Influence of age, breeding state and approach direction on sensitivity to human gaze: a field study on Azure-winged magpies

In predator–prey interactions, various factors affect the prey’s perception of risk and decision to flee. Gaze sensitivity, the ability to react to the presence, direction, or movement of the head and eyes, has been reported in many birds. However, few studies have focussed on variation in sensitivity to human gaze in relation to other risks and potential breeding costs. Here, we studied the influence of human gaze on the escape behaviour of Azure-winged magpies (Cyanopica cyanus) and investigated the effects of breeding state (breeding season and nonbreeding season) and approach direction on gaze sensitivity. In Experiment 1, we tested whether magpies showed different sensitivities to human gaze according to age class and breeding state when approached directly. The results showed that the breeding state could affect the flight initiation distance (FID), with adults in the breeding season having a shorter FID compared to those in the nonbreeding season. Meanwhile, only adults were found to be averse to direct human gaze and juveniles showed no sensitivity. In Experiment 2, we conducted three different gaze treatments on adult magpies in the breeding season under three bypass distances (0 m, 2.5 m, 5 m). The results showed that approach direction had no effect on FID, while the sensitivity to human gaze differed under three bypass distances. Adults could clearly recognise human head and eye direction at a certain bypass distance (2.5 m). Our study reveals the cognitive ability of Azure-winged magpies to human head and eye direction and the effects of age, breeding state and approach direction, which may provide further insights into human–wildlife interactions, especially for birds in urban habitats.


Introduction
As a strong selective force, predation has long been linked to the evolution of various animal physical and behavioural traits (Preisser et al. 2005). All prey animals need to detect and avoid predators swiftly to survive, given that the failure is irretrievable. However, animals are unlikely to spend all their energy and time on increasing vigilance, hiding and other anti-predation behaviour (Kent et al. 2019). Prey should detect and respond to approaching predators, while attending demands for foraging, territorial or courtship activities (Lima and Dill 1990). To maximise fitness, prey must take both predation risk and potential fitness benefit into account when making behavioural decisions (Cooper 2006). In addition, during predator-prey interactions, predation threat may vary with the fluctuating level of fighting behaviour of predators or among predators at different times (Helfman 1989;Jakobsson et al. 1995). As predicted by the threat-sensitivity hypothesis, prey may trade off predator avoidance against other activities and alter their avoidance responses to match the degree of threat posed by the predator (Helfman 1989;Lima and Dill 1990).
Escape is one of the most widespread anti-predator adaptations, given that even species well protected by chemical and morphological defences would commonly resort to escape when confronted by a predator (Stankowich and Blumstein 2005;Tätte et al. 2020). However, escape could be costly as prey need to sustain the possibility of losing feeding, mating opportunities and energy expenditures simultaneously (Cooper and Frederick 2007). Ydenberg  Dill's economic model of escape behaviour focussed on the cost of fleeing and the approaches of the predator. In this model, optimal escape theory states that prey should begin fleeing from an approaching predator when the predator's risk of predation equals the cost of fleeing (Ydenberg and Dill 1986). The distance at which a prey animal is projected to move away from the approaching threat is the optimal flight initiation distance (Ydenberg and Dill 1986). Due to its ease of measurement and association with some critical elements of fleeing behaviour (i.e., alert distance), flight initiation distance has been widely used as a metric in research on optimal escape theory and risk assessment (Cooper and Blumstein 2015). Many experimental examinations of the effects of multiple risk and cost factors on a wide range of prey have corroborated the optimal escape theory (Cooper and Frederick 2010;Stankowich and Blumstein 2005). Among numerous risk factors, the flight initiation distance increases with the level of risk in general, including predator speed, predator size and directness of attack (Stankowich and Blumstein 2005). Other factors involved with potential changes in the cost of fleeing, such as the environment (e.g., distance to refuge), prey physical condition (e.g., group size, body temperature, breeding state) and others, would also directly influence the flight initiation distance (Stankowich and Blumstein 2005). For example, in the breeding season, prey that are involved in defending territories, mating or combating may permit closer approach by predators (Cooper and Wilson 2007;Ota 2018). Previous studies found that many animals in the breeding season may mediate risk aversion (Davidson et al. 2018;Dreber and Hoffman 2010). European mink (Mustela lutreola) became bolder and more explorative in the breeding season (Haage et al. 2013). A captive group of rooks, Corvus frugilegus allowed novel people to approach closely in the breeding season (Greggor et al. 2016). Thus, breeding state may have an impact on the responses to predatory risk. In addition to the conspicuous cues of predatory risk, some subtle indicators of risk, such as gaze direction, have received some empirical attention currently (Davidson and Clayton 2016). The Indian rock lizard (Psammophilus dorsalis) fled earlier when the approaching human was looking directly at the lizard, compared with looking away (Sreekar and Quader 2013). The capacity to respond efficiently to the presence, direction, or movement of the head and eyes is known as gaze sensitivity and it could allow prey to respond efficiently when their potential predators are close (Carter et al. 2008;Davidson and Clayton 2016).
