Differences in vocalizations promote accurate identification of species
during mate choice. The properties of vocalizations vary, however,
and overlap in the properties of different species' calls raises the
possibility of errors in species identification.
A general model of these possibilities is provided by discriminant
function analysis (DFA). To illustrate this use of DFA, we
consider possibilities for error in the classification of advertisement
calls by Hyla ebraccata in a diverse community of neotropical hylid
The analysis used three features of their calls: duration, dominant
frequency, and pulse repetition rate, separately and in combination.
These properties are known to be used for mate choice by female
H. ebraccata and many other species of frogs.
With only one feature, DFA misidentified
12-32% of individuals, either by assigning H. ebraccata calls to
other species (missed detections) or by assigning calls of other species
to H. ebraccata (false alarms). With two call parameters,
DFA committed few or no errors. If the analysis included the
relative abundances of the different species, there were even fewer
Thus DFA can use as few as two features to identify calls of H.
ebraccata almost without error.
The dispersion of the different species in discriminant-function space was
random, not over-dispersed as expected, regardless of the call features
included in the analysis. The lack of overdispersion might be
explained by differences in selection on calls of common and rare species.
This application of DFA indicates several ways in which understanding the
possibilities for errors can advance our understanding of the evolution of
communication in general.