Oecologia (2007) 151:140–149 DOI 10.1007/s00442-006-0553-6 BEHAVIORAL ECOLOGY Network metrics reveal diVerences in social organization between two Wssion–fusion species, Grevy’s zebra and onager Siva R. Sundaresan · Ilya R. FischhoV · Jonathan DushoV · Daniel I. Rubenstein Received: 18 February 2006 / Accepted: 15 August 2006 / Published online: 9 September 2006 © Springer-Verlag 2006 Abstract For species in which group membership fre- associate assortatively by reproductive state in Grevy’s quently changes, it has been a challenge to characterize zebra but not in onagers. The current approach dem- variation in individual interactions and social structure. onstrates the utility of network metrics for identifying Quantifying this variation is necessary to test hypothe- Wne-grained variation among individuals and popula- ses about ecological determinants of social patterns tions in association patterns. From our analysis, we can and to make predictions about how group dynamics make testable predictions about behavioral mecha- aVect the development of cooperative relationships nisms underlying social structure and its eVects on and transmission processes. Network models have transmission processes. recently become popular for analyzing individual con- tacts within a population context. We use network met- Keywords Individual associations · Equids · Animal rics to compare populations of Grevy’s zebra (Equus groups · Social structure grevyi) and onagers (Equus hemionus khur). These closely related equids, previously described as having the same social system, inhabit environments diVering Introduction in the distribution of food, water, and predators. Grevy’s zebra and onagers are one example of many Emergence of social structure from individual relation- sets of coarsely similar Wssion–fusion species and popu- ships is a key organizing problem for the study of ani- lations, observed elsewhere in other ungulates, pri- mal behavior (Couzin and Krause 2003; Hinde 1976; mates, and cetaceans. Our analysis of the population Whitehead and Dufault 1999). In turn, social structure association networks reveals contrasts consistent with shapes the development of social relationships, such as their distinctive environments. Grevy’s zebra individu- dominance or cooperation (Hemelrijk 1999), and als are more selective in their association choices. transmission of ideas (McComb et al. 2001), pathogens Grevy’s zebra form stable cliques, while onager associ- (Keeling and Eames 2005), and genes (Altmann et al. ations are more Xuid. We Wnd evidence that females 1996). Describing variation in individual association choices and social structure is particularly challenging in populations where individuals frequently change associates, as groups form and disintegrate (Chapman Communicated by Marc Mangel. et al. 1993; Whitehead 1997; Whitehead and Dufault W S. R. Sundaresan (&) · I. R. FischhoV · 1999). Many mammals exhibit such ssion–fusion pat- J. DushoV · D. I. Rubenstein terns, including equids (Ginsberg 1987; Rubenstein Department of Ecology and Evolutionary Biology, 1986), chimpanzees (Symington 1990), dolphins Princeton University, Princeton, NJ 08544, USA (Brager 1999), buValos (Cross et al. 2005), and e-mail: [email protected] humans. Among these species, we observe great varia- S. R. Sundaresan · I. R. FischhoV · D. I. Rubenstein tion in ecology and sociality, including group size and Mpala Research Center, P.O. Box 555, Nanyuki, Kenya degree of mixing among individuals. How can we 123 Oecologia (2007) 151:140–149 141 characterize the distinctions in sociality among Wssion– individual association patterns. Ecological variation is fusion populations? Do individuals form preferred known to underlie diVerences across populations in bonds, or do associations reXect chance encounters? social structure (Rubenstein 1994; Rubenstein and How does phenotype shape individuals’ decisions Wrangham 1986). about group membership? Until we characterize the Our study populations diVer in key ecological patterns and processes of associations, it is diYcult to aspects that we expect drive distinct patterns of indi- understand how ecology drives social behavior (Krause vidual association and, therefore, population-level net- and Ruxton 2002; Rubenstein and Wrangham 1986), or work properties. Onagers are found in open desert how population social structure feeds back on social habitat and can therefore easily Wnd other conspeciWcs relationships and constrains transmission processes. if they break an association. In the bushier habitat of Social networks oVer a powerful set of tools for Grevy’s zebra, it may be diYcult to Wnd new associates characterizing and analyzing individual associations after leaving a group. If a Grevy’s zebra individual within