Potential for conspecific pollen transfer
We examined the potential for conspecific pollen transfer as follows.
For each pollinator, we counted the number of pollen grains that were
morphologically identical to grains obtained from the anther slide on
which the pollinator was caught (hereafter ‘conspecific pollen grains’),
from a subsample of 500 pollen grains. To do so, we began examining each
slide at the upper left corner, and we recorded the species of each
grain until either 500 grains were encountered (N = 355 slides) or all
the pollen on the slide had been counted (N = 158 slides; Table S1). We
asked how different our estimate of the proportion of conspecific grains
from a subsample of 500 might be from the proportion calculated using
all grains. To do so, we randomly selected 10 pollinator slides that had
more than 500 total grains and counted and identified all grains (59,189
grains). The correlation of the proportion of conspecific grains
estimated from the subsample of 500 grains and that estimated from all
grains was 0.87 (cor.test function in R; P < 0.001).
We examined if pollinators from natural sites carried a greater amount
of pollen grains that were conspecific (hypothesis 1) using generalized
linear mixed models with a Poisson error structure implemented using the
lme4 package (Bates, 2014) in R (R Core Team, 2013). The number of
conspecific grains was the dependent variable, site type (urban or
natural) was the independent variable, and sampling site was a random
effect. We determined if urban and natural sites differed in species
diversity or richness of pollen on pollinators using linear mixed models
with site type as the independent variable, and sampling site as a
random effect. We used likelihood ratio tests (LR tests) to evaluate the
significance of the independent variable of interest by comparing nested
models with and without that factor. In the results section we report
estimates from the best model chosen via backward model selection
(hereafter “Est.”), and the P-values from likelihood ratio tests
comparing nested models.