Measuring social complexity

SPECIAL ISSUE: SOCIAL EVOLUTION Animal Behaviour xxx (2015) 1e7

Contents lists available at ScienceDirect

Animal Behaviour journal homepage: www.elsevier.com/locate/anbehav

Special Issue: Social Evolution

Measuring social complexity Thore J. Bergman a, b, *, Jacinta C. Beehner a, c a

Department of Psychology, University of Michigan, Ann Arbor, MI, U.S.A. Department of Ecology and Evolutionary Biology, University of Michigan, Ann Arbor, MI, U.S.A. c Department of Anthropology, University of Michigan, Ann Arbor, MI, U.S.A. b

a r t i c l e i n f o Article history: Received 24 November 2014 Initial acceptance 8 February 2015 Final acceptance 16 February 2015 Available online xxx MS. number: SI-14-00953 Keywords: cognition differentiated relationships evolution intelligence social complexity hypothesis

In one of the first formulations of the social complexity hypothesis, Humphrey (1976, page 316, Growing Points in Ethology, Cambridge University Press) predicts ‘that there should be a positive correlation across species between social complexity and individual intelligence’. However, in the many ensuing tests of the hypothesis, surprisingly little consideration has been given to measures of the independent variable in this evolutionary relationship, that is, social complexity. Here, we seek to encourage more rigorous measures of social complexity. We first review previous definitions of this variable and point to two common flaws; a lack of objectivity and a failure to directly connect sociality to the use of cognition. We argue that, rather than creating circularity, including cognition in the definition of social complexity is necessary for accurately testing the social complexity hypothesis. We propose a new definition of social complexity that is based on the number of differentiated relationships that individuals have. We then demonstrate that the definition is both broadly applicable and flexible, allowing researchers to include more detailed information about the degree of differentiation among individuals when the data are available. While we see this definition of social complexity as one possible way forward, our larger goal is to encourage researchers examining the social complexity hypothesis to carefully consider their measurement of social complexity. © 2015 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd.

If you say that someone ‘lacks social skills’, this is an unambiguous insult that reveals a fundamental feature of how we think about social cognition; being social requires certain abilities that are beyond the reach of some people. In addition to informing how we think about other people, this idea forms the basis for evolutionary ideas about social cognition across species. The social complexity hypothesis (independently arrived at by Jolly (1966) and Humphrey (1976)) posits that sociality is cognitively challenging and, consequently, drives cognitive evolution. As an evolutionary hypothesis, the social complexity hypothesis can be tested by making comparisons across species, ideally using comparative methods that control for the pattern of relatedness among the species in the sample (MacLean et al., 2012). As Humphrey (1976, page 316) states, the hypothesis ‘demands that there should be a positive correlation across species between social complexity and individual intelligence’. Indeed, in the nearly 40 years since Humphrey's statement, numerous studies have reported exactly such a relationship. For example, across primate species there is a well-known correlation

* Correspondence: T. J. Bergman, Department of Psychology, University of Michigan, Ann Arbor, MI 48109, U.S.A. E-mail address: [email protected] (T. J. Bergman).

between the size of the social group (used as an indicator of social complexity) and the relative size of the neocortex (used as an indicator of intelligence, Dunbar, 1995). As this example demonstrates, testing the social complexity hypothesis across species requires quantifying (or at least ranking) both the ‘social complexity’ and the ‘individual intelligence’ of species. Certainly, there has been extensive debate about how to measure ‘intelligence’ (or cognition, e.g. Healy & Rowe, 2007), the dependent variable in Humphrey's stated relationship above. Indeed, broad comparisons using crude measures such as brain size are increasingly being replaced or augmented with more targeted comparisons among closely related species that compare experimentally assessed cognitive performance (Bond, Kamil, & Balda, 2003; MacLean et al., 2013). However, despite urgings by Healy and Rowe (2007) and others (e.g. Holekamp, 2007) to improve measures of ‘social complexity’ (the independent variable in the relationship), there has been considerably less progress in this area. Indeed, we are struck that, despite dissatisfaction with the most widely used measure of social complexity (group size; e.g. Shultz & Dunbar, 2007), an alternative has yet to be widely adopted. Moreover, all measures currently employed have flaws, being either too subjective or only tangentially related to social cognition.

http://dx.doi.org/10.1016/j.anbehav.2015.02.018 0003-3472/© 2015 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd.