a population-level social context (Croft et al. wants to be in a group, then it is important for it to 2004; Lusseau 2003; Newman 2003; Wasserman and choose associates with whom it can easily coexist, Faust 1994). Biologists have modeled networks in because opportunities to meet other associates are which individuals are vertices and behavioral interac- infrequent. Thus we expect Grevy’s zebra to prefer tions deWne edges linking individuals. Depending on associates with common needs, and to form cliques the process of interest, an interaction could be deWned through persistent group membership. In onagers, on as shared group membership, sexual contact, aYlia- the other hand, ease of exchanging associates reduces tion, or antagonism. Using networks, we can quantify the beneWt of choosing particular individuals as associ- complementary aspects of sociality: population sub- ates. structure, individual variability, and association prefer- Grevy’s zebra and onagers diVer further in preda- ences. Based on the association matrix that deWnes a tion danger and forage distribution in ways that predict network, researchers have searched for clusters using larger optimal groups in Grevy’s zebra than in onagers. various algorithms (Whitehead and Dufault 1999). Our Grevy’s zebra population faces predation, while Lusseau and Newman (2004) found that key individu- the onagers do not (Moehlman 2002). Compared to als maintain social structure in dolphin networks. Size Grevy’s zebra, the more arid onagers’ environment has and sex can aVect social preferences in dolphins and sparser forage (S. R. Sundaresan, I. R. FischhoV, and guppies (Croft et al. 2005; Lusseau and Newman 2004). D. I. Rubenstein, unpublished data). Once in a group, Network structure has been used to make predictions predation danger is expected to result in high costs for about the course of disease or innovation spread (Keeling Grevy’s zebra of leaving a group by oneself, compared and Eames 2005) to onagers. Furthermore, reduced forage competition We apply network methods to quantify diVerences is expected to result in a low cost of maintaining group between two Wssion–fusion equids having broadly simi- associations in Grevy’s zebra; by comparison, move- lar social organization: Grevy’s zebra (Equus grevyi) ment into sparser forage may prompt onagers to break and onagers (Equus hemionus khur). We deWne bonds associations. Taken together, ecological factors predict in the network based on shared group membership diVerences between the populations in group sizes and because we are interested in individual association the individual association patterns that shape social choices. Grevy’s zebra and onagers are grazers inhabit- structure. ing arid and semi-arid landscapes. Social organization We use network metrics to compare social structure in both species falls under the category of resource between our study populations, and among reproduc- defense polygyny: males seek to control access to tive classes within each population. We extend the util- unstable groups of females by defending areas with ity of network metrics that have previously provided critical resources (Ginsberg 1987; Klingel 1998; Ruben- powerful tools for analyzing variation within a single stein 1994; Rubenstein 1986). In Grevy’s zebra, diVer- population. We use frequency of membership in the ences in water needs of lactating and nonlactating same group as a measure of association strength among females result in partial segregation among females by dyads. We deWne two types of associations: those reproductive status; it is not known how lactation state between individuals seen together at least once and aVects female associations in onagers (Ginsberg and those seen together more than would be expected Rubenstein 1990). Past studies on the two species have through random encounters (Bejder et al. 1998). On a described coarsely similar unstable female groups. But continuum of social bond strength, these two criteria this past research has left unexamined possible represent two points that may be relevant to diVerent diVerences between populations of the two species in social processes. 123 142 Oecologia (2007) 151:140–149 Using recent advances in network theory (Newman distances (hundreds of meters to several kilometers) 2003; Watts and Strogatz 1998), we demonstrate how separating groups. Within each group encountered, we simple network metrics provide richer insights into identify individuals and classify their reproductive sta- association patterns than are possible from examina- tus. In this paper, we use three reproductive classes: tion of group size and association strength distributions males, lactating females, and nonlactating females. We alone. We show that Grevy’s zebra individuals