Humans form agonistic coalitions and alliances in many contexts, but this behavior is thought to be rare in other species. A prominent hypothesis states that coalitions may be under cognitive constraints, but this idea is debated and remains to be tested empirically. In this study, we evaluate the cognitive constraint hypothesis against 3 alternative hypotheses that stress the role of demography, substrate use, and resource competition, for the evolution of male coalitions. A comparative analysis of a unique data set of 86 multimale multifemale groups of 38 nonhuman primate species from all major radiations revealed no evolutionary association of male coalition frequency with cognitive capacity (as indexed by neocortex ratio and endocranial volume). The observed variation was best explained by demography and resource competition in that male coalitions were more likely to occur in species characterized by larger male groups and reduced levels of contest competition (after controlling for phylogeny). These findings suggest that constraints imposed by the socioecological setting, rather than cognition, explain best why some primate species evolved customary coalitionary behavior while others did not. This study presents the first empirical evidence against the long-standing view that cognitive abilities may impose a limit on the use of coalitions in animals.
INTRODUCTION
Male–male coalitionary aggression is ubiquitous in humans (Chagnon and Bugos 1979; Hruschka and Henrich 2006; Flinn and Ponzi 2012), and the evolutionary roots of human coalitionary behavior can be traced by investigating similar behavior in nonhuman animals. Among our closest relatives, the primates, intragroup coalition among males show striking interspecific variation even between closely related taxa (van Schaik et al. 2006; e.g., Henzi et al. 1999). Males living in mixed-sex groups sometimes use coalitions to increase or maintain their rank (e.g., Kutsukake and Hasegawa 2005; Schülke et al. 2010), or increase their access to mates without affecting dominance (e.g., Bercovitch 1988; Bissonnette et al. 2011). This behavior, however, appears to be rare or completely absent in many more species (e.g., Henzi et al. 1999; Paul et al. 2000), a surprising finding given the mating (e.g., Nishida 1983; Bercovitch 1988; Watts 1998; Bissonnette et al. 2011) and reproductive (e.g., Witt et al. 1981; Schülke et al. 2010; Gilby et al. 2013) benefits that this behavior may provide. Although in most primate species the males in a group are not closely related, the absence of kin does not seem to be a serious obstacle to the evolution of male coalitions (van Schaik et al. 2006; e.g., Bercovitch 1988; Schülke et al. 2010; Bissonnette et al. 2011). It is likely that other socioecological or cognitive processes are responsible for driving and keeping coalitions in place (Olson and Blumstein 2009). Given the lack of formal quantification of coalitionary behavior in most species, however, this statement remains speculative. In this paper, we compiled a unique data set of 86 multimale multifemale groups of 38 primate species from all major radiations to test 4 evolutionary hypotheses for the evolution of male coalitions.
Many have argued that coalitions are, at a minimum, a prime example of how social interactions can become complex, and at most, may be the kind of behavior that drove brain evolution generally (Alexander 1989; Harcourt 1992; Connor 2007). Coalitions are thought to be complex because they involve triadic interactions (cf., Kummer 1967) where an individual, in order to be efficient, must take into account not only its direct relationships with others, but also the details of the relationships between other group members (e.g., Kummer 1967; Harcourt 1992; Tomasello and Call 1997; Silk 1999). Coalitions can become cognitively more demanding if individuals use affiliative interactions to cultivate relationships with powerful supporters, to compete for powerful allies, or prevent rivals from forming potentially disruptive alliances (de Waal 1982; Harcourt 1992; Silk 1992; Schülke et al. 2010). The extent to which coalitions may be under cognitive constraints, however, is debated (Dugatkin 1998; Barrett et al. 2007; Jennings et al. 2009; Hemelrijk and Puga-Gonzalez 2012) and remains to be empirically tested (but see Dunbar and Shultz 2007). The cognitive hypothesis assumes that coalitionary behavior is cognitively demanding and that information-processing abilities impose an evolutionary limit on the use of coalitions as a competitive tactic (e.g., Harcourt 1992; Dunbar and Shultz 2007). Accordingly, we predict that male primates with better cognitive abilities use coalitions more often than species with poorer cognitive abilities (Harcourt 1992).
