Asymmetric information in mixed-species mobbing ﬂ ocks: why are leader species special?

Asymmetric information transfer can shape the social structure of animal groups. Birds in mixed-species ﬂ ocks can gain antipredation bene ﬁ ts by relying on both personal and social information, ultimately increasing their survival. However, not all species in a ﬂ ock contribute danger-related information equally, leading to complex patterns of information ﬂ ow between species. Understanding these differences in information ﬂ ow and usage among species is key to exploring the mechanisms underlying the formation of such social assemblages. In this study, we combined mixed-species ﬂ ock surveys with playback experiments to investigate the asymmetric information transfer about danger between leader and follower species in the montane forests of Taiwan. The playback experiments included four treatments: predator calls (personal information), leader and follower mobbing calls (social information) and control. We predicted that leader species would be better at detecting threats than follower species and that the mobbing calls of leader species would elicit stronger antipredator responses in mixed-species ﬂ ocks than those of follower species. As predicted, and more striking than anticipated, leader species were always the ﬁ rst to initiate mobbing calls in response to predator calls, suggesting strong asymmetry in predator detection and/or danger communication between the two species' roles. Moreover, our re-sults showed that birds responded much more strongly to the playback of predator calls than to playback of mobbing calls. Contrary to our prediction, there was not a greater antipredator response to mobbing calls given by leader compared to follower species. Here, we propose that larger conspeci ﬁ c group sizes and associated kin-selected behaviour could drive leader species to develop superior antipredator signalling systems, potentially leading to asymmetric information transfer about predation risk between species. We also suggest that mutual antipredator bene ﬁ ts may be common in mixed mobbing ﬂ ocks, promoting positive interactions among bird species, and ultimately strengthening the cohesion and formation of mixed ﬂ ocks.

Interspecific information flow is a critical factor shaping the social structure of animal communities (Goodale et al., 2010;Martínez et al., 2018).Mixed-species bird flocks (hereafter referred to as 'mixed flocks'), consisting of multiple species that forage and move together in a coordinated manner, provide a typical example of a social assemblage that is influenced by interspecific communication and eavesdropping (Carlson, Healy, et al., 2020;Goodale et al., 2010;Jiang et al., 2020).Information about predators may be a key driver selecting for mixed flocks (Goodale et al., 2010).By gathering both personal and social information about danger, flock members can gain antipredation benefits, including increased foraging efficiency and ultimately survival (Goodale et al., 2010(Goodale et al., , 2020;;Sullivan, 1984a).
Birds in mixed flocks can acquire information about predators either directly from a predator's cues (personal information) or through eavesdropping on social signals (social information) from other flock members' alarm calls (Goodale et al., 2010;Magrath et al., 2007Magrath et al., , 2015;;Martínez & Zenil, 2012).However, social information may not always be reliable (Magrath et al., 2009(Magrath et al., , 2015;;Satischandra et al., 2010), which means that individuals must be selective in using different types of information.This decision to act on various sources of information is mainly influenced by the quality of the information, such as its reliability and relevance (Barrera et al., 2011;H€ am€ al€ ainen et al., 2023;Magrath et al., 2009;McLachlan et al., 2019;Turner et al., 2023).For instance, foliage-gleaning and bark-foraging species in mixed flocks rely more on heterospecific alarm calls to detect predators than do aerially foraging species (Goodale & Kotagama, 2008;Jones & Sieving, 2019;Martínez & Zenil, 2012).This may be because substrategleaning species encounter more visual obstructions that impede their ability to detect predators, making social information more valuable to them (Jones & Sieving, 2019;Martínez & Zenil, 2012).Although social information is important for many reasons, it may not be as reliable as personal information.For example, greater racket-tailed drongos, Dicrurus paradiseus, which are important sentinel species in mixed flocks (Goodale & Kotagama, 2005, 2008), sometimes use 'false' alarm calls to startle other birds and gain foraging benefits through kleptoparasitism (Satischandra et al., 2010).Thus, personal information may be a better indicator of risk (Barrera et al., 2011), as direct predator cues can provide more information about risk, such as accurate spatial information about a predator, compared to social information from either conspecific or heterospecific alarm calls.
