Statistical analysis
In order to identify potential relationships between morphological data (body mass and skull dimensions) on one hand and seed and fruit sizes and seed dispersal distance data on the other hand we used a ”two-block partial least squares” (2B-PLS) approach (Rohlf & Corti 2000). This method makes it possible to quantify the degree of association between two tables of data, recorded for the same species. It is a descriptive multivariate analysis robust to multicollinearity between variables and therefore suitable for the use of morphometric and dietary variables. These analyses generate axes that explain the covariance between two data tables. A PLS correlation coefficient (Rpls) and the covariance percentage for each axis produced are obtained using the function “pls2b” in R from the Morpho library (Schlager 2013). The result comes from a set of 1000 permutations. Next, a sampling distribution of coefficients is obtained by resampling. The P 95‐value is calculated by comparison of the observed PLS coefficient to those obtained after resampling. The significance of each linear combination is assessed by comparing the singular value (PLS coefficient) to those obtained from permuted blocks. If the PLS coefficient is higher than those obtained from permutated blocks, then its associated P 95‐value is considered as significant. For each significant analysis, a graph and histograms of the variables are generated by using the Geomorph library (Adams et al. 2013).
Species share a part of their evolutionary history and therefore cannot be treated as independent data points. Thus, we also conducted these analyses (2B-PLS phylogenetic) with the consideration of phylogeny. We used the ”phylo.integration” function (Adams et al. 2014) in R from the Geomorph library. This function allowed us to quantify the degree of covariance of two data tables but under the Brownian evolution model (Adams et al. 2014). The blocks are phylogenetically corrected and the PLS coefficient (Rpls) between the two blocks is evaluated.
As body mass is known to impact morphology, we ran Spearman correlation tests (non-parametric) and linear regressions of each of the skull dimensions as a function of body mass and extracted the residuals. These were then used for 2B-PLS and phylogenetic 2B-PLS analyses to explore covariation between skull dimensions and fruit and seed size without the confounding effect of body mass.
We used R statistical environment (R core team (2019), version 3.5.2) for these analyses. All data were Log10-transformed before analyses to assure normality and homoscedascity.