A second hypothesis is based on the idea that a complex 3-dimensional arboreal environment may impose a physical constraint on coalition formation by making it harder for animals to coordinate their attack. For example, it has been reported that in savannah baboons “single males with good fighting ability have the advantage as long as the female is in a tree,” whereas “males forming coalitions are relatively better off on the ground where they can surround the consorting male and attack simultaneously” (Noë and Sluijter 1990, p. 149; see also: Smuts 1985; van Noordwijk and van Schaik 2001; Bissonnette et al. 2011). In addition, the energy and risk (e.g., falling down a tree) associated with aggression may be greater in trees than on the ground, hence making coalitions less likely to be achieved in an arboreal environment (Broom et al. 2009). Thus, if the incidence of coalitions is a simple function of the degree of arboreality, we should find that terrestrial species are more likely to form coalitions than arboreal species.
Third, it has long been recognized that demographic factors can influence aspects of primate social structure and social behavior, including coalitions (Chapais and Schulman 1980; Mitani et al. 2002). For example, Henzi et al. (1999) have shown that the noncoalitionary chacma baboon (Papio ursinus) males tend to live in groups that are on average smaller than those of the coalition-forming yellow (P. cynocephalus) and anubis baboon (P. anubis). Living in groups that contain few males may provide less opportunities for coalitionary activity (Henzi et al. 1999), for example, if it decreases the likelihood that appropriate combinations of individual male characteristics are represented for successful coalitions to be formed (Noë 1994; Mitani et al. 2002). This may occur if coalition formation is an alternative mating strategy only employed by certain classes of males (e.g., post-prime aged males) or if coalitions are formed opportunistically based on the combined fighting ability of the allies relative to that of their target (e.g., Bercovitch 1988; Noë and Sluijter 1995; Bissonnette et al. 2009). Thus, if coalitions crucially depend on partner availability, we should find that coalitions are more likely to occur in species characterized by larger male group sizes than species with smaller male group sizes.
The above hypotheses may be best viewed as constraint hypotheses and they are silent about the selective force driving the evolution of male coalitions. A fourth hypothesis states that coalition formation is an evolved response to the selection pressure generated by resource competition (Chapais 1995; van Schaik et al. 2006). For males living in mixed-sex groups, the extent to which male coalitions produce fitness benefits is thought to depend largely on the level of within-group contest competition over access to females (van Schaik et al. 2006). If contest competition is appreciable and fertile matings can be monopolized by one or a few high-ranking males, there is increased selection for males attaining low or no mating success to use alternative mating tactics such as coalitions (van Hooff and van Schaik 1994; van Schaik 1996). Therefore, the incidence of male coalitions should increase with increasing within-group contest competition. However, this prediction ignores additional factors that might influence the relationship between incidence of coalitions and contest potential. Specifically, very high levels of contest competition may hinder coalitionary behavior if males are less tolerant of each other, which hamper cooperation (e.g., Melis et al. 2006; Hare et al. 2007; Olson and Blumstein 2009). Moreover, mounting a coalitionary challenge may be particularly risky at higher contest potential (van Schaik et al. 2006) due to an increased likelihood of retaliation by the target thus making the benefit to cost ratio of coalitions less favorable. If these effects are present and strong enough to override the effects of increasing benefits with increasing contest competition, we should find a negative association between competition levels and the incidence of male coalitions (Olson and Blumstein 2009).
Although the 4 above factors (cognitive capacity, substrate use, demography, and contest potential) are not mutually exclusive, and in fact there is some evidence that each may play at least some role in individual species, their relevance at a broader taxonomic level is still unknown. Thus, we investigated the distribution of male coalitions across the primate phylogeny and the evolutionary association of male coalition frequency with these factors, using phylogenetically based methods.
