Modelling invasion success as a function of phylogenetic and
functional distance.
To test whether phylogenetic distance predicted invasion success, we
asked how mean phylogenetic distance (MPD) and nearest taxon
phylogenetic distance (NTD) affected our two measures of invasion
success by means of generalized linear mixed models (BPMM, hereafter).
We built the models in the R packages MCMCglmm (Hadfield 2010) and BRMS
(Bürkner 2017), which implement Bayesian generalized linear mixed models
by means of a Markov chain Monte Carlo process. To test whether
phylogenetic distance influenced the likelihood that the exotic occurred
in a community, the occurrence of exotic species was modelled by means
of BPMMs with binomial distribution and logit link, including either MPD
or NTD (or related abundance-weighted metrics and metrics based on
functional distances) as fixed predictor and study site, species (as
some species occurred in more than one community) and phylogeny as
random factors. The inclusion of study site as a random effect was
important because it allowed modelling the likelihood that the exotic
was present in a community to be evaluated within each study site,
ensuring thus that the absence of the invaders did not result from the
fact that the species had never been introduced in the region. To test
whether phylogenetic distance influenced the relative abundance of the
invaders within a community, we restricted the analyses to communities
with two or more exotic species and used BPMMs with Gaussian structure
of errors. The fixed and random effects were the same used to model
occurrence with the only difference that study site was replaced by
community as random factor. In this way comparisons were made within
communities instead of across communities. For each BPMM model we ran
two independent runs with 200,000 generations, from which a 10% was
discarded as “burn in”. Convergence and good mixing of each of the
MCMC chains was assessed visually, plotting the traces of each of the
model parameters. We re-ran all models including the interaction between
phylogenetic distance and habitat, and the potential confounds of body
size and native species richness. Based on a model selection approach,
habitat was simplified to two categories according to the degree of
human-related alterations: habitats with high levels of alterations
(highly and moderately urbanized habitats) and habitats with low levels
alterations (little urbanized, rural and natural habitats).
Sampling procedure to study the effect of species richness on
phylogenetic distances. We randomly selected pairs of species without
replacement for different sample sizes (from 10 to 150 pairs) and
extracted the nearest phylogenetic distance from the matrix of
phylogenetic distances for the entire avian phylogeny. We repeated the
sampling 1,000 times, and represented the results as means and standards
errors of the mean. The sampling was conducted for all birds, as well
for the subset of species that were introduced in regions outside their
native range and those from the subset that became established at least
once (data from GAVIA (Dyer et al. 2017)). The approach was also
used to show that whether introduced and established species are a
non-random subset of all avian species.
Ecological validation of phylogenetic distance. We evaluated the
assumption that phylogenetic distance is a good surrogate of functional
distance in two ways. First, we used the package “geomorph” (Adams &
Otárola-Castillo 2013)to estimate the overall phylogenetic signal by
means of a multivariate version of Blomberg’s K (Adams 2014).
Major axes of niche variation were described by means of a Principal
Coordinates Analysis (PCoA), based on Gower’s generalized distance.
Second, we used Mantel correlations to compare phylogenetic distances
with functional distances (Mouquet et al. 2012). This analysis
was conducted both on the entire dataset and splitting it in the main
clades where exotics were represented. Finally, we used a multi-mantel
test to test whether, besides describing the resource niche,
phylogenetic relatedness also contained information on the degree of
similarity in response traits (i.e. relative brain size, resource niche
breadth and brood value).