Relationship between diversity and precipitation
We compiled precipitation
data at the plot level using the mean annual precipitation from locally
interpolated climate data (Giambelluca et al., 2013). To relate
diversity to precipitation, we used generalized linear models with gamma
family and log link, as diversity metrics are skewed and always
positive. We added an interaction term between mean annual precipitation
and island to test the effect of island age on the model. We fitted the
models using the base function in R ‘glm’: Diversity ~
log(precipitation) * island, family = Gamma(link = ”log”), where
diversity is taxonomic, phylogenetic and functional alpha diversity of
the different q orders (0, 1, 2) and precipitation is mean annual
precipitation (mm/year). We tested the relationship between betaC
diversity and the absolute difference in mean annual precipitation using
generalized linear models with gamma family and log link, with an
additional interaction term between island and precipitation difference.
We determined whether the interaction between island and precipitation
was significant for all models in two ways. First by plotting the model
estimates and their uncertainty (85% and 95% confident intervals),
using the function ‘plot_summs’ in the jtools package (Long, 2022),
which reports the results of regression models. Second, we tested the
statistical significance of the interaction using the Likelihood ratio
test, which compares the goodness of fit of a simpler (no interaction)
versus a more complex (interaction) model. We favor the complex model
with interaction if the Likelihood ratio test shows a p-value smaller
than 0.05. We used the ‘lrtest’ function from the lmtest package
(Zeileis & Hothorn, 2002) to compute the Likelihood ratio test. Lastly,
because the range of mean annual precipitation across islands differs,
we subset precipitation to a common range for all islands (MAP
> 800 & MAP < 4000) to test the consistency of
the model’s results for all diversity metrics. All analyses were
performed in R version 4.1.0 (R Core Team, 2021).