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018

2

T. J. Bergman, J. C. Beehner / Animal Behaviour xxx (2015) 1e7

Therefore, our goal in this manuscript is primarily to encourage more objective and systematic measurements of social complexity. After reviewing previous measures, we propose a new measure of social complexity based on the number of differentiated relationships that members of a species have with conspecifics. We demonstrate how this definition can be expanded to facilitate finer-grained comparisons when more detailed data are available. Finally, we illustrate how this definition can be applied using three examples. TERMINOLOGY There are several different terms for evolutionary hypotheses that connect sociality and cognition: ‘the social complexity hypothesis’ (e.g. Connor, Smolker, & Richards, 1992) , ‘ the social brain hypothesis’ (e.g. Dunbar, 1998), ‘the social intelligence hypothesis’ (e.g. Kummer, Daston, Gigerenzer, & Silk, 1997) and ‘the Machiavellian intelligence hypothesis’ (e.g. Byrne & Whiten, 1988). Although research under each hypothesis may emphasize different aspects of sociality or cognition (e.g. only the Machiavellian hypothesis emphasizes deception as a key component; Byrne & Whiten, 1988), they are nevertheless all similar in that they propose a causal evolutionary link from sociality to cognition. Here, we use the term ‘social complexity hypothesis’ because it is broadly applicable and because it places emphasis where we feel it should be placed; that is, on the independent variable in the relationship, sociality. This emphasis avoids logically confounding the independent (sociality) and dependent (cognition) variables. In practice, it can be difficult to completely separate the independent and dependent variables because over shorter timescales the causation can go in the other direction: greater cognitive ability may enable more complex social interactions. COMPLEXITY It is perhaps also worth considering the broader issue of complexity. Although easy to intuit, the term ‘complexity’ itself is difficult to quantify. Most attempts to quantify complexity focus on either the number of parts in a system, the relationship between the parts, or both. For example, the definition of McShea and Brandon (2010, page 7) captures both of these concepts: ‘the number of part types or degree of differentiation among parts’. Definitions of social complexity have also tended to focus on these two concepts (more parts, or more differentiation between parts) with varying degrees of explicitness. For example, the widely used variable ‘group size’ directly measures the number of parts in the system. PREVIOUS MEASURES OF SOCIAL COMPLEXITY Social Features as Indicators of Complexity A common way to compare social complexity across species is to use particular aspects of sociality as indicators of complexity. For example, a species with a linear dominance hierarchy may be more socially complex than a species that lacks a dominance hierarchy (Maclean, Merritt, & Brannon, 2008). Other features that have been used to indicate social complexity include: pair bonds (Shultz & Dunbar, 2007), complex alliances (Connor, 2007), flexible nesting strategies (Tibbetts, 2004), foraging as a permanent flock (Bond et al., 2003), social transactional interactions (Burish, Kueh, & Wang, 2004), a lack of reproductive skew (Pawłowskil, Lowen, & Dunbar, 1998), tactical deception (Byrne & Corp, 2004) and
ches (Krasheninnikova, Bra €ger, & Wanker, 2013). Although this cre approach appears an objective one, the use of arbitrary features can be quite subjective. With a limited number of species, it is nearly

always possible to identify one particular social feature that corresponds to greater (or lesser) cognitive ability. Thus, in the absence of any a priori reason to consider a social feature as cognitively challenging, this method makes it difficult to falsify the hypothesis. This problem is exacerbated if the social feature is not clearly defined. For example, as pointed out by Beauchamp and Fern andezJuricic (2004), the use of ‘transactional’ interactions as an indicator of complex bird societies by Burish et al. (2004) does not clearly relate to published accounts of each species' sociality. Instead, the variable appears to allow large-brained (but mostly solitary) birds such as woodpeckers (Picidae) to count as socially complex ndez-Juricic, 2004). Furthermore, the social (Beauchamp & Ferna features typically used in such comparisons are often only tangentially (if at all) related to cognition. For example, while
ches sets the stage for more complex social interactions forming cre in parrots (Krasheninnikova et al., 2013), there is little social
che formation; it merely requires attraccognition inherent in cre tion to other nests and tolerance of other offspring. Without some direct link to cognition, comparative patterns linking social features to cognition may be entirely spurious. Moreover, noncognitive definitions of sociality put researchers at risk of missing actual correlations between sociality and cognition in their taxa. Qualitative Comparisons of Social Systems Rather than identifying a particular social feature as an indicator of social complexity, some researchers have made qualitative comparisons across different types of social systems. For example, fissionefusion societies have been suggested to be more complex than other vertebrate societies (Amici, Aureli, & Call, 2008; Aureli et al., 2008). Fissionefusion societies with multiple levels may be even more complex (Couzin, 2006). Conversely, in bats, stable groups may be more complex than unstable groups because they are more likely to involve cooperative behaviours (Wilkinson, 2003). Within ungulates, monogamy has been proposed as more complex than alternative mating patterns (Shultz & Dunbar, 2006). Conversely, in primates, multimale systems may be more complex than the alternatives (Shultz & Dunbar, 2007). Yet, this approach is also subject to the aforementioned problems; (1) researchers may be tempted to retro-fit the social system ranking to the cognitive data in hand, and (2) they may do so with no grounding in cognition. As such, sociocognitive relationships may not be detected accurately. Quantitative Comparisons of Social Systems Many researchers rely on quantitative comparisons of social features across species, comparing species based on the number of X, with X being some social variable. The most common such variable is the number of individuals in a social unit (group size, e.g. Dunbar, 1995), but other variables include grooming clique size (Kudo & Dunbar, 2001), colony size (Wilkinson, 2003), or the number of demographic roles (Blumstein & Armitage, 1998). Quantitative measures are a dramatic improvement over other measures because they are objective and avoid issues of circular reasoning. Quantitative comparisons have an inherent order to them because complexity increases with more ‘parts’. Thus, quantitative comparisons are less fungible than qualitative ones. However, quantitative comparisons are still susceptible to finding spurious correlations; or, equally problematic, can obscure true relationships because the measured social variable has no direct relationship to cognition. For example, the use of group size has been criticized because it obscures underlying variation in how animals are interacting within those groups (Shultz & Dunbar, 2007).

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018

T. J. Bergman, J. C. Beehner / Animal Behaviour xxx (2015) 1e7

3

Our primary goal in this review is to encourage the use of objective measures of social complexity. Whether the definition of social complexity is applied within a particular taxon or across many taxa, an objective definition is necessary to ensure the integrity and repeatability of research under the social complexity hypothesis. Because (admittedly) our intuitions about social complexity can be quite strong, we may feel it is not necessary to objectively measure it. However, clear a priori criteria for assessing complexity are necessary to avoid any (surely, unconscious!) post hoc retrofitting of data to hypotheses.