MATERIALS AND METHODS
Data collection
An extensive survey of the primate literature was undertaken to obtain reports of coalition behavior among adult males (principal key words for search: coalition, alliance, intervention, aid, support). Because published information on male coalitions is relatively rare, we further contacted 45 researchers known to have conducted extensive behavioral field research (at least 1000 observation hours) with a given species and asked them to answer a questionnaire on male coalitions (see Supplementary Data in the Electronic Supplementary Material [ESM]). Criteria for the inclusion of a study were as follows. We only considered groups with at least 3 adult males. We excluded 2 species in which mating season mobility has been reported (Cercopithecus mitis and Erythrocebus patas). In these species a uni-male unit is the modal social structure over long periods and groups are only multimale during the mating season when several males temporarily immigrate (Cords 2000; Isbell et al. 2009; and references therein). A study group was included if a reliable assessment of the frequency of male coalitions could be made (see below). We gave preference to those studies of wild over nonwild groups (e.g., for Macaca sylvanus, we preferred data from Morocco over Affenberg Salem; see Supplementary Data [ESM] for additional data). Altogether the coalition data set consists of 86 multimale groups from 38 species including lemurs (2), New World monkeys (12), Old World monkeys (21), and apes (3).
Definitions and data selection
Male coalitions
We defined a male–male coalition as: “simultaneous aggression by at least two adult males against another adult male” (Harcourt and de Waal 1992) with all males residing in the same group. Data on the actual rates of male coalitions or the percentage of male aggression that is coalitionary were limited to a few species (N = 12). Consequently, we classified the frequency of male coalitions into 4 categories (modified from Whiten et al. 1999): customary (coded 3), for which coalitions are routinely observed in all or most adult males of a group or subgroup (e.g., among post-prime males in P. anubis and P. cynocephalus); habitual (coded 2), for which the behavior is not customary but has been observed repeatedly across several adult males or has occurred repeatedly under specific social circumstances (e.g., when a male challenges the alpha position in M. fascicularis); present (coded 1), for which the behavior is neither customary nor habitual but has been clearly identified, usually only once or a few times in thousands of observation hours; absent (coded 0), for which the behavior has never been observed in adult males, and this was not because of inadequacy of relevant observational opportunities.
All codings were assigned based on experts’ evaluations and cross-checked whenever possible with detailed information from the literature. In case of multiple study groups for a species, the values from different groups were averaged (rounded to the nearest integer). For P. ursinus and Alouatta paliatta male coalitions were observed in only 1 out of 8 and 5 groups, respectively, and thus the average coalition frequency for both species was zero. We ran a second set of analyses with the coalition value for both species set to 1 (i.e., “present”) and obtained very similar results (data not shown).
Cognitive capacity
The most appropriate measure of cognitive capacity is controversial with no single preferred measure (Deaner et al. 2000; Healy and Rowe 2007). We considered 2 widely used measures: 1) Male endocranial volume (ECV, measured from museum’s skeletal material) and the associated body mass to control for allometric effects derived from Isler et al. (2008), and 2) neocortex ratio (i.e., volume of the neocortex/volume of the rest of the brain-neocortex), calculated based on actual brain tissue volumes as given by Stephan et al. (1981). If no published data on neocortex volumes were available, we estimated neocortex ratio from male ECV using the equations given by Kudo and Dunbar (2001). We validated these equations by comparing the real neocortex measures of 9 species in the database with their estimated values and found that these measures were highly correlated (Pearson r = 0.88, P = 0.002). All measures were log-transformed.
Substrate use
The degree of arboreality was defined as the percentage of daylight hours spent in trees. Whenever possible, data on substrate use came from the same study group or population as the coalition data. We used the mean if multiple estimates were available for a species.
Demography
The number of adult males and females were preferentially taken from the same study group and period as the coalition data. The values from different groups were averaged in case of multiple study groups for a species. In species with fission–fusion dynamics (Pan troglodytes, P. paniscus, and Ateles spp.; Aureli et al. 2008), we considered the number of adult males and females in the community rather than the party because males can potentially form coalitions with other community males and mate with any community female.
