Observations were carried out from August 2003 to April 2004, following four years of serious drought, in which reproduction was reduced from 180 nestlings and fledglings, ringed in 1997, to the low number of about 30 ringed in each of the years 2000, 2001 and 2002. As a result the population was old and very stable ( Zahavi, 1990), enabling us to select a large number of similar groups, each comprising two adult males (Alpha male, M ₁; beta male, M ₂) and one adult female (F ₁). Some of the groups included one or two independent youngsters fledged in 2003 ( Table 1). The males were not related to the females and thus competed for reproduction, except in one group (MTE), in which the beta male was the son of the breeding female. In another group (HNC), there was acute competition between father and son culminating in eviction of the father leaving a group of two. Observation of this group ceased following the eviction. Competition over breeding was also intense in the SAL group, in which M ₂ was observed to copulate with the female and on several occasions interrupted the copulation of the dominant pair by following them ( Perl, 1996).

All the birds in this study were colour-banded for individual recognition and tame to the extent that they readily accepted food from the hand of the observer. None of the groups inhabited a territory bordering directly on agriculture.

Data on group composition, their ages and relatedness among the males are presented in Table 1.

a. Guarding. When the group is searching for food, or occupied in some other activity, one individual will stop feeding and climb onto a high place (e.g. a tree), inspecting the neighbourhood. It is usually easy to distinguish the sentinel from an individual that is in the tree for another reason, such as feeding, resting, auto-preening, etc. The height and extent of exposure of the sentinel depends on environmental conditions: in the darkness of dawn or during danger, it perches on the edge of the canopy. With increasing light it takes a more exposed perch, climbing to the top of the canopy, or often onto a dry branch devoid of leaves high above the canopy, from where it can see and be seen for a great distance. Many groups have a preferred guarding perch in their territory, highly exposed above the canopy ( Figure 1). Sometimes the sentinel flies directly to its guard post, but more often it climbs the tree, scanning around, then climbs to a higher and more exposed position and sometimes flies off to another tree. There is usually only a single sentinel, but when a predator or an alien babbler is sighted the whole group may perch for a while on the top of the tree. At the end of the guarding bout the sentinel either flies down directly or moves slowly into the canopy. A new individual starts its guarding bout at variable intervals after the previous sentinel had left its post. However, about 15% of the guarding bouts are started by the replacement of an active sentinel see below.

In contrast to many other species ( Bell et al., 2010; Gaston, 1977), adult Arabian babbler sentinels do not emit regular calls, except for “alarm calls”, or shouts when interacting with distant babblers ( Sommer, 2011). Young sentinels, however, often produce soft sub-songs while they guard. These vocalizations attract attention to the youngster, advertising its activity to the group.

We do not know the proximate cause for a babbler to assume guarding and we do not discuss this issue here. However, we do know that the presence of foreign babblers in the territory increases guarding activity by the residents. The presence of an alien female in the territory greatly increases guarding by both the breeding female and the other females ( Ben-Mocha, 2013). Food availability too has a profound effect: on a rainy day little guarding was observed, but on the following day a maximum duration of almost 45 minutes/h was reached, probably due to the availability of a large number of insects that had drowned in the flooding ( Anava et al., 2002).

b. Replacements of sentinels. We observed several methods of sentinel replacement: it could be direct, with the replacer flying directly to the guarding bird, with or without some food item, and the previous sentinel then leaving its perch; or indirect, with the replacer taking up a guarding position on another perch. If the replacing bird was a subordinate, it often perched below the dominant or on another, usually lower, tree, and climbed to the final guarding position only after the dominant had vacated it. Eventually, mostly within one minute, one of the birds would abandon its position. In most cases, if the original sentinel did not leave its perch the second one retreated. The conflict between the two was often manifested by both of them nervously preening themselves.