The evolution of spatial cognition in seed-caching birds makes this point more directly. Scatter hoarding of seeds is thought to select for greater spatial cognition but only because cognition is used in recovering seeds (Balda, Kamil, & Bednekoff, 1996; Clayton, 1998; Shettleworth, 1990). If the hoarders relied on scent or random searches to recover the seeds (rather than a memory of the location the seed was placed), there would be no reason to expect an evolutionary relationship between seed-caching and spatial memory ability. Similarly, if cognition is not needed to cope with social complexity, there is no reason to expect an evolutionary relationship between social complexity and cognition. As for circularity, we feel that this only becomes an issue if our measures of cognition are the same as our measures of social complexity. For example, if we only measured prey speed when they are being pursued by predators, this creates logical problems. Faster predators might push prey into faster chases resulting in a mechanistic, rather than evolutionary, relationship between predator and prey speed. The logic behind both hypotheses is that the selection pressure (i.e. predator speed, or in our case, social complexity) leads to an ability (i.e. prey speed, or in our case, cognitive ability) with utility that extends beyond the situation where selection occurs. In the prey speed example, testing for the ability might be accomplished by placing a prey species on a treadmill and measuring their maximum running speed. In the seed-caching example, this means testing seed-caching birds on generic spatial memory tasks rather than (or in addition to) the exact task of recovering cached seeds (Balda et al., 1996). So, too, should social cognition be measured in a different context than the measure of social complexity.

Incorporating Cognition in Definitions of Social Complexity

New Definition: Number of Differentiated Relationships

Additionally, we propose that the criteria one uses for assessing social complexity must relate directly to cognition. That is, there should be some reason for thinking the aspect of sociality being measured is cognitively challenging. Otherwise, we are subject to finding spurious correlations or missing true relationships between sociality and cognition. Spurious correlations are a particular problem in this area because social variables are often correlated with ecological ones (the basis of socioecology; Terborgh & Janson, 1986). Thus, it can be particularly difficult to tease apart social and ecological drivers of cognitive evolution, underscoring the importance of having a direct link between the social measure and the use of social cognition. One could argue that including cognition in a definition of social complexity is creating precisely the same type of circularity that we are trying to avoid. That is, if social complexity is measured cognitively, is it any surprise then, that it is correlated with cognition? We disagree and describe two analogies that we hope will clarify our position that incorporating cognition in the definition of social complexity is necessary. As it may be easier to contemplate the evolution of physical traits rather than cognitive ones, we first describe a hypothesis relating to the running speed of herbivores (i.e. prey). The hypothesis is that the running speed of prey is driven by the speed of the carnivores (i.e. predators) that feed on them (Bro-Jørgensen, 2013). This is directly analogous to the social complexity hypothesis because the independent variable (predator running speed) is a selective pressure driving the evolution of the dependent variable (prey running speed). The hypothesis is supported by comparative data with one important caveat: only the speed of pursuit predators (and not ambush predators) relates to the speed of the prey (BroJørgensen, 2013). In other words, the hypothesis only holds in cases where speed is used in avoiding capture by predators. Similarly, we would only expect the social complexity hypothesis to hold when social cognition is used in coping with social complexity.

Our definition builds on the core idea that the use of cognition is a critical component of social complexity. Mechanistically, more ‘parts’ can create greater complexity; however, more individuals do not necessarily create greater social complexity (i.e. individuals may not be differentiating among each other). For example, in a migratory herd of tens of thousands of wildebeest, Connochaetes taurinus, it is unlikely that animals on one end of the herd are even aware that animals on the other end are there, much less using social cognition to monitor them. Therefore, adding additional members to the herd does not require any additional cognition. If individuals do not recognize or categorize conspecifics surrounding them, the actual number of individuals in the herd (or group) is meaningless (with respect to cognition). Therefore, we propose that social complexity should be measured as the number of differentiated relationships that individuals have. By differentiated relationships, we simply mean the number of relationships that can be distinguished by an observer; that is, the number of consistently different interactions that are seen. If members of a species treat all conspecifics exactly the same, the number of differentiated relationships is 0. If they treat every individual they regularly encounter differently (as in many primate groups where interactions are differentiated based on factors such as relative rank, kinship, consort history and age; Cheney & Seyfarth, 2008), then the number of differentiated relationships is the number of individuals they regularly encounter (i.e. N
1, where N is the group size for such species). Most species will fall in between these two extremes because they may, for example, treat their offspring (or mate, or close kin, or nestmates) differently from ‘the rest’, but ‘the rest’ are treated the same. This may be why group size/brain size relationships hold in primates (where the number of differentiated relationships often approaches the size of the group) but not in other taxa (where the number of stable interactions indicative of relationships is often lower than the size of the group; Shultz & Dunbar, 2007). Below, we provide some examples of how to apply this definition.

A recent quantitative definition of social complexity by Freeberg, Dunbar, and Ord (2012, page 1787) is the most detailed to date: ‘Complex social systems are those in which individuals frequently interact in many different contexts with many different individuals, and often repeatedly interact with many of the same individuals over time’. The definition appears quantitative because terms such as ‘frequently’ and ‘many’ lend themselves to enumeration. We feel this definition avoids some of the problems with previous definitions, although it fails to incorporate cognition (cognition being only implicit in the last part of the definition). Additionally, despite its more quantitative treatment of social interactions, this definition is difficult to distill down to a single value that can be used in comparative studies. IMPROVEMENTS TO MEASURES OF SOCIAL COMPLEXITY Objective Measures

