Multiple Motivations: Agonistic Coalitions and Interventions in Blue Monkeys
ABSTRACT
Primates are known for forming agonistic coalitions, but most data come from species in which agonism occurs frequently and rank predicts fitness. We analyzed coalitions and interventions in wild blue monkeys (Cercopithecus mitis), in which both agonism and third-party involvement are relatively rare, and in which rank does not predict fitness. Data came from a long-term study in the Kakamega Forest, Kenya, spanning 12 years and 12 groups. Intervening animals both supported winners and defended losers, and coalition partners nearly always prevailed over their opponent. Adult females were joiners and juveniles were coalition-recipients disproportionately, while opponents were disproportionately adults, especially males. A multivariate analysis confirmed these patterns and also showed that joiners were most likely to support the original contestant who was winning (vs. unclear outcome or losing) and the one to whom they were more closely related. A subset of the data showed higher odds of joining the higher- versus lower-ranking original opponent. In high-risk interventions (coalition recipient losing, joiner smaller than opponent), the preference for more related opponents was magnified. Blue monkeys intervening in agonistic disputes appear to take sides in ways that minimize costs by supporting the winner, while maximizing inclusive fitness benefits by preferring the more closely related contestant, especially when intervention is risky. Their additional tendencies to support young individuals versus older ones, all else equal, suggest an additional motivation to protect vulnerable group-mates. Coalitions of smaller-bodied groupmates may contribute to the social peripheralization of the group's adult male.
Summary
-
Wild blue monkeys provide agonistic support rarely but may side with either the winner or the loser when they do.
-
Interventions disproportionately involve adult females aiding peers or juveniles against adults (especially adult males).
-
Intervening animals favor the original contestant who is winning and more closely related to themselves, with these effects amplified when they take on more risk.
1 Introduction
In some social animals, cooperation can take the form of agonistic support, when two or more group-mates join forces in an aggressive interaction against a third party (Harcourt and de Waal 1992; Smith et al. 2010). While some such coalitions are directed to out-group animals, many also occur within the group. Most commonly, one individual intervenes in an ongoing dispute by taking the side of one of the original opponents, but in addition, two animals may jointly initiate an aggressive action against a third (de Waal and Harcourt 1992; Bissonnette et al. 2015). These coalitions are often successful in that the coalition partners prevail over their target; however, the act of joining forces may also cause animals to incur costs, as starting or joining a contest takes energy and seems likely to raise the risk of injury relative to merely standing by (van Schaik, Pandit, and Vogel 2006; Smith et al. 2010). Accordingly, one expects animals to form coalitions strategically, making choices about when to intervene and who they will support in ways that minimize their risks and costs while maximizing potential benefits.
There is evidence for such strategic decision making (Bissonnette et al. 2015). For example, primates and other animals often and disproportionately provide agonistic support to close kin, potentially increasing inclusive fitness (Silk 2002; Smith 2014; Smith et al. 2010; Leonardo et al. 2021). A bias toward close kin may be particularly strong when the animal forming a coalition takes on more risk, for example by challenging a groupmate that is larger or higher ranking than itself and thus increasing the chance of retaliation by the target (Kaplan 1978; Langergraber 2012). Intervening animals also often take the side of the original contestant who is already winning or who is higher-ranking, thus reinforcing a supportive relationship with the recipient while minimizing the chance and costs of defeat (Cheney 1977; Harcourt and Stewart 1989; Smith et al. 2010; Langergraber 2012; Leonardo et al. 2021). When body size indexes resource-holding potential which in turn predicts winning, this choice would mean intervening on behalf of the larger-bodied opponent, although a tendency to protectively defend younger (and smaller) close relatives would counter such a pattern (Cheney 1977; Smuts 1987; Watts 1997).
Most reports of coalitions in primates come from a limited set of taxa (papionin monkeys, capuchins, great apes; Bissonnette et al. 2015) in which rank predicts behavioral or direct indicators of fitness and/or in which coalitions are relatively common. There is an expected connection between the importance of rank relations and the occurrence of coalitions, as coalitions may contribute to agonistic power and reinforce status relationships (Chapais 1995; Silk, Alberts, and Altmann 2004). Most species whose coalitionary behavior has been described also live in groups with multiple males, and coalitions between males, often unrelated and competing for high social status or for access to an indivisible resource (i.e., a mate), raise interesting questions about the evolution of such cooperative behavior (e.g., van Schaik, Pandit, and Vogel 2006; Ionica, Li, and Berman 2007; Berghänel, Schülke, and Ostner 2010; Freeman et al. 2016; Watts 2018; Toyoda et al. 2022). Coalitions also involve females and juveniles (Smith et al. 2022), however, where they are part of maternal or other animals' protective or typically kin-biased supportive behavior (Bernstein and Ehardt 1985; Watts 1997, Ferreira, Izar, and Lee 2006), and where—in some species—they may be central to the process by which females socially “inherit” and maintain their dominance rank, with consequences for competitive and reproductive success (Broom, Koenig, and Borries 2009; Langergraber 2012; Strauss and Holekamp 2019).
