Social bonds and rank acquisition in raven nonbreeder aggregations

Abstract

Complex social life has been characterized as cognitively challenging and recently, social relationships such as long-term social bonds and alliances have been identified as key elements for brain evolution. Whereas good evidence is available to support the link between social relations and cognition in mammals, it remains unsatisfying for birds. Here we investigated the role of avian social bonds in a nonbreeder aggregation of ravens, Corvus corax, in the Austrian Alps. We individually marked 138 wild ravens, representing approximately half of a population that uses the area of a local zoo for foraging. For 2 years, we observed the dynamics of group composition and the birds' agonistic and affiliative interactions. We identified two levels of organization: the formation of an unrelated local group and the individuals' engagement in social bonds of different length and reciprocity pattern. Whereas belonging to the local group had no significant effect on conflicts won during foraging, the individual bonding type did. Birds that engaged in affiliative relationships were more successful when competing for food than those without such bonds. Bonded birds did suffer from aggression by other bonded birds and, probably as a consequence, most of the ravens' social relations were not stable over time. These results support the idea that social bonding and selective cooperation and competition are prominent features in nonbreeding ravens. Proximately, bonding may qualify as a social manoeuvre that facilitates access to resources; ultimately it might function to assess the quality of a partner in these long-term monogamous birds.

Highlights

► Big brains correlate with the formation of strong nonsexual bonds in mammals. ► We examine formation and use of bonds in birds, namely wild nonbreeder ravens. ► Individuals engage in different social bonds and benefit from them in conflicts. ► Active investment in nonsexual strong bonds for rank acquisition in a big-brained bird.

Social living has evolved many times in the animal kingdom and reached all imaginable forms from huge anonymous aggregations to small cohesive groups structured by differentiated relationships. Differences in social structure emerge from ultimate and proximate factors, such as underlying ecological pressures and physiological properties such as morphology and hormonal constitution (Whitehead 1997; Goodson et al. 2012). The social structure sets the platform upon which the members of a population meet, and hence determines the degree to which they need to interact and communicate. It has a strong potential to drive the evolution of sociocognitive skills (Byrne & Whiten 1988; Dunbar 1998) and more complex forms of sociality should be accompanied by sophisticated social cognition. This is well documented in mammals (Dunbar & Bever 1998; Connor et al. 1999; Kudo & Dunbar 2001; McComb et al. 2001; Byrne & Corp 2004), but leaves many open questions in birds (Beauchamp & Fernandez-Juiric 2004; Iwaniuk & Arnold 2004; Emery et al. 2007).

A critical point concerns the indices used to measure social complexity. Mammal social complexity initially was based on group size: primates living in larger social groups have larger brains, supposedly because they have to deal with an increased number of dyadic relationships compared to species living in smaller groups (Dunbar 1992, 1998; Barton 1996). In birds and some other mammals, group size does not consistently correlate with social complexity: aggregations might be large without any social relevance for the individual, as in anonymous herds (Pérez-Barbería et al. 2007), or conversely, affiliative interactions may extend to members of other groups with overlapping ranges, as in dispersed societies (Smuts et al. 1987; Byrne & Bates 2007; Randic et al. 2012). Other indices for social complexity seem to be rather primate-centric and not applicable to a lot of other taxa, for example frequency of tool use and tactical deception (Reader & Laland 2002; Byrne & Corp 2004) or grooming clique size (Kudo & Dunbar 2001). Individual network size of strongly bonded relationships, in contrast, turns out to be, for many species, a good proxy for social complexity (Kudo & Dunbar 2001; Dunbar & Shultz 2007); as a consequence, the ability to build social bonds is in the process of becoming a key concept for social complexity. In vertebrates pair bonding probably initiated a qualitative shift from loose aggregations of individuals to complex negotiated relationships which are generalized to all social partners in only a few taxa, for example in anthropoid primates (Shultz & Dunbar 2007).