Information transfer about predation risk is often asymmetrical among species in mixed flocks, as flock members can differ in ecological, morphological and behavioral traits, which could affect information transfer (Goodale et al., 2010;Jones & Sieving, 2019;Nolen & Lucas, 2009).For instance, aerial insectivores like flycatchers are better at detecting predators than are other species (Jones & Sieving, 2019), as their perch-based scanning for prey provides a wider visual field that enhances their ability to spot predators (Goodale & Kotagama, 2008;Jones & Sieving, 2019).Similarly, gregarious bird species like orange-billed babblers, Argya rufescens, are more likely to communicate potential dangers in mixed flocks (Goodale & Kotagama, 2005, 2008), due to their superior antipredator signalling systems facilitated by kin-selected behaviour (Goodale & Beauchamp, 2010;Sridhar et al., 2009).Additionally, species such as greater racket-tailed drongo and New Holland honeyeater, Phylidonyris novaehollandiae, elicit especially strong antipredator responses to their alarm calls because of their usually higher reliability in conveying information about dangers (Goodale & Kotagama, 2008;Magrath et al., 2009Magrath et al., , 2015)).These examples suggest that species differ in their capacity for information transfer and do not contribute information equally to mixed flocks.Understanding differences in information flow and usage between species is fundamental for exploring the mechanisms behind the formation of social assemblages.
In mixed flocks, species often have distinct roles that can be categorized based on features such as frequency of participation, conspecific group size, specific behaviours and spatial positions in the flock (Greenberg, 2000;Hutto, 1994;Mangini et al., 2023;Sridhar et al., 2009).Generally, 'leader' species are the first species to move in a particular direction, 'follower' species are those that follow the leaders and 'occasional' species may only temporarily join the flock when it passes through their territories (Goodale et al., 2020;Greenberg, 2000;Hutto, 1994;Mangini et al., 2023;Sridhar et al., 2009).Of these roles, leader species are considered crucial for the formation and maintenance of flock cohesion (Goodale et al., 2020;Greenberg, 2000;Sridhar et al., 2009).Information commonly flows from leaders to other species in the flock, providing foraging and/or antipredation benefits to the other flock members (Goodale et al., 2010;Goodale & Kotagama, 2008;Munn & Terborgh, 1979;Pagani-Núñez et al., 2018).For example, woodpecker species, which are typical followers, show significantly lower vigilance rates inside mixed flocks, suggesting that they receive antipredator benefits by joining flocks and making use of heterospecific alarm calls (Limparungpatthanakij et al., 2019;Sullivan, 1984aSullivan, , 1984b)).However, in some cases, the flow of information during mobbing can be from followers to leaders.For example, the white-breasted nuthatch, Sitta carolinensis, a common follower in tit flocks, more often responded first to playback calls of the eastern screech-owl, Megascops asio, than did the leader species (Nolen & Lucas, 2009).It appears that the flow of information in mixed flocks is more complex than often thought and may vary depending on specific conditions.
Mobbing calls are an important source of social information about dangers in avian communities, produced in response to predators that do not pose an immediate threat (Curio, 1978;Dutour et al., 2017).Previous studies have indicated that mobbing behaviour is also a vital antipredator behaviour in mixed flocks, with participants often provoked by the mobbing calls of other members (Carlson, Greene, et al., 2020;Coppinger et al., 2020;Jiang et al., 2020;Zhou et al., 2021).For example, red-breasted nuthatches, Sitta canadensis, a typical follower species, eavesdrop on subtle variations in heterospecific chickadee mobbing calls and are able to respond appropriately in mixed winter flocks (Templeton & Greene, 2007).They even vary their own mobbing calls in response to personal or social information, which suggests that nuthatches are sensitive to the source and reliability of information (Carlson, Greene, et al., 2020).Although both mobbing behaviour and mixed-species flocking are generally regarded as adaptations to reduce predation risk, their connections have rarely been experimentally investigated (Zhou et al., 2021).In particular, the potential consequences of interspecific asymmetries in information transfer in mixed mobbing flocks are not well understood (Carlson, Greene, et al., 2020;Carlson, Healy, et al., 2020;Nolen & Lucas, 2009).
