Can you trust comparative trait data based on singleton species?

Abstract

Trait measures are affected by intra- and interspecific variability. Most studies aggregate species-level data to averages to analyze patterns of interspecific variation. Reliable per-species averages require data for many individuals per species, which leads to insurmountable measuring effort when studying species-rich assemblages. Here we argue that across a large number of species, patterns and relationships can be precisely recovered even if they are based on measures from a single individual per species. While these deviate to an unknown degree from the true per-species averages, randomly distributed errors will level out across many species. We used subsamples of body size data along elevational gradients for moths and small mammals (dozens to several hundred species per dataset), as well as simulated species assemblages, to illustrate this effect and explore some of its consequences. Single-individual measures correlated well with “true” (i.e., full data) averages. Furthermore, single-individual measures recovered the same conclusions on elevation-body size relationships as true data. Randomly removing individuals per species recovered true elevation-body size relationships very well, while randomly dropping species quickly led to high random variability in relationships across subsampling runs. Simulated species assemblages illustrated how the ratio of intraspecific to interspecific variability affects the correlation between singleton-based and true data. We conclude that trait measures based on single individuals are a viable alternative to multi-individual averages when analyzing assemblages of medium to high species richness. Reducing study effort by limiting the measurement of individuals per species, while retaining all species, is a much more reliable approach than restricting the number of species included in a study.