Testing for non-randomness in sizes and habitats of West Indian lizards: choice of species pool affects conclusions from null models

Several null models are proposed for testing whether size or habitat differences in West Indian Anolis lizards are greater than expected 'by chance'. The models differ primarily in choice of the pool from which species are sampled to form random communities. Regardless of choice of pool, size differences in the Lesser Antilles are greater than null models predict; the pool using species on the known source (Puerto Rico) gives a greater variance in ratios but about the same mean ratio (for males), or a greater mean ratio (for females), compared with the pool composed of species on the islands being tested (the Strong et al., 1979, Galapagos procedure). On satellite islands of the Greater Antilles, sizes do not differ more than expected from null models. Pools composed of mainland-source species give null communities with more small or more large ratios than those composed of island species, depending upon whether four-species islands are included or excluded, respectively. Colwell and Winkler's unmodified 'Narcissus' hypothesis is contradicted by these results in procedures where species not likely to be able to occur on small islands are included in the species pool. Using the most biologically reasonable, but not other, choices of source pool, species on satellite islands of the Greater Antilles differ more in structural habitat then expected 'by chance'. In contrast to some of the results on size, here mainland-source pools are more likely to produce a statistically significant difference between real and random communities, as predicted by the 'Narcissus' hypothesis. However, exclusion of structural habitat categories not found on satellite islands is necessary to achieve this significance.