Statistical analyses
For the pollination treatments, a generalized linear model (GLM) with binomial distribution and logistic-link function was used to detect the effects of the treatments on the seed set (with seed number as event variable, total ovule number as trial variable and different treatments as factors) and fruit set (with fruit number as event variable, total treated flower number as trial variable, and different treatments as factors). We also performed a GLM with normal distribution and identity-link function to test for differences in the visitation rates (with visitation rate as the dependent variable, and pollinator species as factors) and pollination efficiency (with pollination efficiency as the dependent variable, and pollinator species as factors) of various pollinator species. To compare pollen removal and pollen deposition among pollinator species, data were analyzed with a GLM with Poisson distribution and loglinear-link function (with pollen number as a dependent variable and different treatments as factors). We carried out a Pearson correlation analysis to determine the relationship between the body size and the pollination efficiency of various pollinator species in H. mutabilis . All statistical analyses were performed in SPSS V. 19.0 (SPSS Inc., USA).