A Practical Guide to the Study of Social Relationships

Evolutionary Anthropology 22:213–225 (2013)

ARTICLE

JOAN SILK, DOROTHY CHENEY, AND ROBERT SEYFARTH

Behavioral ecologists have devoted considerable effort to identifying the sour- raised several related questions: How ces of variation in individual reproductive success.1–5 Much of this work has are relationships among individuals focused on the characteristics of individuals, such as their sex and rank. How- patterned? What are the proximate ever, many animals live in stable social groups and the fitness of individuals mechanisms that mediate the impact depends at least in part on the outcome of their interactions with other group of social interactions on individuals? members. For example, in many primate species, high dominance rank enhan- What are the long-term adaptive conse- ces access to resources and reproductive success.4,5 The ability to acquire and quences of social relationships for indi- maintain high rank often depends on the availability and effectiveness of coali- viduals? What do primates know about tionary support.6 Allies may be cultivated and coalitions may be reinforced by their own relationships and the rela- affiliative interactions such as grooming, food sharing, and tolerance.7,8 These tionships of others? findings suggest that if we want to understand the selective pressures that The absence of an established toolkit shape the social behavior of primates, it will be profitable to broaden our focus for describing the properties of social from the characteristics of individuals to the properties of the relationships that relationships complicates efforts to they form with others. The goal of this paper is to discuss a set of methods that answer the first question, and this can be used to quantify the properties of social relationships. inevitably affects our ability to answer the others. Thus, our primary goal is to present a set of procedures for quanti- Although primatologists have long on individuals for several reasons. fying the properties of social relation- emphasized the importance of social First, we are used to thinking about ships. Our efforts are strongly relationships, in most studies of how natural selection shapes behav- influenced by Jeanne Altmann’s influ- social behavior the individual, not ioral traits, and natural selection ential efforts to provide a rigorous the dyad, is the unit of analysis. acts on individuals, not dyads. Sec- foundation for behavioral data collec- Researchers have focused primarily ond, dyadic data present a statistical tion. We also discuss experimental headache because dyads are not methods that have been developed for independent. Although powerful new probing primates’ perception of their relationships, as well as noninvasive Joan Silk is Professor of Anthropology in statistical procedures enable us to the School of Human Evolution and deal with this problem, not all of us procedures that have begun to illumi- Social Change at Arizona State Univer- are up to speed on these methods. nate the proximate physiological mech- sity. For most of her career, she has anisms linking behavior to adaptive studied the evolution of social behavior Third, behavioral data are not always outcomes. We then explore the associa- and reproductive performance of female dense enough to support dyadic macaques and baboons. Email: tion between the structure of social [email protected] analyses; in these cases, researchers relationships and fitness outcomes. We Dorothy Cheney and Robert Seyfarth may prefer to combine information are, respectively, Professor of Biology end with a discussion of questions that across individuals within specified and Psychology at the University of may direct future work. Pennsylvania. They have pioneered the categories such as close kin and use of naturalistic experiments to investi- peers. Moreover, there is no estab- gate vocal communication and social CONCEPTUAL FOUNDATIONS cognition in vervet monkeys and lished consensus about how measure baboons. the properties of social relationships. Current interest in social relation- The goal of this paper is to provide a ships builds on conceptual founda- Key words: observational methods; behavioral practical guide to the study of social tions established by Robert Hinde, analysis; methods; dyadic relationships; social relationships. We begin with a brief who pioneered the study of relation- bonds discussion of the conceptual founda- ships in primates.9,10 Hinde considered tions for the study of relationships that relationships to be the outcome of a

VC 2013 Wiley Periodicals, Inc. was originally developed by Robert contingent series of interactions DOI: 10.1002/evan.21367 Hinde and subsequently expanded by between two individuals. and empha- Published online in Wiley Online Library (wileyonlinelibrary.com). Hans Kummer. Hinde and Kummer sized the importance of constructing 214 Silk et al. ARTICLE descriptions of relationships from ob- tions of relationships from observa- dyads in which one partner becomes servational data on social interactions. tions of social interactions. Such attractive to the other for a limited In studies of mother-infant relation- descriptions would include informa- period (for example, a female in ships among rhesus macaques, Hinde tion about the content, quality, and estrus), and such relationships might and his colleagues showed how temporal sequence of interactions.10 be characterized as opportunistic or changes in the behavior of both Researchers have spent thousands of instrumental. mother and infant influenced the hours conducting focal observa- Some researchers have used the amount of time they spent in close tions15 that generate high-quality rate of affiliative behavior or the proximity during the infant’s matura- data about who does what to whom, absence of agonistic behavior as a tion.11–13 Hinde’s work showed that in how often, for how long, and in proxy for relationship quality.18 In many cases the behavior of an individ- what sequence. However, we do not general, dyads with high rates of ual was best predicted by the proper- have a well-developed consensus affiliation and/or low rates of aggres- ties of its relationship with others, not about how to quantify the character- sion are categorized as having good, by its own properties. For example, the istics of dyadic relationships.16 strong, or intense social relation- behavioral consequences of a brief sep- Any two animals that recognize ships. Rates of affiliation may be an aration of mother and infant were one another as individuals and meet important component of social rela- best predicted by the characteristics of repeatedly have some kind of relationships, but they may not fully the relationship before separation tionship; the empirical task is to capture their complex dynamics. occurred rather than by individual describe its characteristics. We sug- Our multi-dimensional view of attributes such as the infant’s age, sex, gest that relationships or social social relationships is similar in or the mother’s experience. Infants bonds occupy a multidimensional some ways to Cords and Aureli’s19 who showed the greatest distress were space, and that we can use behav- proposition that relationships can be those who, before the separation, had ioral data to map the contours characterized by three components: been relatively more active than their within that space. There are various compatibility, security, and value. mothers in maintaining physical con- dimensions along which relation- They defined compatibility as the tact. Jeanne Altmann’s classic mono- ships may differ.17 For example, two degree of tolerance within a dyad, it graph, Baboon Mothers and Infants, individuals may rarely interact or is therefore