Null models and randomization processes: some applications in community ecology

Abstract

Therefore, a hypothesis about a causal process often must be tested by comparing the observed patterns with the ones which would be expected in the absence of the postulated process - the patterns corresponding to the null hypothesis. A null model is the statistical procedure (usually involving simulation) by which the pattern corresponding to the null hypothesis is obtained, Null models are increasingly popular in community ecology, ever since Connor & Simberloff used a null model of CO-OCCUlTences (presence / absence) to deny that coexistence of bird species on islands differed of random patterns, as it had been claimed by J. Diamond. Connor and Simberloff's model has limitations, just as the alterative model proposed by Diamond & Gilpin to rescue their own conclusions. As the limitations of both models are different and complementary, it seems justified to use hybrid models which bring together the advantages of both. Perhaps more important than the models by themselves, however, was their role in raising an intense philosophical discussion on hypothesis testing in community ecology. More recent applications of null models are not restricted to binary data, allowing the use of quantitative data, modeling patterns of co-abundance (Sale & Stell 's method); the degree of organization in plant communities can also be evaluated (Zobel & Zobel's method). It is proposed the use of null models to objectively determine the number of ecologically significant factorial dimensions in multivariate ordinations. Such procedure is exemplified using the results of an ordination of a community of marine fish.