Null model analyses of specialization in plant-pollinator interactions

Recent studies have suggested that plant-pollinator interactions may be less specialized than previously thought. We contrasted patterns of specialization observed in five plant-pollinator interaction webs with predictions based on null models. In the five data sets, the observed number of extreme specialists and extreme generalists was significantly higher than the null expectation. This pattern was mostly due to a positive correlation between species frequency of interaction (f) and their estimated degree of generalization (s). After accounting for this association, the expected frequency distribution of degree of specialization generated by the null model closely matched the observed frequency distribution in the five data sets. A second null model, which explicitly incorporated the correlation between f and s, also generated expected frequency distributions of specialization that closely resemble those observed in the data sets. To make progress in understanding the distribution of degree of specialization in pollination systems it will be necessary not only to improve the quality of the data and to refine methods used to quantify specialization, but also to answer the question of why more frequently interacting species appear to be more generalized.