Here, we combine mixed flock surveys with playback experiments in the montane forests of Taiwan to investigate asymmetric information transfer about danger between leader and follower species in mixed mobbing flocks.The specific objectives of this study were to (1) investigate whether leaders are more likely to be the first to respond to danger, (2) test whether personal information (from hearing a predator calling) provokes a greater response than social information (from hearing other species giving mobbing calls) and (3) examine whether the leaders' mobbing calls elicit stronger antipredator responses than those of followers.We predicted that leaders would be faster to communicate about predator cues than followers, based on the highly gregarious nature of the leader species in our flocking system and their potentially better antipredator systems (Chen & Hsieh, 2002;Liao et al., 2022).We predicted that flocking birds would be more responsive to direct predator cues than indirect mobbing calls, as direct cues from predators may provide more information about risk than indirect signals from conspecifics or heterospecifics.Finally, we predicted that leaders' mobbing calls would elicit stronger antipredator responses than followers' mobbing calls, as followers commonly depend more on information from leaders in mixed flocks.

Study Sites and the Flocking Systems
Fieldwork was carried out in two locations in central Taiwan, from October 2021 to January 2022.One was in the Alishan area (23 30 0 36 00 N, 120 47 0 24 00 E; elevation 1300e2000 m), and the other was in the Zhushan area (23 40 0 48 00 N, 120 44 0 24 00 E; elevation 950e1700 m).
Both study sites were dominated by coniferebroadleaf mixed forests and bamboo forests.Extensive trails within each study site were used to search for mixed flocks on foot.There were four (total length: 12.54 km; range 1.37e4.5 km) and three (total length: 6.36 km; range 1.31e3.35km) main trails in the Alishan and the Zhushan areas, respectively.The trails were each located in a large patch of forest and at least 500 m apart from one another to avoid repeated encounters with the same flock in a day.
Mixed flocks regularly form during the nonbreeding season when cold weather prevails in Taiwan.Most of these mixed flocks are led by one or two numerically dominant bird species that account for a major portion of the overall flock size (Chen & Hsieh, 2002;Liao et al., 2022).For example, in our study sites, blackthroated tit, Aegithalos concinnus, and Morrison's fulvetta, Alcippe morrisonia, are the most common leader species in canopy and understory flocks, respectively, often constituting up to two-thirds of the individuals in a mixed flock (Chen et al., 2022;Chen & Hsieh, 2002).On the other hand, most follower species are found in smaller numbers within mixed flocks, typically one to three individuals.Due to the evenly distributed vegetation layers in the forests at our study sites, particularly in bamboo forests, canopy and understory flocks frequently meet and form larger 'mega-flocks' (Liao et al., 2022).Sometimes, such combined flocks can consist of more than 100 individuals and include more than 10 bird species.
In this study, a mixed-species flock was defined as a roving group of three or more individuals, consisting of at least two species, which moved together in the same direction for at least 5 min while foraging (Morse, 1970).We defined 'leader' species as those found at the forefront of flocks, initiating movement and being followed by other species as the group travelled from one area to another, 'follower' species as those that consistently followed leader species in a mixed flock and 'occasional' species as those that temporarily joined a mixed flock when it passed through their territories but remained in the mixed flock during a playback trial.Before initiating a playback, we classified each bird species' role within a mixed flock based on these definitions.Due to the diverse composition and structure of each mixed flock, the same species could potentially assume any of the three roles depending on the particular mixed flock (see Appendix, Table A6).Sometimes, a mixed flock may include more than one leader species, such as when canopy and understory mixed flocks converge to form a single mixed flock (Greenberg, 2000;Liao et al., 2022;Munn, 1985;Poulsen, 1996).In such cases, species roles are generally maintained within each subunit of the mixed flock (Liao et al., 2022;Munn, 1985).Thus, in this study, we considered bird species that initially led their respective types of mixed flocks to be leader species, allowing for the possibility of two leader species within a single mixed flock.

Study Species
We chose three focal bird species for our playback experiments that are commonly found in mixed flocks in the mid-elevation mountains of Taiwan.The black-throated tit is a cooperative breeding bird (Li et al., 2012) that commonly forms large conspecific flocks of a dozen or more birds in the forest canopy (Severinghaus et al., 2010).During the nonbreeding season, the black-throated tit is also a dominant leader species in mixed flocks (Chen et al., 2022;Liao et al., 2022).They usually produce soft contact calls while foraging but become highly vocal when threatened.Green-backed tits, Parus monticolus, are often active in mixed flocks with black-throated tits and can be considered as an obligate follower species (Liao, 2015).They usually produce loud, clear vocalizations, particularly alarm calls.Steere's liocichla, Liocichla steerii, a typical understory bird, temporarily participates in mixed flocks when they pass through its territory (Liao et al., 2022).This species produces common, loud and nonthreatening vocalizations throughout the year, making their calls ideal for use as a control playback treatment.