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018

4

T. J. Bergman, J. C. Beehner / Animal Behaviour xxx (2015) 1e7

We see several potential criticisms of this definition. First, some may argue that this definition is circular. However, this approach is only circular if the ability to differentiate social partners was our measure of cognition. In such a case, a positive correlation between social complexity and cognition would be a fait accompli. But, under the social complexity hypothesis, sociality is expected to favour the evolution of cognition that can be applied beyond specific social interactions. Thus, as long as the cognitive measure is different from the social measure, the issue of circularity is avoided. Alternatively, one could argue that a measure of social complexity should be more cognitive. That is, we should start with specific tasks with known cognitive components (e.g. a single neuron mediates the unconditioned stimulus in honeybee olfactory learning; Hammer, 1993) and build up from there to understand the cognition required for different social interactions (Lihoreau, Latty, & Chittka, 2012). Such a bottomeup approach would certainly allow us to make more specific predictions about the cognitive correlates of sociality while also potentially addressing puzzles about why large-bodied animals need so many neurons to do what insects can do far more efficiently (Chittka & Skorupski, 2011). However, our goal here is to facilitate comparative studies across a broad range of taxa. Therefore, we prefer a topedown approach based on observable social interactions; differentiated relationships are both observable in the wild and linked to social cognition. Although incorporating more cognition in the measure of social complexity might be an improvement, it would make it harder to apply. In the next section we describe ways this definition can be expanded in this direction when the data are available. One final criticism might be that this measure of social complexity will be more difficult to apply than previous measures because it requires having more detailed information about social interactions in a set of taxa than many previous measures (such as group size). In answer to this, we argue that this difficulty is more than overcome by the following four advantages that are gained from this definition. (1) It is objective. Once the number of differentiated relationships has been assessed, there is a clear criteria for comparing complexity: more relationships means more complexity. (2) It directly links sociality and cognition. Differentiated relationships require cognition. Distinguishing among individuals requires discrimination, and consistent interactions across time requires memory. Thus, this measure focuses on the aspects of sociality that themselves might be cognitively challenging. (3) It is widely applicable, allowing for comparisons across broad taxonomic categories. Differentiated relationships have been observed in many animal phyla (e.g. many Vertebrata, including lizards: Carazo, Font, & Desfilis, 2008; and fish: Riggio, 1985; Insecta: Tibbetts, 2002; Mollusca: Tricarico, Borrelli, Gherardi, & Fiorito, 2011; Crustacea: Gherardi & Tiedemann, 2004; Arthropoda: Whitehouse, 1997). (4) It is flexible. As we discuss in the next section, the number of differentiated relationships is a useful starting point for more finegrained ways to measure social complexity. Importantly, this definition captures the main parts of Freeberg et al.’s (2012) definition given above. The number of differentiated relationships combines the points: ‘… individuals frequently interact…with many different individuals…often repeatedly interact with many of the same individuals over time’ into a single quantity. As discussed below, the only remaining section of their definition (‘… in many different contexts…’) can be included in elaborations of our definition. Elaborations on the Measure Our secondary goal in this review is to offer a simple measure of social complexity that can be readily applied in comparative studies

that is analogous to using group size, while, at the same time, demonstrating a stronger connection to cognition. In some systems, however, it may be useful to have a more detailed measure of social complexity. In this section, we discuss ways to expand on this core definition when more detailed social data are available. First, we may wish to consider changes in relationships across time. In most cases, we envision that the number of differentiated relationships will be a static ‘snap-shot’ in time (as with previous measures, such as group size). However, if individuals differentiate among group members and group members are continuously changing, then it is reasonable to expect that learning new individuals should require additional cognitive abilities. This may be one reason why longevity has recently emerged as a correlate of brain size in mammals (Gonzalez-Lagos, Sol, & Reader, 2010). All else being equal, longer-lived animals will have to differentiate among more individuals across their lifetime than will shorter-lived animals. Thus, in such dynamic systems, it may be useful to consider the number of differentiated relationships in a lifetime (or some proxy for lifetime). This adds additional difficulty to the measure (requiring estimates of both longevity and relationship turnover rate), and thus, may not be as broadly applicable. Second, it may be useful to consider the degree of differentiation among relationships. In many animal societies, relationships may be differentiated along a single dimension such as dominance rank. However, in some species, such as chacma baboons, Papio hamadryas ursinus, and Japanese macaques, Macaca fuscata, there is evidence that relationships simultaneously vary across both dominance rank and kinship (Bergman, Beehner, Cheney, & Seyfarth, 2003; Schino, Tiddi, & Di Sorrentino, 2006). In such cases, researchers may need to consider both the number of dimensions along which relationships differ and the number of differentiated relationships (e.g. one could simply sum the number of relationships across dimensions). Additionally, dynamic fissionefusion societies have a temporal dimension to conspecific interactions that stable societies lack (Amici et al., 2008). If this temporal dimension involves cognition (that is, if temporal overlap is diagnostic of a relationship), then fissionefusion societies would ‘count’ as more complex than stable societies of the same size. Additionally, it may be useful to consider how differentiated relationships are, with greater differentiation indicating greater complexity. That is, if all of an individual's relationships cluster together in some measure of sociality (while still being discernable), this is arguably less complex than if all of an individual's relationships are spread out along the social dimension. For example, a primate with 10 social relationships that differ slightly in how often they affiliate (e.g. groom) is presumably less complex than a primate with 10 social relationships that range from highly antagonistic to highly affiliative; the latter requires a broader array of responses to be linked to the same number of stimuli. Furthermore, in the latter case, if five partners cluster at the antagonistic end and the other five cluster at the affiliative end, this is arguably simpler than if the 10 partners are spread throughout the range, again because a greater diversity of responses are required in the second scenario. Together, considering both the number of dimensions and how differentiated relationships are, can be thought of as the volume of relationship space that an individual's social network occupies. Obviously, such detailed data will rarely be available in comparative studies, but these may be useful issues to consider, particularly when comparing closely related species. Finally, researchers have argued that specific interactions like complex alliances (Connor, 2007) and consolation (de Waal & Aureli, 1996) are indicators of extreme social complexity. Such interactions can be incorporated into the above definition of social complexity by considering that differentiated relationships can be context dependent. That is, an individual might consistently treat