similar to what we have took Hinde’s approach to the field.10 interact often (frequency); they may termed the tenor of social relation- Hans Kummer integrated Hinde’s interact in a limited number or in a ships. Secure relationships are those conceptual formulation with the func- diverse range of behavioral contexts that are predictable and consistent tional perspectives of behavioral ecol- (diversity); their directional interac- over time, and thus partially overlap ogy. Kummer14 proposed that tions may be highly one-sided or with the dimension that we have relationships are long-term invest- evenly balanced (symmetry); their labeled consistency. Cords and Aureli ments that generate benefits for the interactions may range from mostly define relationship value in terms of members of the dyad. Kummer rea- hostile to mainly friendly (tenor); the benefits that individuals derive soned that the value of relationships they may be tense or relaxed when from their relationships. Our scheme would depend on the intrinsic qualities they are together (tension); they may has no direct analog for relationship of the members of the dyad, such as behave toward one another in a con- value, partly because the value of rank and sex; their tendency to act in sistent or inconsistent manner (pre- social interactions is so difficult to ways that increased or decreased each dictability); and they may interact at quantify. other’s fitness; and their availability to high rates for short periods or con- The procedures that we describe one another, which would affect their sistently at high rates over long peri- below provide a starting point for ability to act on their behavioral ten- ods (stability). This is a preliminary, efforts to quantify the dimensions of dencies. He proposed that individuals probably incomplete, list of the dyadic relationships. Many of these use social interactions to shape partic- dimensions of social relationships methods were developed for analyses ular aspects of their relationships. For that can be assessed with behavioral of baboon behavior, but ought to be example, one monkey might groom data. appropriate for a range of species. another to enhance the likelihood that The distribution of relationships We have not attempted to provide an the recipient would be motivated to along these types of continua can exhaustive review of all methods that maintain proximity and thus be avail- help us to characterize relationship have been used for assessing social able to provide coalitionary support. quality. Relationships that fall behavior, but focus on ones that Kummer assumed that individuals toward the right side of the continua seem to be particularly useful for selectively invested in relationships in Figure 1 could reasonably be cate- characterizing the quality of social that ultimately enhanced their fitness. gorized as close social bonds, relationships in sizable groups. For roughly similar to what most people some dimensions of relationships, QUANTITATIVE DESCRIPTIONS OF would think of as friendships. Rela- we currently lack appropriate meth- tionships that show high levels on ods. We fully expect that the proce- SOCIAL RELATIONSHIPS some dimensions but not others dures that we describe will be Hinde emphasized the importance might be characterized in different modified, improved, and expanded of constructing systematic descrip- ways. For example, there might be by others over time. ARTICLE A Practical Guide to the Study of Social Relationships 215

the duration of state variables is available, the proportion of time devoted to the activity can be computed by dividing the summed dura- tions by the amount of time observed. It is slightly more complicated to measure the frequency and duration of social interactions and proximity for species like chimpanzees, that live in fission-fusion groups, where opportunities to interact are influenced by both observation time and party composition. Thus, the number of interactions between two individuals, A and B, would be divided by the amount of time that A was observed when B was present in the same party and the amount of time that B was observed when A was present in the same party. Figure 1. The dimensions of social relationships. For animals that live in fission- fusion groups, the decision to join or leave a particular subgroup may be Observational Methods conspicuousness of behaviors, ad based on general preferences for libitum sampling remains an impor- being with others (gregariousness) or Analyses of social relationships are tant part of the observational toolkit preferences for particular individuals constructed from raw behavioral because it provides information within the subgroup. Pepper, Mitani, data. The advantages and disadvan- 20 about important social interactions and Watts devised a procedure to tages of various data collection that are uncommon or unpredict- distinguish between associations that schemes were first laid out in Jeanne able. For example, most observers arise from general gregariousness Altmann’s classic paper,15 which rely on ad lib observations of conflict and those that arise from preferen- remains the primary reference on this and competition to assess domi- ces for particular partners. The pair- topic. Altmann distinguished between nance relationships and to provide wise affinity index can be expressed behavioral events, which are instanta- information about the patterning of as: neous, and behavioral states, which coalitionary support. have a measurable duration. I n s s 21 Focal sampling, in which a single AB Ã 1 ið i Þ A s 21 B s 21 individual is observed for a predeter- ið i PÞÃ ið i Þ mined length of time and behavior is DIMENSIONS OF DYADIC P P continuously monitored, is particu- SOCIAL RELATIONSHIPS where IAB equals the number of larly useful for the study of dyadic Frequency of Events and times that individuals A and B were relationships because focal data can present in the same party, n repre- provide information about the form, Duration of States sents the number of parties frequency, and sequence of social The frequency of interaction is observed, and s represents party size. events, such as approaches and often used to evaluate relationship The summation term in the numera- vocalizations and the duration of quality. It is straightforward to tabu- tor represents the total number of social states, such as grooming or late the frequency of events from “neighbors” that individuals had in proximity. Observers can keep track focal data. However, it is often all parties that were formed. In a of which individual initiated social important to correct for differences party of size s, the number of possi- events or states and which individual in observation time, which can arise ble neighbors is equal to s(s21). This was responsible for ending them. if individuals die or disappear, or are value is computed for each party and Continuous monitoring of activity not sampled on the same schedule. summed across all parties observed. state and location during focal sam- Frequencies of event variables are The first term in the denominator ples can also provide useful informa- often converted to rates for this rea- represents the total number of neigh- tion about the context in which son.15 Rates are computed by divid- bors that A had in all of the parties interactions occur. ing the number of events in which that A was observed in. This is equal Despite its many problems, includ- an individual participated by the to the product of the number of ing biases due to differences in the amount of time that individual was times that A was in a party of size s observability of individuals and the observed. When information about and the number of neighbors in a 216 Silk et al. ARTICLE party of size s, or