Statistical analysis of food webs

The numbers of species in different trophic levels of a food web are modeled as a trinomial random vector, with cell probabilities potentially depending on the total number of species in the web. A maximum likelihood method is developed to test the hypothesis that the fractions of species in different levels are independent of the total number of species. The method is applied to a data set whose properties have been debated in the literature, and it is shown to be a powerful alternative to the simple linear regression approach used in previous analyses.