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018

T. J. Bergman, J. C. Beehner / Animal Behaviour xxx (2015) 1e7

another individual differently depending on who else is around (the basis of strategic alliance partner choice; Perry, Barrett, & Manson, 2004; Silk, 1999) or what has happened recently (the basis of postconflict behaviours like reconciliation and consolation; de Waal & Aureli, 1996). Therefore, an individual can have multiple ‘relationships’ with the same individual that intricately depend on the immediate social context. In such situations, the number of differentiated relationships can increase exponentially, leading to extreme scores for social complexity in a species. Of course, the risk here is that the well-studied taxa may tend to have a higher complexity score than the less well-studied taxa. Therefore, it will be important to consider duration of contact time in natural settings to accurately assess such context-dependent differences. EXAMPLES We do not provide a single method for detecting differentiated relationships because the nature of interactions and the data available varies so widely across taxa. Measures such as association patterns, proximity, aggression, affiliation, grooming, joint displays, behavioural coordination, co-feeding or food sharing may be useful indicators of relationships in some species but not others. The main concern is that the data used to detect differentiated relationships are objective and repeatable. Here we apply our measure of social complexity to three different situations to illustrate the broad utility of the definition as well as some of the issues that may arise when going from behavioural data to the number of differentiated relationships. (1) Geladas, Theropithecus gelada Geladas are Old World monkeys with a multilevel society, fissionefusion dynamics, and very large group sizes (the largest aggregations can number over 1100 individuals; Snyder-Mackler, Beehner, & Bergman, 2012). Gelada society comprises dozens of ‘reproductive units’: a single, dominant ‘leader’ male, related adult females (N ¼ 1e13) and their immature offspring, natal juveniles, and zero to three subordinate ‘follower’ males (typically about 20 individuals in all, Dunbar, 1984). These units congregate to form three higher social levels with varying degrees of overlap across time: (1) ‘teams’ (two to three units, together more than 90% of the time; note that teams are rare, and less than one-third of units are in a team); (2) ‘bands’ (5e20 units, together more than 50% of the time); and (3) the ‘community’ (>50 units, together >0% of the time). The association patterns are discontinuous, with clear clusters of associations at the team and band level (Snyder-Mackler et al., 2012). Given these different levels of society, how socially complex are geladas? Based on previous qualitative comparisons of social systems, they appear very complex: they exhibit fissionefusion dynamics and multiple social levels, two indicators of social complexity. Furthermore (depending on which social level is considered the ‘group’), previous quantitative measures such as group size also peg geladas as being very complex (indeed, geladas have among the largest natural aggregations of any nonhuman primate; Snyder-Mackler et al., 2012). However, let us now consider the complexity of gelada society using the proposed method of the number of differentiated relationships. First, within units, both males and females exhibit dominance hierarchies and social interactions among females are strongly biased towards close kin (Johnson, Snyder-Mackler, Beehner, & Bergman, 2014), indicating differentiated relationships. However, outside of units, males do not recognize the other members of their band (Bergman, 2010; le Roux & Bergman, 2012). Furthermore, there are no consistent interactions across units in the same band (e.g. we do not see

5

dominance relationships or affiliation between units, Bergman & Beehner, n.d.). In other words, the level of differentiated relationships an individual could have with other community members does not reflect the number of differentiated relationships they do have with one another (Bergman, 2010). Rather than indicating a true relationship, different degrees of social overlap (which is how we delineate gelada bands) may simply result from different degrees of shared home ranges. Thus, for geladas, we would argue that the number of differentiated relationships is only the number of individuals in the reproductive unit (or perhaps the team). Teams are ambiguous because most units are not in a team (at the time of the playback experiment, we only had a single team among the 17 units that we followed), so we have been prevented from drawing strong conclusions by small sample sizes, a problem we continue to address. Our measure of gelada social complexity corresponds to what Dunbar (1995) called the ‘cognitive group size’ of geladas. It also places gelada social complexity below that of their close relatives, the baboons (Papio spp.), whose single-level, but highly differentiated, societies often number more than 80 individuals. Note that this lack of social complexity in geladas is only in relation to social cognition; gelada society itself is undoubtedly a multilevel society with fissionefusion dynamics, the causes and consequences of which are only beginning to be understood (Snyder-Mackler, Alberts, & Bergman, 2014). (2) Pair Bonds in Nonprimate Mammals In one of the broadest tests of the social complexity hypothesis to date, Shultz and Dunbar (2007) found very different relationships between sociality (measured by social organization categories) and cognition (measured as relative brain size) across a wide array of taxa. Specifically, in carnivores, ungulates and bats, pair bonds are associated with the largest relative brain size, and species with multimale groups have small relative brain sizes (although see Finarelli & Flynn, 2009, for a critique of the carnivore work). By contrast, in primates, pair-bonded species score low on relative brain size and multimale groups score the highest. To explain this difference between primate and nonprimate systems, Shultz and Dunbar demonstrated that primate social relationships are more ‘intense’ than those in other mammals, and they suggest that this drives the difference. That is, in primates, multimale groups contain many strong bonds (essentially a pair bond extended to every member of the group), whereas in other mammals, pair bonds represent a strong social bond that stands in sharp contrast to the loose aggregations seen in multimale societies. We agree with this interpretation. However, we also suggest that using the number of differentiated relationships would further allow us to quantify these differences. For example, in the nonprimate taxa, a pair bond represents a differentiated relationship that loose aggregations of males and females in multimale societies do not have. Thus, the pair-bonded societies are more complex than the others. In primates, differentiated relationships generally extend to the whole group such that multimale groups have many differentiated relationships (e.g. a male dominance hierarchy, a female dominance hierarchy, mate preferences and kin-structured relationships) and are more complex than pair-bonded primates that have fewer differentiated relationships in their comparatively small groups. (3) Social Insects and Castes Much research has been devoted to accurately describing the social complexity of eusocial insects (e.g. Anderson & McShea, 2001). The degree of differentiation between castes and the extent of the division of labour are indicators of social complexity

