Publication bias
The funnel plots were symmetric for body size and pollination service provision (Fig. S7, Table S5). Although the funnel plots of pollinator abundance and activity length were asymmetric (Fig. S7), the regression estimates using the Trim and Fill method did not change. Pollinator richness changed from significant to not significant; however, the fail-safe number (N = 12 053) is much higher than the one required (N = 745), indicating that publication bias can be safely ignored (Table S5). We interpreted asymmetry in funnel plots carefully given the small sample sizes e.g., for activity length, or the lack of bidirectional outcomes for the effects of urbanisation on some variables, e.g., pollinator richness, which have been found to decrease across a lot of study systems, and thus will inevitably lead to a biased plot.