*2.5. Hypothesis Testing*

To test hypothesis *i*, that the abundances and diversities of woody plants and large herbivores differs across habitats, we compared data from 33 sites. To estimate diversity, we used species richness, *ses*MPD, and *ses*MNTD at sites. To estimate abundance, we used the total number of plant stems and the summed RAI of all herbivore species. We focused on total stems as an indicator of thicket density and sightline obstruction, which inform large-herbivore habitat use [35]. However, different trees have different branching architectures—even within species, shorter trees can have more short stems—and heavy damage can promote stem proliferation where megaherbivores are abundant [36]. Thus, interpretations of total stem counts as a measure of thicket density and basal area as a measure of plant biomass may differ. Habitat comparisons were made using ANOVAs and Tukey's HSD. We also compared total plant-species richness across sites within each habitat using sample-based rarefaction based on the Bernoulli product model [37].

To test hypothesis *ii*, that local plant and herbivore diversities were positively correlated, we compared abundance and diversity both within and between habitats using ANCOVA. We constructed linear models in R using the herbivore-community characteristic as the dependent variable and the corresponding vegetation characteristic as the independent variable, with habitat types as the covariate (plateau vs. low plain). Because we had a small sample size of slope sites (N = 3), we only included the major plateau (N = 14) and low plain (N = 16) habitats in these linear models.

Finally, we evaluated hypothesis *iii*, that plant and animal communities are spatially linked. First, we tested for significant differences in PCD between habitats using permutational multivariate analyses of variance (perMANOVA) [38]. Second, we tested for significant correspondence between plant and animal PCD using Mantel tests. To account for the possibility that community similarities arise from spatial proximity, we used partial Mantel tests to evaluate correlations while accounting for distance between sites. Finally, to identify species-specific habitat associations, we performed indicator species analyses using the point biserial correlation coefficient based on Pearson's φ statistic with 999 bootstraps in *indicspecies* [39]. Pearson's φ ranges from −1 to 1, indicating strong avoidance or preference, respectively. The analysis was based on species' relative abundances using Hellinger transformation, corrected for unequal sampling across habitats, reported with *P*-values calculated independently across species.