Here we describe coalitions in a cercopithecine primate, the blue monkey (Cercopithecus mitis stuhlmanni), in which coalitions have been described as rare (< 1% of agonistic interactions, Cords 2000 vs. the 19% mean for other primates, Smith et al. 2010), and in which rank-related advantages to female reproductive success (i.e., fertility and survival) are not evident (Roberts and Cords 2013; Thompson and Cords 2018). These arboreal guenons live in groups comprising females and their young, and typically one but sometimes multiple adult males (Lawes, Cords, and Lehn 2013; Gao and Cords 2020). Adult males compete for status as the sole resident of a group, and seldom live together; not surprisingly, adult males almost never form agonistic coalitions with each other. Like most other cercopithecines, females are philopatric, remaining in their natal groups for life, so groups comprise multiple matrilines. Female dominance relations are related to kinship, with adult matriline members generally ranking adjacent to each other in a group-wide linear hierarchy (Klass and Cords 2015). However, there is evidence that coalitionary support by kin contributes little to this rank structure in the group: specifically, the availability of kin as potential coalition partners has little effect on the process of rank acquisition (Donabedian and Cords 2021). We were motivated to see if factors like kinship, rank, size, and age explained coalitionary patterns in blue monkeys generally, as these factors are important in other species where coalitions are more common and rank has clearer effects on behavior and fitness (Bissonnette et al. 2015).
We begin by describing the incidence of coalition formation. Then, focusing on interventions where one animal joined in an ongoing aggressive dispute, we evaluated their success in terms of the coalitionary pair prevailing over their target, as well as the way that members of different age and sex classes were involved. Next, we examined partner choice by the intervening individual, hypothesizing that it would choose to support one of the two original contestants in a strategic way, both maximizing its own benefits and minimizing costs. Accordingly, we predicted that the interventions would favor the original contestant that was in a bigger size class (so more physically powerful), that was winning at the time of the intervention, and that was more closely related to the intervening individual. Among a subset of animals with known rank, we expected intervening individuals to support the higher-ranking of the two original opponents. Finally, when an intervening animal took on more risk (being smaller than the opponent, and aiding a recipient who was losing), we predicted that its relative kinship to the original contestants would be especially important in its choice of partner, with an even stronger bias to support the more closely related contestant.
2 Methods
Field research was carried out under the Columbia University IACUC supervision and according to the laws of Kenya. Our field protocols adhered to the American Society of Primatologists Principles for the Ethical Treatment of Non-Human Primates and followed its Code for Best Practices in Field Primatology.
2.1 Study Population
Data came from a long-term study of blue monkeys inhabiting the Isecheno study site in the Kakamega Forest, western Kenya (0°19′ N 34°52′ E; elevation 1580 m). This rainforest habitat (mean annual rainfall 2006–2018: 2239 ± SD 282 mm) supports a relatively dense population of blue monkeys (2.9–5.0 groups/km², Fashing et al. 2012). These monkeys have a diverse diet, but fruits are preferred foods that supply most energy (Takahashi et al. 2019), and agonism occurs disproportionately during feeding, especially when feeding on fruit (Cords 2000; Pazol and Cords 2005). Agonism is rare (total rate for adult females giving and receiving from all group members is 0.39–0.60 acts/h, about half of which are spatial displacements), but as noted above, linear matrilineally structured hierarchies are evident in the pattern of wins and losses (Pazol and Cords 2005; Klass and Cords 2015). Higher-ranking females and those in larger groups experience higher rates of agonism (Klass and Cords 2015).
Data for this study came from a period of 12.6 years (June 1, 2006 thru December 31, 2018), during which near daily observations were carried out on 4–8 study groups at a time. The number of groups grew over this period because of four group fissions (Table 1). All animals in the study groups were habituated to observers, as more than 10 years of near-daily group monitoring preceded the observations we report. Observers trained for 2–4 months and did not collect data until they could identify all group members individually five times in a row without error, and their coding of behavioral elements matched (≥ 95%) those of the most experienced team members.
| Group | Dates studied | Mean group size (range) | Mean N adult ♂♂ | Mean N adult ♀♀ | N obs days |
|---|---|---|---|---|---|
| GS | 06/01/2006–10/02/2008 | 47.2 (39–55) | 1.3 | 16.4 | 852 |
| GSB | 10/03/2008–12/31/2018 | 9.2 (6–12) | 1.0 | 2.9 | 2659 |
| GSA | 10/03/2008–10/10/2009 | 32.6 (29–34) | 1.1 | 9.0 | 371 |
| GSAA | 10/11/2009–12/31/2018 | 35.3 (22–58) | 1.1 | 11.9 | 3302 |
| GSC | 10/11/2009–12/31/2018 | 28.0 (9–45) | 1.0 | 10.2 | 3253 |
| GN | 06/01/2006–07/23/2016 | 45.8 (36–61) | 1.2 | 14.6 | 3700 |
| GNN | 07/24/2016–12/31/2018 | 9.7 (7–14) | 1.0 | 3.1 | 854 |
| GNS | 07/24/2016–12/31/2018 | 49.0 (44–57) | 1.0 | 16.0 | 889 |
| TWN | 06/01/2006–12/31/2018 | 27.3 (19–42) | 1.2 | 9.0 | 4558 |
| TWS | 06/01/2006–07/15/2018 | 54.6 (46–76) | 1.6 | 19.2 | 4421 |
| TPA | 07/16/2018–12/31/2018 | 44.9 (43–50) | 2.1 | 18.8 | 168 |
| TPB | 07/16/2018–12/31/2018 | 12.4 (11–17) | 1.4 | 5.0 | 158 |
2.2 Data Collection
As part of long-term monitoring, observers surveyed the group on every observation day, recording all animals present. We assumed that immatures and adult females were present even on days when they were not observed as long as they were seen in the group before and afterward. For adult males, whose membership in groups is more dynamic, we did not make this assumption, and considered them present only on days when they were observed. These data allowed us to calculate daily group size and composition (Table 1).
Observers also collected information on all observed within-group agonistic interactions, both ad libitum and during systematic focal animal samples of adult females, which were part of the long-term monitoring protocol. These records included the identity of the individuals and behavior involved: spatial displacements (avoid, supplant), threats (stare, lunge, growl), chases, contact aggression (push, hit, bite), and signs of submission (flee, cower, gecker, scream, trill), if present. These records did not include cases of juvenile males harassing copulations, or responses to such acts.