As reviewed by Davidson and Clayton (2016), the aversive escape response to gaze, known as "gaze aversion", has been reported in mammals, birds, reptiles and fish. Individuals may react more nervously to direct gaze than averted gaze because forwards-facing eyes are connected with predator attacks (Davidson and Clayton 2016). The aversion to gaze may be present at birth, which means that fearful responses to direct gaze cues could emerge from a very early age, as suggested in studies of African jewel fish (Hemichromis bimaculatus) and chickens (Gallus gallus) (Coss 1979;Jones 1980). However, current evidence argues that the development of gaze aversion may also be influenced by experience (Davidson and Clayton 2016). Bobwhite quail (Colinus virginianus) was not sensitive to directed human gaze only if they were exposed to human caretakers after hatching (Jaime et al. 2009).
The sensitivity to subtle indicators could also be affected by other risks, including predator and social information (Davidson and Clayton 2016). There is often interplay between multiple predator information and the features of the gaze cues available (Davidson et al. 2014). As reviewed by Davidson and Clayton (2016), predator identity, facial expression, approach direction and other predator information had significant effects on gaze sensitivity. For example, a study on American crows (Corvus brachyrhynchos) showed that regardless of human facial expression, crows fled sooner when an approaching human looked directly at them (Clucas et al. 2013). A tangential approach may not immediately imply predatory intent, so the birds may notice more subtle cues, such as direct gaze, compared to a direct approach (Bateman and Fleming 2011;Cooper et al. 2003;Stankowich and Coss 2006). Some environmental cues such as landscape characteristics and human presence may also be considered when investigating predator gaze cues (Ydenberg and Dill 1986). For instance, hadeda ibises (Bostrychia hagedash) could distinguish the direction of an approaching human's attention, no matter whether there were multiple other humans nearby (Bateman and Fleming 2011). More research on how animals integrate gaze cues with additional cues is needed for an in-depth understanding of how and when animals use information from gaze cues to optimise their decisions.
Here, we investigated the aversion to human gaze (with the congruent and incongruent movements of eyes and head direction being considered separately) of Azurewinged magpies (Cyanopica cyanus) and the effects of the breeding state and approach direction on a university campus. As a common urban species, Azure-winged magpies would assess the degree of threat posed by humans, who could be both potential predators and resource competitors. Azure-winged magpies are cooperative breeders and have flexible helping behaviour, with nearly 50% of nests being assisted by helpers at the nest (Valencia et al. 2003). The helpers mainly contribute to the provision of the incubation female and the brood, as well as the defence of the nest (Ren et al. 2016). The breeding season of Azurewinged magpies in this area started from late April to late August. We first tested the effects of breeding state and age class (i.e., adults in the nonbreeding season, adults in the breeding season and juveniles in the breeding season) on their sensitivity to human gaze. The juveniles are the birds that have just fledged the nest. If gaze aversion is acquired through age (i.e., potentially experience), as corroborated by the previous studies mentioned above, then adults should have shorter flight initiation distances in the breeding season and show a greater sensitivity to the difference between the three gaze treatments compared with juveniles. Meanwhile, if additional cost could be incurred by the breeding state of adults and flight initiation distance would decrease with increasing cost, as proposed by optimal escape theory, then a significantly shorter FID was expected to be seen in adults in the breeding season than in those in the nonbreeding season. The other objective of this study was to test the changes in gaze aversion of adult magpies in the breeding season under additional risk (approach direction). Thus, two risk factors, human gaze and approach direction, were given simultaneously. If gaze aversion could be affected by the other risk factor, then magpies would show different sensitivities to human gaze under direct and tangential approaches.