s21. The second the rate of behavior i for dyad xy; across dyads, such as grooming, term in the denominator represents and f i is the mean rate of behavior i have a particularly important impact the same calculation for B. Observed across all dyads in the subset of on relationship quality. If so, it values of the pairwise affinity index interest (for example, all group mem- would not be useful to standardize can be compared against values that bers or all adult females). The values the behavioral variables. At this would be obtained if individuals of this index range from 0 . point, we suggest that users compute !1 were randomly assigned to parties. High values of the Composite Social- both versions of the CSI and com- The randomization procedure retains ity Index (CSI) represent dyads that pare their predictive power. information about the observed dis- have more frequent and/or longer tribution of party sizes and the num- lasting affiliative interactions than ber of times that particular the average dyad in their group. Low Exploratory Principal Components individuals were observed in parties, values represent dyads that have less Analysis but reshuffles the composition of frequent and/or shorter affiliative parties to generate values expected interactions than the average dyad. Another approach is to use princi- on the basis of chance. pal components analysis (PCA) to reduce a set of correlated behavioral Composite Sociality Index variables to a smaller set of variables ...it is not clear whether that vary orthogonally and are thus There are often positive correla- independent. For example, in a study tions in the frequency with which par- composite measures of captive chimpanzees, Fraser, ticular dyads participate in different based on standardized Schino, and Aureli30 used explora- types of interactions. For example, or unstandardized varia- tory PCA to analyze inter- dyads that have high rates of associa- relationships across nine behavioral tion may also have high rates of bles are more biologi- variables. They found that these grooming. It is problematic to analyze cally meaningful. It is behaviors formed three distinct com- inter-correlated behaviors separately ponents, which they interpreted as because multiple tests increase the possible that behaviors relationship value, compatibility, and risk of committing Type 1 errors and with very uneven fre- security. the interactions are not independ- This is an intuitively appealing 21 ent. One solution is to combine dif- quencies across dyads, approach, but may not be entirely ferent types of interactions together such as grooming, have suitable for some behavioral data- into a single variable, but this is not a sets. PCA is most reliable when the useful solution if the absolute rates of a particularly important sample size is large and the ratio of behaviors differ (for example, impact on relationship observations to behavioral variables approaches are much more common is high.31 However, there are no firm than grooming initiations). Under quality. guidelines for sample size or the these conditions, aggregate measures ratio of subjects to variables. will be strongly biased by the most Another potential drawback of common types of behaviors. PCA is that it can be difficult to To address this problem, Sapolsky, interpret the factors that emerge 21 Alberts, and Altmann created a com- This basic procedure has been used from the analysis. For example, four posite measure based on the relative to characterize dyadic relationships behavioral variables (grooming, frequencies of positively correlated among baboons,22–24 several species proximity, support, and food shar- nonaggressive social interactions. of macaques,25–28 and wild horses.29 ing) loaded strongly on the first com- This measure was originally devel- One potential shortcoming of the ponent in Fraser, Schino, and Aureli’s oped to measure the degree of social CSI is that it does not account for analysis. They labeled this component integration of individuals. It was sub- differences in the amount of varia- as “value,” but others might assign a sequently modified to measure the tion in behavioral variables (R. Mun- different label, such as relationship strength of dyadic affiliative relation- strength or affiliativeness. 22 dry, personal communication). The ships. This measure, which we label contribution of variables to the com- the Dyadic Composite Sociality Index posite measure will be correlated Diversity of Behaviors to distinguish it from the individual- with their coefficient of variation. based measure, is computed with the For all behavioral variables to con- Observational data also provide following formula: tribute equally, they must be stand- information about the types of inter- ardized. However, it is not clear actions that dyads participate in and d fixy i 1 whether composite measures based the kinds of vocalizations that they ¼ f i DSIxy ¼ P d on standardized or unstandardized direct to one another. Some dyads variables are more biologically might exchange many different kinds Here, d is the number of behaviors meaningful. It is possible that behav- of interactions, while others might that contribute to the index; fixy is iors with very uneven frequencies participate in a narrower range of ARTICLE A Practical Guide to the Study of Social Relationships 217 interactions. The fact that rates of This measure is similar to van dence that grooming is highly unbal- various types of affiliative interac- Hoof and Wesling’s Directional Con- anced within the dyad. Values may tions are positively correlated across sistency Index (DCI),34 which was approach 1 if samples are large and dyads suggests that variation in the originally designed to assess the grooming is evenly balanced within range and diversity of interactions directionality of interactions at the the dyad or if samples are small and does exist. group level. The DCI, which has binomial probabilities of events are Measures of the number of been adapted to assess consistency therefore similar. It is possible to “behavioral diversity” factors are in the direction of interactions distinguish between these alterna- likely to be sensitive to observation within dyads,35 is computed as tives by computing the statistical time. Individuals observed often will (Fij2Fji)/(Fij 1 Fji), where Fij is equal power of the binomial probability have more opportunities to interact to the frequency of interactions estimate. than individuals that are observed directed by i to j, Fij equals the fre- The reciprocity index can be less often.32 This means that investi- quency of interactions directed by j rescaled to provide information gators need to assess how the diver- to i, and Fij > Fji. The DCI ranges about the pattern of skews in the dis- sity of interactions is related to from 0 when behaviors are perfectly tribution of events within dyads. If d observation time within their study balanced within dyads to 1 when > s, the adjusted reciprocity index, groups and account for this in their behaviors are unidirectional. One RI , equals 1 2 (0.5 3 RI). If d s, adj analyses. limitation of both these measures is RIadj 5 0.5 3 RI. RIadj approaches 1 that they are likely to be unreliable when d is much greater than s, and Symmetry when samples are small. approaches 0 when s is much greater than d. If d and s represent contribu- For directional social interactions, tions by the dominant and subordi- Reciprocity Index such as grooming, it is possible to nate members of a dyad, then when measure the contributions of each Skews in small samples are likely RIadj approaches 1, the dominant member of a dyad, and this variation to occur by chance when the behav- member of the dyad would be may be meaningful. For example, iors of interest occur