Modelling network metrics in relation to environmental gradients
We performed four separate analyses using H2’,
H2’ z-score, dsi*ants, and dsi*plants as response
variables, the environmental predictors as fixed effects (latitude,
elevation, and anthropogenic disturbance), and study identity as a
random effect (full model in supplementary, appendix 4 & 5). Linear
mixed effects models (lmer function in the lme4 package)
were used to predict changes in H2’ and
H2’ z-score, fitted using maximal log-likelihood (REML =
F) in order to compare models with different fixed effects. Generalized
linear mixed effects models with beta error distribution (glmmTMBfunction in the glmmTMB package) were used to predict dsi*ants,
and dsi*plants, as dsi* varies between 0 and 1. Model selection was
based on the corrected Akaike Information Criterion (AICc), calculated
for the complete set of models using the dredge function in theMuMIn package. The global model included all main predictor
variables and first order interactions between network type and each of
the other three predictors. The model with the lowest AICc score was
selected. For categorical predictor variables in the final model, to
understand which factor levels differed significantly, we inspected test
statistics and p-values for pairwise contrasts between levels, using therelevel function where necessary to explore all pairwise
comparisons. Note that family level type 1 error rate is retained at
0.05 for these comparisons, so no further correction for multiple
testing is required. This was also done for interactions involving
categorical predictors. Lastly, because only foraging networks were
present above 1500 m a.s.l. (Figure 3), we removed these four outliers
in terms of predictor variables, and repeated our full analyses to
explore whether these data were driving any of the patterns we observed.
We found that the model from all response variable stayed the same but
with higher AICc (see supplementary, appendix 6 & 7). Hence, for the
remainder of the paper, we present the analysis of the full dataset.