Contest potential
Among primate species living in mixed-sex groups, contest potential is determined mainly by the variation in the number of females mating simultaneously (receptive synchrony) and secondarily due to the variation in the number of males per group (Ostner et al. 2008; Port and Kappeler 2010). We used the synchrony index as one estimate of contest potential. The synchrony index assumes that the expected level of estrous overlap between females is a function of the average number of mating days by individual females per ovulatory cycle, the duration of the mating season, and the number of females in the social group (Nunn 1999). The probability of Y females mating simultaneously was calculated using a binomial distribution:
where P(Y) is the probability of Y females mating simultaneously, k is the number of females (rounded to the nearest integer), and p is the probability that any given female is mating. The variable p is calculated as 2 times the duration of mating divided by the length of the mating season. The expected female overlap is thus P(Y ≥ 2) = 1 − P(Y < 2) (Nunn 1999). Mitani et al. (1996), van Schaik and van Noordwijk (1999), and van Noordwijk and van Schaik (2001) were the primary source of information for mating season length and mating duration. Additional data were taken from the sources listed in Supplementary Data (ESM). We use the species average from coalition studies for the number of adult females (see above).
The measure of receptive synchrony described above has been criticized, because it reduces the information on receptive females to a dichotomous variable (more or less than 2 receptive females). In fact, based on the priority-of-access model (Altmann 1962), it may be argued that the actual number of receptive females might affect the share of paternity that the alpha male can monopolize (Gogarten and Koenig 2013). Thus, we built on the synchrony index and added the effect of the number of females into a new index that we called the “expected share of matings in the alpha male” index (AS). It is calculated as follows:
where P(Y) is the probability of Y females mating simultaneously (calculated using the binomial distribution shown above), k is the number of females (rounded to the nearest integer), and m is the number of males in the group (rounded to the nearest integer). When the number of receptive females exceeds that of males (i.e., i ≥ m + 1), the alpha’s share is given by the number of males (i.e., 1/m). If the number of receptive females is smaller than that of males (i.e., i < m) the alpha’s share is given by the number of females (i.e., 1/i). For example, if there are 2 males and 6 receptive females, the alpha’s share is 1/2 or 50%. In contrast, if there are 2 males and 1 receptive female, the alpha’s share is 100%.
In our data set, we found a very high, negative correlation between the 2 above indices (Pearson: r = −0.98, P < 0.001, N = 33), that is, the alpha male’s share decreases as female receptive synchrony increases. Thus, we only used the expected alpha’s share of matings as our proxy for contest potential in the following analyses. The expected share of matings in the alpha male was calculated for 33 of 38 species.
Statistical and phylogenetic analyses
All statistical analyses were conducted in R 2.11.0 (R Development Core Team). Three packages, APE (Paradis et al. 2004), GEIGER (Harmon et al. 2008), and CAR (Fox and Weisberg 2011) were used to conduct the analyses.
We tested phylogenetic signal in each variable using Pagel’s lambda (λ; Pagel 1999), which is a tree transformation metric that varies between 0 (no phylogenetic signal) and 1 (trait distribution matches a Brownian model of evolution). Of the 6 variables in our data set, 4 contained a phylogenetic signal significantly greater than zero (lambda value: neocortex ratio: 0.96, P < 0.001; ECV: 0.76, P < 0.001, arboreality: 0.61, P < 0.001; expected alpha’s share: 0.56, P = 0.01), whereas 2 did not (coalition frequency: 0, P = 1, male group size: 0.18, P = 0.25).
As the dependent variable—coalition frequency—was categorical, we tested our predictions using a simulation-based phylogenetic analysis of covariance (Garland et al. 1993). For each independent variable we used the corresponding t value as test statistic. We used phylogenetic simulations to generate distributions of these test statistics under the null hypothesis that none of independent variables had any effect on the dependent variable. To this end we first fitted a phylogenetic model to the observed coalition frequencies (“fitDiscrete” in Geiger). Then we used the obtained model to perform 10000 simulations (“simChar” in Geiger) of the evolution of coalition frequencies. For each simulated set we calculated t values using the observed values of the independent variables and the simulated values of the dependent variable. For each obtained null distribution we calculated 95% confidence intervals (CIs). A variable was considered to be significant if the t value in the observed data was outside the 95% CI of the simulation-based null distribution.