infrequently or mainly responsible for grooming; Seyfarth’s33 hypothesis that females some pairs of individuals seldom when RIadj approaches 0, the subor- will trade grooming for support from interact. One solution to this prob- dinate individual would be mainly higher ranking individuals predicts lem is to limit the analysis to pairs responsible for grooming. Variables that low-ranking females will groom of animals that interact more often d and s could also represent older higher-ranking females more than than a threshold value, but it is not and younger members of dyads or they are groomed in return. To clear where this threshold should be males and females in mixed-sex assess this kind of prediction, we set. An alternative method, first dyads. As for the reciprocity index, need to assess the distribution of developed to analyze grooming extreme values of RIadj are easier to grooming within the dyad. among female baboons,36 is based interpret than intermediate values. Grooming Equality Index on the binomial probability of Again, a power analysis can help to observing x events in a sample of interpret the meaning of intermedi- Grooming can be evenly balanced x1y events, in which x and y are the ate values. within a dyad or one member of the contributions of each member of a pair can contribute more grooming dyad. The directional reciprocity Hinde Index than the other. The extent of groom- index is computed as the ratio ing equality within dyads can be between the cumulative binomial The Hinde Index, originally devel- assessed with the following probability of the smaller of the two oped for assessing the nature of 22 formula: values and the cumulative binomial mother-infant relationships among probability of the larger of the two rhesus macaques,37 provides an values. When these values are equal, alternative method for assessing the 12 Gij2Gji = Gij1Gji j½ð Þ ð Þj the probabilities are the same and extent of symmetry in social interac- the ratio is equal to 1. The reciproc- tions. The Hinde Index is calculated Here, Gij is the amount of time that ity index ranges from 0 when inter- as: individual i groomed individual j and actions are highly skewed to 1 when Gji is the amount of time that j interactions are perfectly balanced Ci Bi groomed i. The Grooming Equality within dyads. 1003 2 C 1C B 1B Index ranges from 1, when grooming It is important to point out that it i m i m is evenly balanced within the dyad, is more straightforward to interpret to 0 when grooming is completely values of the reciprocity index that In the case of physical contact one-sided. This measure could also are near 0 than values that approach between mothers and infants, Ci is be used to assess other types of 1. Values near 0 occur only when the number of times the infant directional interactions, such as the there is a substantial difference in makes contact with the mother, Cm distribution of food transfers or ini- the binomial probabilities of the two is the number of times the mother tiation of proximity. events and thus provide strong evi- makes contact with the infant, Bi is 218 Silk et al. ARTICLE

the number of times the infant Fij An alternative measure may be breaks contact with the mother, and F more useful. Perry, Barrett, and H 40 Bm is the number of times the 0 ij1 Manson divided the number of mother makes contact with the H focal samples that included friendly infant. Values of the Hinde index @ A or cooperative behavior by the num- where F is the rate of friendly inter- range from 2100 to 1100. High val- ij ber of focal samples that contained actions between i and j, F is the ues are generally interpreted to friendly, cooperative, or agonistic average rate of friendly interactions mean that i is mainly responsible for behavior. This index ranges from 0 across dyads, H is the rate of hostile maintaining contact with m; low val- ij in dyads that have exclusively hostile interactions between i and j, and H ues are interpreted to mean that m is interactions to 1 in dyads that have is the average rate of hostile interac- mainly responsible for maintaining exclusively friendly or cooperative tions across dyads. One problem contact with i. The variables i and m interactions. We used a very similar with this measure is that it is a ratio, could also represent male and female procedure, based on the rates of and ratios are ungainly. In this case, consort partners, high- and low- affiliative and aggressive interac- ranking members of a dyad, and so tions, to assess the tenor of relation- on. Similarly, in analyses of respon- ships among female baboons.23 sibility for proximity, C would represent approaches to within a given Analyses of the Tension distance and B would represent sequence of interactions leaves. Pairs of individuals may be relaxed Brown38 has pointed out that val- and the conditional or tense when they are together. In ues of the Hinde Index are difficult probability of particular some primate species, the rate of to interpret because different prox- self-directed behaviors (SDB), such imity patterns can generate the same outcomes could provide as scratching, provides a reliable values. A value of 0 could occur an index of predictabil- behavioral measure of stress and anxiety.41 While the rate of SDB has because i was responsible for an ity. For example, equal number of makes and breaks mainly been used to assess individuals’ response to conflict and reconcil- of contact or because m was respon- baboons and macaques 42 sible for all makes and breaks of iation, it can also be used to assess sometimes grunt as they the quality of relationships. For contact. To avoid this ambiguity, 43 Brown devised an alternative mea- approach others, and example, Castles found that female sure, which is calculated as: baboons displayed 40% higher rates these grunts seem to of self-directed behaviors when their serve as a predictable nearest neighbor was higher-ranking C 1B 1003 i i than themselves than when their C 1B 1C 1B signal that the caller will i i m m nearest neighbor was lower-ranking, not behave aggres- suggesting that proximity to higher- 46–48 The values of Brown’s index range sively. The use of ranking individuals caused anxiety. from 0 to 100. High values of this these kinds of signals index indicate that i is responsible Predictability most of the changes in contact and might vary across dyads low values indicate that m is respon- and provide a measure One of the benefits that individu- sible for most of the changes in als derive from close relationships is contact. of the predictability of greater predictability and sense of control.44,45 However, to our knowl- aggression. edge, no methods for assessing the Relationship Tenor predictability of social interactions It is common for pairs of individu- have been developed. Analyses of the als to exchange both affiliative and sequence of interactions and the con- aggressive behaviors. One way to ditional probability of particular out- assess the general tenor of a relation- the measure is undefined if the comes could provide an index of ship is to evaluate the relative fre- denominator is zero. This means predictability. For example, baboons quency of affiliative and hostile that it is possible to assess the qual- and macaques sometimes grunt as behaviors within the dyad. Weaver ity of relationships in dyads that they approach others, and these and de Waal39 devised a measure never engaged in affiliation (numera- grunts seem to serve as a predictable that is based on the ratio of the rela- tor equals 0) but did engage in con- signal that the caller will not behave 46–48 tive rate of affiliative to aggressive flict. However, it is not possible to aggressively. The use of these interactions. Their measure, which assess the quality of relationships kinds of signals might vary across they referred to as relationship qual- that engaged in affiliation but not dyads and provide a measure of the ity, is calculated as: conflict (denominator equals 0). predictability of aggression. ARTICLE A Practical Guide to the Study of Social Relationships 219

