Statistical analyses
The three regions differ greatly in climate, topography, and altitude
range (Table 1). Furthermore, which regions are occupied by a species is
the outcome of many factors. Therefore, in the
statistical analyses we treat region
as a synoptic trait that encompasses many unspecified environmental
factors. The three regions are contiguous, and we assume that, over the
evolutionary time scale represented by the diversity of the 9,370
species in the data set, there is little constraint caused by dispersal;
this assumption is confirmed by the analyses (see ”Results”).
To partition the relative contributions of climate region, plant growth
form, and phylogenetic conservatism on the variation in fruit type, we
used the R package “rr2” (Ives & Li 2018; Ives 2019) to calculate
partial R2s for phylogenetic logistic regression
models fit using the phyloglm function in the R package ”phylolm”
(Tung Ho & Ané 2014). The phyloglm function allows for the
calculation of only one type of partial R2 in the
rr2 package,
R2lik, that is based on the
likelihood.
R2likis the appropriate R2 when comparing models according
to the statistical significance of differences between them. The partial
R2lik for each factor was calculated
by comparing the full model with reduced models in which a given factor
was removed, and measuring the consequent reduction in explained
variance. The full model is a logistic regression model with fruit type
(fleshy and dry) as the dependent variable, climate region (tropic,
subtropical, and temperate) and growth form (woody and herbaceous) as
independent variables, and phylogeny as covariances in the residual
variation.
To give more information about the pattern of phylogenetic conservatism,
we performed the analyses not only with the full time-scaled phylogeny,
but also a phylogeny in which relatively recent phylogenetic
diversification was removed. Specifically, we created phylogenies with
reduced ”recent” phylogenetic structure by collapsing nodes above a
given threshold together to form a ”star”. For example, for a threshold
of 0.67, any node above the 66.7% (2/3) mark on the phylogeny was
collapsed so that evolution above this threshold was assumed to occur
independently among species (Fig. 1). The specific threshold of 0.67
corresponds roughly to the taxonomic scale of families. To determine the
importance phylogenetic patterns across the entire depth of the
phylogeny, we performed the same analysis using thresholds from 1 down
to 0.2. Re-analyzing the data across this range of thresholds addresses
the relative importance of ”recent” versus ”ancient” species
relationships in explaining phylogenetic conservatism.
The logistic regression showed that fruit type was poorly explained by
climate region and growth form, but the residual variation was
well-explained by the phylogeny. Because we had expected climate region
and growth form to be good predictors of fruit type, we investigated all
three variables – fruit type, climate region, and growth form –
separately to determine whether they all show similar patterns of
phylogenetic conservatism. If climate region and/or growth form show
less phylogenetic conservatism than fruit type, this would suggest that
climate region and/or growth form are phylogenetically more labile and
therefore might not be expected to predict fruit type. If climate region
and/or growth form show the same phylogenetic conservatism as fruit
type, then this, in combination with the low explanatory power they have
of fruit type, would imply that their evolution is uncorrelated to fruit
type.