Detailed knowledge of a species' social structure is needed to classify them correctly according to their social complexity. So far, however, we are lacking data on the relevance and gradation of social relationships in many taxa, particularly in birds. On the one hand, the high mobility of birds and seasonal variability of social structure in a lot of them complicate an otherwise easy comparison to the mammalian systems (Seed et al. 2009). Critically, social monogamy provides the basic hormonal frame of bonding in mammals (Broad et al. 2006). Because of egg incubation and the lack of lactation, it is a very common social strategy in birds, raising the possibility of totally different bonding patterns in birds and mammals. Relative brain size (measured by a regression of brain volume against body weight), on the other hand, shows a correlation with social structure in birds, with those aggregating in well-arranged groups of 5–30 individuals and those with a long-term social monogamous or cooperative mating system having the largest brains (Emery et al. 2007). Brain structures in birds evolved in convergence with those in mammals, indicating specializations in social problem solving in some bird families (Burish et al. 2004). Furthermore, corvids, one of the biggest brained bird families, display bonding patterns that have no immediate reproductive relevance, but rather social ones: rooks, Corvus frugilegus (Emery et al. 2007), jackdaws, Corvus monedula (de Kort et al. 2006) and ravens (Lorenz 1931; Gwinner 1964; Heinrich 1999; Fraser & Bugnyar 2010a) form strong bonds with group members of both sexes while still sexually immature. Ravens remember those bonds for years, even after becoming territorial breeders (Boeckle & Bugnyar 2012). Similar to those in primates (Fraser et al. 2008), raven bonds can be characterized as valuable relationships, whereby male–male and male–female relations tend to be more compatible and secure than female–female relations (Fraser & Bugnyar 2010a). Bonded birds are likely to reconcile conflicts (Fraser & Bugnyar 2011), actively support one another in conflicts with others (Gwinner 1964; Fraser & Bugnyar 2012) and console one another after severe conflicts with others (Fraser & Bugnyar 2010b). Social bonds thus seem to be critical in achieving and maintaining (high) dominance rank (Gwinner 1964).

All these studies, however, have been conducted on captive ravens, whereas in the wild, they are thought to assemble in anonymous aggregations in which they have no affiliations until they are sexually mature and mate for life. From then on they defend large areas (much more than 10 km²) year round (Heinrich 1989; Rösner & Selva 2005) and, owing to their long life span of up to 30 years, often for decades (Haffer 1993). Independent offspring are not tolerated in the parental territory and join nonbreeder aggregations for foraging and roosting. Throughout the Holarctic region, ravens are scavengers that compete with potentially dangerous predators for their prey (Heinrich 1989; Stahler et al. 2002). They need to cooperate with conspecifics to overcome predators or even the dominance of a pair of territorial breeders (Marzluff & Heinrich 1991). Yet, to date, there is hardly any evidence for social bonding within nonbreeder groups of wild ravens. Marked birds disperse randomly (Huber 1991; Heinrich et al. 1994; Marzluff et al. 1996) and genetic relatedness within foraging groups is not higher than between them (Parker et al. 1994), which would be expected if siblings joined the nonbreeders together.

Here we made a new attempt at investigating social bonds in wild ravens. Unlike other studies (Huber 1991; Heinrich et al. 1994), we focused on one study site at which nonbreeders assemble on a daily basis; in addition, we significantly increased the percentage of marked birds (50% of the ravens were marked in our study, compared to 10–20% in previous studies) as well as observational effort. We examined the social dynamics of raven groups, notably when social bonds were formed, how long they lasted and what they were used for. Assuming that the birds' sophisticated sociocognitive patterns shown in captivity are of relevance under field conditions, we hypothesized that (1) social bonds should play a role in the birds' acquisition and maintenance of dominance rank, (2) social bonds should form independent of immediate reproductive reasons in several age classes, but (3) higher ranking ravens could intervene in the formation of bonds and complicate its maintenance. In contrast, if social bonding follows reproductive reasons in an anonymous setting, then (1) they should gradually change with maturation and (2) once formed, adult bonds should hold until a territory becomes available.

Methods

Field Site and Study Animals

The study was conducted on a wild population of common ravens roaming in the Northern Austrian Alps and regularly using the Cumberland Wildpark, a small local zoo in the inner valley of the river Alm, 6 km from the village of Grünau. Groups of ravens visit the park year round, mainly for foraging, but also for playing and socializing (Drack & Kotrschal 1995). They consist mostly of nonbreeders (i.e. sexually immature birds and sexually mature but not reproductively active birds) that roost communally in a side valley adjacent to the Wildpark. Over the day, nonbreeding ravens disperse over a relatively restricted area of the valley, which contains four enclosures in the park (Przewalski's horse, Equus ferus przewalski, wolf/bear Canis lupus/Ursus arctos, wild boar, Sus scrofa, and red/fallow deer, Cervus elaphus/Dama dama) and a cliff right outside the park's border (but observable from within the park; Fig. 1). The rest of the inner part of the valley (approximately 100 km², including half of the park) is occupied by seven territorial breeding pairs, which aggressively defend their territories from being used by nonbreeders but frequently join the groups in the park for foraging (Drack 1994). All ravens are well habituated to the presence of human observers. In general, they can be studied from a close distance (<25 m) with the help of binoculars and video cameras. For the present study, the main observer (A.B.) invested 3 months to ensure that the ravens were fully habituated to her person, allowing her to walk around and move between enclosures and observation sites without causing any disturbance or flight responses.

Figure 1.