The three species used in playback experiments are vulnerable to similar predators, including the collared owlet, Glaucidium brodiei, and the besra, Accipiter virgatus (Severinghaus et al., 2010).For our experiment, we chose the collared owlet to simulate predator presence.This is a diurnal owl that preys on small passerines and is commonly found in forests (Severinghaus et al., 2010).With a body length of 15e17 cm and extended foraging time, the owlet poses a significant threat to small birds.The owlet's distinct four-note whistled call can be heard during both day and night, eliciting strong mobbing responses in forest birds (Severinghaus et al., 2010).
In this study, we played back the vocalizations of three treatment species (i.e.black-throated tit, green-backed tit, and Steere's liocichla) and one predator species (i.e.collared owlet) to the mixed flocks.However, our focus extended beyond merely observing the behavioural responses of the three treatment species.Instead, we quantified the behavioural responses to the playbacks from all species present in each mixed flock.In total, we found 29 bird species participating in the mixed flocks (N ¼ 64; Appendix, Table A6).Among these species, four were classified as leader species in at least one flock according to our definitions: the blackthroated tit, Morrison's fulvetta, Taiwan yuhina, Yuhina brunneiceps, and rusty laughingthrush, Pterorhinus poecilorhynchus (Appendix, Table A6).Some commonly found follower species included the rufous-faced warbler, Abroscopus albogularis, green-backed tit and rufous-capped babbler, Cyanoderma ruficeps (Appendix, Table A6).

Playback Experiment and Recordings
We used a playback experiment to test our hypotheses about information flow in these mixed-species flocks.The experiment included four playback treatments: (1) leader mobbing calls, from black-throated tits; (2) follower mobbing calls, from green-backed tits; (3) predator calls, from collared owlets; and (4) control songs, from Steere's liocichlas (Fig. 1).Previous studies have demonstrated that mobbing behaviour can be elicited by using the playback of a predator's calls in mixed flocks (Jiang et al., 2020;Zhou et al., 2021).Using playback rather than predator models also ensured that all individuals had the opportunity to simultaneously detect the predator's cues.Each playback treatment lasted 30 s and was made with Raven Pro 1.6 software (Cornell Laboratory of Ornithology, Ithaca, NY, U.S.A.).
Mobbing and control playbacks were prepared from recordings of species at the study sites during the nonbreeding season.We prompted mobbing calls from the two tit species by playing back a 10 s recording of collared owlet calls.Recordings were made from a distance of 3e5 m using a Sennheiser ME67 directional microphone and a Marantz PMD660 digital recorder, sampling wave files at 48 kHz and 16 bits.Specifically, we placed a camouflaged JBL Go 3 Bluetooth speaker 1.5e2 m above the ground, within 10 m of a focal individual or single-species flock.We then stood about 5 m from the speaker, broadcast the owlet calls, recorded any calls and noted the behavioural responses from the target birds.The birds all exhibited typical mobbing responses, including (1) stopping foraging and looking around, (2) producing calls and (3) approaching the speaker.Additionally, we recorded the Steere's liocichla songs at a distance of 3e5 m from a singing individual.Playback exemplars for these three species were prepared from high-quality recordings without calls from other bird species or distinct background noise.Using these recordings, we made five different 30 s exemplars for each species by trimming or repeating them as needed.We filtered out sounds below the frequency of the calls: below 2 kHz for leader mobbing calls and below 1 kHz for both follower mobbing calls and control calls.
We were unable to record collared owlets calling, so had to use recordings from the Macaulay Library (https://www.macaulaylibrary.org/).We selected five high-quality wave files of owlet recordings, which included recordings from Taiwan (2 tracks: 24.9 s and 19.1 s), Malaysia (1 track: 14.4 s), Thailand (1 track: 9.5 s) and India (1 track: 9.7 s).These recordings were characterized by prominent owlet calls with little or no background noise and were ranked as having the best signal-to-noise ratio (5-or 4-star).Despite the calls being selected from different locations, they were acoustically similar, each consisting of a series of three-to four-note 1 kHz tones (peak frequency range 937.5e1033.6 kHz; N ¼ 5).At present, there is no evidence to suggest that these owlets have geographical variation in call structure.We made five different exemplars from these recordings by repeating tracks from the beginning to achieve a total duration of 30 s each.We filtered these exemplars at 0.3 kHz to remove sounds below the frequency of the owlet calls.