To generate a phylogeny for the taxa in our data set, we used a Bayesian consensus tree from the 10kTrees project (Version 3), available from http://10ktrees.fas.harvard.edu/ (Figure 1). We used an appropriate sister taxon to place one species that was not available in the phylogeny and to infer branch length (i.e., Brachytheles arachnoides was used as sister taxon for B. hypoxanthus). In addition, we extracted a sample of 100 trees from the 10kTrees website to assess the impact of uncertainty in phylogenetic tree structure on the results (see Supplementary Data [ESM]).
Complete phylogeny for all 38 species used in the analysis (branch lengths proportional to absolute time: for more information see the 10kTrees documentation). The frequency of male coalitions is indicated by the color of the box (see main text for explanations). Taxonomy follows Groves (77).
Complete phylogeny for all 38 species used in the analysis (branch lengths proportional to absolute time: for more information see the 10kTrees documentation). The frequency of male coalitions is indicated by the color of the box (see main text for explanations). Taxonomy follows Groves (77).
Male group size showed a significant, negative correlation with expected alpha’s share of matings (Pearson: r = −0.73, P < 0.001, N = 31). We nevertheless included expected alpha’s share of matings as a separate predictor from male group size because we proposed a different, independent mechanism by which these factors may affect male coalitions. We adopted a conservative approach and conducted multiple regression analyses of neocortex ratio (or brain size and body mass), male group size, arboreality, and alpha’s share on male coalition frequency, as well as partial regressions including all covariates but male group size or expected alpha’s share of matings. The Pearson correlation coefficients among the predictor variables are presented in the Supplementary Data (ESM).
RESULTS
Figure 1 shows the distribution of male coalitions across the primate tree. Male coalitions are more widespread than we had anticipated: they are observed in almost all major radiations except in Malagasy prosimians and in almost all genera that form multimale groups. The frequency of male coalitions, however, varies widely across species and even within families or genera and this behavior is customary in only 23.7% of the sampled species (or 9/38). The absence of coalitions does not seem to be concentrated anywhere in the anthropoid tree.
The frequency of male coalitions was significantly related to both male group size and expected alpha’s share in partial regression models (Table 1). Male coalitions were more likely to be customary with increasing male group size and decreasing alpha’s share of matings. The multivariate regression including both covariates, however, found nonsignificant coefficients, thus indicating that their effects cannot be reliably untangled statistically. No effect of neocortex size or ECV or degree of arboreality was found.
Results of simulation-based phylogenetic ANCOVAs
| Dependent variable: frequency of male coalitions | ||||
|---|---|---|---|---|
| Predictor variables | Standardized estimate | t value | 95% CI | |
| (1) | Neocortex ratio | 0.173 | 1.096 | −2.005, 2.019 |
| Male group size | 0.416 | 1.926 | −2.071, 2.028 | |
| Degree of arboreality | −0.120 | −0.680 | −2.049, 2.046 | |
| Alpha’s share | −0.121 | −0.535 | −2.075, 2.045 | |
| (2) | Neocortex ratio | 0.152 | 1.043 | −1.992, 2.036 |
| Male group size | 0.510 | 3.504 | −2.051, 2.046 | |
| Degree of arboreality | −0.163 | −1.048 | −2.024, 2.047 | |
| (3) | Neocortex ratio | 0.204 | 1.236 | −1.994, 2.017 |
| Degree of arboreality | −0.147 | −0.796 | −2.060, 2.079 | |
| Alpha’s share | −0.420 | −2.463 | −2.054, 2.057 | |
| (4) | ECV | 0.301 | 1.027 | −2.001, 2.009 |
| Male group size | 0.432 | 1.954 | −2.056, 2.023 | |
| Degree of arboreality | −0.124 | −0.599 | −2.060, 2.040 | |
| Alpha’s share | −0.126 | −0.528 | −2.069, 2.028 | |
| (5) | ECV | 0.272 | 1.012 | −2.007, 1.999 |
| Male group size | 0.525 | 3.536 | −2.064, 2.049 | |
| Degree of arboreality | −0.186 | −1.063 | −2.066, 2.058 | |
| (6) | ECV | 0.306 | 0.993 | −2.006, 2.009 |
| Degree of arboreality | −0.136 | −0.623 | −2.053, 2.034 | |
| Alpha’s share | −0.451 | −2.486 | −2.067, 2.066 | |
| Dependent variable: frequency of male coalitions | ||||
|---|---|---|---|---|
| Predictor variables | Standardized estimate | t value | 95% CI | |
| (1) | Neocortex ratio | 0.173 | 1.096 | −2.005, 2.019 |