Temporal Stability grooming, and the stability of social equality index for each female with bonds among male chimpanzees. As all of her partners. In some cases, Methods for assessing the tempo- more information about the proper- however, different methods are ral stability of social relationships at ties of social relationships is accu- appropriate. the dyadic level are not well- mulated, we will be able to developed. One procedure to assess determine whether there are consist- Sociality Index this dimension of relationships ent patterns in their collective focuses on relationships with pre- Sapolsky, Alberts, and Altmann21 properties. ferred partners.23 In a study of the devised an individual measure of duration of close social bonds among sociality, or social integration, which female baboons, values of the dyadic Individual Measures of Sociality was subsequently modified to create CSI were first used to rank each the dyadic Composite Sociality female’s partners in each year. The Individuals may vary in the kinds Index. The individual Sociality Index, partners with the three highest CSI of relationships they form with SI, is calculated as: values were then categorized as the others. For example, some individuals may be more gregarious than female’s “top partners.” We tabulated d others. Similarly, some individuals bix the number of consecutive years in i 1 bi might allocate all of their grooming SIx ¼ which particular females appeared ¼ P d among the focal females’ three top to a small number of partners, while others might distribute grooming partners. For example, if female A where b is the rate of behavior i for across a larger number of partners. ix was observed over 5 years and individual x, and b is the median Thus, it is also useful to develop i female B was among her top three rate of behavior x for all individuals partners in years two, three, and in the subset of interest (for exam- four, their close social bond lasted 3 ple, all group members or all adult years. A 1-year gap between females), and d is the number of dif- “consecutive” years was allowed. The ferent types of behavior measured. rationale for allowing this gap was In some cases, individual The values of this index range from based on the arbitrariness of the measures of sociality can 0 . High values of the index rep- !1 threshold for identifying top partners resent individuals who are more and the realization that there might be derived directly from socially integrated than other are be temporary fluctuations in the dyadic data. For exam- members of their group; low values nature of females’ relationships from ple, to determine whether represent individuals who are less year to year. However, this decision socially integrated than other mem- has been criticized because it artifi- females vary in how well- bers of their group. cially inflates estimates of the tempo- balanced their grooming ral stability of social bonds.49,50 In Duration such analyses, it is important to con- relationships are, we There are several ways to assess trol for partner availability, as oppor- could compute the aver- the temporal stability of relation- tunities to form close social bonds ships at the individual level. Correla- are constrained by coresidence. age value of the groom- tions in partner rank orders between Mitani51 used a similar procedure to ing equality index for successive time periods can be com- assess the stability of social bonds puted using matrix correlation statis- among male chimpanzees. each female with all of tics.51,52 This procedure requires her partners. In some separate tests for each pair of suc- AGGREGATING INFORMATION cases, however, different cessive years, which raises the likeli- ABOUT THE PROPERTIES OF hood of committing Type 1 errors. methods are appropriate. To evaluate the consistency of part- SOCIAL RELATIONSHIPS ner rank orderings over longer times Once we have measured the vari- or more time periods, a nonparamet- ous dimensions of relationships, we ric Friedman’s test53 can be used. can explore the correlations among However, these methods cannot them. For example, we have discov- methods for assessing sociality at the accommodate changes in group ered that pairs of female baboons individual level. membership from one time period to having high CSI values also tend to In some cases, individual measures the next, which limits their have well-balanced grooming rela- of sociality can be derived directly usefulness. tionships and to maintain their rela- from dyadic data. For example, to We devised two alternative proce- tionships over time.23,24 Similarly, determine whether females vary in dures to assess variation in the Mitani51 documented correlations how well-balanced their grooming stability of social bonds and between the frequency of pairwise relationships are, we could compute to accommodate demographic associations, the symmetry of the average value of the grooming changes.54 For each individual in 220 Silk et al. ARTICLE

2S2U each year, partners are rank ordered. PPI EXPERIMENTAL STUDIES OF For each individual, the number of ¼ 2S2S2X RELATIONSHIP QUALITY AND different partners that the individual where S is the number of partner FUNCTION had within a given range of partner rank slots being evaluated, U is the ranks (for example, rank orders 123, Playback experiments using ani- number of different partners that the 426, 729) across time periods is mals’ natural vocalizations have individual had in those years, and X tabulated. The following formula is mainly been designed to plumb pri- is the number of top partners in the computed: mates’ knowledge about their social first time period who were not pres- world, but they can also provide valu- ent in the next time period. Individu- able insights about the nature and NS2U als with high values on the Partner PSI function of social interactions and ¼ NS2S Preference Index are ones that main- relationships. For example, we tained the same partners from one observed that female baboons some- where N is the number of years time period to the next; individuals times grunted to their former oppo- being considered, S is the number of with low values on the Partner Pref- nents after conflicts. We hypothesized partner-rank slots being evaluated, erence Index are ones who switched that the grunts served a reconciliatory and U is the observed number of dif- partners from one time period to the function.46 To test this hypothesis, a ferent (unique) partners that occupy next although their former partners series of playback experiments were positions within the S partner-rank were still present in the group. conducted in which victims of aggres- slots across years. The value of the Partner Diversity sion heard the grunts of their former PSI varies from 1 for females who opponent (test condition) or another had the same S partners in each year If the amount of time and energy female (control condition) shortly to 0 for females who had a com- that animals can devote to socializ- after conflicts had ended. 