Sketch of study site in the inner valley of the River Alm, Austrian Alps. White areas a–d represent our observation sites in the Cumberland Wildpark (a: Przewalski's horse; b: wolf/bear; c: wild boar; d: red/fallow deer) and area e represents the cliff along the park's border; ‘X’ marks the exact observation point in each of the areas. Grey areas (I–IV) symbolize the breeding territories surrounding the study site. Dark grey area symbolizes the River Alm as a natural border of the Wildpark.

All experiments and bird manipulations comply with the current laws of Austria, and were authorized by the Central Administration of Upper Austria.

Trapping and Marking

During the study period we caught a total of 120 wild ravens with drop-in traps (Stiehl 1978) with a mean rate of 6 ± 6 ravens per month. Traps (3 × 3 × 3 m) were located in the Wildpark; they were equipped with perches and positioned in shady areas next to the enclosures the ravens use for foraging. Traps were regularly checked during the day so that the time ravens spent in captivity was reduced to a minimum (<2 h). Trapped birds were photographed, weighed and measured. Age class was determined by mouth and feather colour and assigned to juvenile (within their first year) if the mouth linings were pink and the plumage colour brown, subadult (second and third year), if the mouth linings were pink with black spots and plumage was black, and adult (from third year on), if mouth linings and plumage were black (Heinrich & Marzluff 1994). To analyse sex and genetic relatedness we took 50–200 μl of blood from the alar vein. Birds were marked with a patagial wing tag (Caffrey 2000) and an individual combination of colour rings and a numbered metal ring from Vogelwarte Radolfzell on the legs. Additionally, we released 15 captive-bred but raven-raised juvenile birds (seven females, eight males; marked in the same way as the wild-caught birds) during the course of the study period. The release of captive-bred birds followed the procedure of reintroduction programmes for ravens in Germany (Koch et al. 1986) and always resulted in the integration of the released birds into the nonbreeder flock within a month. Three wild birds were already marked with colour rings from a previous study, in which the scrounging tactics and movement coordination of wild ravens were observed (Bugnyar & Kotrschal 2001, 2002b). Altogether 138 ravens (73 females, 65 males) could be individually identified over the course of the study.

Data Collection

Over 2 years, from August 2008 to July 2010, the ravens were observed on a total of 392 days during the morning feedings of the wild boars and the bears/wolves. Presence at feedings served as a reference for flock composition; a marked bird was observed in the valley without being present in the morning feeds only twice. Feedings took place between 0700 and 0900 hours, during which data were collected for 30 min at the wild boar enclosure and 15 min at the bear/wolf enclosure. These observations in the morning were supplemented by 183 observation rounds during the day, which were conducted between 0900 and 1100 hours (morning), 1100 and 1300 hours (mid-day) or 1600 and 1800 hours (evening); a round consisted of five 15 min protocols taken from fixed positions that provided a good overview of the five enclosures regularly used by the ravens (Fig. 1). In all protocols, we used a combination of scan and behavioural sampling (Martin & Bateson 1986). The number of ravens present and the identity of marked birds were recorded every 5 min. Additionally, all agonistic and affiliative interactions (Table 1) involving a marked bird were recorded on an ad libitum basis, which means that all interactions were recorded unless more than two were occurring at once (in these rare cases, the focus was on those interactions occurring first). Affiliative interactions were counted only once per day to avoid pseudoreplication.

Table 1.

Definition of behavioural parameters

Behavioural category	Behaviour	Description
Agonistic interactions	Approach–retreat	Approached bird retreats without contact
	Forced retreat	Approached bird retreats after being threatened
	Fight	Contact aggression involving both individuals hitting each other
	Chase flight	One individual flies in pursuit of another trying to grab it; no playful vocalizations
	Show off	Acoustic and visual display of dominance, without contact
	Subordination	Defensive acoustic and visual display without contact
Affiliative interactions	Contact sitting	Sitting within one body's length of a partner
	Allopreening	Preening the plumage of a partner
	Inviting preening	Bowing towards partner, with fluffed head feathers and/or movements of inner eyelids
	Food/object offering/sharing	Approaching with an object held towards the partner and passing it over or manipulating object jointly
	Displaying	Female and male courtship displays and joint show off against others

A total of 37 marked ravens (28% of the total number of marked birds) were observed on more than two-thirds of the days sampled (all data corrected for the time span when the birds were marked), indicating that these birds were using the valley on a regular basis; these birds are called locals hereafter (Table 2). Twenty-two ravens (17% of the total number of marked birds) were present on approximately half of the days sampled and are called frequent visitors hereafter. Finally, 73 ravens (55% of the total number of marked birds) were present on fewer than one-third of the days sampled and are called infrequent visitors hereafter. In Table 2 we list the number of individuals contributing to the data sets of either affiliative or agonistic interactions, separated according to their presence in the valley.

