Supplementary material

Data collection was conducted at the Pied Babbler Research Project, Kuruman River Reserve (26˚58’ S, 21˚ 49’ E) in the semi-arid acacia savanna of the southern Kalahari, South Africa, between December 2003 and May 2008 (for habitat and climate details see Raihani & Ridley 2007).

Mixed species flocks aggressively mob perched raptors, giving repeated loud calls lasting up to 15 minutes, to which pied babblers respond strongly, frequently joining the mobbing. We opportunistically recorded 5 mobbing events that did not involve pied babblers (3 directed at pearl-spotted owlets, Glaucidium perlatum, one at a giant eagle-owl, Bubo lacteus, one at a gabar goshawk, Melierax gabar – all species to which pied babblers are known to give alarm calls and which pied babblers have been observed to mob). Using these recordings, we constructed 10 one-minute playback tracks. No section of recording was used on more than one track; all tracks contained the same number of calls from the same species (fork tailed drongo, Dicrurus adsimilis; common scimitarbill, Rhinopomastus aterrimus; crimson-breasted shrike, Laniarius atrococcineus; southern yellow-billed hornbill, Tockus leucomelas and Cape glossy starling, Lamprotornis nitens). As controls, we recorded the same species giving context-neutral calls (territorial or sexual calls) and created 10 one-minute tracks with the same number of calls. Pied babblers were never observed to respond to the context-neutral calls of heterospecifics.

To perform the playbacks, we waited until there had been no disturbance and no predator alarm for at least 10 minutes, before recording group behaviour for 10 minutes. We then placed a speaker in a concealed position within vegetation, 2m above ground, 10m from the centre of the group and at least 5m from the nearest bird, and played back either mobbing calls or control calls (at least 5 days between playbacks at the same group; half groups received mobbing calls first, half received control calls first). After the playback, we recorded group behaviour for another 10 minutes.

Predator presentations were conducted from March-May 2008. We used puff adders as our predator model. Primarily ambush predators, they prey on a variety of small terrestrial vertebrates, (Mehrtens 1987). Our puff adder model replicated the average girth, length and colour of puff adders found in the area. We placed the puff adder model on the ground in the direction that the group was foraging, 10 m away from the foraging group. Responses to the model were recorded as soon as the observer had moved 10 m away from the model snake. We placed the model so it was initially hidden from the group to prevent any association between observer and model. Groups never spotted the snake while the observer was placing it on the ground or moving away from it.

Paired comparisons were performed in SPSS Statistics 17.0. All tests were 2-tailed and data were tested for normality before appropriate parametric tests applied. Because the LMM analysis of sentinel behaviour following natural predator alarm calls involved repeated observations of sentinel activity at the same groups, we included a random term (group identity) that allowed the analysis to take account of repeated measures, estimating the variance components using the Restricted Maximum Likelihood (REML) method. Model simplification using backward elimination was adopted. Terms were systematically removed from the model and only added back if their removal resulted in a significant loss of explanatory power. Null hypotheses were rejected at α < 0.05.