Low-amplitude song predicts attack in a North American wood warbler

Many animal signals given during agonistic interactions are thought to convey levels of aggressive motivation of signallers, but few studies have experimentally tested whether such signals actually predict subsequent aggressive behaviour. In songbirds, the predictive content of most putatively aggressive vocal signals remains largely unresolved. Prior studies of agonistic vocal signalling in birds have generally assessed the reliability of vocal displays using indirect response measures such as approach to a playback speaker. However, this method does not allow aggression to be assessed directly, because subjects have no opportunity to engage in actual attacks. In this study, we assessed the aggressive content of vocal signals in black-throated blue warblers, Dendroica caerulescens, using a direct measure of aggression. We presented territorial males with a brief period of conspecific song playback, recorded their vocal responses, and then revealed a taxidermic mount, giving subjects an opportunity to attack. We found that one song feature, low-amplitude ‘soft’ song, was an extremely reliable predictor of attack; a logistic regression model in which the occurrence of soft song was the sole predictor correctly classified 93% of males as attackers or nonattackers. By contrast, four other vocal behaviours (total number of songs, rate of song type switching and numbers of two nonsong vocalizations) failed to predict whether birds would attack the mount. These data provide evidence that soft song in black-throated blue warblers conveys extremely reliable information about impending aggression. We suggest that the reliability of soft song observed in this study and recent studies in emberizid sparrows may be stabilized by vulnerability costs because the limited distance with which it can be transmitted forces signallers to place themselves at risk of retaliatory attack.