46,47 After pletely different set of partners in ing varies, then they can apportion hearing the grunts of their former each year. social interactions evenly among opponents, females seemed to be less One useful feature of this procedure potential partners or focus their anxious about becoming the target of is that observed values of the PSI can attention on a subset of the group. redirected aggression from their for- 56 be compared using simulation analy- Cheney used the Shannon-Weiner mer opponents46 and were also more sis against values that would be Diversity Index, originally developed likely to approach and interact with expected if females chose their part- as an information theoretic measure, their former opponents.47 ners. It is also possible to evaluate dif- which has been widely used in ecol- Playback experiments are also use- ferent possible ranges of rank ogy, to assess how evenly grooming ful because they can provide inde- orderings (for example, 1–2, 2–4...; is distributed across group members. pendent evidence that a particular 1–4, 5–8...), and to compare observed The index is calculated as: kind of relationship exists between values of the PSI against simulated two individuals. For example, several values. When we applied this method R playback studies have shown that H p log p to female baboons, we found that i i adult male baboons respond more ¼ i 1 females showed stable preferences for X¼ strongly to the distress calls of their their top three partners but not for where pi is the relative proportion of female friends’ infants than to the lower-ranked partners.54 grooming directed toward individual calls of other females’ infants,58,59 There are some limitations to this i. H is maximized when a female suggesting that males have a special procedure. For instance, it does not interacts with all possible partners kind of relationship with the offspring deal well with multiple tied ranks, with equal frequency and minimized of their female friends. which often arose at the bottom when a female focuses all of her Playback experiments, using the of females’ preference orders. In interactions on one partner. calls of a single individual or a addition, this procedure may be less To standardize the measure for sequence of calls from several individ- useful for species that live in comparison with values obtained in uals, can be designed to simulate nor- small groups because it will be studies of groups of different sizes, it mal social interactions or to mimic an harder to demonstrate that observed is useful to compute the ratio anomalous event such as an apparent patterns of preferences differ from between the observed diversity index, rank reversal. Most playback experi- patterns expected to occur by H, and the maximum possible value ments follow a within-subject design; chance. of the diversity index for a group of all include a test and at least one con- The PSI does not account for size N: Hmax 52ln(N 2 1). trol. Although some playback experi- changes in partner rankings that were One strong caution in using this ments can be done opportunistically, due to the death, disappearance, or method is that the distribution of others are dependent on a prior event. dispersal of a partner from one grooming across dyads is sensitive to For example, in a study designed to time period to the next. To account sampling effort and variation in the determine whether baboons treat for this, we created an alternative rate of interaction across individu- vocal signals by an opponent’s close measure, the Partner Preference als.57 Alternative measures of partner relative as a form of alliance support, Index55,56: diversity are needed. playbacks were conducted only after ARTICLE A Practical Guide to the Study of Social Relationships 221 naturally occurring agonistic interac- a systematic basis. As a result, it is additional light on how animals tions were observed.60 If the critical now possible to monitor with some respond to social events. For example, prior events are uncommon, these precision how individuals respond to it has been known for some time that kinds of experiments can take months social and environmental events. For the neuropeptide oxytocin is impor- to complete. a more complete discussion of the tant in social bond formation and To retain their credibility, play- methods of involved in collecting affiliation.71 Recently, it been shown back experiments must be done at and processing hormone samples, that the peripheral release of oxytocin very low rates. Ideally, experimenters see Beehner and Whitten,65 Hodges in urine also correlates with affiliative should also use as dependent varia- and Heistermann,66 and Anestis.67 behaviors.72,73 Studying captive mar- bles behaviors other than orienting Information derived from hormo- mosets, Snowdon and coworkers72 responses, such as the subject’s sub- nal analyses is beginning to provide found that elevated levels of oxytocin sequent interactions with the caller, insight about the impact of social were correlated with different behav- because orienting responses can be relationships on individuals. For iors in males and females, but that difficult to interpret. example, among chacma baboons, the highest mean levels of oxytocin Because they use natural stimuli the presence of immigrant, poten- were found in the most strongly that can be combined in novel ways tially infanticidal males produces ele- bonded male-female pairs. In other and presented under controlled con- vated GC levels among lactating words, supporting Hinde’s original ditions, playback experiments can females with vulnerable infants.68,69 observation, oxytocin levels were best offer an independent test of hypothe- However, lactating females who have predicted by properties of the pair’s ses generated from observation. For relationship, not by any single prop- more details about the design and erty of either individual. Similarly, execution of field playback experi- Crockford and colleagues74 found ments, see Zuberbuhler€ and Wittig.61 that in wild chimpanzees urinary con- Hormone metabolites centrations of oxytocin were higher extracted from fecal after individuals groomed with a METHODS FOR ASSESSING samples generally closely bonded partner than after PROXIMATE MECHANISMS THAT they groomed with a less closely MEDIATE THE EFFECTS OF SOCIAL reflect responses to bonded partner. Hormone metabolites extracted BEHAVIOR events during the past from fecal samples generally reflect Just as playback experiments offer 24-48 hours, whereas responses to events during the past a means to ‘interview’ subjects about urine samples reflect 24-48 hours, whereas urine samples what they know about each other’s reflect responses to events during the relationships, new advances in the responses to events dur- previous 1-3 hours.66 When there is a noninvasive extraction of hormones ing the previous 1-3 time lag, a urine sample is much provide information about how indi- 66 more likely to be influenced by a sin- viduals perceive the events they expe- hours. When there is a gle social event, such as a fight or a rience. Robert Sapolsky, who time lag, a urine sample grooming interaction, than is a fecal pioneered endocrinological studies of sample. Thus, if observers are inter- free-ranging primates, developed is much more likely to ested in hormonal responses to par- darting techniques that minimized be influenced by a sin- ticular events, urine is more useful anticipatory stress before the proce- than feces. However, this also means dure and avoided loss of habitua- gle social event, such as that the observer must know what tion.62 He showed that dominant a fight or a grooming has happened to the animal over the male baboons generally had lower interaction, than is a last few hours. The observer will glucocorticoid metabolite (GC) levels therefore have to have monitored the than subordinate males, but that fecal sample. individual continuously for several dominant males who rarely groomed hours before and after the collection females or interacted with infants of a urine sample in order to make had elevated GC levels.63,64 This use of it. work suggested that social relationships may buffer the effects of