c. Allofeeding. Adult babblers sometimes allofeed ( Kalishov et al., 2005). This often happens when the feeder seeks to replace the sentinel. We occasionally observed how a babbler that intended to replace another by allofeeding, first looked at the sentinel, and then started searching for a suitable food item. Upon finding one, it flew directly to the sentinel, fed it and usually replaced it. The sentinel either accepted the food (acceptance), sometimes crouching in a begging position (accepting like a fledgling), or rejected it. Finally it either did or did not leave its post. Often the sentinel, which may well be aware of the intention of the feeder - left its post before the feeder reached it. On rare occasions, when the sentinel did not accept the food item, the replacer (always a dominant) aggressively pushed the food into its beak. On several occasions a subordinate babbler that had just refused food offered by the dominant, immediately approached the observer to take a small crumb of bread, suggesting that despite being hungry it had refused to accept the "gift" from another babbler. Such interactions are not common but we have witnessed several over the years ( Zahavi & Zahavi, 1997).

d. Social phases. We have defined three social phases: a. non-breeding; b. courtship, beginning when sticks are collected and lasting until the last egg has been laid; and c. incubation and feeding the young until independence. Breeding cycles were often aborted for various reasons.

Every group was visited on average every 7–10 days. However, visits were not equally distributed – during mate guarding and egg laying a group could be visited daily until the first day of incubation, when frequency of visits usually dropped.

Observations started with first light, usually before the babblers had left their night-roost tree. For the first 2–3 hours they were followed without any interference by the observer. The data presented in this paper were collected during that period. Following these observations the babblers were offered some bread tidbits and water. This was done in order to induce allopreening, which was the main subject of the study ( Dattner, 2005). In one group (MTE) an entire day of observations without interference was conducted once a month. Of those days, only observations from the first three hours of the morning are included in the data presented here. We recorded the time of ascent and descent of the sentinel, its identity, the way by which it took up its perch, whether there was another sentinel at the time, whether it was replaced, the identities of the replacer and the replaced, type of replacement (direct or indirect or with allofeeding, acceptance or rejection of the food). The observations were written up on cards, noting the exact time and circumstances, and were later transferred to Excel spreadsheets. The social phase of the group in relation to the breeding cycle was also recorded. Table 2 presents the number of observation hours for each group, at the different social phases.

Weight was measured occasionally by other researchers who monitored the groups on alternate days, but not on our "observation days". The babblers were weighed in the morning as soon as they left the night roost. The birds were lured into mounting a scale (Moznei Shekel) by placing tiny tidbits of bread on it. The data presented in Table 3 are averages of several measurements taken over several months. A few individuals were not weighed because they either refused to mount the scale or were absent for some reason.

The data presented in this paper summarize a total of 637 h of observations in the mornings without interference by the researchers: 212 h in the non-breeding phase, 322 h in the courtship phase (179 h during nest building and 143 h during mate-guarding), and 103 h while incubating and feeding the young. When not otherwise noted, numbers are averages per hour of observation.

Statistical analysis: Paired t-test was used to compare the behaviour of the alpha male, the beta male and the female in the group (when needed, Bonferroni-type adjustments for multiple comparisons were used). The P values are for a two-tailed alternative. Effect size calculations were also carried out using effect size calculators ( http://www.polyu.edu.hk/mm/effectsizefaqs/calculator/calculator.html).

a. Daily activity. Foraging for food was the main activity in the morning. Later in the day, when the birds were partly satiated, other activities dominated, such as allopreening ( Dattner, 2005), play ( Pozis-Francois et al., 2004) and, mainly in summer, sleep. Activity in the afternoon was mixed. This pattern of activity is reflected in the number of guarding bouts during the day ( Supplementary material 1). In the first three hours the number of bouts was maximal, between 8–9 bouts per group per hour. Later, the number fell below 5, and averaged less than one bout per hour in the middle of the day (and down to zero in the hot hours of the summer). Sentinel activity was resumed together with the other activities in the afternoon, but did not reach the same level as in the mornings.

b. Guarding. The mean (±SE) duration in which a sentinel was present in the morning was 22.75±1.3 ( n = 10 groups) minutes per hour of observation, meaning that even in the morning, at the time of maximal sentinel activity, a sentinel was present only for about one-third of the time. In every group there was great daily variability in the amount of sentinel activity, ranging from a minimum of 6.5 minutes to a maximum of 50.5 minutes per hour of observation ( Supplementary material 2).