Our analysis focused on those agonistic interactions in which there was a coalition, which we identified when two (or more) individuals took the same side in an agonistic dispute. In some cases, the coalition was “in progress” when observers detected the agonism, whereas in others they noted the identities of the original two contestants and the individual who intervened (the “joiner”) to support one of them (“recipient”) by directing aggressive behavior to its opponent. We refer to these cases in which a joiner could be identified as interventions. In other primates, recipients use specific recruitment behavior to attract particular coalition partners (Smuts 1987; Range and Noë 2005; Perry 2012); such recruitment behavior was not conspicuous before blue monkeys formed coalitions and was not recorded.
Some interactions involved more than three individuals, with multiple joiners participating simultaneously or sequentially. In these cases, we deconstructed the interaction into multiple unique triadic sets of coalition partners. For example, if individual A originally directed aggression to B, and then first C and subsequently D joined in to side with B against A, we recognized three interventions: C siding with B against A, D siding with B against A, and D siding with C against A. For the analyses of interventions we report here, however, we focused on primary interventions, that is, on the first single animal to join an initial dyadic dispute, as this is a simpler decision point involving a choice between just two contestants. Except to quantify the number of joiners on each side of the initial dispute, we excluded cases in which multiple animals were first joiners simultaneously.
2.3 Data Analysis
We first estimated how often dyadic agonism led to coalition formation, examining 2 years of agonism data (2010 and 2017, chosen as representative of the study period) for all groups. We calculated the percentage of independent agonistic interactions (> 30 s without aggressive or submissive behavior separated subsequent interactions of the same individuals) that involved coalitions, including cases in which a coalition was “in progress” when the agonism was first detected and those in which one (or more) intervention/s were witnessed.
Focusing on primary interventions, that is, on the first animal to intervene in a dyadic contest, we then described the proportion that involved supporting a recipient who was winning in the original dyadic contest versus defending a recipient who was losing versus taking the side of a recipient whose status as winner or loser was unclear when the intervention occurred. This analysis, and those following, used data for the entire study period.
To gauge success of joint aggression, we again focused on primary interventions and how they influenced the course of the agonistic interaction. We scored a success if: (1) the joiner supported a recipient who was already winning (i.e., receiving submission from the opponent) and the coalitionary team continued to win; (2) the joiner defended a recipient who was losing (behaving submissively) to its opponent, and the opponent stopped behaving aggressively and/or showed signs of submission after the joiner intervened; or (3) the intervention on behalf of a losing recipient ended with the joiner and/or recipient threatening the opponent who did not respond, terminating the aggressive episode. We scored an intervention as unsuccessful if: (1) the joiner defended a losing recipient but failed to stop the aggression it was receiving, or (2) the joiner supported a winning recipient who ended up receiving aggression from its opponent. To see whether the success of interventions was related to the outcome of the dyadic dispute when the joiner intervened, we ran a logistic mixed effects regression predicting success as a function of whether the recipient was winning, losing or neither, including joiner identity as a random factor.
Using the data on primary interventions, we next examined whether certain age or age-sex classes were disproportionately represented as joiners, recipients and opponents. For this purpose, we distinguished infants (first year of life), juveniles (first birthday until natal emigration for males, or until first birth for females), and adults (parous females, males that had emigrated from their natal group; both of these events occur at age 7 years on average; Ekernas and Cords 2007; Bronikowski et al. 2016). Ages were known from long-term monitoring of the population, in which near daily group surveys allowed us to pinpoint dates of birth. In our data set, birth dates were known to the day for 66% of 280 individuals, to the week for 33% of individuals, and to periods of 2–4 months for 1% of individuals. We derived random expectations from the frequencies with which members of different age-sex classes were involved in primary interventions, regardless of their role (joiner, recipient, opponent). Specifically, we created a hypothetical pool of 5000 individuals in which the proportion of different age-sex classes matched their representation in the primary intervention data set. From this pool, we first randomly selected two individuals without replacement, and randomly assigned them as recipient and opponent. We then randomly selected a joiner from those remaining in the pool. We repeated this entire process 10,000 times to derive expectations for how often members of each age-sex class should take each role “randomly.” We compared observed participation in relation to these expected values.
We modeled the partner choice of the joiner who intervened in an ongoing dispute to identify characteristics of the original opponents that attracted the joiner's support. For this analysis, we first randomly assigned either the recipient (who the joiner supported) or the opponent as Animal 1, and the other individual as Animal 2, and we then modeled whether the joiner chose to support Animal 1 (against Animal 2) as a binary response in a logistic mixed effects regression model. As fixed effects, we included the age-sex classes of Animal 1 and Animal 2, and three additional variables based on classifying Animals 1 and 2 in terms of: (i) their relative size (specifically, was Animal 1 bigger than Animal 2), (ii) whether Animal 1 was winning the dyadic encounter with Animal 2 before the intervention occurred, and (iii) the difference in their relatedness to the joiner. Assessment of relative size followed from the age-sex class designations above, with adult males being the biggest of all age-sex classes, followed by adult females and immatures in order of age (large (≥ 5 years old), medium (3–4 years), and small juveniles (1–2 years), infants (< 1 year)). If the original opponents were in the same category, we classified them as same-sized. We scored Animal 1 as winning the dyadic dispute (i.e., before the intervention) if it supplanted, was avoided by or received submissive behavior (flee, cower, gecker, scream, trill) from Animal 2. We scored Animal 1 as losing if Animal 2 met any of these criteria. If Animal 1 was neither winning nor losing, we coded this variable as “unclear.” For relatedness, we computed the difference between the joiner's relatedness (R-value) to Animal 1 and to Animal 2: positive values indicated that the joiner was more closely related to Animal 1 than to Animal 2, negative values that the joiner was less closely related to Animal 1, and a value of 0 that the joiner was equally related to Animals 1 and 2. We assigned R-values using qualitative relatedness categories, based on maternal kinship only: for example, maternal siblings were assigned an R value of 0.25. The most distant relatedness category we identified in our data set was second (half) cousins once removed. We assumed a resident male was unrelated (R = 0) to the adult females in his group (given that males disperse from their natal groups before sexual maturity), and unrelated to infants born into a group before he became that group's resident male.