All playback treatments were played at 60 dB at 5 m, which is within the natural range of all three focal species, as measured at that distance using a TES-1350A sound level meter, with Aweighting and fast setting (Appendix, Table A1).The same level meter and settings were used to measure playback amplitude.We kept playback amplitude constant to control for the possibility that amplitude alone might account for differences in response among playback treatments.For that reason, we also broadcast owlet calls at the same amplitude, even though we could not confirm natural levels.

Playback Experiment and Response Measures
Two investigators searched for mixed flocks at each study site between 0700 and 1700 hours, averaging 3 days per week.We avoided using the same trail twice within a 24 h period to minimize the possibility of encountering the same flock more than once.When we encountered a flock, we identified all species, determined their species roles (e.g.leader, follower, occasional species) and counted with binoculars the number of individuals of each species.Before starting a playback experiment, we followed the flock from a distance of about 5 m for at least 5 min to let the flock habituate to our presence.After a single playback, we continued walking along the trail, looking for a new flock to receive playback.The average (± SD) distance between two playbacks was 492.93 ± 280.07 m (range 75.09e1610 m; N ¼ 47).Furthermore, we were able to distinguish between flocks based on their unique composition and structure.
Playbacks were broadcast from wave files on an iPhone 13 through a JBL Go 3 Bluetooth speaker positioned on a branch at a height of about 1.5e2 m.Two investigators positioned themselves about 5 m away from the speaker, each at a different angle, and stood in a direction away from the movement of the mixed flocks.One investigator (C.-C.Liao) initiated playback when the flock was 5e8 m from the speaker.Throughout the playback trial, the investigators positioned themselves at different angles relative to the mixed flock.This allowed for a broader view when scoring the behavioural responses of the birds and minimized the potential for visual obstructions that could arise from viewing from a single angle.Playbacks were initiated when (1) the flocking birds were foraging, (2) there was no sign of predators and (3) there was no evidence of antipredator behaviour by the flocking birds.The playback exemplar and order were randomized for each flock to control for potential effects of specific exemplars and order effects.We recorded 30 s of background sounds before playback and then audio-recorded the vocal responses of birds during playback and afterwards until all birds had moved more than 10 m from the speaker.The following behavioural responses were also noted by both investigators: (1) responding species (those giving mobbing alarm calls and/or approaching the speaker); (2) how closely each bird species came to the speaker (<1 m, 1e5 m, 5e10 m and >10 m).
Audio recordings were analysed to create an index of mobbing call intensity within the 30 s period after playback.For each 1 s interval, we noted the presence or absence (0/1) of alarm calls for each bird species in the flock.Since mobbing events often involve multiple species producing intense and loud alarm calls simultaneously, calls from different species may overlap, making it difficult to distinguish bird species based solely on their alarm calls.However, to improve the accuracy of our data, two researchers (C.-C.Liao and J.-C.Wu) with over 20 years of experience birding in Taiwan independently scored the audio recordings.Out of a total of 10 470 1 s samples, there was a 1.19 % difference in the two data sets.The discrepancy mainly occurred in the predator treatment due to (1) some species' calls being masked by louder calls from other species, (2) the difficulty in detecting high-frequency and soft alarm calls, such as those from rufous-faced warblers and flamecrests, Regulus goodfellowi, and (3) some alarm calls slightly crossing two 1 s sample periods, resulting in different interpretations between the two researchers.After cross-checking, our final data set included a total of 2143 s with alarm calls across 64 mixed flocks.These data were then used to calculate 'calling intensity', as the proportion of 1 s intervals that contained mobbing calls.
The behavioural and audio data were used to create five response variables for quantifying the effects of the playbacks.From the behavioural data, we identified (1) the species that responded by giving mobbing calls and (2) the species that responded by approaching the speaker during each playback trial in the mixed flocks.We defined 'approaching' as birds moving within 10 m of the speaker.This distance range was chosen because accurately identifying species and measuring distances beyond 10 m is challenging, particularly in dense forest environments.By analysing the audio data, we identified (3) the initiator species, which is the species that first gave an alarm call in response to the playbacks.We also measured (4) the latency to call as the time from the onset of the playback to the time when a bird initiated an alarm call and (5) the overall calling intensity of each mixed flock.

Statistical Analysis
All statistical analyses and figure plotting were performed in R version 4.0.3(R Core Team, 2020).