Study site and species
We carried out this study on free-living Azure-winged magpies at Nanjing University Xianlin Campus in Nanjing, Jiangsu, China (32° 7′ 12" N, 118° 57′ 11" E), during the nonbreeding season (November 2020-March 2021) and breeding season (May-August 2021). Five flocks were studied on campus, which have been tracked all day long in advance and corroborated to have discrete territories. The number of individuals within each flock ranges from 29 to 42 (36 on average), with the newborn birds not being accounted. Each flock was sampled at one site at least 500 m from the others. At each site, a feeder constructed from a metal board (40 cm × 60 cm) affixed to a 1.5 m tall metal pole on the ground was set up. No flock was spotted to cross its sites and foraged on other feeders over the entire experiment, so it is reasonable that each site represented a unique flock. We did not identify individuals within each site, as birds were not captured and marked in this study. Therefore, the within-site pseudorepeatability could not be avoided and the precise breeding state of each bird (whether they had chicks in the breeding season) remained unknown as well. Each feeder was baited with bird food (Jiangsu Synergetic Pharmaceutical Bioengineering Co., Ltd., Nanjing, China) every day for at least one week before the experiment and then early in the morning on every data collection day.

Procedure
We measured the FID of each magpie as a primary response variable. During each trial, the experimenter used a Shendawei ranging telescope SW-600A laser rangefinder (Senwei Electronics Co., Ltd., Guangdong, China) to measure FID. An observer positioned approximately 50 m away used a Cannon legria hf r806 camcorder (Canon China Co., Ltd., Beijing, China) to film the trials, in which way the observer's gaze interference could be avoided. The starting distance was standardised at 25 m ± 0.5 m because starting distance is known to affect FID significantly in other birds (Goumas et al. 2020). Upon locating a magpie on the feeder in the camcorder lens, the observer instructed the experimenter to mark the starting point with a small bean bag (diameter: 3.5 cm) and walked forwards at a constant speed of 0.8 m/s. The instructions were conducted through Bluetooth headsets with no extra acoustic disturbance to the bird. In the same way, the experimenter was instructed to stop, locate the ending point and measure the distance when the focal bird fled from the feeder. The trial would be disregarded if there was any uncertainty about the distance or if the bird was disturbed by a passing automobile or other provoking conditions.
In Experiment 1, to investigate the effects of breeding state and age class on gaze sensitivity, we conducted gaze sensitivity trials on three types of Azure-winged magpies (adults in the nonbreeding season, adults in the breeding season and juveniles in the breeding season). Three human gaze treatments were set in gaze sensitivity trials: (1) "Eye toward and head toward" = the experimenter approached with face and eyes directed towards the focal bird on the feeder; (2) "Eye down and head toward" = the experimenter approached with face directed towards the focal bird but eyegaze directed towards the ground; (3) "Eye down and head down" = experimenter approached with head drop and eyes directed towards the ground in front of him. For each bird type, the experimenter measured the FID under three human gaze treatments with a direct approach to the feeder. Two age classes of Azure-winged magpies were distinguished based on plumage and the top of the head (Harada 1997). The targeted adults older than one year had a glossy black top to the head and brighter plumage. The targeted juveniles have fled the nest in this breeding season and have white-mottled heads and duller plumage. Each human gaze treatment on adults in the breeding season was replicated 15 times at each site and in the nonbreeding season. However, owing to the limited number of juveniles in each flock, each human gaze treatment on juveniles was replicated 26 times in all (i.e., less than six times on average in each site).