| Male group size | 0.416 | 1.926 | −2.071, 2.028 | |
| Degree of arboreality | −0.120 | −0.680 | −2.049, 2.046 | |
| Alpha’s share | −0.121 | −0.535 | −2.075, 2.045 | |
| (2) | Neocortex ratio | 0.152 | 1.043 | −1.992, 2.036 |
| Male group size | 0.510 | 3.504 | −2.051, 2.046 | |
| Degree of arboreality | −0.163 | −1.048 | −2.024, 2.047 | |
| (3) | Neocortex ratio | 0.204 | 1.236 | −1.994, 2.017 |
| Degree of arboreality | −0.147 | −0.796 | −2.060, 2.079 | |
| Alpha’s share | −0.420 | −2.463 | −2.054, 2.057 | |
| (4) | ECV | 0.301 | 1.027 | −2.001, 2.009 |
| Male group size | 0.432 | 1.954 | −2.056, 2.023 | |
| Degree of arboreality | −0.124 | −0.599 | −2.060, 2.040 | |
| Alpha’s share | −0.126 | −0.528 | −2.069, 2.028 | |
| (5) | ECV | 0.272 | 1.012 | −2.007, 1.999 |
| Male group size | 0.525 | 3.536 | −2.064, 2.049 | |
| Degree of arboreality | −0.186 | −1.063 | −2.066, 2.058 | |
| (6) | ECV | 0.306 | 0.993 | −2.006, 2.009 |
| Degree of arboreality | −0.136 | −0.623 | −2.053, 2.034 | |
| Alpha’s share | −0.451 | −2.486 | −2.067, 2.066 | |
Statistical models include neocortex ratio or ECV and degree of arboreality in combination with male group size and alpha’s share (models 1 and 4) or male group size only (2 and 5) or alpha’s share only (3 and 6). A variable was considered to be significant if the t value in the observed data was outside the 95% CI of the simulation-based null distribution (marked bold, see main text for explanations). Standardized estimates were obtained by z-transforming all predictor and response variables. All models with ECV included log body mass as a covariate. ANCOVA, analysis of covariance.
Results of 3 additional analyses confirm the robustness of the results presented in Table 1. These analyses include: 1) nonphylogenetic statistical models, 2) analyses where species with male philopatry were excluded, and 3) analyses that considered uncertainty in the phylogenetic tree (see Supplementary Data [ESM]).
DISCUSSION
Despite the long-standing view that cognitive abilities may impose a limit on the use of coalitions in animals (Harcourt 1992; Tomasello and Call 1997; Dunbar and Shultz 2007), no support for this hypothesis was found in the current study. Instead, our results indicate that coalitionary behavior was more likely to be the product of socioecological factors that promote larger male group sizes and reduce female monopolizability (see also Olson and Blumstein 2009). The strong, negative association between male group size and contest potential (as indexed by the expected alpha’s share of matings), however, made it impossible to disentangle their independent effects in the present comparative analysis.
There are a number of (nonmutually exclusive) reasons why living with more males may provide more opportunities for coalitionary activity and thus stronger selection pressure on that behavior (Henzi et al. 1999). Coalitions can only form if a sufficient number of animals are in a position to do so, and the chance of finding a coalition partner is expected to increase the more that potential partners are in close spatial proximity (e.g., Vogel et al. 2007; Higham and Maestripieri 2010). Whether or not a coalition is successful is mostly determined by the combined strength of the allies relative to their target (e.g., Noë 1994; Noë and Sluijter 1995; Bissonnette et al. 2009). Because average power differentials among males tend to decrease with increasing male group size, an increase in male group size can lead to an increase in the proportion of triads in which 2 weaker males can form a successful coalition against single stronger male (Noë 1994). A group size effect could also arise if coalition formation is an alternative reproductive strategy that is employed only by a subset of coresident males. In this case, larger groups are more likely to contain enough males that engage in extensive coalitionary activity (e.g., Alberts et al. 2003; Bissonnette et al. 2011). Finally, larger male group size may lead to improved partner choice where unsatisfactory partners can be abandoned and replaced by more reliable ones, which should positively influence the emergence and stability of coalitions (Noë and Hammerstein 1994; Barrett et al. 2007). Future within-group studies should help determine whether the incidence of male coalitions is a facultative response to current group size, or whether they represent an evolved (inflexible) response reflecting the taxon’s evolutionary ecology (e.g., Henzi et al. 1999).