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018

6

T. J. Bergman, J. C. Beehner / Animal Behaviour xxx (2015) 1e7

that invite interesting comparisons to human societies (Sterelny, 2007), yet they are not applicable to other animal societies. To apply our definition of social complexity to eusocial insects requires us to expand the definition of ‘differentiated relationships’ from individuals to castes. It is also possible to have a differentiated relationship with a class or a group of individuals, treating all members of a class the same while differentiating among classes. In eusocial insects, interactions between individuals are structured by their caste membership. These interactions do not require recognition in the same way that individual relationships do because caste membership can be assessed from the phenotype of the individual at the time of the interaction. Treating members of different casts differently does, however, require discrimination and some way to guide the behavioural response, without which it would be impossible to respond appropriately to the phenotypic cues of caste membership. Thus, caste-level relationships have many of the cognitive components of differentiated individual relationships, and we would categorize the number of differentiated relationships in such societies as the number of castes. Note that it is still possible to have differentiation at the group level but also differentiate relationships within that group, adding to the complexity. For example, honeybees treat some members of the same caste differently (Herb et al., 2012), suggesting more differentiated relationships than just the number of castes. Furthermore, chacma baboons simultaneously make group-level (based on matriline) and individual (based on rank) distinctions when interacting with others (Bergman et al., 2003). Thus, group-level differentiation can also create additional complexity beyond the total number of individuals in a group. SUMMARY Our main objective was to point out the need to bring more rigour and standardization to measures of social complexity. We first reviewed the previous ways researchers have measured social complexity and identified two major flaws in these measures: (1) a lack of objectivity and/or (2) a failure to link the measure of social complexity to cognition. Our second objective was to provide a versatile definition of social complexity based on the number of differentiated relationships that individuals have. A key point that we hoped to make is that including cognition in the definition of social complexity (since maintaining differentiated relationships requires cognition) does not create problems of circularity for the social complexity hypothesis. Rather, it ensures a more accurate appraisal of the evolutionary relationship between sociality and cognition. We then demonstrated the flexibility and broad applicability of this new definition. While we do not expect that all future researchers will adopt this definition, we do hope that this discussion will encourage more systematic measures of social complexity. Acknowledgments We thank Robert Seyfarth and Dorothy Cheney for the opportunity to contribute to this special issue. This work was supported by the University of Michigan. For our gelada research, we are grateful to the Ethiopian Wildlife Conservation Department and the wardens and staff of the Simien Mountains National Park for granting us the permission to conduct research. Funding for the gelada project was provided by the Wildlife Conservation Society (SSF grant number 67250), the National Geographic Society (grant number 8100-06), the Leakey Foundation (multiple grants), the National Science Foundation (grant number BCS-0715179, BCS0962118, IOS-1255974) and the University of Michigan. The gelada research was approved by the University Committee on Use and