In Experiment 2, we used a 3 × 3 factorial design ("approach direction" × "human gaze") to examine the effects of approach direction on the gaze sensitivity of adult azure-winged magpies in the breeding season. Three treatments of approach direction were set, including a direct approach (bypass distance: 0 m) and two tangential approaches (bypass distance: 2.5 m and 5 m). The direct approach is the same as the approach in Experiment 1. The bypass distance was the closest feasible distance between the focal feeder and the linear path of the approaching experimenter (Fig. 1). Under these bypass distances (0 m, 2.5 m and 5 m), we conducted the same three different gaze treatments as explained in Experiment 1, including "Eye toward and head toward", "Eye down and head toward" and "Eye down and head down". The approach was always in a straight line, either direct or tangential, from a direction so that no trees would prevent the magpies from seeing the experimenter. Each human gaze treatment was replicated 15 times at each site when the experimenter approached directly (bypass distance of 0 m) and 10 times at each site under bypass distances of 2.5 m and 5 m. The focal bird and each treatment were selected randomly, so it could not be guaranteed that the same bird was tested for all three treatments.
In each trial, the initiation posture of the focal bird (i.e., front, back, profile) and the maximum number of Azurewinged magpies, which can be observed at a single time within the 5 m radius of the focal feeder, were recorded. Magpies on the feeding station with the body oriented toward the experimenter were recorded as "front". Magpies stood with their back towards the experimenter and were recorded as "back". Magpies stood with one side of the body oriented toward the experimenter and were recorded as "profile". As the humans and automobiles present may affect FID in previous studies (Goumas et al. 2020;Stankowich and Coss 2006), we recorded the passing humans and automobiles within an estimated 15 m radius of the focal feeder during each trial as well.
Although we controlled the proportion of replicate sets to the flock size and tracked the movement of magpies after testing at the same sites, the possibility of testing the same individual repeatedly could not be avoided. Therefore, the psuedoreplication still existed and the results should be treated with caution.

Statistical analyses
We used a generalised linear mixed model (GLMM) to evaluate the effect of bird type, human gaze direction and bypass distance on FID. We have the FID data log-transformed to meet the assumptions of the generalised linear mixed models, as the raw data did not follow a normal distribution, checked by the Kolmogorov-Smirnov one sample test. All tests were two-tailed unless specified otherwise. A Gaussian distribution was used for the log-transformed FID. Flocks were entered as a random factor to account for repeated measures. In Experiment 1, we included gaze direction (i.e., "Eye toward and head toward"; "Eye down and head toward"; "Eye down and head down"), bird type (i.e., adults in the nonbreeding season; adults in the breeding season; juveniles in the breeding season) and the interaction between gaze direction and bird type as fixed factors to test whether age and breeding state affected sensitivity to human gaze. The maximum number of conspecifics, passing humans and automobiles, as well as the focal bird's initial posture (i.e., front, back, profile), were also entered as additional fixed factors. In Experiment 2, we included gaze direction, bypass distance (i.e., 0 m; 2.5 m; 5 m) and the interaction term between human gaze direction and bypass distance as fixed Fig. 1 Arrangement of Experiment 2. The bypass distance (0 m, 2.5 m, 5 m) was the closest feasible distance between the focal feeder and the linear path of the approaching experimenter factors to test whether bypass distance affected sensitivity to human gaze. Other additional parameters were the same as in Experiment 1. The number of humans and cars was classified as 0 (absent) or 1 (present) since both variables were highly heteroscedastic with multiple zero counts. We reported the results of the full model, with all terms retained. Statistical analyses were performed using SPSS 26.0 (IBM Corp., Armonk, NY, U.S.A.) and Origin 9.0 (OriginLab Corp., Northampton, MA, U.S.A.) was used for plotting.