To this basic model, we added interactions of the kinship variable with the other four predictors, in case relative kinship of the joiner to Animal 1 versus Animal 2 influenced how the other variables predicted the joiner's choice. We used likelihood ratio tests (LRT) to compare models with versus without the interaction term, and dropped nonsignificant interaction terms to improve interpretability of main effects (Engqvist 2005).
To examine how dominance ranks related to coalitions, we used a subset of all coalitions (both interventions and “in progress”) in which ranks were known for all three contestants to characterize coalitions (including those in progress when detected) and primary interventions (first single joiner witnessed) as “all up” (coalition partners both rank lower than opponent), “bridging” (partners straddle the opponent's rank) or “all down” (partners both rank higher than opponent; Bissonnette et al. 2015). These data included only adult or large juvenile (as of the year in which they reached their fifth birthday) females, as it was only for these individuals that we consistently had a reasonable amount of data to assess rank. To compute ranks, we first compiled for each calendar year all records of dyadic agonistic interactions with a decided outcome (i.e., in which one and only one of the opponents behaved submissively). We then used Domicalc (Schmid and deVries 2013) to identify an optimal rank order from the win:loss matrix. In the rare cases when Domicalc identified more than one optimal order (which usually involved swapping positions of two adjacently ranked individuals), we averaged the results. We rescaled final ordinal ranks on a 0–1 (low to high) scale, using the formula: rank = 1−((ordrank–1)/(N−1)), where rank is the rescaled rank, ordrank is the original ordinal rank (1 = highest, N = lowest), and N is the number of individuals in the hierarchy. We also used a mixed effects logistic regression to evaluate whether “all up” interventions were associated with kinship between joiner and recipient.
To model how joiners intervened in an ongoing contest in relation to the contestants' relative dominance ranks, we used data on interventions in which the ranks of both recipient and opponent (but not necessarily the joiner) were known: here, the recipients and opponents were large juvenile or adult females, but the joiner could belong to any age-sex class. We used the difference in ranks of Animal 1 and Animal 2 (rank1–rank2) as a predictor in a model of the joiner's choice, along with the difference in their relatedness to the joiner.
Finally, to evaluate more closely the interventions in which the joiner appeared to assume the greatest risk, we selected a different subset of the primary intervention data in which (1) the recipient was losing (before the joiner intervened) and (2) the joiner was smaller than the opponent. We modeled joiner's choice as a function of its relatedness to Animal 1 versus Animal 2.
In all models of joiner's choice, we used the joiner's identity as a random factor. We examined variance inflation factors (GVIFs) to check for multicollinearity. We carried out all modeling in R (version 1.2.5033) using the package lme4.
3 Results
3.1 Characteristics of Coalitions
Coalitions were rare: only 1.8% of dyadic agonistic encounters (N = 11,877 from all groups in 2010 and 2017) led to the formation of a coalition. If we limited the count to agonistic encounters that occurred during focal sampling, a smaller percentage led to coalition formation (0.8%, N = 3708), suggesting that there might be a bias toward detecting non-focal agonism that involves multiple parties.
For dyadic agonism that included at least one intervention (i.e., one or more joiners were witnessed taking a side in an initially dyadic dispute), the vast majority of cases (99%, N = 1193) involved animals intervening on only one side. In these cases, the average number of intervening individuals was 1.3 individuals (range: 1–7, N = 1182; we excluded 11 cases in which the number of joiners was coded as ‘multiple’ because an exact count was not possible). When joiners intervened on the opponent's side as well, the average number was 1.5 (range: 1–3, N = 11).
Intervening animals (first joiners only) both defended the original contestant who was losing (37.4% of 971 first-joiner interventions) and supported the one who was winning (41.4%). In another 21.2% of cases, the original contest did not have a clear winner or loser.
Interventions were almost always (99%, N = 971 by first joiners) successful, meaning that the receiver did not receive further aggression from the opponent. This was true even when the opponent was an adult male, who was bigger than all other group members (97% successful, N = 134). Interventions in which the first joiner defended a partner who was losing its dyadic contest were more often unsuccessful (8/363, 2.2%) than those in which there was no clear winner (2/206, 1.0%) or those in which the joiner supported an already winning partner (1/402, 0.2%). A GLMM predicting success as a function of whether the recipient was winning, losing or neither when the joiner intervened in the original dyadic dispute differed significantly from a null model with only the random effect (joiner identity; LRT: χ2 = 8.26, p = 0.02, N = 971). The odds of success were 67 times higher when the recipient was winning versus losing (odds ratio 67.0, 95% CI: 0.5–8293.3, Wald's p = 0.09), but indistinguishable for either winning or losing versus an unclear outcome (odds ratio for winning vs. unclear: 11.1, 95% CI: 0.08–1597.1, Wald's p = 0.34; odds ratio for losing vs. unclear: 0.17, 95% CI: 0.01–2.1, Wald's p = 0.17).
Animals of various age-sex classes were differentially and nonrandomly represented as joiners, recipients, and opponents (Figure 1). Specifically, adult females acted as joiners 1.3 times more often than expected based on their overall presence in the coalition data set, while all other age-sex classes (and especially infants and adult males) were represented less than expected. Juvenile females and juvenile males were recipients 1.4 times more often than expected and infants were recipients 2.6 times more than expected. Adults were represented less than expected as recipients, especially adult males. Adults were opponents more often than expected, however, especially adult males (2.3 times more often) but also adult females to a lesser extent (1.1 times). All immatures were opponents less often than expected.
