Factors contributing to non-randomness in species Co-occurrences on Islands

There has been dispute whether patterns of species co-occurrence on islands are largely random. We present a new method for testing this question; this method lets one not only examine whether a whole fauna is non-randomly structured, but also identify in which direction and by how much each particular species combination deviates from expectations based on randomness. Application of this method to the whole Bismarck and New Hebridean avifaunas, and to two particular guilds, shows that some pairs of species have more exclusive distributions than expected for random placement of species, because of competition, differing distributional strategies, or different geographical orgins. Other pairs of species have more coincident distributions than expected, because of shared habitat, single-island endemism, shared distributional strategies, or shared geographical origins. Much of the information about non-random co-occurrence is contained in the incidence graphs for occurrence of individual species. Finally, our present understanding of assembly rules is summarized.