Data analysis
All statistical analysis were performed in R v3.5.2 (R Core Team, 2018)
using packages car (Fox & Weisberg, 2011), iNEXT (Chao et al., 2014),
MASS (Venables & Ripley, 2002) and mvabund (Wang, Naumann,
Eddelbuettel, & Warton, 2018). A significance level of α = 0.05 was
considered. To test for sexual size dimorphism, we compared all the
adult bird’s measurements (wing, 3rd primary, tail,
tarsus, weight, bill length, depth and width) using a MANOVA and
subsequent univariate tests. Dietary analysis and comparisons were all
done at 3 taxonomic levels: highest prey resolution (all prey items to
the most resolved possible taxonomic levels, which varied across
taxonomic groups), family and order. To compare the average number of
prey taxa detected per dropping of males and females, we used a GLM with
a Poisson error distribution. The overall richness of prey ingested by
both sexes was estimated using Hill numbers with the double of the
reference sample size to avoid extrapolation bias (Chao et al., 2014).
We compared the estimated richness considering sample coverage and not
sample size (Chao & Jost, 2012). Instead of comparing the 95%
confidence interval, a very conservative approach, we considered that
differences were significant if the 84% confidence interval (a proxy
for α = 0.05) of both estimates did not overlap (MacGregor-Fors &
Payton, 2013). Finally, we also compared the diet composition between
sexes using Generalized Linear Models for Multivariate Abundance Data
with a binomial distribution (manyglm and anova.manyglmfunctions). We did not include in diet analysis possible confounding
variables as sampling day or sample collection localization, because
they do not differ between sexes (sampling day (1stApril = day 1), GLM with negative binomial distribution: LR Chisq =
1.066, df = 1, p = 0.302; latitude, GLM with Poisson distribution: LR
Chisq = 2.149, df = 1, p = 0.143; longitude, GLM with negative binomial
distribution: LR Chisq = 2.056, df = 1, p = 0.152).