Table 2.

Number of individuals contributing to the data for the total of birds present over the time of data collection, the observed affiliations or the conflicts

	Birds present		Affiliative interactions		Agonistic interactions
	Males	Females	Males	Females	Males	Females
Local ravens	22	15	20	15	17	13
Frequent visitors	9	13	8	8	8	9
Infrequent visitors	30	43	7	8	14	10
Total number	61	71	35	31	39	32

The data are divided according to the residency of the birds.

Observations were carried out using binoculars and a digital voice recorder. To ensure that the observations were not affected by differences in habituation of the birds, we regularly checked that marked birds that were observed feeding in the morning could be found during the day as well.

Data Processing and Statistical Analysis

To account for possible seasonal differences (Gwinner 2003), we divided a year into three phases, each lasting 4 months: Phase 1 (from July to October) reflects the time when juvenile ravens leave their families and integrate into the nonbreeder groups; Phase 2 (from November to February) reflects the time with high group densities and high levels of recruitment to ephemeral food such as carcasses; Phase 3 (from March to June) reflects the breeding season, with territorial ravens exerting pressure on nonbreeders by strongly defending food sources and displaying their superiority in resource-holding potential.

Each bird was classified in one of four bonding classes for each phase: unbonded: the bird was never seen in any affiliative interaction with another bird; casually bonded: the bird was observed in affiliative interactions with others; however, interactions were not directed towards the same individuals for longer than 1 month, or if longer, were always initiated by the same individual; closely bonded: the bird was observed in friendly interactions with a specific partner over more than 1 month, and interactions were initiated by both partners; territorial: the bird was paired, defending a territory and was not observed in the game park except at feeding times.

We used a generalized linear mixed model (GLMM) to test which factors influenced the outcome of agonistic encounters, with the outcome of an encounter (won/lost) as response variable, identity of the opponents as random factors and the age and bonding class for the specific phase, the sex and residency of the birds as fixed factors. Furthermore, we performed two linear mixed models (LMM) to test (1) the effect of age and sex on the propensity of individuals to switch bonding classes or partners, using number of switches as response variable, sex and age class as fixed factors and identity of subject as a random factor; (2) to reveal the effects of bonding classes and sex on the likelihood of winning within-sex agonistic interactions, using percentage won interactions as response variable, bonding class and sex of the individuals as fixed factors, and identity of the individual as a random factor. For all mixed models, we used a step-up strategy whereby fixed factors were added to the model sequentially. Akaike's information criterion (AIC) values were used to assess all possible candidate models and if the difference in AIC values between models was less than two, we averaged the best models (Pinheiro & Bates 2000; Burnham & Anderson 2004). Only the effects of fixed factors in the best models are presented.

Mann–Whitney U tests were applied to compare the percentage of marked ravens within the flock at the beginning and the end of the study and the distribution of bonding classes within the group of local birds with the distribution found within the group of visitors. For the latter, we normalized the data by converting them into percentages. A Spearman correlation was used to test for a relationship between the length of time an individual was marked and the length of time that individual was recorded present in the valley. The distribution of bonding classes over the phases was compared using a Kruskal–Wallis test, and the competitive ability of individuals when changing bonding status was compared with a paired t test. All analyses were conducted in SPSS version 19 (SPSS Inc., Chicago, IL, U.S.A.). Data conformed to normality whenever parametric tests were used. All tests were two tailed and α was set to 0.05.

Results

Residency and Feeding Flock Composition

A mean ± SD number of 40 ± 18 ravens were present in the valley per day. Out of those, 20 ± 9 ravens were individually marked; the percentage of tagged birds slightly increased during the study period, from 42 ± 17% in the first half year to 59 ± 19% in the last half year of the study (U = −6.03, N₁ = N₂ = 73, P < 0.01). Thus, from the total number of 138 marked birds, only a small proportion could be observed in the valley per day. The majority of birds showed up in the valley rather infrequently (see Table 2). There was no correlation between the time a bird was marked and the time it was present in the valley (r_S = −0.14, N = 132, P = 0.11).

Foraging groups of ravens were composed of all age classes (Fig. 2). As expected, sexually immature birds were in the majority (marked juveniles within their first year of life: 25 ± 2% = 4 ± 3 individuals; subadults in their second and third year: 54 ± 7% = 8 ± 4 individuals); however, also 21 ± 5% = 5 ± 3 sexually mature birds were present, out of which 3 ± 2% = 1 ± 1 birds could be classified as local territory-holding breeders. Groups were not cohesive units, that is, birds were coming and going during the feedings of the zoo animals, which lasted 10–40 min/enclosure. On average, marked ravens were recorded as ‘present’ at a feeding site in 59 ± 17% of the 5 min scans taken during that feeding. The rest of the time, birds were scattered over the surrounding area, carrying off food to cache or eat in private.