We used chi-square tests to examine whether leader species were more likely than follower species to initiate alarm calls upon detecting threat signals.Specifically, we used chi-square goodnessof-fit tests with Monte Carlo simulations with 2000 replicates to calculate the P value, using the 'chisq.test()'function from the 'stats' package.We used Monte Carlo simulations because of the small expected values in some cells.
We used linear mixed effect models (LMMs) to examine the effects of playbacks on the timing and intensity of responses to playback.In each case, we constructed LMMs with normal distributions and identity link functions, using the 'lmer()' function of the 'lme4' package.Latency to call (s) and the index of calling intensity were entered separately as response variables into LMMs.The playback type, total flock size and total flocking species were entered as fixed effects in the model and the identity of the playback track, trail and study site were entered as random effects.
We used generalized linear mixed effects models (GLMMs) to examine the effects of playbacks on the dichotomous responses of whether a species gave mobbing calls or approached the speaker within 10 m.The playback type, conspecific flock size and total flocking species were entered as fixed effects and the identity of the playback track, trial and study site were entered as random effects in our models.As flocking birds never approached the controls (Steere's liocichla songs) in any playback experiment, these were excluded from the GLMM analysis to avoid issues associated with complete separation of data (see Results).The GLMMs were constructed with binomial distributions and logit link functions, using the 'glmer()' function of the 'lme4' package.We carried out pairwise comparisons using 'emmeans()' function of the 'emmeans' package.Conditional R 2 values of models were calculated using the 'r.squaredGLMM()' function of the 'MuMIn' package.The full model with all factors of interest was fitted before likelihood ratio tests were used to identify significant fixed effects by removing them individually from the model and assessing the change in deviance.
As predicted, birds responded more frequently and strongly to playback of the predator's calls than they did to playback of mobbing calls.In the predator treatment, birds were both more likely to give mobbing calls and approach the speaker compared to the other treatments (alarm calling: GLMM: c 2 3 ¼ 43.73, P < 0.0001; Fig. 4a, Appendix, Table A4; approaching: GLMM: c 2 2 ¼ 26.74, P < 0.0001; Fig. 4b, Appendix, Table A4).Similarly, in the 30 s period after playbacks, alarm call intensity was higher in the predator treatment than in the other treatments (LMM: c 2 3 ¼ 55.59, P < 0.0001; Fig. 4c, Appendix, Table A5).Our results also demonstrated that conspecific flock sizes influenced birds' alarm calling and approaching behaviour, with larger conspecific groups in mixed flocks showing a higher probability of giving alarm calls and approaching the speaker (Appendix, Table A4).Additionally, there was a slightly lower likelihood of approach when there was a greater total number of species in the mixed flock (Appendix, Table A4).
Contrary to our prediction, the leader's mobbing calls did not provoke stronger responses than the follower's mobbing calls.There was no significant difference in the latency to give the first alarm call between the leader (mean ± SE ¼ 8.3 ± 2.5 s) and follower (10.1 ± 2.5 s; Fig. 3) treatments.Similarly, the probability of alarm calling (Fig. 4a), approaching (Fig. 4b), and even alarm call intensity (Fig. 4c) showed no significant differences between leader and follower treatments, with the weakest or no response found in the control (Fig. 4).

DISCUSSION
Our findings reveal asymmetries between leader and follower species in the response to information about danger, yet we found no evidence that a leader's mobbing calls were more evocative than those of followers.Leader species were more responsive to playbacks than follower species, and in particular always reacted first to  A3).N ¼ 14 flocks in follower treatment (data from two flocks excluded because no birds gave alarm calls), N ¼ 16 flocks in both leader and predator treatments.The proportion of alarm initiation by species' roles according to playback treatments.For the playbacks: follower ¼ green-backed tit mobbing calls; leader -¼ black-throated tit mobbing calls; predator ¼ collared owlet calls.Specific leader and follower species responding to playbacks are listed in the Appendix (Table A2) and were defined by their roles in flocks, as explained in the text.The results of statistical analyses are presented in the text and in the Appendix (Table A2).N ¼ 14 flocks in follower treatment (data from two flocks excluded because no birds gave alarm calls), N ¼ 16 flocks in both leader and predator treatments.
predator calls, revealing a strong asymmetry in predator detection or communication between the two roles within mixed flocks.Birds were more responsive to direct cues from the predator than to mobbing calls from flock members, showing that personal information was important in the response to danger.Social information nevertheless played a crucial role in amplifying antipredator behaviour, with birds responding equally to mobbing calls from both leader and follower species.Our results highlight a complex dynamic in interspecific information flow about danger that will contribute to maintaining flock cohesion and promoting the formation of mixed flocks.