stress. established friendships with resident Darting is invasive and potentially males have significantly lower GC lev- 69 dangerous, so researchers welcomed els than females who have not. Sim- LONG-TERM CONSEQUENCES OF new methods for the extraction of ilarly, females who lose a close VARIATION IN THE QUALITY OF hormone metabolites from feces. relative to predation experience sig- SOCIAL RELATIONSHIPS Feces can be collected without han- nificant increases in GC levels com- dling animals directly. Also, it is rela- pared with females who have not.70 Systematic descriptions of the tively straightforward to collect Developments in the extraction of properties of social relationships samples from known individuals on hormones from urine promise to shed derived from behavioral data, 222 Silk et al. ARTICLE experiments, or physiological analy- because males preferentially support We do not fully understand the ses, can be used to assess the adapt- preferred partners in agonistic inter- mechanisms that underlie the rela- ive value of social bonds. The first actions. Also, coalitionary support tionship between sociality and fit- effort to address this possibility drew plays an important role in the acqui- ness in primate groups. Data from on long-term data on the social sition of top-ranking positions and female baboons suggest that the behavior and reproductive histories top-ranking males sire the majority strength of social bonds enhances of female yellow baboons.75 For each of infants within their groups.80 females’ ability to cope with stress female, an individual-based sociality Moreover, top-ranking males may and protects them and their off- index (SI) was constructed. These selectively tolerate their allies’ efforts spring from the deleterious effects of data were matched against informa- to mate with receptive females.81 chronic elevation of the stress tion about offspring survival to one However, researchers have not yet response. It is also possible that year of age, an important component directly linked the quality of social females with stronger and more of variation in the lifetime fitness of secure social bonds may be less vul- females. Females with higher SI val- nerable to predation. Crested maca- ues had higher survivorship among ques respond more strongly to alarm their offspring than females with calls produced by individuals with lower SI values. This effect was inde- We do not fully under- whom they have a close social bond pendent of the effects of female dom- stand the mechanisms than to alarm calls produced by inance rank and the presence of kin. other group members.28 The alarm These data suggest that sociality that underlie the rela- calls alert others that a predator is enhances female reproductive suc- tionship between social- nearby and prompt collective mob- cess. But it is also possible that the bing behavior. Social bonds may also causal arrow goes in the other direc- ity and fitness in primate buffer feeding competition and tion: Females that have surviving groups. Data from enhance foraging efficiency.82 infants may become more sociable.54 Female baboons are strongly female baboons suggest attracted to newborn infants,76,77 so that the strength of FUTURE DIRECTIONS it is possible that the relationship between female sociality and infant social bonds enhances Primatologists have made consid- survival is an artifact of this “natal females’ ability to cope erable progress in addressing the attraction.” Therefore, in a subse- fundamental questions raised by quent study of a different baboon with stress and protects Hinde and Kummer often using the population, we restricted the analy- them and their offspring observational methods advocated by ses to observations of adult females from the deleterious Jeanne Altmann. At the same time, when they did not have young however, this research has generated infants and computed dyad-specific effects of chronic eleva- new questions that may guide future Composite Sociality Index (CSI) val- tion of the stress research. We outline a few of these ues for females with each of their questions below. adult female partners. The offspring response. It is also possi- 1. What mechanisms underlie the of females with higher CSI values ble that females with relationship between sociality and lived significantly longer than did fitness outcomes? In humans, social the offspring of females with lower stronger and more support has important effects on CSI values. This pattern was inde- secure social bonds health.83 Social ties seem to buffer pendent of the effects of dominance the short-term effects of stress and rank.78 Taken together, these analy- may be less vulnerable may enhance the efficacy of anabolic ses strongly suggest that the quality to predation. processes involved in somatic main- of females’ social bonds influences tenance and repair. This may affect the survival of their offspring. both females and their offspring. Females that had the strongest and Maternal exposure to environmental most stable bonds with their most and social stressors can have detri- preferred partners also lived longer mental impacts on their offspring’s than other females.79 bonds among male chimpanzees to growth rates, longevity, physiology, There is also some evidence that male reproductive success. Such and behavior.84–87 It is important to the strength of social bonds influen- links have been established in one determine whether the same mecha- ces the reproductive success of pri- group of Assamese macaques: the nisms operate in humans and other mate males. For example, male strength of males’ social bonds was primates. chimpanzees form strong and endur- linked to coalitionary support, their 2. Why do levels of sociability dif- ing bonds with other males in their ability to acquire and maintain high fer among individuals? A growing group.51 These relationships may ranking positions, and paternity body of evidence suggests that social affect male reproductive success success.25 bonds enhance the fitness of ARTICLE A Practical Guide to the Study of Social Relationships 223 individuals, but there is considerable researchers to describe and interpret REFERENCES variation in the levels of sociability complex social processes in primate 1 Koenig WD, Sheng-Feng S, Krakauer AH, among individuals. In baboons, groups. However, the full interpreta- et al. 2009. Reproductive skew in avian soci- some variation is linked to the pres- tion of our results would not have eties. In: Hager R, Jones CB, editors. Reproduc- tive skew in vertebrates: proximate and ence of preferred categories of part- been possible if detailed analyses of ultimate causes. Cambridge: Cambridge Univer- ners, such as close kin and peers, ... behavior at the level of the dyad sity Press. p 227–264. but a substantial amount of variation had not been conducted previously.” 2 Rubenstein DI, Nunez~ CM. 2009. Sociality remains unexplained. 4. How widespread and strong are and reproductive skew in horses and zebras. In: Hager R, Jones CB, editors. Reproductive skew We have recently explored the link- the effects of sociality on fitness? A in vertebrates: proximate and ultimate causes. age between sociality and personality growing body of evidence from a Cambridge: Cambridge University Press. p 196– dimensions among female wide range of taxa suggests that the 226. 88 3 Holekamp KE, Engh AL. 2009. Reproductive baboons. Females with the highest correlation between social bonds and skew in female-dominated mammalian soci- scores on the Composite Sociality reproductive success is not limited to eties. Hager R, Jones CB, editors. Reproductive Index were significantly more likely primates. Similar patterns have been skew in vertebrates: proximate and ultimate causes. Cambridge: Cambridge University to be characterized as “nice,” which also been detected in female Press. p 53–83. 