FST models
We began our phylogenetic multiple regressions analyses of factors
affecting genetic structure by constructing a null model with the
sampling-scheme variables. We sequentially added the life history traits
to this null model, checking whether each addition improved model fit of
a multiple regression based on Akaike Information Criterion (AIC) scores
(Akaike, 1974). Mating system and growth form were added together as
there is ample evidence of their effect on FST (Duminil
et al., 2007; Hamrick & Godt, 1992). We then added pollination mode and
seed dispersal mode, to check whether either, or both together, improved
the previous model. After finding the best model explaining
FST with life history traits (Q1), we compared this
model to one that included latitudinal region as an additional factor
(Q2). We assessed the variance explained by each model with the R
package rr2 and the function ‘R2.pred’ (Ives, 2018; Ives & Li, 2018).
We further evaluated the best-fit model through a backward stepwise
model selection with the function ‘phylostep’ in the phylolm package.
The functions ‘phylostep’ and ‘phylolm’ were congruent in finding the
same best model.
We then evaluated the importance of each variable in this best-fit model
(Q3). We used the R package rr2 and the function ‘R2.lik’ to obtain the
unique contribution of each factor in terms of the amount of
FST variance explained by comparing the best-fit model
with a reduced model not including the factor of interest.