Phylogenetic uncertainty
We acknowledged phylogenetic uncertainty throughout our analyses, which
result in high overlap of parameter estimates (Figs. 2-4, S8-S10). Upham
et al. (2019) reported that building their rodent phylogeny was
especially challenging due to missing genetic data and topological
uncertainty producing polytomies. These uncertainties were further
propagated across our estimates of tip-based metrics. Thus, our
inference of traits was based on the collective evidence provided by the
phylogenies of Upham et al. 2019.
Results suggest two major implications of phylogenetic uncertainty. The
first is the error possible when estimating the average value of the
tip-based metrics. For example, consider the density plot in the middle
of Figure 3 where we have two peaks of stasis time. If you choose
calculating tip-based metrics using only one phylogeny it is very likely
that you would have an estimate of either ~2 or
~3 ma, but would err by at least ~1 ma.
The second implication is the error we can make when estimating the
effect size of variables. Again, considering the density plot in the
middle of Figure 3, at ~2 ma of stasis time the
assemblages from the Atlantic Rainforest have higher stasis time than
assemblages located outside Atlantic Rainforest. However, no difference
between these regions can be found at ~3 ma. By
propagating uncertainty in our estimates, we infer that the more likely
value of average stasis time is ~2.3 ma (inset gray
boxplot, Fig. 3), and the more likely effect of Atlantic Rainforest
ecoregions is a positive deviation from average aST (inset black
boxplot, Fig. 3). Therefore, it is highly desirable that, when
available, we use a set of phylogenies rather than only one to test
hypothesis about the evolution of ecological traits (Rangel et al.
2015).