Figure 2.

Composition of raven foraging groups according to age class and breeding status. Box plots represent median, 25% and 75% quartiles, whiskers indicate 10% and 90% range, and circles represent outliers. *P < 0.05.

Affiliation Patterns and Structure of Social Bonds

During the course of the study, 472 affiliative interactions involving marked birds (N = 68, of which 62 were nonbreeders and six were territorial birds) were recorded. These comprised the following behaviours: ‘contact sitting’ (28% of cases), ‘allopreening’ (27% of cases), ‘inviting allopreening’ (23% of cases), ‘joint object play’ (8% of cases) and ‘pair displays’ (14% of cases). Affiliative interactions could occur between the same individuals over an extended time period (58% of cases, involving N = 32 marked ravens) or involve several individuals over shorter time periods (42% of cases, involving N = 60 marked ravens). The resulting relationships were characterized as closely bonded (same individuals, extended period) and casually bonded (short periods, see Methods). Significantly more birds were either unbonded or closely bonded than casually bonded (H₂ = 7.64, P = 0.02; Fig. 3). The distribution of bonding classes was similar in the subgroup of local birds (N _unbonded = 9, N _{casually bonded} = 10, N _{closely bonded} = 14, N _territorial = 3) and in the subgroup of regularly visiting ravens (N _unbonded = 7, N _{casually bonded} = 4, N _{closely bonded} = 7, N _territorial = 1; U = −0.29, N₁ = N₂ = 4, P = 0.77). Infrequent visitors showed up in the valley too rarely to determine their bonding status.

Figure 3.

Number of marked birds belonging to the four bonding classes. Box plots represent median, 25% and 75% quartiles, whiskers indicate 10% and 90% range, and circles represent outliers. Box plots with the same capital letter at the bottom are not significantly different. Territorial ravens were not included in the statistics, as they do not belong to the nonbreeders. Arrows and sample sizes refer to total number of switches that occurred between the classes.

With the exception of territorial breeders (which were considered a special case of closely bonded birds), most ravens did not maintain a given affiliation constellation over the entire study period. In fact, 48 marked ravens (71% of ravens in affiliations) switched partners in the course of the study and/or shifted between bonding classes between the three phases of a year and/or between years (see Fig. 3 for visualization). There was no effect of sex on the propensity to switch either partners (F _33.36 > 0.01, P = 0.96) or bonding class (F _34.69 > 0.01, P = 0.94). However, there was an age effect: subadult ravens switched partners significantly more often (mean ± SE = 1 ± 0.17 switches) than both juveniles (0.03 ± 0.21 switches; all post hoc analyses of pairwise comparisons are Bonferroni corrected: df = 42.78, P < 0.01) and adults (0.35 ± 0.25 switches; df = 65.37, P = 0.09). This also holds if we only compare yearling ravens (within their second year; 0.99 ± 0.21 switches) with juveniles (within their first year; 0.03 ± 0.21 switches; df = 32.87, P = 0.01) to control for the longer time ravens are classified as subadults as compared to juveniles. The propensity to switch bonding classes showed similar age effects with yearling ravens switching classes significantly more often (1.27 ± 0.20 switches) than juveniles (0.40 ± 0.20 switches; df = 33.33, P = 0.01), and no difference between the other age classes, either subadult (1.17 ± 0.17 switches) and adult ravens (0.71 ± 0.24 switches; df = 65.15, P = 0.35), or juveniles (0.38 ± 0.20 switches) and adults (df = 65.94, P = 0.89). On average, nonbreeding ravens of all age classes (N = 62) interacted with at least 3 ± 2 affiliation partners in the course of the 2-year period. This number is relatively conservative since unmarked ravens involved in affiliative interactions with marked birds were counted as one bird.

We had individual information for both partners for 60 affiliation constellations. In most of these constellations (78%) one partner was older than the other (66% older females, 34% older males) and in 92% partners were of opposite sex: 8% concerned male–male combinations. All male–male combinations (N = 5) were seen only once, and affiliative interactions between females were never observed.

Agonistic Interaction Patterns and Dominance Hierarchy Structure

At the feeding sites, a total of 1513 conflicts between marked competitors (N = 103) were recorded. Of these, 19% were directly about food; however, the majority (81%) were not food related and involved approach–retreat interactions of low and high intensity (17% retreats; 37% forced retreats), subordination displays (10%) or dominance displays (5%), fights (4%) and unresolved conflicts (8%). The outcome of an agonistic encounter was significantly influenced by the relative sex (F ₄ = 8.71, P < 0.01), age (F ₄ = 11.03, P < 0.01) and bonding class (F ₁₄ = 2.47, P < 0.01), but not the residency of both combatants (F ₃ = 0.15, P = 0.93). In general, males dominated females, older birds dominated younger ones, and birds with affiliation partners dominated singletons.