Care of Animals (UCUCA) at the University of Michigan and adhered to the laws and guidelines of Ethiopia. References Amici, F., Aureli, F., & Call, J. (2008). Fission-fusion dynamics, behavioral flexibility, and inhibitory control in primates. Current Biology, 18(18), 1415e1419. http:// dx.doi.org/10.1016/j.cub.2008.08.020. Anderson, C., & McShea, D. W. (2001). Individual versus social complexity, with particular reference to ant colonies. Biological Reviews, 76(2), 211e237. http:// dx.doi.org/10.1017/s1464793101005656. Aureli, F., Schaffner, C. M., Boesch, C., Bearder, S. K., Call, J., Chapman, C. A., et al. (2008). Fissionefusion dynamics. Current Anthropology, 49(4), 627e654. Balda, R. P., Kamil, A. C., & Bednekoff, P. A. (1996). Predicting cognitive capacity from natural history. Current Ornithology, 13, 33e66. ndez-Juricic, E. (2004). Is there a relationship between Beauchamp, G., & Ferna forebrain size and group size in birds? Evolutionary Ecology Research, 6(6), 833e842. Bergman, T. J. (2010). Experimental evidence for limited vocal recognition in a wild primate: implications for the social complexity hypothesis. Proceedings of the Royal Society B: Biological Sciences, 277(1696), 3045e3053. Bergman, T. J., & Beehner, J. C. (n.d.). [Individual focal data on inter-unit social interactions]. Unpublished raw data. Bergman, T. J., Beehner, J. C., Cheney, D. L., & Seyfarth, R. M. (2003). Hierarchical classification by rank and kinship in baboons. Science, 302(5648), 1234e1236. Blumstein, D. T., & Armitage, K. B. (1998). Life history consequences of social complexity: a comparative study of ground-dwelling sciurids. Behavioral Ecology, 9(1), 8e19. Bond, A. B., Kamil, A. C., & Balda, R. P. (2003). Social complexity and transitive inference in corvids. Animal Behaviour, 65, 479e487. http://dx.doi.org/10.1006/ anbe.2003.2101. Bro-Jørgensen, J. (2013). Evolution of sprint speed in african savannah herbivores in relation to predation. Evolution, 67(11), 3371e3376. Burish, M. J., Kueh, H. Y., & Wang, S.-H. (2004). Brain architecture and social complexity in modern and ancient birds. Brain, Behavior and Evolution, 63(2), 107e124. Byrne, R. W., & Corp, N. (2004). Neocortex size predicts deception rate in primates. Proceedings of the Royal Society B: Biological Sciences, 271(1549), 1693. Byrne, R. W., & Whiten, A. (Eds.). (1988). Machiavellian intelligence: Social expertise and the evolution of intellect in monkeys, apes, and humans. Oxford, U.K.: Clarendon Press. Carazo, P., Font, E., & Desfilis, E. (2008). Beyond ‘nasty neighbours’ and ‘dear enemies’? Individual recognition by scent marks in a lizard (Podarcis hispanica). Animal Behaviour, 76(6), 1953e1963. Cheney, D. L., & Seyfarth, R. M. (2008). Baboon metaphysics: The evolution of a social mind. Chicago, IL: University of Chicago Press. Chittka, L., & Skorupski, P. (2011). Information processing in miniature brains. Proceedings of the Royal Society B: Biological Sciences, 278(1707), 885e888. Clayton, N. (1998). Memory and the hippocampus in food-storing birds: a comparative approach. Neuropharmacology, 37(4), 441e452. Connor, R. C. (2007). Dolphin social intelligence: complex alliance relationships in bottlenose dolphins and a consideration of selective environments for extreme brain size evolution in mammals. Philosophical Transactions of the Royal Society B: Biological Sciences, 362(1480), 587e602. Connor, R. C., Smolker, R. A., & Richards, A. F. (1992). Two levels of alliance formation among male bottlenose dolphins (Tursiops sp.). Proceedings of the National Academy of Sciences of the United States of America, 89(3), 987e990. Couzin, I. D. (2006). Behavioral ecology: social organization in fissionefusion societies. Current Biology, 16(5), R169eR171. Dunbar, R. I. M. (1984). Reproductive decisions: An economic analysis of gelada baboon social strategies. Princeton, NJ: Princeton University Press. Dunbar, R. I. M. (1995). Neocortex size and group size in primates: a test of the hypothesis. Journal of Human Evolution, 28(3), 287e296. Dunbar, R. I. M. (1998). The social brain hypothesis. Evolutionary Anthropology, 6, 178e190. Finarelli, J. A., & Flynn, J. A. (2009). Brain-size evolution and sociality in Carnivora. Proceedings of the National Academy of Sciences of the United States of America, 106(23), 9345e9349. Freeberg, T. M., Dunbar, R. I. M., & Ord, T. J. (2012). Social complexity as a proximate and ultimate factor in communicative complexity. Philosophical Transactions of the Royal Society B: Biological Sciences, 367(1597), 1785e1801. http://dx.doi.org/ 10.1098/rstb.2011.0213. Gherardi, F., & Tiedemann, J. (2004). Chemical cues and binary individual recognition in the hermit crab Pagurus longicarpus. Journal of Zoology, 263(1), 23e29. Gonzalez-Lagos, C., Sol, D., & Reader, S. (2010). Large-brained mammals live longer. Journal of Evolutionary Biology, 23(5), 1064e1074. Hammer, M. (1993). An identified neuron mediates the unconditioned stimulus in associative olfactory learning in honeybees. Nature, 366(6450), 59e63. http:// dx.doi.org/10.1038/366059a0. Healy, S. D., & Rowe, C. (2007). A critique of comparative studies of brain size. Proceedings of the Royal Society B: Biological Sciences, 274(1609), 453e464. http://dx.doi.org/10.1098/rspb.2006.3748.

