Pollinator and plant traits
For pollinators, we collected several functional trait data (Table 1). Activity length and seasonality (i.e., abundance and richness at different seasons) data were collected for all pollinator groups. For body size, we used the inter-tegular distance (ITD) for bees and the wingspan for Lepidoptera. Hymenoptera pollinators were further categorized based on (1) their nesting behaviour: above ground (tree, wood, stem, above ground cavity) or below ground (within existing tunnels or excavators), (2) sociality: social, solitary or parasitic and (3) diet specialization; polylectic or oligolectic depending on if they feed on various or a particular plant taxon (Michener 2007). Due to data limitations, functional traits were collected mainly for Hymenoptera and Lepidoptera (for details, see Table 1).
We categorized each flower as non-radial (e.g., Lotus spp.) or radial (e.g., Asteraceae) to test if flower morphology mediates the effects of urbanisation on plant reproductive success.
Effect size calculation andhierarchical meta-analysis
We used Hedges’ d , weighted by sample size, as our effect size.r -type and statistic values -type data were transformed into Cohen’sd and then into Hedges’ d using standard mathematical formulas (Koricheva et al. 2013; Borenstein et al. 2021; Table S1). Hedges’ d was calculated using the R statistical software (R Core Team 2021). In all cases, a negative value of Hedges’d reflects negative effects of urbanisation on pollinator communities. Effect sizes were considered statistically significant if their 95% bias-corrected bootstrap confidence intervals (CI) did not overlap with zero (Koricheva et al. 2013; Borenstein et al. 2021).
Some publications provided more than one effect size, which may result in pseudoreplication, so we carried out a hierarchical meta-analysis that allows nesting effects within papers/studies (Tuck et al.2014). We included a publication-level random effect as a nesting factor to incorporate this non-independency. We first performed a random effects meta-analysis to calculate the overall mean effect size of urbanisation on pollinator abundance and richness, flower abundance and richness, functional traits, and pollination separately. Second, we incorporated moderators, including the climatic region of the study, the origin of the species, the taxonomic group of pollinators, and the symmetry of the flowers. To assess the levels of the heterogeneity of effect sizes, we calculated the P-value of the Qtstatistics. When they were statistically significant (P < 0.05), the influence of the moderators on the effects of urbanisation was examined using Qm.
To test whether a change in floral diversity (Hedges’d for flower abundance and richness) could predict a change in pollinator diversity (Hedges’ d for pollinator abundance and richness), we fitted maximum likelihood meta-regression models (Filazzolaet al. 2020). The adjustment of Knapp and Hartung was then used to account for uncertainty in the variance between studies, with overall model significance against an F-distribution (Knapp & Hartung 2003). All analyses were conducted in R statistical software using themetafor package (Viechtbauer 2010; R Core Team 2021).