Stable individualized dominance hierarchies are characterized by dyadic resolved relations, in which the same individual from a dyad keeps winning the conflicts (unidirectional dyad). Only 35 of the marked dyads had more than the six agonistic interactions that are necessary to calculate the dyadic asymmetry. With fewer interactions and a binomial probability of P = 0.5 it is not possible to reject the null hypothesis, which is that the dyad is symmetric. All 35 dyads were unidirectional, with conflicts being won in 97 ± 6% of the cases by the same individual. Of the three dyads for which age, sex and bonding class were matched (as these affect the outcome of an encounter, see above), all were fully unidirectional (i.e. conflicts were always won by the same individual).

Benefits and Limitations to Bonding

Combining information on the ravens' affiliation constellation and behaviour in conflicts allows us to structure foraging groups hierarchically and separately for each of the sexes (Fig. 4). An LMM revealed significant effects of bonding class (F _3.428 = 21.57, P < 0.01) and the interaction between sex and bonding class (F _3.428 = 7.64, P < 0.01) on the percentage of within-sex agonistic interactions won. Splitting the data according to the sex still gave significant effects of the bonding classes on the competitive ability of the individuals in both females (F _3.102 = 8.00, P < 0.01) and males (F _3.340 = 28.08, P < 0.01; Fig. 4). In females, unbonded birds had lower competitive ability compared to the other bonding classes [unbonded (N = 9)–casually bonded (N = 6): 29 ± 7% versus 62 ± 8% won conflicts: P = 0.02; unbonded–closely bonded (N = 11): 29 ± 7% versus 58 ± 6% won conflicts: P = 0.02; unbonded–territorial birds (N = 1): 29 ± 7% versus 100 ± 0%: P = 0.05], whereas all other bonding classes had the same competitive ability (all P > 0.05). In males, on the other hand, unbonded birds (N = 17) did not differ from casually bonded (N = 6) birds (31 ± 4% versus 15 ± 6% won conflicts: P = 0.24), but had significantly lower competitive ability than closely bonded or territorial birds [unbonded–closely bonded (N = 12): 31 ± 4% versus 63 ± 4%: P < 0.01; unbonded–territorial (N = 3): 31 ± 4% versus 97 ± 9%; P < 0.01; casually bonded–closely bonded: P < 0.01; casually bonded–territorial: P < 0.01; closely bonded–territorial: P = 0.01; all Bonferroni corrected]. Thus, females appeared to benefit in conflicts from any type of affiliation, whereas males only benefited when they were closely bonded. It is worth mentioning that the affiliation partners did not actively help or were not near the target of aggression (<2 m) in most of these conflicts. It is thus likely that the mere presence of affiliated/bonded individuals at the foraging site is enough for the effect observed. In those cases in which the bonding status of birds changed from closely to casually bonded or unbonded, respectively (N = 17), the competitive ability dropped significantly (t test: t₁₆ = 2.92, P = 0.01). Hence, the observed differences in competitiveness were mainly caused by bonding status and not so much by physical properties of the birds.

Figure 4.

Percentage of won interactions for different bonding classes plotted separately for females and males (including only within-sex agonistic interactions). Numbers in the bars indicate the number of individuals per category; only individuals with at least six observed interactions were included. Box plots represent median, 25% and 75% quartiles, whiskers indicate 10% and 90% range, and circles represent outliers. Box plots with the same capital letter at the bottom are not significantly different.

Third parties separating interventions could be regularly observed during affiliation bouts, that is, two birds that were sitting in contact and/or engaged in allopreening were separated by a third bird that was previously not involved in the interaction. Of the total of 472 affiliative interactions, 34 were interrupted by a third party this way, resulting in the displacement of one or both affiliation partners. Individuals acting as interveners were always affiliated themselves, mostly exclusively (85% of interventions by eight closely bonded birds and four territorial breeders) and sometimes casually (15% of cases by three casually bonded birds). Unbonded birds were never observed to intervene in others' affiliative behaviours. Targets of interventions could be identified in 23 cases: some interventions were directed at birds that were engaged in an affiliative interaction with the bonding partner of the intervener (13% of the cases, N _{constellations} = 3), but in most of the cases (87%, N _{constellations} = 20), both targets had no obvious relationship to the intervener, but were in a casual or closely bonded relationship themselves.

