Community-wide herbivory measurements
We recorded foliar herbivory for each species following the methods outlined in Chen et al. (2019). For each plant species in a quadrat, we randomly sampled 25 leaves from at least five individuals. For those species with less than 25 leaves, we sampled all leaves available. To quantify herbivory severity, we predefined six damage categories according to how much of the leaf was consumed: 0%, <5%, 5–10%, 10–20%, 20–50% and >50% (Scherber et al. , 2010; Ness et al. , 2011). For needlelike leaves (e.g. , fromKobresia myosuroides ), all incidences of leaf wounding were placed into the <5% category as field observations revealed that insect predators only ever consumed the tip of the leaves.
We calculated the overall herbivory percentage (ranging from 0% to 100%) for each plant species by first multiplying the number of leaves in each damage class by the median removal value for that class (i.e. , 0%, 2.5%, 7.5%, 15.0%, 35.0% and 75.0%). Obtained values were then summed across all damage classes and the sum divided by the total number of leaves (including undamaged ones) in a sample.
We calculated community-wide herbivory (hereafter, CWM herbivory) as the summed species’ (herbivory) means weighted by their abundance (Chenet al. , 2019) using the following formula:
\begin{equation} CWM\ herbivory=\frac{\sum_{i=1}^{S}{a_{i}h_{i}}}{\sum_{i=1}^{S}a_{i}}\nonumber \\ \end{equation}
where S is the total number of plant species in a quadrat,ai is the aboveground biomass of plant speciesi , and hi is the herbivory on plant species i .
Using the method proposed by Lepš et al . (2011) to decompose effects into their direct and indirect components (i.e. , “intraspecific variability” and “turnover”, respectively), we subtracted species’ turnover effects from the total CWM herbivory to obtain the intraspecific variability in herbivory among sampling sites. Turnover effects represent the expected CWM herbivory (based on constituent host plant species) independent of the actual herbivory measured in a given plot (Chen et al., 2019).