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018

T. J. Bergman, J. C. Beehner / Animal Behaviour xxx (2015) 1e7 Herb, B. R., Wolschin, F., Hansen, K. D., Aryee, M. J., Langmead, B., Irizarry, R., et al. (2012). Reversible switching between epigenetic states in honeybee behavioral subcastes. Nature Neuroscience, 15(10), 1371e1373. Holekamp, K. E. (2007). Questioning the social intelligence hypothesis. Trends in Cognitive Sciences, 11(2), 65e69. Humphrey, N. K. (1976). The social function of intellect. In P. P. G. Bateson, & R. A. Hinde (Eds.), Growing points in ethology (pp. 303e317). Cambridge, U.K.: Cambridge University Press. Johnson, E. T., Snyder-Mackler, N., Beehner, J. C., & Bergman, T. J. (2014). Kinship and dominance rank influence the strength of social bonds in female geladas (Theropithecus gelada). International Journal of Primatology, 35(1), 288e304. Jolly, A. (1966). Lemur social behavior and primate intelligence: step from prosimian to monkey intelligence probably took place in a social context. Science, 153(3735), 501e506. http://dx.doi.org/10.1126/science.153.3735.501. Krasheninnikova, A., Br€ ager, S., & Wanker, R. (2013). Meanseend comprehension in four parrot species: explained by social complexity. Animal Cognition, 16(5), 755e764. Kudo, H., & Dunbar, R. (2001). Neocortex size and social network size in primates. Animal Behaviour, 62(4), 711e722. Kummer, H., Daston, L., Gigerenzer, G., & Silk, J. (1997). The social intelligence hypothesis. In P. Weingart, S. D. Mitchell, P. J. RIcherson, & S. Maasen (Eds.), Human by nature: Between biology and the social sciences (pp. 157e179). Hillsdale, NJ: L. Erlbaum. Lihoreau, M., Latty, T., & Chittka, L. (2012). An exploration of the social brain hypothesis in insects. Frontiers in Physiology, 3, 442. http://dx.doi.org/10.3389/ fphys.2012.00442. MacLean, E. L., Matthews, L. J., Hare, B. A., Nunn, C. L., Anderson, R. C., Aureli, F., et al. (2012). How does cognition evolve? Phylogenetic comparative psychology. Animal Cognition, 15(2), 223e238. http://dx.doi.org/10.1007/s10071-011-0448-8. Maclean, E. L., Merritt, D. J., & Brannon, E. M. (2008). Social complexity predicts transitive reasoning in prosimian primates. Animal Behaviour, 76, 479e486. http://dx.doi.org/10.1016/j.anbehav.2008.01.025. MacLean, E. L., Sandel, A. A., Bray, J., Oldenkamp, R. E., Reddy, R. B., & Hare, B. A. (2013). Group size predicts social but not nonsocial cognition in lemurs. PLoS One, 8(6), e66359. McShea, D. W., & Brandon, R. N. (2010). Biology's first law: The tendency for diversity and complexity to increase in evolutionary systems. Chicago, IL: University of Chicago Press. Pawłowskil, B., Lowen, C., & Dunbar, R. (1998). Neocortex size, social skills and mating success in primates. Behaviour, 135(3), 357e368. Perry, S., Barrett, H. C., & Manson, J. H. (2004). White-faced capuchin monkeys show triadic awareness in their choice of allies. Animal Behaviour, 67, 165e170. http:// dx.doi.org/10.1016/j.anbev.2003.04.005. Riggio, R. J. (1985). Acoustically mediated individual recognition by a coral reef fish (Pomacentrus partitus). Animal Behaviour, 33, 411e416.

7

le Roux, A., & Bergman, T. J. (2012). Indirect rival assessment in a social primate, Theropithecus gelada. Animal Behaviour, 83(1), 249e255. Schino, G., Tiddi, B., & Di Sorrentino, E. P. (2006). Simultaneous classification by rank and kinship in Japanese macaques. Animal Behaviour, 71, 1069e1074. Shettleworth, S. J. (1990). Spatial memory in food-storing birds. Philosophical Transactions of the Royal Society B: Biological Sciences, 329(1253), 143e151. Shultz, S., & Dunbar, R. I. M. (2006). Both social and ecological factors predict ungulate brain size. Proceedings of the Royal Society B: Biological Sciences, 273(1583), 207e215. Shultz, S., & Dunbar, R. I. M. (2007). The evolution of the social brain: anthropoid primates contrast with other vertebrates. Proceedings of the Royal Society B: Biological Sciences, 274(1624), 2429e2436. http://dx.doi.org/10.1098/ rspb.2007.0693. Silk, J. B. (1999). Male bonnet macaques use information about third-party rank relationships to recruit allies. Animal Behaviour, 58, 45e51. http://dx.doi.org/ 10.1006/anbe.1999.1129. Snyder-Mackler, N., Alberts, S. C., & Bergman, T. J. (2014). The socio-genetics of a complex society: female gelada monkey relatedness patterns mirror association patterns in a multi-level society. Molecular Ecology, 23, 6179e6199. http:// dx.doi.org/10.1111/mec.12987. Snyder-Mackler, N., Beehner, J. C., & Bergman, T. J. (2012). Defining higher levels in the multilevel societies of geladas (Theropithecus gelada). International Journal of Primatology, 33(5), 1054e1068. Sterelny, K. (2007). Social intelligence, human intelligence and niche construction. Philosophical Transactions of the Royal Society B: Biological Sciences, 362(1480), 719e730. http://dx.doi.org/10.1098/rstb.2006.2006. Terborgh, J., & Janson, C. (1986). The socioecology of primate groups. Annual Review of Ecology and Systematics, 17, 111e136. Tibbetts, E. A. (2002). Visual signals of individual identity in the wasp Polistes fuscatus. Proceedings of the Royal Society B: Biological Sciences, 269(1499), 1423e1428. Tibbetts, E. A. (2004). Complex social behaviour can select for variability in visual features: a case study in Polistes wasps. Proceedings of the Royal Society B: Biological Sciences, 271(1551), 1955e1960. Tricarico, E., Borrelli, L., Gherardi, F., & Fiorito, G. (2011). I know my neighbour: individual recognition in Octopus vulgaris. PLoS One, 6(4), e18710. de Waal, F. B., & Aureli, F. (1996). Consolation, reconciliation, and a possible cognitive difference between macaques and chimpanzees. In A. E. Russon, K. A. Bard, & S. T. Parker (Eds.), Reaching into thought: The minds of the great apes (pp. 80e110). Cambridge, U.K.: Cambridge University Press. Whitehouse, M. E. (1997). Experience influences maleemale contests in the spider Argyrodes antipodiana (Theridiidae: Araneae). Animal Behaviour, 53, 913e923. Wilkinson, G. S. (2003). Social and vocal complexity in bats. In F. B. de Waal, & P. L. Tyack (Eds.), Animal social complexity (pp. 322e341). Cambridge, MA: Harvard University Press.

Please cite this article in press as: Bergman, T. J., & Beehner, J. C., Measuring social complexity, Animal Behaviour (2015), http://dx.doi.org/ 10.1016/j.anbehav.2015.02.018