Discussion

This is, to our knowledge, the first study that demonstrates different structural levels within raven nonbreeder aggregations: first, a loose organization composed of local birds, regular visitors and infrequent visitors and, second, the individuals' engagement in consecutive social bonds. In all age classes, birds tried to bond with other individuals (often older ones and of opposing sex) by offering and exchanging affiliative behaviours. When bonded, they won more conflicts in feeding situations; yet, birds rarely maintained the same bonding status over the study period of 2 years. Ravens, which live in long-term monogamy as adult breeders, thus showed considerable flexibility in forming and using affiliative relationships as nonbreeders.

Local group formation in ravens could have been expected, as studies on raven roosting behaviour regularly describe troops of 5–30 birds arriving and departing together (Marzluff et al. 1996; Wright et al. 2003), or even subroost formation (Dall & Wright 2009). However, the group of local nonbreeders in our field site is not a discrete, cohesive group. It has a loose organizational pattern that becomes apparent through longer observation periods and probably resembles a geographical distribution, for example, dialect regions in humans, rather than independent social units. It is, however, the main pool of individuals from which our marked members chose their bonding partners and we cannot exclude the possibility that the group forms among individuals that prefer each other's company. Analysis of the kinship pattern will be an important next step.

Results from winter flock formation in other songbirds show that individuals with prior residency, that is, those birds that were earlier in a flock, have an advantage in agonistic encounters (magpie, Pica pica: Eden 1987) that even outweighs the effects of body size and age (great tit, Parus major: Sandell & Smith 1991; willow tit, Poecile montanus: Koivula et al. 1993) and enhances the individual's chances of survival through its first winter (marsh tit, Poecile palustris: Nilsson & Smith 1988). The group of locals within our nonbreeders would thus be expected to have prior residency advantages over visiting ravens, but this was not the case. Local birds did not win more agonistic conflicts at the foraging sites than birds that were using the valley only infrequently. It therefore seems that the effects of age and bonding status are stronger predictors for status than prior residency in ravens.

The conditions at our study site may be considered as artificial in respect to the predictability of food availability, causing a proportion of ravens (our ‘local’ birds) to give up their vagrant life. One might also argue that this artificial stability in food supply could have triggered increased competition at feeding sites and thus changed the settings for social bonding. Consequently, our findings may appear to be less applicable to other raven populations. However, as in our game park setting, ravens living in relatively undisturbed environments tend to forage together with large predators such as wolves and bears (Stahler et al. 2002). In central Europe, ravens have recently started to exploit these species held in captivity. In fact, the foraging conditions encountered by ravens in our study now represent a typical scenario for the Alpine area (Koch et al. 1986; Huber 1991). Note that ravens generally display high flexibility in adapting to human-shaped environments, using human-made resources for hundreds to thousands of years, from scavenging on carcasses at battle fields to feeding on human remains (Heinrich 1999). Furthermore, stable raven roosts are known from all over the Holarctic, with numbers regularly exceeding our records by many individuals (>2000 ravens: Engel et al. 1992; up to 2000: Wright et al. 2003; >350: Blázquez et al. 2009). Similarly, our mean number of ravens at feeding locations ranges well within the typical number reported for nonbreeder flocks from naturalistic settings (Bialowieza National Park, Rösner et al. 2005). Finally, nonbreeders are typically found together with territorial breeders at food sources (Marzluff & Heinrich 1991; Heinrich 1994). Likewise, the mixed structure of all age classes in foraging groups is already known from previous studies (Huber 1991). So the agonistic and affiliative interaction patterns observed in this study are presumably unaffected by the relatively constant foraging conditions.

In aviary situations, corvids typically form linear dominance hierarchies (Heinrich 1994; Izawa & Watanabe 2008; Scheid et al. 2008). If groups are large and open, however, linear dominance hierarchies are thought to be rare in the animal kingdom (Drews 1993) because the mechanism to form them cannot rely solely on intrinsic factors (e.g. body size, age, sex or confidence) but has to build on the memory of consecutive encounters within the specific dyad (Barnard & Burk 1979). In accordance with this, linear dominance hierarchies in wild corvids are only reported from species that form small, cohesive groups such as the cooperatively breeding carrion crows, Corvus corone (Chiarati et al. 2010) and Florida scrub jays, Aphelocoma coerulescens (Woolfenden & Fitzpatrick 1977). In ravens the picture is contradictory: on one hand, nonbreeder aggregations are large, open groups with unstable group composition, making linear dominance hierarchies unlikely. On the other hand, ravens spend years as nonbreeders, and aggregations may, according to our current results, comprise a number of long-term core members. This would support the formation of linear dominance hierarchies, at least in certain subgroups. Our sample size of agonistic encounters between marked ravens is too small to draw conclusions for the whole study group of nonbreeders. Yet, in all dyads that provided enough data for statistical analysis, we found stable dominance relationships with unidirectional dyads. Results from studies under similar conditions (Huber 1991) support this trend: if dyads have repeated conflicts, the same individual will be dominant over the other for a long time period (over 1 month in the study by Huber, up to 2 years in this study). Note that in those analyses the individuals' sex, age and bonding status need to be considered since all these factors determine the outcome of an encounter. Older birds, which are dominant over younger ones, males that dominate females and affiliated/mated individuals that are dominant over single birds, describe a structure of dominance hierarchy that is typical for birds (Gauthreaux 1978; Lamprecht 1986; Piper & Haven Wiley 1989; Moore et al. 2003).