These same patterns are evident if one examines how different age-sex classes occupy the three roles—joiner, partner, and opponent—simultaneously in an intervention (Table 2). Adult females sided with juveniles against an adult male 6.2 times more often than expected by chance, and sided with a female peer against adult males 2.5 times more than expected. Females sided with a juvenile against another female 2 times more than expected and against another juvenile 1.5 time more than expected. Other relatively common triad combinations occurred about as often or less often than expected (Table 2).
| Joiner | Recipient | Opponent | Frequency | % observed | % expected | obs:exp ratio |
|---|---|---|---|---|---|---|
| fem | juv | fem | 163 | 17.5 | 8.8 | 2.0 |
| fem | juv | juv | 154 | 16.5 | 11.2 | 1.5 |
| juv | juv | fem | 143 | 15.3 | 11.6 | 1.3 |
| juv | juv | juv | 135 | 14.5 | 14.0 | 1.0 |
| fem | juv | male | 82 | 8.8 | 1.4 | 6.2 |
| fem | fem | fem | 52 | 5.6 | 7.5 | 0.7 |
| juv | fem | fem | 50 | 5.4 | 9.4 | 0.6 |
| juv | fem | juv | 45 | 4.8 | 11.3 | 0.4 |
| fem | fem | juv | 35 | 3.7 | 8.8 | 0.4 |
| fem | fem | male | 23 | 2.5 | 1.0 | 2.5 |
| male | fem | fem | 9 | 1.0 | 1.1 | 0.9 |
- Note: Only the most common combinations, comprising 95% of primary interventions (891 of 934), are shown. % observed and % expected values are based on all 934 interventions. Expected values reflect presence of different age-sex classes in primary interventions, regardless of their role as joiner, recipient, or opponent, as explained in text.
3.2 Models of Joiner's Choice of Partner
To understand joiner's choice based on the entire data set, we focus on a model with 5 main effects only (Table 3, Figure 2), as this model was not improved by adding interaction terms with the kinship variable (Table S1). This model showed that the joiner's likelihood of intervening on behalf of Animal 1 (vs. Animal 2) was related to Animal 1's success in the original contest versus Animal 2. Specifically, the joiner was least likely to support Animal 1 if it was losing the initial dyadic encounter, more likely to support Animal 1 if the outcome was unclear, and most likely to support Animal 1 if Animal 1 was already winning. The joiner's choice of Animal 1 (vs. Animal 2) was not clearly associated with the size difference between Animals 1 and 2 although joiners appeared somewhat more likely to intervene on behalf of Animal 1 if it was the smaller contestant (Figure 2). The joiner's relative relatedness to Animals 1 and 2 had a stronger effect on partner choice than relative size. The joiner was more likely to intervene on behalf of the original opponent to which it was more closely related, and a larger differential in relatedness of the two opponents increased the joiner's likelihood of choosing the closer relative. Lastly, as expected based on the previous analysis, the age-sex class of both Animal 1 and its opponent (Animal 2) influenced the likelihood that the joiner intervened on Animal 1's behalf. Specifically, the joiner was least likely to intervene on behalf of Animal 1 if it was an adult male, and most likely to do so if it was a small (1–2 years old) or medium (3–4 years old) juvenile or infant (< 1 year old). The joiner was also most likely to intervene on behalf of Animal 1 if its opponent (Animal 2) was an adult male, and least likely to do if the opponent was a small or medium juvenile or infant; opponents who were adult females or large juveniles increased the probability of support to an intermediate degree.
| Variable | Odds ratio | 95% CI | LRT (χ2, p) |
|---|---|---|---|
| A1winninga | 94.2, < 0.001 | ||
| Unclear | 2.4 | 1.2–4.7 | |
| Yes | 13.0 | 7.1–23.8 | |
| A1 vs. A2 relative sizeb | 1.6, 0.448 | ||
| Same size | 1.2 | 0.5–3.1 | |
| A1 smaller | 2.2 | 0.6–8.6 | |
| Difference in A1 versus A2's relatedness to joiner | 4116.4 | 657.1–2578.9 | 180.1, < 0.001 |
| A1 age-sex classc | 46.3, < 0.001 | ||
| Adult female | 20.2 | 5.3–76.6 | |
| Large juv | 18.1 | 4.0–83.0 | |
| Medium juv | 77.1 | 13.3–446.8 | |
| Small juv | 252.3 | 31.8–2005.9 | |
| Infant | 306.9 | 23.3–4035.7 | |
| A2 age-sex classd | 35.4, < 0.001 | ||
| Adult male | 80.8 | 6.5–996.8 | |
| Adult female | 7.1 | 0.9–53.8 | |
| Large juv | 8.2 | 1.3–53.0 | |
| Medium juv | 2.0 | 0.3–11.9 | |
| Small juv | 0.9 | 0.1–5.6 | |
| (Intercept) | 0.001 | 0.0001–0.016 |
- a Reference class: A1 not winning (i.e., A1 losing).
- b Reference class: A1 larger than A2.
- c Reference class: adult male.
- d Reference class: infant.
