Individual contributions to the CS
To test whether the obtained complementary information on lemur and plant traits influence the magnitude of the CS, we fitted linear mixed models with a lognormal distribution, using a Bayesian approximation (details in Supplementary Material S9). We fitted separated models for lemurs and plants, and separated models for each interaction type where CS was significant (see Supplemental Material for details).
For lemur traits, we had complete information for all lemur species included in our analyses. We fitted four models, one for each interaction type: frugivory, folivory, florivory and granivory. In each model, we included activity pattern (diurnal, nocturnal, cathemeral) and feeding guild (nectarivorous, omnivorous, folivorous-frugivorous, folivorous, frugivorous, insectivorous) as independent variables, and Procrustean residuals calculated by the PACo function as response variables. Abundant species tend to be generalists, so to account for the variation in geographic distribution area among lemur species, we used geographic range as a proxy of their density, given the lack of abundance data for each specific site. This approach is supported by strong associations between species distribution ranges and population sizes (Estrada et al. 2017). Data on lemur distribution range (in km2) were extracted from IUCN maps (IUCN 2022) and added as random effect to models. To account for the possible influence of sample size of interactions, the total number of studies reporting each interaction was incorporated as a random effect to the models. To control for the possible influence of phylogenetic relatedness, we added lemur phylogeny as a covariate in our models.
For plant traits, we had information for only few species, and each trait was available for different sets of plants (Table 1). In order to take better advantage of the available information, we fitted separated models for each trait, correlating it to the respective interaction type. To account for the possible influence of sample size of interactions, the total number of studies reporting each interaction was considered as random effect in the models. To control for the possible influence of phylogenetic relatedness, we added plant phylogeny as a covariate in our models.