Experiment 1 human gaze direction and bird type
Human gaze direction (i.e., "Eye toward and head toward", n = 176; "Eye down and head toward", n = 176; "Eye down and head down", n = 176) and bird type (i.e., adults in the nonbreeding season, n = 225; adults in the breeding season, n = 225; juveniles in the breeding season, n = 78) had a significant effect on FID (Table 1). A two-by-two comparison test showed that Azure-winged magpies had shorter FIDs when the experimenter approached with "Eye down and head down" compared to "Eye down and head toward" and "Eye toward and head toward" (see supplementary material, Table S1). However, FID did not differ between "Eye toward and head toward" and "Eye down and head toward" treatments (see supplementary material, Table S1). Adults in the nonbreeding season had longer FIDs than adults in the breeding season and juveniles in the breeding season. However, there was no significant difference between adults and juveniles in the breeding season (see supplementary material, Table S1).
Some environmental factors such as humans present within a 15 m radius and the number of magpies within a 5 m radius during the trials significantly affected FID. Azure-winged magpies had shorter FIDs in the area where humans were present (Table 1). They could be approached more closely before fleeing with the increased number of magpies around the feeding station (Table 1). The initial posture of the focal bird also had a significant effect on FID. A two-by-two comparison test showed that the experimenter could approach magpies more closely when the initial posture of magpies was "back" and "profile" compared to the initial posture "front" (see supplementary material, Table S1).
A significant interaction between gaze direction and bird type was found for sensitivity to human gaze (Table 1). Adults were sensitive to human gaze, while juveniles could hardly distinguish human gaze direction (Fig. 2). A two-bytwo comparison test showed that adults in the breeding season could distinguish the experimenter head direction, with significantly shorter FIDs in "Eye down and head down" treatments compared to "Eye toward and head toward" and "Eye down and head toward" treatments ( Fig. 2, see supplementary material, Table S2). Nevertheless, adults in the breeding season did not distinguish the experimenter's eye direction ("Eye toward and head toward" vs "Eye down and head toward", Fig. 2, see supplementary material, Table S2). Adults in the nonbreeding season responded differently to human gaze direction, with a significant difference between "Eye toward and head toward" and "Eye down and head toward" instead ( Fig. 2, see supplementary material, Table S2). No significant difference was found in FIDs of juveniles in the breeding season under the three human gaze treatments (Fig. 2, see supplementary material, Table S2).

Experiment 2 human gaze direction and bypass distance
Human gaze direction (i.e., "Eye toward and head toward", n = 175; "Eye down and head toward", n = 175; "Eye down and head down", n = 175) had a significant effect on FID of adults Azure-winged magpie in the breeding season with different bypass distances (Table 2). Bypass distance (i.e., 0 m, n = 225; 2.5 m, n = 150; 5 m, n = 150) did not have a significant effect on FID ( Table 2). The paired comparison test between bypass distances was shown in Table S3 in supplementary material. FIDs were also not significantly affected by an interaction between gaze direction and bypass distance (Table 2). However, the paired comparison test showed that adult magpies could clearly distinguish both human head and eye direction at the 2.5 m bypass distance. FIDs were significantly greater in "Eye toward and head toward" treatments and shorter in "Eye down and head down" treatments ( Fig. 3, see supplementary material, Table S4). Compared to the 2.5 m bypass distance, adult magpies were less sensitive to human gaze at the 0 m bypass distance, with no difference in FIDs between "Eye toward and head toward" and "Eye down and head toward" treatments ( Fig. 3, see supplementary material,  Table S4). Meanwhile, they behaved even worse at the 5 m bypass distance, with only FIDs in "Eye down and head down" treatments significantly shorter than those in "Eye toward and head toward" treatments ( Fig. 3, see supplementary material, Table S4). The presence of cars had a significant effect on FID ( Table 2). The experimenter could approach adult Azure-winged magpies more closely in areas where cars were present compared to areas where cars were absent. The number of adult Azure-winged magpies nearby also slightly affected the FIDs (Table 2).

Discussion
The aversive response of animals to direct gaze known as gaze aversion has been widely reported in birds (Davidson and Clayton 2016). Reports of gaze sensitivity across different bird species have demonstrated differential sensitivity towards gaze cues such as eye and head direction (Carter et al. 2008;Hampton 1994). In addition, gaze sensitivity has been corroborated to be affected by additional predator and social information (Davidson and Clayton 2016).