The alpha male guarded for longer duration (a) and more frequently (b) than the other two adults ( Figure 2). (a). The mean duration of guarding was different between M ₁ (the alpha male) and M ₂ (the beta male) and between M ₁ and F ₁ (the female) ( t-test for paired observations, t ₉ = 5.157 P = 0.0012 and t ₉ = 5.642 P = 0.0006, respectively – P values are given after Bonferroni's correction). In both cases, the mean values for M ₁ were larger than for M ₂ and for F ₁. (b). The mean frequency of guard bouts was different between M ₁ and M ₂ and between M ₁ and F ₁ ( t-test for paired observations, t ₉ = 4.583 P = 0.0026 and t ₉ = 6.239 P = 0.0003, respectively – P values are given after Bonferroni's correction). In both cases, the mean values for M ₁ were larger than for M ₂ and for F ₁. Effect sizes (Es) for the differences in the duration of guarding and for the number of bouts between M ₁ and M ₂ were both large (see Dataset, Figure 2). In 80% of the 241 observation days the alpha male guarded more than any other individual in the group. Only on 20% of the days did the beta male or the female guard somewhat more than the alpha. This often happened when the latter was busy incubating ( Figure 11). There were no consistent differences in the number of bouts or in the duration of guarding between a beta male and a female.

As soon as a subordinate male became dominant – following the eviction or disappearance of the alpha male – its guarding increased to the level of dominant ( Figure 3). (Before: 1.7211±1.1953 (mean±SD) After: 3.8738±1.1465. After minus Before: t-test for paired observations, t ₉ = 9.378 P = 0.0026). One could argue that the former M ₂ was now guarding against the former M ₁, which might still be present at the border of the territory. However, in group BOT1 the eviction of the M ₁ occurred early on in the observation period (with the group thus becoming BOT2), and the former M ₂ maintained a high rate of guarding until it was itself evicted. All other cases of eviction occurred towards the very end or after the end of the present study.

c. Relationship of guarding to body mass is presented in Figure 4. Within each category (M ₁, M ₂ and F ₁) there was no significant correlation between the babbler's weight and its guarding effort (Linear regressions: R = –0.310 for M ₁, R = 0.089 for M ₂, R = –0.116 for F ₁ – all are non-significant).

In five of the seven groups for which the weight of both males was known M ₂ was heavier than M ₁ ( Table 3) and in all of them M ₁ guarded much more than M ₂ (see Dataset, Figure 4). As soon as an M ₂ became dominant, its guarding increased to match that of the previous dominant ( Figure 3), without a corresponding change in its weight.

d. Replacements of the sentinel. Replacements of a sentinel occurred on average about once per hour of observation. Thirty-percent of the replacements (32%±1.57, n = 14) occurred within 0–1 minutes after the original sentinel had started its guarding bout. When not interrupted, only about 10% (±1.24, n = 14) of the bouts lasted less than 1 minute. There was a significant difference between interrupted and uninterrupted bouts ( P < 0.01, W = 0, n = 14, Es: large.) ( Supplementary material 3).

The number of replacements/h as well as the percentage of guarding bouts starting with the replacement of an active sentinel is presented in Figure 5. Mean frequency of replacements by M ₁ is significantly larger than by M ₂ or by F ₁ ( t-test for paired observations with Bonferroni's correction, t ₇ = 4.829 P = 0.0038 and t ₇ = 7.657 P = 0.0002, respectively, Es: large), whereas mean frequency of replacements by M ₂ and by F ₁ are not significantly different.

In about 20% of its guarding bouts M ₁ started its session by replacing an active sentinel. M ₂ did so in 11% and an F ₁ in 18% of their guarding bouts respectively ( Figure 5b.) (Es: large).