If the variation in the incidence of male coalitions crucially depends on the availability of competent partners, it is nevertheless the case that males should form coalitions only if this form of cooperation increases their reproductive success (van Hooff and van Schaik 1992, 1994; Chapais 1995; van Schaik 1996; van Schaik et al. 2004). Males are expected to use alternative reproductive strategies such as coalitions to obtain access to females when contest competition is appreciable and some individuals tend to be excluded from mating (van Hooff and van Schaik 1994; van Schaik 1996). When they succeed, coalitions should be profitable because of improved rank (e.g., Witt et al. 1981; Schülke et al. 2010) or improved access by coalition partners to fertile females (e.g., Bercovitch 1988; Noë and Sluijter 1990; Berghänel et al. 2010; Bissonnette et al. 2011; Young et al. 2013), unless they are prohibitively costly in terms of time, energy, and/or death (van Schaik et al. 2006). Although the degree to which participation in coalitions is costly has been debated and is probably variable (Mesterton-Gibbons et al. 2011), there is some suggestion that coalitionary costs may be prohibitive for certain species characterized by high contest potential. The extreme rarity or complete absence of male coalitions in the tufted capuchin monkey, for example, has been suggested to reflect extremely high costs based on the observation that, when there is an overthrow of the alpha male, the alpha and the challenger(s) usually are severely injured or killed (e.g., C. nigritus, Ramirez-Llorens et al. 2008, Tiddi B, Wheeler B, personal communication; C. apella, Janson, C, personal communication). However, the closely related white-faced capuchin (C. capucinus) exhibits a high mortality risk in male–male competition together with customary coalitionary behavior (Perry 2012). Thus, additional factors, such as lesser reproductive skew due to long alpha male tenure, are needed to explain why coalitions are far more common in the white-faced capuchins than in the tufted capuchin species (see Perry 2012 for a discussion). It is also possible that high contest potential species are characterized by a steeper distribution of fighting abilities (van Schaik et al. 2004, 2006). In that case, the number of feasible triads in which 2 weaker males can beat a single stronger one will always be low, thus exacerbating the effect of small cohort size. At the proximate level, intense contest competition may hinder coalitionary behavior through reduced tolerance, a necessary factor in cooperation and coalition formation (e.g., Melis et al. 2006; Hare et al. 2007; Olson and Blumstein 2009). This argument has been recently put forward by Olson and Blumstein (2009) to explain why “complex” male coalitions are less likely to be found in those mammalian species where females can be easily monopolized by males. A high degree of competition may increase the risk of aggression between the coalition partners thus making it more difficult for this behavior to emerge (cf., Fragaszy and Visalberghi 1990), unless it also leads to enhanced social mechanisms to cope with aggression (Preuschoft and Paul 2000). Future studies quantifying the benefits and costs of coalitions as well as male tolerance levels in relation to contest potential would be important for understanding the existing variation and potential constraints on male coalitions.