In this study, we investigated how Azure-winged magpies responded to human gaze and the effects of age, breeding season, and approach direction. Whereas, the pseudoreplication could not be avoided, as individuals were not marked. Therefore, the results should be interpreted with caution, and a more carefully designed study is needed in the future. In the breeding season, we found that adult Azure-winged magpies were sensitive to human gaze, while juveniles showed no sensitivity to human gaze. This result implied that gaze aversion might not be present at birth. In addition, the sensitivity to human gaze differed under three bypass distances. Adult magpies in the breeding season could significantly recognise human head and eye direction at the 2.5 m bypass distance, compared to the 0 m and 5 m bypass distances. In all trials, we controlled the starting distance and initial Fig. 2 Box plots of the effect of human gaze ("Eye toward and head toward", "Eye down and head toward" and "Eye down and head down") and bird type ("A(b)"; "J(b)"; "A(n)") on the raw FID data of Azure-winged magpies. "A(b)": adults in the breeding season; "J(b)": juveniles in the breeding season; "A(n)": adults in the nonbreeding season. The box plots show the median and 25th and 75th percentiles; the whiskers indicate the values within 1.5 times the interquartile range, and the square within each box represents the mean value. **P < 0.01; ***P < 0.001 height of the focal bird from the ground, which have been shown to have significant effects on flight initiation distance in previous studies (Goumas et al. 2020;Stankowich and Coss 2006). In both experiments, FID was strongly affected by gaze direction, with direct gaze eliciting flight at a further distance. This result suggests that Azure-winged magpies perceive direct gaze as an indicator of greater risk and would increase their distance from humans to reduce the threat level. This aversion to human gaze has been reported not only in birds but also in ungulates and lizards (Sreekar and Quader 2013;Stankowich and Coss 2006). A predator gaze may provide accurate information for prey to decide when to flee (Davidson et al. 2014). Prey may eventually benefit from more foraging opportunities or more frequent nest visits by precisely perceiving a predator's direct gaze (Carter et al. 2008;Davidson et al. 2015).
Although bird type had a strong influence on FID in Experiment 1, the paired comparison revealed no significant differences in the FID of adult and juvenile Azurewinged magpies (see supplementary material, Table S1). The result implies that the ability to detect the risk of humans directly approaching is inherent, with adult and juvenile magpies showing a similar level of response. In contrast, the aversion to human gaze was suggested to be acquired posteriorly, according to the result of a significant interaction between gaze direction and bird type. Adults in the breeding season were able to respond differently to the head direction treatments, while juveniles were not sensitive to the difference between the three gaze direction treatments (Fig. 2, see supplementary material, Table S2). Young ravens (Corvus corax) started to follow the gaze of conspecifics at six days after fledging, coorienting with human gaze at eight weeks later, and gazing behind obstacles at six months of age (Schloegl et al. 2007). Therefore, sensitivity to gaze could be a developmental process which would appear later in the life of Azure-winged magpies either dependent or independent of experience. Gaze sensitivity could be learnt from extensive experience with neighbourhood or human activity. Such posteriority has been found in northern bobwhite (Colinus virginianus), of which the gaze aversion developed following early experience with human faces (Jaime et al. 2009). In the current study, Azure-winged magpies are frequently seen feeding around garbage cans or other human activity areas where food was dried. The procedure for driving the bird away involves a human looking directly at individual birds, and magpies may learn from the human gaze stimulus. In our study, we did not control the early experience and followed the future performance of certain juveniles. Therefore, whether exposure to and learning of human gaze at an early age is necessary for the development of Fig. 3 Box plots of the effect of human gaze ("Eye toward and head toward", "Eye down and head toward" and "Eye down and head down") on the raw FID data of adult Azure-winged magpies in the breeding season under three bypass distances (0 m, 2.5 m and 5 m). The box plots show the median and 25th and 75th percentiles; the whiskers indicate the values within 1.5 times the interquartile range, the square within each box represents the mean value. *P < 0.05; **P < 0.01; ***P < 0.001 gaze-aversion responses in Azure-winged magpies remains to be investigated.