Figure 6 and Figure 7 show which individual replaced which other individual. Figure 6 presents the total number of replacements and number of direct and indirect replacements of M ₂ or F ₁ by M ₁ and of M ₁ or F ₁ by M ₂. In over 50% of encounters M ₁ replaced M ₂ directly. M ₂ never replaced M ₁ directly but they did replace the females directly (see Dataset, Figure 6). Total: M ₁/M ₂ more than M ₂/M ₁ ( t-test for paired observations, t ₆ = 7.994 P = 0.0002.). If we break the Total into its components (Direct and Indirect), we still have M ₁/M ₂ significantly more than M ₂/M ₁ for each component. (Direct: t ₆ = 5.943 P = 0.0010 Indirect: t ₆ = 8.357 P = 0.0002)

In Figure 7 replacements of the young are also shown. In this figure replacements are presented as a percentage of the guarding bouts of the babbler being replaced. This was done to compensate for the low number of bouts by the young. Only groups that included young birds are presented in this figure.

The figure clearly reveals that M ₁ replaced the sentinels much more often than M ₂. M ₁ also replaced its adult partners as sentinels much more often than it replaced the young inexperienced birds ( p < 0.05, W = 0, n = 5, Es: large). M ₂ rarely replaced M ₁ but it did replace females and young.

e. Allofeeding. ( Figure 8, Table 4). On average, allofeedings during replacements were observed about once per observation session. Data in Figure 8 clearly show that M ₁ replaced M ₂ with allofeedings significantly more than it replaced and allofed the young ( t-test for paired observations, t ₄ = 9.418 P = 0.0007, Es: large), although the young were surely more in need of it. It is also apparent that the youngsters allofed each other more than they received from the adults. We never observed an M ₂ allofeeding an M ₁, although they do, on rare occasions, as was observed in other studies (unpublished report). M ₂ allofed both the females ( Table 4) and the young ( Figure 8). M ₂, and to a lesser extent the females, sometimes received the food in a crouching position, like fledglings. However, the food was also sometimes refused ( Table 4).

f. Social phase. The breeding phase had a profound effect on all aspects of guarding, especially on that of the alpha male ( Figure 9– Figure 11). The number of guarding bouts by M ₁ increased significantly during the courtship phase but dropped back as soon as incubation started. Frequency of guarding bouts by M ₁ during Breeding is significantly larger than during Non-breeding: t ₅ = 4.413 P = 0.0139. Breeding vs. Incubation: t ₅ = 2.088 P = 0.1823, which is not significantly different (Bonferroni's correction has been applied). (Es: large). In some of the groups the M ₂ also slightly increased its guarding bouts ( Figure 9) in the courtship phase but it did not drop during incubation and feeding of the young.

The number of replacements of M ₂ by M ₁ as well as the number of allofeedings of M ₂ by M ₁ also increased significantly during mate-guarding and then declined at the start of incubation: Replacements/h: Breeding vs. Non-breeding: t ₅ = 5.945 P = 0.0038, Breeding vs. Incubation t ₅ = 1.579 P = 0.3504.

Allofeedings/h: t ₅ = 5.458 P = 0.0056 Breeding vs. Non-breeding: Breeding vs. Incubation: t ₅ = 4.670 P = 0.0110.

During the first days of incubation the breeding-pair often monopolized incubation. On those days the guarding by M ₁ declined and those of M ₂ increased significantly in all groups ( P < 0.05 W = 0; n = 5 Figure 11).

M ₁ Mate Guarding vs. M ₁ Incubation: t ₄ = 3.740 P = 0.0403

M ₂ Mate Guarding vs. M ₂ Incubation: t ₄ = –6.043 P = 0.0076 (Please note a reversed direction).

M ₁ Mate Guarding vs. M ₂ Mate Guarding: t ₄ = 3.054 P = 0.0757

M ₁ Incubation vs. M ₂ Incubation: t ₄ = 3.848 P = 0.0367

(Bonferroni's correction has been applied). Note that the first, the second and the fourth comparisons (but not the third) are significant.

Different types of replacements: indirect, direct and with allofeedings can be visualized in this video clip: https://youtu.be/H_EdXAbRu2g