Our study raises a fundamental question regarding the degree to which complex cognitive capacities are required to engage in coalition formation (Hemelrijk and Puga-Gonzalez 2012). It is often assumed that coalitionary behavior is cognitively complex because it relies on the ability to recognize interactions and relationships in which the observer is not directly involved (Harcourt 1992; Tomasello and Call 1997). However, the extent to which coalitionary decisions are based on “smart thinking” in the here and now (Barrett et al. 2007) remains unclear. There is now extensive evidence of at least a rudimentary form of third-party recognition in many species of animals, including those lacking large brains (review in Cheney 2011). Even in primates where the understanding of tertiary relations is pervasive (Tomasello and Call 1997), not all species make use of this knowledge when they form coalitions. In fact, it is often the case that cognitive shortcuts (i.e., evolved rules-of-thumb, cf., Gigerenzer et al. 1999) can account for the patterns observed (e.g., Range and Noë 2005; Bissonnette et al. 2009). For example, it has been suggested that males form differentiated affiliative bonds with other males and as a special rule-of-thumb preferentially form coalitions with closely bonded partners instead of making cognitively demanding decisions every time they go into a polyadic conflict (e.g., Berghänel et al. 2011). Given that a wide range of mammalian species, and some birds, are now known to engage in coalitions as defined in the current study (reviewed by Smith et al. 2010; Bissonnette et al. in review, Behaviour), the absence of consistent effect of brain size on the incidence of coalition in primate males is not entirely unexpected. Thus, our study suggests that most primates, and probably other mammals as well, are above the minimal cognitive capacity threshold required to perform this behavior (cf., Bowles and Gintis 1976).
It is still possible that animals with larger brains use coalitions more efficiently or in more varied ways than those with smaller brains (cf., Harcourt 1992). For example, studies in chimpanzees suggest that males compete for powerful partners in several ways: they perform “separating interventions” to keep rivals apart and may even adopt coercive tactics by supporting others against those who have previously formed coalitions against them (e.g., de Waal 1982; Nishida 1983; Nishida and Hosaka 1996). What has been called “political” coalitionary behavior in chimpanzees (de Waal 1982; see also Schülke et al. 2010) may hinge on more sophisticated cognitive skills than, for example, the “gang-attack” coalitions described in rhesus macaque males (Higham and Maestripieri 2010). Detailed empirical data are needed from both natural and laboratory studies to investigate how much cognition is required to implement different coalitionary tactics. For example, empirical data on some processes that may have high cognitive demands, such as active communication between partners about the process of coalition formation, the degree of coordination among the participants, perhaps even postconflict punishment of uncooperative partners, are needed to elucidate the cognitive requirements for successful coalition formation. Since this kind of data is currently unavailable for most species, we can conclude at the very least that coalitionary behavior does not necessarily rely on advanced cognition (Dugatkin 1998; Jennings et al. 2009; Hemelrijk and Puga-Gonzalez 2012).
It is likely that additional factors, such as the variance in male relatedness within a group (as distinct from philopatry in itself, which is strongly female biased in group living primates, Broom et al. 2009) or the reliance on females as primary coalition partners (e.g., Surbeck et al. 2011), may also affect male coalitionary behavior. These factors, however, were not available for a sufficient number of species and only additional data will allow testing of these suggestions. For now, we can conclude that the variation in the incidence of male coalitions in primates has more to do with constraints imposed by the socioecological setting than with any major cognitive differences among the taxa. These processes are probably general to many social systems and we expect the patterns that we report here to apply to a wide range of mammalian species living in mixed-sex groups (Olson and Blumstein 2009).
SUPPLEMENTARY MATERIAL
Supplementary material can be found at Supplementary Data
FUNDING
This work was supported by the Max Planck Society; the Funds of the German Initiative for Excellence to the University of Göttingen; the Alexander von Humboldt Foundation to A.B.; and the Fonds de recherche Société et culture of the province of Québec (FQRSC) to A.B.
We are grateful to all the primatologists listed in the Supplementary Data (ESM) as well as M. Bowler, J. Chism, R. Hilgartner, E. Huchard, L. Isbell, L. Matthews, and T. O’Brien, for kindly accepting to provide data for their study species. For training in phylogenetic comparative methods, we thank C. Nunn and the AnthroTree Workshop, which was supported by the National Science Foundation (BCS-0923791). We thank C. van Schaik and C. Dubuc for inspiring discussions at an earlier stage of this work, and M. Kiebs for helping with the literature search. We are grateful to L. Barrett and S. Bowles for providing insightful comments on an earlier version of the manuscript. We also thank the Editor and 2 anonymous reviewers for their comments.