3.3 How Dominance Rank Relates to Coalitions and Joiner Choice
We identified 372 coalitions (i.e., those detected in progress, as well as those with a clear first joiner, including cases in which two or more were simultaneously first joiners) involving adult and large juvenile females in which dominance ranks were known for all three individuals. Most coalitions (320, or 86%) were “all down,” 12% (44) were bridging, and only 2% (8) were “all up.” Of the eight all-up coalitions, six cases involved unrelated coalition partners, whereas two involved a mother supporting her adult or large juvenile daughter. The opponents of these all-up coalitions were unrelated to either coalition partner. Results were very similar for 106 primary interventions (i.e., excluding coalitions that were in progress when detected, and cases in which two animals joined simultaneously): 83% all-down, 13% bridging, and 4% all up. We found no evidence that joiner-recipient dyads from the same matriline were more likely than unrelated dyads to have all-up or to have all-down interventions (Table S2). The intervention data set allowed us to distinguish two kinds of bridging interventions: cases in which the joiner out-ranked both the opponent and the partner versus those in which the joiner was lower-ranking than the two others. In most bridging interventions (11/14), joiners out-ranked opponents and recipients.
A model of joiner's choice (based on the subset of data in which the ranks of recipient and opponent were known) showed that when group members intervened in contests involving two animals of known rank (namely adult or large juvenile females), they showed a clear preference for (higher odds of) supporting the higher-ranking and the more closely related of the original two contestants (Table 4).
| Variable | Odds ratio | 95% CI | LRT (χ2, p) |
|---|---|---|---|
| Difference in A1 versus A2 rank | 91.5 | 22.1–379.3 | 69.0, < 0.001 |
| Difference in A1 versus A2's relatedness to joiner | 3499.8 | 43.3–28273.5 | 27.7, < 0.001 |
| (Intercept) | 1.3 | 0.8–1.9 |
3.4 Highest Risk Interventions
There were 146 high-risk primary interventions, i.e. cases in which the recipient was losing its original dyadic interaction and the joiner was smaller than the opponent. Most opponents were adults, 65% adult males and 27% adult females. Joiners and recipients were close relatives in more than half of the high-risk interventions: mothers intervened on behalf of their offspring in 32% and siblings aided each other in another 22% of the cases. Another third (33%) of the high-risk interventions involved a joiner who defended an unrelated recipient, most often against an adult male (28/48, or 58%) or an adult female (12/48, or 25%).
The model of joiner's choice for these high-risk cases showed that joiners were more likely to intervene on behalf of the more closely related of the two original contestants (odds ratio for difference in joiner's relatedness to A1 vs. A2: 6.61 e + 06, 95% CI: 1174.1–3.724 e + 10, LRT: χ2 = 109.7, p < 0.0001, N = 146), as was true in the full data set. We note, however, that the coefficient in the model of high-risk interventions is much bigger than in the model of the full data set, suggesting that in high-risk interventions, relative relatedness of the opponents to the joiner predicts the joiner's choice more strongly, increasing the odds of choosing Animal 1 versus Animal 2.
4 Discussion
Coalitions in blue monkeys are rare, occurring in < 1% of agonistic interactions, which are rare to begin with (an adult female averages one agonistic encounter every 1.7–2.5 h, about half of which are approach-retreat interactions; Cords 2000, Pazol and Cords 2005; Klass and Cords 2015). The proportion of agonistic interactions that involve coalitions also appears to be considerably lower than other primate (and non-primate) taxa, for which Smith et al. (2010) summarized values ranging from 2% to 69% (mean: 19% for 25 primate species). There may well be other primates with similarly low proportions as blue monkeys, but coalition rates (per agonistic conflict) are probably under-reported in taxa where coalitions are rare (for an exception, see Borries 1993). Previous comparative studies of captive or provisioned populations have suggested that guenons generally have low rates of coalition formation, especially in comparison to several of their papionin relatives (Kaplan 1987). Our findings confirm this suggestion with data from wild living guenons, and also indicate low rates relative to those of capuchins, reported more recently (Ferreira, Izar, and Lee 2006). We also acknowledge, however, that coalitions are rare in some papionins (Wittig et al. 2007).
Coalitions involving females have generally been linked to the potential for contest competition (e.g., Sterck, Watts, and van Schaik 1997). Blue monkeys do engage in contests, and females do so disproportionately when feeding, and particularly when feeding on preferred fruits (Cords 2000; Pazol and Cords 2005), a major source of energy (Takahashi et al. 2019). However, dominance rank does not predict longevity or fertility of adult females (Roberts and Cords 2013; Thompson and Cords 2018), and thus it appears that contest competition, while behaviorally apparent, is weak in blue monkeys. The relative rarity of coalitions is thus consistent with the proposed link between contest competition—as evidenced by unequal fitness—and coalitionary aggression. Based on a modeling study, Ekanayake-Weber, O'Connor-Coates, and Koenig (2024) recently proposed that asymmetries in agonistic interactions and their hierarchical organization (i.e., the existence of “steep” hierarchies) need not co-occur with rank-related skew in fitness because different factors drive these behavioral patterns versus fitness skew. In particular, the heterogeneous spatial distribution of resources is a major determinant of rank-based fitness skew only. The relative rarity of coalitions in blue monkeys, despite their moderately steep hierarchies (Klass and Cords 2015), suggests a possible extension of these results to agonistic cooperation. Specifically, coalition formation, like skew in fitness, need not coincide with the behavioral asymmetries that allow one to recognize clear hierarchies. Blue monkeys illustrate this disconnect, with clearly hierarchical behavior but relatively rare coalitions.
