Hypothesis testing
To test whether community size scales identically with elevation on the two islands, as predicted if communities are locally saturated, we first regressed local community size (total number of individuals derived from mark-resight models) against elevation. Similarly, to assess whether greater regional diversity enhanced community-level species diversity across the same environmental gradient, we regressed resight-model-derived plot-level species richness against elevation. In both cases we used posterior means from the resight model to represent the best-estimate of true values.
We evaluated trajectories of abundance and species richness along elevation using linear mixed-effects models (LMMs). Our full model predicted the response as a function of linear and quadratic elevation terms, with full interactions with island identity. Models included a random intercept of landscape to acknowledge that multiple plots occurred in the same geographic area. We iteratively dropped terms from these models, comparing the full model with the nested model, until a likelihood-ratio test (evaluated against a chi-square distribution) indicated that all remaining terms were significant (alpha = 0.05). We square-root transformed resight-model-derived abundance and species richness estimates to meet model assumptions of residual normality and homoscedasticity (log transformation—more typical of richness and abundance data—failed residual normality and homoscedasticity tests).
Next, to test whether greater macroevolutionary diversity facilitates the evolution of environmental specialists, thereby permitting greater partitioning of communities across space, we calculated abundance-weighted pairwise Bray dissimilarities between sites within each island using the betapart package (Baselga & Orme 2012). We further examined abundance-weighted phylogenetic and morphological dissimilarity to assess whether differences in turnover had deeper evolutionary underpinnings, or resulted in ecomorphologically different community structures, respectively. To do so we used the time-calibrated phylogeny and measurements of ecologically-relevant morphological traits for Greater Antillean anoles presented in Mahler et al. (2013). We calculated phylogenetic dissimilarity of plots using abundance-weighted unifrac from the weighted_unifrac function within the abdiv package (Bittinger 2020). For morphological dissimilarity we calculated abundance-weighted mean pairwise distances between the species of all community pairs on each island in a morphospace of ecologically-relevant traits (the four-dimensional PC-space of Mahler et al. (2013), which represents 93% of total morphological variation across Greater Antillean Anolis lizards). In all cases we assessed the rate at which pairwise community dissimilarities changed as a function of pairwise elevation differences between sites on the same island. We used linear regression models including both linear and quadratic effects of elevation as well as an effect of island identity, and an island-by-elevation interaction.
Given that sub-regional faunas formed to differing degrees on the two islands, we evaluated whether similar environments led to whole-fauna morphological convergence at both island and sub-region levels, gauged using community mean morphology. We applied a cut-off of 700m to split lowland and highland elevational zones into subregions. This elevation corresponded to the location of maximum turnover between highland and lowland species (but results are not sensitive to exact cutoff value). We assessed the abundance-weighted mean morphology along the four ecomorphological PC axes described above. To test whether morphology between faunas (or subfaunas) was more similar than expected by chance, we shuffled species morphologies 10,000 times, and examined where the observed Euclidean distance between fauna-mean morphologies fell relative to the permuted null distribution (we obtained two-tailed p-values by multiplying one-tailed p-values by two).
Finally, given the distinctiveness of the high-elevation fauna, we examined whether these species drove patterns of diversity within local communities across the elevational gradient. To do so we eliminated from the community dataset all highland species (species whose abundances are maximized above 700m, none of which occur at sea level). We then reassessed how both abundance and species richness of lowland species changed across the elevational gradient of each island using the same LMMs and model selection process described above.