Referring to optimal escape theory, cost factors affect FIDs (Ydenberg and Dill 1986). In Experiment 1, adults in the breeding season could be approached more closely than adults in the nonbreeding season (see supplementary material, Table S1), which implies that the alternation of the response of adult Azure-winged magpies to human approach may result from additional costs incurred by the breeding state. Rodgers et al. (1997) found that some waterbirds exhibited greater flushing distances during foraging or loafing compared to nesting. This is because cost of failing to protect the nest is greater than the cost of failing to defend a loafing site or temporary source of food (Rodgers and Smith 1997). Azure-winged magpies, as a cooperatively breeding species, may act as helpers and assist more than one nest during the breeding season even if they do not have their breeding (Valencia et al. 2003). Therefore, the Azure-winged magpies may also allow the predator to get closer to maximise the feed intake to cope with the large amount of energy needed in the breeding season.
There was a significant interaction between gaze direction and bird type, indicating that adults in the nonbreeding season, adults in the breeding season and juveniles show different sensitivities to human gaze (Fig. 2, see supplementary material, Table S2). Some previous gaze sensitivity empirical fieldwork measured anti-predator responses to either human eye-gaze direction or head orientation merely (Carter et al. 2008;Goumas et al. 2020;Hampton 1994). Different from previous fieldwork, we set three gaze treatments including "Eye toward and head toward", "Eye down and head toward" and "Eye down and head down" to investigate the different sensitivities to eye direction and head direction. We found that adult Azure-winged magpies in the breeding season are more attentive to head direction, while adults in the nonbreeding season are more attentive to eye direction. During the breeding season, many bird species would experience physiological changes (i.e., stress hormones) (Pdulka et al. 2004). Levels of stress hormones linked to individual personalities could influence the boldness of individuals and their response to the risk of the novel object (neophobia) (Baugh et al. 2013;Greggor et al. 2016). Rooks (Corvus frugilegus) were less risk averse in the breeding season, allowing novel people to approach more closely compared to the nonbreeding season (Greggor et al. 2016). Another study on black-legged kittiwake (Rissa tridactyla) showed that the birds may also become increasingly bold in the breeding season (Collins et al. 2019). Thus, animals may change their anti-predator tactics depending on whether they are breeding. Meanwhile, a study on roe deer (Capreolus capreolus) suggested that individuals would mainly be focussed on avoiding the higher risk when given two risks of different levels (Lone et al. 2014). Therefore, we supposed that the changes in hormones incurred by the breeding state might make Azure-winged magpies decrease vigilance to humans and only be attentive to head direction rather than subtle risk indicators, such as eye-gaze.
Some environmental factors, such as the presence of humans, the number of magpies and the bird's initial posture, had a significant impact on the FID of Azure-winged magpies. We discovered that Azure-winged magpies had shorter FIDs when humans were present, which is consistent with previous research on herring gulls (Larus argentatus) and hadeda ibises (Bostrychia hagedash) (Bateman and Fleming 2011;Goumas et al. 2020). In this situation, birds may become accustomed to human presence and permit closer approaches by humans (Blumstein 2016). Alternatively, the presence of other people might have served as a distraction from the approaching experimenter, impairing the Azure-winged magpies' capacity to assess risk. In this study, Azure-winged magpies could be approached more closely before fleeing if accompanied by more conspecifics. As suggested by Stankowich (2005), species using coordinated defence against predators would feel safer and could be approached more closely in larger groups. For example, minnows (Phoxinus phoxinus) have been reported to have smaller flee initiation in larger groups (Magurran et al. 1987). Azure-winged magpies, a cooperatively breeding corvid, exhibit proactive prosociality and frequently mob predators (Canário et al. 2004;Horn et al. 2016). When surrounded by conspecifics, Azure-winged magpies may feel safer and respond to humans more slowly. The counter intuitive finding is that the initial posture of the focal bird also has a significant effect on FID. Birds would flee at a longer distance when their initial posture is "front" compared to the posture "profile" and "back" (see supplementary material, Table S1). We supposed that this finding may be due to the visual fixation strategy. Four head movement-based fixation strategies, including monocular gaze-locking, monocular alternating, binocular alternating and binocular gaze-locking, have been proposed for birds previously (Butler et al. 2018). The visual fixation strategy that Azure-winged magpies used may help them gather information about predatory risk more quickly under the initial posture "front" compared to other postures.