Despite their rarity, blue monkey coalitions—and particularly interventions, where we could discern who first joined an initially dyadic agonistic dispute—exhibit several clear patterns. First, over half the interventions involve defending the victim of the original aggression (37%) or interceding when the outcome was unclear (21%), although support of winners accounted for the remaining 41% of cases. This observation contrasts with Kaplan's (1987) summary of guenon fight interference, which claimed that supporting aggressors predominated, in contrast to the main pattern in papionin monkeys. With the exception of vervets, none of the studies Kaplan summarized were of wild living animals, however, and the number and size of groups was generally small. It is possible that the cost:benefit ratio of interventions in confined spaces is quite different from what occurs in wild living groups, where animals are more spread out; such spacing facilitates avoidance of escalated conflicts, and reduces opportunities for what Rowell (1971) described as “chaining” (piling on top of a victim), also observed in captive groups (Smuts 1987). Nevertheless, we note that when other factors were controlled for (Table 3), blue monkeys had higher odds of choosing the original opponent who was winning, not the loser. Comparing raw frequency data (in which supporting the winner did not predominate) to the results of a multivariate model (which indicated a preference for supporting the winner) highlights the importance of simultaneously considering multiple predictors of a joiner's choice of partner.
A second clear pattern was that interventions were almost invariably successful, in that the recipient of coalitionary aid either continued to prevail against or received no further aggression from the opponent. Success has been measured in various ways across studies, but high success rates, or higher success rates in comparison to dyadic agonistic encounters, have also been reported in non-primate taxa (Frigerio et al. 2005; Leonardo et al. 2021); among primates, there are also exceptions to near-universal success (Bissonnette et al. 2015). We found that interventions defending a losing partner were slightly less likely to be successful than those supporting the original aggressor, but even so, the success rate was over 97% when defending the losing opponent. This figure is somewhat higher than what Cheney (1977) reported (87%) for defensive interventions on behalf of juvenile baboons predominantly carried out by adult females, or what Kaplan (1978) reported (84%) for defensive interventions of rhesus macaques. The difference might reflect the fact that blue monkey males, who were disproportionately the targets of interventions, are more likely than baboon or macaque opponents to be intimidated by coalitions of smaller-bodied group-mates (females and juveniles). Adult males fleeing the collective aggression of females has been noted previously in this species (Kaplan 1987; Struhsaker and Leland 1979).
A third clear pattern was the disproportionate involvement of different age-sex classes in interventions. Adult females were disproportionately the animals that intervened (joiners), and joiners supported juveniles disproportionately. The targets of coalitions were disproportionately adults, and especially adult males. Female primates are not always disproportionately involved in coalitions (Silk, Alberts, and Altmann 2004; Ferreira, Izar, and Lee 2006), but their role in providing agonistic support to group-mates has been highlighted in taxa other than blue monkeys (Bernstein and Ehardt 1985; Smuts and Smuts 1993; Koenig et al. 2022). In addition, supporting group-mates against adult male opponents is widespread (though not universal, Link, Di Fiore, and Spehar 2009; Swedell 2011) among primates, and can serve as a source of power for adult females in relation to larger adult males who often dominate females in one-on-one encounters (Packer and Pusey 1979; Smuts 1987; Perry 2003; Newton-Fisher 2006; Tokuyama and Furuichi 2016; Smit et al. 2022). Group members forming coalitions against resident males has been noted in guenons particularly, with researchers contrasting this behavior to its relative absence in closely related papionin monkeys (Hall 1965; Wolfheim 1977; Struhsaker and Leland 1979; Rowell 1988). For example, in vervets (Chlorocebus pygerythrus), which are part of the guenon clade, group members that intervened in disputes between an adult female and an adult male supported the female against the male 90% of the time (Saccà et al. 2022), a figure that matched our study exactly. Smuts and Smuts (1993) highlighted the fact that in female-bonded groups, the coalitions formed by females against adult males are less strikingly kin-biased than those targeting other group members; however, we did not find this trend in blue monkeys, where there was no evidence that the age-sex class moderated the effect of kinship in the joiner's choice of which original contestant to support.
The role of adult males in coalitions of blue monkeys, where they were disproportionately targets of coalitions by smaller group-mates, contrasts with reports of their roles in other primates' coalitions. In tufted capuchins (Sapajus spp., Ferreira, Izar, and Lee 2006), for instance, males act disproportionately as joiners, protecting often young members of their groups, whereas blue monkey males intervened less often than expected, and only 1% of primary coalitions involved males siding with juveniles. Similarly, in Assamese macaques (Macaca assamensis), 50% of adult interventions involved an adult male supporting a juvenile (Minge et al. 2016), whereas the comparable figure for blue monkeys is 3%. In mountain gorillas (Gorilla beringei beringei) and olive baboons (Papio anubis), males intervene aggressively in ongoing disputes among females, either taking one side or simply bringing the conflict to an end for both females (Watts, Colmenares, and Arnold 2000). Watts et al. viewed these “policing” interventions as benefitting females and thus cultivating their “loyalty” to the male joiner. Male blue monkeys only rarely joined dyadic aggression between adult females, however, and almost always (7 of 9 occasions) took the side of the already-winning contestant. Evidently male blue monkeys do not use agonistic support as a strategy to foster female loyalty. It is noteworthy in this context that females often prefer their resident as a mate anyway (Schembari et al. 2024), although residents lose 40% of paternities to other males (Roberts, Nikitopoulos, and Cords 2014).
A fourth clear pattern was that intervening animals acted in their own self-interest in certain respects, preferentially siding with closer kin, which may provide inclusive fitness benefits, and minimizing their risk of injury by preferentially supporting the original contestant who was winning. A preference for closer kin was enhanced in high-risk interventions (see below). Agonistic support of kin is a widespread pattern in social mammals (Smith et al. 2010). We also note here, however, that there was a sizeable proportion (36%) of interventions in which the joiner was unrelated to both of the original antagonists, so while they prefer closer kin, not all agonistic support benefits relatives. A tendency to support the opponent who is winning is implied in many reports of other species in which joiners take the side of the aggressor or the higher-ranking opponent (e.g., Chapais 1995; Perry 2003; Range and Noë 2005; Silk, Alberts, and Altmann 2004; Smith et al. 2010). For the subset of our data in which ranks of the original contestants were known, we also could show this latter trend. Supporting the winner or higher-ranking opponent (as opposed to the loser or lower-ranking one) probably makes interventions relatively low-risk. In such cases, intervening animals may be opportunistically exercising their own self-interest by reinforcing their own status vis a vis the opponent, demonstrating solidarity with their partner, or both (Chapais 1995; Smith et al. 2010; Langergraber 2012).