What characterizes raven aggregations, however, is the structure and plasticity of the social bonds birds possess before sexual maturation and the beginning of breeding, respectively. Similar to primates (e.g. de Waal & van Roosmalen 1979; Noë & Hammerstein 1995; Henzi & Barrett 1999), these bonds are formed through investment by one or both individuals and differ in terms of duration and degree of reciprocity between the dyads. Individuals who have a bonding partner clearly benefit while foraging, whereby females already profit from unidirectional, casual relations. Presumably, the ability to assert oneself is dependent on the bonding status (and is stronger in females than in males), since most of the conflicts are initiated by bonded birds. Notably, the effect of winning conflicts is lost again if a pair breaks up. These results do not fit the alternative explanation of bonded ravens reflecting a particular phenotype, which combines both competitive ability and attractiveness (i.e. high-quality individuals being bonded and therefore winning conflicts). Instead, older partners of the opposite sex are preferred by both sexes, indicating the attractiveness of status and/or resource-holding potential, because age correlates with rank. In contrast to primates (e.g. Kummer et al. 1974; Cheney et al. 1986), ravens do not seem to establish and maintain a social network of several affiliates but focus on one bonding partner at a time. However, birds do switch bonding partners and, presumably as a consequence, their affiliation status regularly changes across seasons and years. All these components support the idea that the bonds of nonbreeding ravens serve as a social manoeuvre, which is decoupled from direct reproductive goals. The findings fit those obtained from captive groups of ravens, where birds also form valuable relationships with a few individuals only (Fraser & Bugnyar 2010a) and selectively use them in aggressive encounters (Fraser & Bugnyar 2010b, 2011).

Finally, the question remains why this bonding tactic is not adopted by all nonbreeders since bonded individuals gain easier access to resources. There are a couple of possible, mutually not exclusive, explanations: first, subordinate ravens can gain access to food by means other than interference competition, that is, by the selective pilfering of food caches (Heinrich & Pepper 1998; Bugnyar & Kotrschal 2002a). Ravens are capable of observational spatial memory, remembering the location of food caches they see others make (Bednekoff & Balda 1996; Heinrich & Pepper 1998). Given that they can also learn to withhold their intention and judge the appropriate timing of pilfering (Bugnyar & Heinrich 2005, 2006), they may rely on cognitive skills rather than resource-holding potential which would be enhanced by bonding. Second, the age asymmetry seen in most of the bonding attempts suggests that preferred partners are higher-ranked than the selecting individual. However, as most pairings were of opposite sex, the hierarchical distance between the partners is not a direct one. Females may have a difficult time achieving acceptance by older males, which could easily displace them; attempts of young males were often also ignored by older females. Obviously, being attracted to older opposite-sex partners limits the opportunities as the older partner would have to choose against his/her preferences. Third, bonded ravens regularly face harassment by other bonded ravens, including territorial breeders. This might be because strongly bonded birds are relatively dominant at feeding sites and territorial pairs tend to attack the most dominant of the nonbreeders first (Marzluff & Heinrich 1991). However, during the formation of bonds, which generally occurs outside of feeding, others may already intervene regularly and severely. For instance, the most enduring chase flights, lasting in full speed for up to 10 min, were observed between territorial females chasing other bonded females (A. Braun, unpublished data). If the extra costs of this harassment by others outweigh the benefits of a given bond, we would expect it to cease. A critical point may be whether or not the new bonding partner joins the conflict and actively helps or, at least, provides postconflict affiliation, as demonstrated under aviary conditions (Fraser & Bugnyar 2010b).

Overall, we have shown that a large-brained bird species, which has the time, through delayed maturation, and the need, through strong competition, to operate socially, does so. The evolutionary function of nonreproductive bonding in ravens is most likely to assess the quality of a partner before deciding on life-long monogamy. Away from that, however, nonreproductive bonding in ravens may qualify as a social manoeuvre facilitating access to resources and increasing status.

MS. number: 11-00863R