In reality, prey commonly encounter multiple risks at the same time. When investigating the effect of approach direction on the gaze sensitivity of Azure-winged Magpies, we found that the approach direction did not affect the FID of Azurewinged Magpies, which differs from the typical assumption in studies of human disturbance that wildlife show more aversion to direct approaches. In a review of animal fear, the researchers suggested that the predator's intent to attack or bypass the focal animal is strongly indicated by the predator's direct path of approach (Stankowich and Blumstein 2005). Previous studies in Columbian black-tailed deer and Bonaire whiptail lizards have reported similar results (Cooper et al. 2003;Stankowich and Coss 2006). However, the study site was on campus, which is a part of urban habitat with large crowds of people, rather than wild habitat where humans are rarely present. A previous study that gathered FID data from 62 species of birds from urban and rural habitats found that birds in urban habitats have no significant difference in FID between direct and tangential approaches (Møller and Tryjanowski 2014). Therefore, to adapt to the unique environment and increase fitness, Azure-winged magpies may be habituated to frequent and dense human proximity to a large extent and would not distinguish the approach direction.
Although Azure-winged magpies on the campus may have adapted to frequent human approaches, they were still sensitive to directed attention in our study. The predicted interaction between the gaze direction and approach direction was not observed. Instead, we found that Azure-winged magpies in the breeding season showed different sensitivities to human gaze under three bypass distances (see supplementary material, Table S4), which was consistent with our original hypothesis. They were sensitive to both human head and eye direction at the 2.5 m bypass distance but failed to distinguish subtle eye direction at the 0 m bypass distance. It is possible that direct approaches make magpies less receptive to subtle cues because they focus their attention on other, more salient signs of risk, such as the direction and speed of approaching humans. As a result, when approached tangentially, the magpie could save its attention in perceiving the direction of human eyes and head because the tangential body would present less interference, as previously reported (Bateman and Fleming 2011;Sreekar and Quader 2013). However, tangential approaches could influence the probabilities of risk detection at the same time (Bulova 1994). Longer bypass distances would reduce the detectability of risk, so Azure-winged magpies may struggle to detect the human risk cue at a longer distance, as they demonstrated the lowest sensitivity to human gaze at a 5 m bypass distance in our experiment (Fig. 3). A similar result was previously reported in Bonaire whiptail lizards (Cnemidophorus murinus), which found that most lizards (46 of 48) did not pay attention to human risk at longer bypass distances (Cooper et al. 2003). A link between bypass distance and risk detectability can be found primarily in the visual fields (Fernández-Juricic et al. 2004). The magpie's capacity to perceive human gaze risk may be diminished because of the increased bypass distance, which would put the approaching human outside of its range of view.

Conclusion
Our study found that Azure-winged magpies were averse to direct human gaze and this gaze aversion was not present at birth. Meanwhile, the breeding state and approach direction may affect the gaze sensitivity of Azure-winged magpies. The breeding state may decrease the flight initiation distance of adult magpies, as predicted by the optimal escape theory, but it does not enhance their sensitivity to human gaze. In Experiment 2, although there was no interaction between the gaze direction and approach direction, the sensitivity to human gaze differed under three bypass distances. Adult magpies in the breeding season could discriminate human head and eye direction clearly at a certain bypass distance (2.5 m), which indicated that magpies would alter their anti-predator responses to fluctuant predation risk in an adaptive way. The changes in anti-predator responses may provide a deeper understanding of human-wildlife interactions. Thus, further studies on how Azure-winged magpies integrate multiple risks posed by humans into anti-predator decision making and the underlying mechanisms are needed.