Diamond's method. It is not possible to tell whether the negative associations are due to autecology, biogeography, or to interactions between species.