Data from: Uneven sampling and the analysis of vocal performance constraints

Studies of trilled vocalizations provide a premiere illustration of how
performance constraints shape the evolution of mating displays. In trill
production, vocal tract mechanics impose a trade-off between syllable
repetition rate and frequency bandwidth, with the trade-off most
pronounced at higher values of both parameters. Available evidence
suggests that trills that simultaneously maximize both traits are more
threatening to males or more attractive to females, consistent with a
history of sexual selection favoring high-performance trills. Here, we
identify a sampling limitation that confounds the detection and
description of performance trade-offs. We reassess 70 data sets (from 26
published studies) and show that sampling limitations afflict 63 of these
to some degree. Traditional upper-bound regression, which does not control
for sampling limitations, detects performance trade-offs in 33 data sets;
yet when sampling limitations are controlled, performance trade-offs are
detected in only 15. Sampling limitations therefore confound more than
half of all performance trade-offs reported using the traditional method.
An alternative method that circumvents this sampling limitation, which we
explore here, is quantile regression. Our goal is not to question the
presence of mechanical trade-offs on trill production but rather to
reconsider how these trade-offs can be detected and characterized from
acoustic data.