Rhesus macaques disrupt a keystone rodent-plant mutualism through behaviorally mediated higher-order interactions

Seed survival was modeled as a binomial response with a logit-link function (i.e., 1 for survival and 0 for failure). A GLMM with a Gamma error distribution was used to analyze variation in dispersal distance in relation to macaque presence. Macaque presence (present/absent) was treated as a fixed effect, while seed station and species were included as random effects.

Additional GLMMs were used to assess interannual differences in seed fate and the influence of macaque visit timing (measured as the day of the first visit after seed release), with Tukey’s post hoc tests for pairwise comparisons. To evaluate how macaque presence and visit timing influenced final seed survival, we employed a path-analytic framework that explicitly represented the sequential seed-fate process. Based on our conceptual model, macaque presence was hypothesized to influence final seed survival through three distinct pathways: (1) a direct effect, (2) an indirect effect mediated by scatter-hoarding, and (3) an indirect effect mediated sequentially through seed predation (the proportion of seeds eaten) and then scatter-hoarding.

We first fitted the model using all five years of data (including all seed species). Variance inflation factors (VIFs) were calculated for all predictors, and all VIF values were < 3, indicating no multicollinearity. To test path significance while mitigating potential small-sample or separation bias, we used bias-reduced binomial GLMs (brglmFit).

Standardized path coefficients and total effects were derived from the piecewise path models, and bootstrap resampling with 1,000 iterations was applied to obtain 95% confidence intervals for all parameter estimates. To assess whether macaque effects on seed fate varied among species in a trait-dependent manner, we extended the path-analytic framework to each of the 15 seed species monitored for two or more years. For each species individually, we applied the same analytical approach used for the pooled dataset (including all seed species) to evaluate how macaque presence and visit timing influenced seed predation, scatter-hoarding, and final seed survival.

Specifically, we derived standardized total effects of macaque presence and visit timing on each of these three response variables for each species. We then used linear regression models to examine whether these species-specific effect sizes could be explained by three key seed traits: seed mass, tannin content, and endocarp thickness. Separate univariate linear regressions were fitted for each trait-effect combination.

All analyses were conducted in R v.4.4.3. GLMMs were fitted using lme4 (1.1-36); Wald chi-square tests in the GLMMs were performed with car (3.1.3); Tukey’s test for post hoc pairwise comparisons was performed with emmeans (1.11.1); piecewise path models were constructed using piecewiseSEM (2.3.0.1); and bootstrap resampling was performed with boot (1.3-31).</p>