As predicted, but even more striking than anticipated, leader species were always the first to initiate mobbing calls in response to predator calls.This pattern means that there is a strong asymmetry in information flow, from leader to follower species, in these Taiwanese mixed flocks.We suggest that this pattern arises because the leader species were more gregarious than follower species within these mixed flocks, which is likely to reduce the costs and increase the benefits of mobbing (Goodale et al., 2010;Sridhar et al., 2009).For example, black-throated tits and Morrison's fulvettas, the two most common leader species, had mean conspecific flock sizes of 16.4 and 14.8 individuals, respectively, whereas follower species typically had one to three individuals (Appendix, Table A6).Previous studies suggest that larger groups may dilute an individual's risk of being attacked or reduce predator attack efficiency by confusing the predator (Lima, 1995;Roberts, 1996).Hence, due to their gregarious nature, leader species may bear lower risks and costs than follower species when initiating mobbing calls.Another potential explanation is the greater presence of kin within larger groups, prompting individuals to rapidly produce alarm calls to alert their kin (Goodale & Beauchamp, 2010;Goodale & Kotagama, 2008;Sridhar et al., 2009).Indeed, black-throated tits are cooperative breeders (Li et al., 2012) that flock with their siblings and offspring in mixed flocks (C.-C.Liao and C.-C. Chen, personal communication).Previous studies also indicated that Morrison's fulvettas likely flock with their mates and offspring during nonbreeding seasons (Chen & Hsieh, 2002;Lin et al., 2003).Therefore, our findings suggest that initiation of mobbing calls in response to predator calls by these leader species may be primarily influenced by their sociality, facilitated by a risk dilution effect and kin-selected behaviour.
Another explanation for the faster initiation of mobbing by leader species is that their larger numbers within mixed flocks means that they simply detect predators more quickly, but we consider this less likely because predator cues were auditory.The 'many eyes' effect is a prevalent adaptive explanation for enhancing antipredator benefits in mixed flocks (Morse, 1977).It suggests that collective vigilance in larger groups enables prey to detect predators more quickly, especially via visual cues (Lima, 1995;Olson et al., 2015;Pulliam, 1973).Having more eyes looking for danger is likely to be a benefit, because different individuals can look in different directions and from different angles.However, in our study, we provided only auditory cues of danger, so flock members should have received cues of danger almost simultaneously, so the number of individuals may make little difference to the ability to detect danger.Therefore, it seems more likely that the faster response by leaders to predator calls in our study related to communication about danger (above), rather than detection of danger.However, if auditory cues became difficult to perceive, for instance due to lower amplitude or degradation, any one individual may have a low chance of detecting danger and group size may make detection quicker, an effect more like visual detection.Future studies could examine the antipredator benefits of group size according to sensory channel and the ease of detection.That approach could offer valuable insights into the mechanisms underlying asymmetric information transfer in intricate mixed-species animal groups.
Consistent with our second prediction, based on the principle that direct information is more reliable than social information, birds responded much more strongly to the playback of predator calls than to playback of mobbing calls.Specifically, our results showed that birds reacted to predator calls over twice as quickly as they did to either leader or follower mobbing calls (Fig. 3).Furthermore, the probability of alarm calling and approaching and the intensity of calling in response to predator calls were all more than three times greater compared to the responses elicited by mobbing calls (Fig. 4).The greater response to predator calls may be because they not only directly indicate the predator's presence, but also provide more accurate spatial information about the predator's position compared to mobbing calls (Barrera et al., 2011), which in turn can elicit the most intense mobbing.In addition, mobbing calls appear to represent the caller's perceived threat rather than the actual threat, potentially limiting their reliability or accuracy and so the response by others (Carlson, Greene, et al., 2020;Carlson, Healy, et al., 2020).For example, red-breasted nuthatches, a follower species, exhibited stronger mobbing behaviour in response to direct high-threat predator calls than to indirect heterospecific mobbing calls (Carlson, Greene, et al., 2020).Similarly, aerial insectivores, such as flycatchers and drongos, tend to rely more heavily on personal information than social information while foraging in mixed flocks (Goodale & Kotagama, 2008;Jones & Sieving, 2019).To sum up, our results indicate that flocking birds are more likely to rely on personal information (from predator calls) than potentially less reliable social information (from mobbing calls).This might help flocking birds avoid the costs of mistaking the degree of threat and help prevent error cascades (Carlson, Greene, et al., 2020).
