Phylogenetic ANOVA
As seen in Table 2, the 49 species of aquatic birds in this study,
excluding the darter and dippers, were assigned to nine foraging niches
(independent groups) in accordance to Pigot et al. (2020). Twelve land
bird species were included for the purpose of similarity comparisons.
These twelve species were divided into two foraging niches (ground
feeders and aerial/sally). Darters were excluded on account of their
exceptional feather microstructure (see Discussion). The datasets of the
two dippers were incomplete and have not been included in the
calculations (reported here for archival purpose).
All statistical analyses were conducted using the R statistical analysis
software (version 3.6.0). Significance for all analyses was recognized
for values of p < 0.05. Normality of datasets was calculated
using the Shapiro-Wilk test of normality with the shapiro.testfunction from the R base functions. It was shown that the deflection
parameter dataset was not normally distributed (W = 0.824, p< 0.001). Significance of differences of deflection parameters
between groups were calculated to determine which foraging niche
represented higher/lower deflection parameter values. These results were
used in comparison with the phylogenetic ANOVA results to substantiate
the influence of phylogeny. Significance of the differences of
deflection parameter values between aquatic and land bird species was
calculated using the Mann-Whitey U Test with the wilcox.testfunction. Significance of differences between foraging niches of aquatic
birds was calculated using the Kruskal-Wallis H Test with thekruskal.test function.
For both aquatic birds and land birds, phylogenetic ANOVA was used to
determine whether feeding niches explain differences in feather
microstructure while accounting for phylogenetic relationships. Two
independent phylogenetic trees, consisting of 49 aquatic and 12 land
bird species, were obtained from www.birdtree.org (Jetz and Thomas
2014). A 1000 trees were generated for both land and aquatic birds and
representative trees were constructed using the maxCladeCredfunction from the phangorn package (version 2.5.3). Phylogenetic
trees depicting the phylogenetic relationships between bird species as
well as placement of groupings in the different feeding niches are
illustrated in Fig. 1 and Fig. 2.
Group aggregation of the bird groups on the phylogenetic trees was
calculated using the two.b.pls function from the geomorphpackage (version 3.1.2). An R-value of 1 was indicative of total group
aggregation and a value of 0 indicated no group aggregation. The
foraging niches listed in Table 2 were regarded as the independent
variable. Data on feather microstructure with regard to deflection
parameter were considered as dependent variables.
In this study, a randomizing residuals in a permutation procedure (RRPP)
phylogenetic ANOVA approach was used as described by Adams and Collyer
(2018). This analysis was performed in 1000 iterations using theprocD.pgls function from the geomorph package (version
3.1.2). This method is beneficial since it has demonstrated the
importance of accounting for group aggregation in phylogenetic ANOVA.
Moreover, it has shown that group differences can be detected, if they
exist, in a phylogenetic context more accurately than in phylogenetic
simulation models. RRPP is also more appropriate for highly multivariate
datasets.