Finally, despite the self-interest evident in aspects of their partner choice, joiners generally favored siding with younger immature age-classes. This trend suggest a motivation to protect the most vulnerable of the original two contestants. It is noteworthy that this tendency appears to be independent of the joiner's relative kinship to the original opponents, as kinship was a separate predictor in the model and thus controlled for, and the interaction term with kinship did not significantly improve the model. In short, both self-interest and protecting the most vulnerable characterize blue monkey coalitions.
These same principles apply to the highest-risk interventions, cases where the recipient of coalitionary support was losing its original contest at the time of the intervention and in which the intervening individual (joiner) was smaller than the original opponent and thus at a competitive disadvantage. In these high-risk cases, joiners defended a vulnerable (i.e., losing) contestant by definition, and they also showed a magnified tendency to support the more closely related of the two original opponents. A stronger preference for close kin may arise because taking on higher risk does not pay off otherwise.
The relationship of relative dominance rank to the structure of coalitions in blue monkeys also resembles other primate species. A large majority (86%) of coalitions among large juvenile and adult females (the animals for whom we could assign ranks) involved coalition partners who both outranked their common target (“all down”), as is common in other taxa, especially for coalition partners that are unrelated (Bissonnette et al. 2015). For blue monkeys, the proportion of “all down” interventions was the same whether coalition partners were related or not. Still, this is a somewhat less overwhelming percentage than the 96% reported for female white-faced capuchins (Perry 2012). In addition, the proportion of “bridging” coalitions was about three times greater than in these capuchins. Finally, in contrast to both white-faced capuchins (0%) and yellow baboons (0.5%, Silk, Alberts, and Altmann 2004), blue monkeys showed more (though still few) “all up” coalitions (2%) or interventions (4%). All-up coalitions involved both kin and non-kin as partners. Overall, these comparisons suggest that rank may explain less about coalition participation in blue monkeys, where protection of vulnerable individuals is evidently also a strong motivation guiding the intervening individual's choice of partner.
5 Conclusion
We would generally emphasize how blue monkeys use interventions to selfishly support the vulnerable. The vulnerable are often immature animals, so the overall patterning of agonistic support highlights a protective motivation. This motivation appears less evident, however, in coalitionary triads of older juveniles and adult females, where joiners side with the higher-ranking of the original two opponents. In general, then, multiple characteristics of the potential recipients of an intervention appear to influence the joiner's choice of where to direct their support, which suggests that this decision is not a simple one. Future studies could carry these analyses further by examining the decision of whether to intervene at all: we did not have the fine-grained data that would be needed to control for the opportunity to intervene, an important control when animals are widely dispersed.
We also highlight the unanticipated pattern that adult males were frequently the target of coalitions by group members. As adult males are larger than all other age-sex classes, and as coalitions were generally successful in halting aggression, it is clear that this form of cooperation is a way for physically less powerful individuals to increase their social power. It would be interesting to see how frequently interventions occur when adult males aggress other group members, which our study could not address. It is tempting to hypothesize that the social and spatial peripheralization of adult male guenons (Kaplan 1987; Struhsaker and Leland 1979; Cords 1987; Isbell 2013) may to some extent reflect the coalitions formed by their group-mates against them. Peripheralization may in turn limit the role of males as joiners (Watts, Colmenares, and Arnold 2000).
Author Contributions
Kyle Rotter: conceptualization (supporting), data curation (equal), formal analysis (lead), investigation (equal), methodology (equal), validation (equal), visualization (lead), writing–original draft (equal), writing–review and editing (supporting). Marina Cords: conceptualization (lead), data curation (equal), formal analysis (supporting), funding acquisition (lead), investigation (equal), methodology (equal), project administration (lead), supervision (lead), validation (equal), visualization (supporting), writing–original draft (equal), writing–review and editing (lead).
Acknowledgments
We are grateful to the Government of Kenya (National Commission for Science, Technology and Innovation), Kenya Forest Service, and Kenya Wildlife Service for authorization to conduct this research, and to the University of Nairobi Zoology Department, Institute for Primate Research (National Museums of Kenya), Moi University Department of Wildlife Management, and Masinde Muliro University of Science and Technology (Center for Kakamega Tropical Forest Studies) for local affiliation. We warmly acknowledge many who helped gather field data for this report, including long-termers M. Atamba, B. Brogan, C. Brogan, O. Cords, A. Fulghum, K. Gaynor, C.B. Goodale, F. Hardy, M. Hirschauer, S. Khamusini, J. Lucore, K. MacLean, L. McGee, C. Mitchell, J. Munayi, C. Nunez, C. Oduor, L. Pollack, S. Roberts, R. Settele, E. Shikanga, D. Shilabiga, and E. Widava, as well as P. Richardson for help in programming a distribution of random expectations. We also thank reviewers for their comments on a previous version. Funding for the long-term project came from the National Science Foundation (SBE 95-23623, BCS 98-08273, DGE 03-33415, BCS 05-54747, DGE 09-66166, BCS 10-28471), Ford, Leakey, Wenner-Gren and H.F. Guggenheim Foundations, AAAS-WISC, and Columbia University.
Open Research
Data Availability Statement
The data used for joiner choice models are openly available in Dryad (https://doi.org/10.5061/dryad.fxpnvx11q).