Contrary to our third prediction, there was not a greater response to mobbing calls given by leader compared to follower species.There were no substantial differences in any of the four response variables (latency to call, probability of calling, approaching, calling intensity) between leader and follower treatments (Figs 3e4).In mixed flocks, leader species are often considered important information producers (Goodale et al., 2010;Pagani-Núñez et al., 2018;Zhou et al., 2021), providing foraging and/or antipredation benefits to the other flock members (Greenberg, 2000;Martínez et al., 2018;Sullivan, 1984b).However, the dynamics of information flow among flocking species are rather complex.For instance, the orange-billed babbler and the greater racket-tailed drongo, two leader species, have been documented responding to each other's aerial alarm calls in a Sri Lankan rainforest (Goodale & Kotagama, 2008).Furthermore, vocal information in mobbing was also observed to flow from nuthatches (a follower species) to chickadees and titmice (leader species) in response to eastern screech-owls (Nolen & Lucas, 2009).Here, our results show that more than 50 % of leader species initiated alarm calls in response to follower species' mobbing calls (Fig. 2), indicating that information about danger flows not only from leader to follower species but also vice versa (Nolen & Lucas, 2009;Zhou et al., 2021).Thus, mutual antipredator benefits can occur between the species roles in mixed mobbing flocks, promoting positive interactions among species and thus enhancing the cohesion and formation of mixed flocks (Goodale et al., 2020;Sridhar et al., 2012).
Distinguishing among different types of asymmetries between species within mixed-species social groups is a significant challenge.In mixed flocks, certain species are better at detecting dangers, some are more likely to communicate threats and some others have alarm calls that prove to be more reliable.These speciesspecific variations construct a complex dynamic of information flow within mixed flocks (Goodale et al., 2010).Other factors, such as the type of predator, urgency level of information and kin relationships can also indirectly influence these communication networks (Nolen & Lucas, 2009;Sridhar et al., 2009).Our results highlight a strong asymmetry in responses between leader and follower species under more dangerous conditions (i.e.hearing predator calls).However, in the case of lower-urgency mobbing calls from flock members, birds showed similar antipredator responses, indicating no asymmetry.This implies that different types of interactions within mixed flocks can lead to different symmetry in information flow.Our results suggest that sensory cues from different predators, and alarm calls of different urgency, might affect the symmetry of interspecific information within mixed flocks.Future studies that delve into these differences in information flow and utilization among species can help unravel the evolutionary mechanisms driving the formation of complex interspecific animal societies.

Figure 3 .
Figure3.Latency to call (s) to playbacks.The latency was the first bird to call, regardless of its role in the flock.For the playbacks: follower ¼ green-backed tit mobbing calls; leader ¼ black-throated tit mobbing calls; predator ¼ collared owlet calls.Predicted means ± SEM shown.Different letters indicate significant differences among means as assessed by post hoc Tukey HSD test (P < 0.05).The results of statistical analyses are presented in the text and in the Appendix (TableA3).N ¼ 14 flocks in follower treatment (data from two flocks excluded because no birds gave alarm calls), N ¼ 16 flocks in both leader and predator treatments.

Figure 4 .
Figure 4. Response to playbacks: (a) probability of calling, (b) probability of approaching the speaker and (c) calling intensity.For the playbacks: control ¼ Steere's liocichla songs; follower ¼ green-backed tit mobbing calls; leader ¼ black-throated tit mobbing calls; predator ¼ collared owlet calls.Predicted means ± SEM shown.Different letters indicate significant differences among means as assessed by post hoc Tukey HSD test (P < 0.05).The results of statistical analyses are presented in the text and in the Appendix (Tables A4eA5).N ¼ 16 flocks in each treatment.

Table A6
Twenty-nine bird species were recorded participating in 64 mixed-species flocks in this study Flocking frequency (%), mean number (±SD) and percentages of species roles in the mixed flocks of each species are presented.
Clements et al. (2022)owsClements et al. (2022).bPercentage of species roles, including leader species (L), follower species (F) and occasional species (O).For definitions of the species roles, please refer to the Methods section.1 Migratory birds in Taiwan.