29 91 means that they had higher rates of horses, bottlenose dolphins, wild 4 Alberts SC. 2012. Magnitude and sources of friendly contact (excluding groom- house mice,92 laboratory rats,93 and variation in male reproductive performance. In: 94 Mitani JC, Call J, Kappeler PM, et al., editors. ing), higher rates of grunting when rock hyrax. However, we know lit- The evolution of primate societies. Chicago: they approached lower ranking tle about the properties of social University of Chicago Press. p 412–431. females, and were less likely to be bonds in these taxa, and it is not yet 5 Pusey A. 2012. Magnitude and sources of var- alone. In contrast, females that were clear whether particular features of iation in female reproductive performance. In: Mitani JC, Call J, Kappeler PM, et al., editors. characterized as “loners” (females close bonds recur across taxa. The evolution of primate societies. Chicago: that were often alone and grunted University of Chicago Press. p 343–366 mainly when they approached higher 6 Watts DP. 2010. Dominance, power, and poli- CONCLUSIONS tics in non-human and human primates. In: ranking females with infants) had Kappeler PM, Silk JB, editors. 2010. Mind the uniformly low Composite Sociality By collecting fine-grained data on gap. Berlin: Springer. p 109–138. Index values and higher stress levels. the behavior, social knowledge, and 7 Schino G, Aureli F. 2008. Grooming recipro- These findings suggest that at least physiology of known individuals over cation among female primates: a meta-analysis. Biol Lett 4:9–11. some of the variation in sociality extended times, primatologists are in a position to ask detailed questions 8 Schino G, Aureli F. 2009. Reciprocal altruism may be caused by individual differ- in primates: partner choice, cognition and emo- ences in personality. Further efforts about the selective pressures that tion. Adv Study Behav 39:45–69. to identify the linkage among per- shape the evolution of social behav- 9 Hinde RA. 1976. Interactions, relationships and social structure. Man 11:1–17. sonality, sociality, and fitness are ior and cognition in complex societies. To make the most of our data, 10 Hinde RA. 1977. Mother-infant separation clearly needed. and the nature of inter-individual relationships: 3. Is there meaningful structure in we need systematic procedures for experiments with rhesus monkeys. Proc R Soc primate groups above the level of the assessing and describing the behav- Lond B 196:29–50. ior of individuals and the properties 11 Hinde RA, Atkinson S. 1970. Assessing the dyad? There is growing interest in roles of social partners in maintaining mutual using social network analyses to of their relationships with others. proximity, as exemplified by mother-infant describe social structure in primate We can use the kinds of methods relations in rhesus monkeys. Anim Behav 18: 169–176. groups.89 Network analyses are par- described here to test hypotheses about the selective forces that shape 12 Hinde RA, Spencer-Booth Y. 1968. Review ticularly useful because they can lecture: the study of mother-infant interaction reveal structure above the level of the behavioral strategies and to con- in captive group-living rhesus monkeys. Proc R dyad. There are good reasons to struct comparisons across individu- Soc Lond B 169:177–201. als, groups, or taxa. Additional 13 Hinde RA, White LE. 1974. Dynamics of a believe that indirect (nondyadic) con- relationship: rhesus mother-infant ventro-ven- nections can have important impacts methods for characterizing social tral contact. J Comp Physiol Psychol 86:8–23. on information transfer and disease relationships are clearly needed. We 14 Kummer H. 1978. On the value of social transmission within groups, but it is hope that this paper stimulates relationships to nonhuman primates: a heuris- efforts to develop them. tic scheme. Soc Sci Information 17; 687–705. less clear how indirect connections 15 Altmann J. 1974. Observational study of would influence the patterns or con- behavior: sampling methods. Behaviour 49: 227–266. sequences of social behavior. Brent ACKNOWLEDGMENTS and colleagues90 have shown that 16 Dunbar RIM, Shultz S. 2010. Bondedness and sociality. Behaviour 147:775–803. We thank John Fleagle for the invi- proximity reach, one measure of indi- 17 Silk JB. 2012. The adaptive value of social- rect connectedness, is associated with tation to write this paper, Kevin Lan- ity. In: Mitani JC, Call J, Kappeler PM, et al., lower glucocorticoid levels among gergraber for help with the pair-wise editors. The evolution of primate societies. Chi- cago: University of Chicago Press. p 552–564. high-ranking female rhesus maca- affinity index, and Roger Mundry for 18 Aureli F, Fraser ON, Schaffner C, et al. ques. However, we do not believe that sharing his concerns about proce- 2012. The regulation of social relationships. In: network analyses can replace dyadic dures for computing the DSI. John Mitani JC, Call J, Kappeler PM, et al., editors. Mitani and several anonymous The evolution of primate societies. Chicago: analyses. As King, Clark, and Cow- University of Chicago Press. p 531–551. 82 reviewers provided helpful comments lishaw concluded, “our results show 19 Cords M, Aureli F. 1999. Reconciliation and how network analyses can enable on the manuscript. relationship qualities. In: Aureli F, de Waal 224 Silk et al. ARTICLE

FBM, editors. Natural conflict resolution. 37 Hinde RA, Spencer-Booth Y. 1971. Towards 58 Lemasson A, Palombit RA, Jubin R. 2008. Berkeley: University of California Press, p 177– understanding individual differences in rhesus Friendships between males and lactating 198. mother-infant interaction. Anim Behav 19:165– females in a free-ranging group of olive 20 Pepper JW, Mitani JC, Watts DP. 1999. Gen- 173. baboons (Papio hamadryas anubis): evidence eral gregariousness and specific social preferen- 38 Brown GR. 2001. Using proximity measures from playback experiments. Behav Ecol Socio- ces among wild chimpanzees. Int J Primatol to describe mother-infant relationships. Folia biol 62:1027–1035. 20:613–632. Primatol 72:80–84. 59 Moscovice LR, Heesen M, Di Fiore A, et al. 2009. Paternity alone does not predict long- 21 Sapolsky RM, Alberts SC, Altmann J. 1997. 39 Weaver A, de Waal FBM. 2002. An index of term investment in juveniles by male baboons. Hypercortisolism associated with social subor- relationship quality based on attachment Behav Ecol Sociobiol 63:1471–1482. dinance or social isolation among wild theory. J Comp Psychol 116:93–106. baboons. Arch Gen Psychiatry 54:1137–1143. 40 Perry S, Barrett HC, Manson JH. 2004. 60 Wittig RM, Crockford C, Seyfarth RM, et al. 2007. Vocal alliances in chacma baboons (Papio 22 Silk JB, Altmann J, Alberts SC. 2006. Social White-faced capuchin monkeys show triadic hamadryas ursinus). Behav Ecol Sociobiol 61: relationships among adult female baboons awareness in their choice of allies. Anim Behav 899–909. (Papio cynocephalus) I. Variation in the 67:165–170. strength of social bonds. Behav Ecol Sociobiol 41 Maestripieri D, Schino G, Aureli F, et al. 61 Zuberbuhler€ K, Wittig RM. 2011. Field 61:183–195. 1992. A modest proposal: displacement activ- experiments with nonhuman primates: a tuto- rial. In: Setchell JM, Curtis DJ, editors. Field 23 Silk JB, Alberts SC, Altmann J. 2006. Social ities as an indicator of emotions in primates. nd Anim Behav 44:967–979. and laboratory methods in primatology, 2 ed. relationships among adult female baboons Cambridge: Cambridge University Press. p 207– (Papio cynocephalus) II: Variation in the quality 42 Aureli F, van Schaik CP, van Hooff J. 1989. 224. and stability of social bonds. Behav Ecol Socio- Functional aspects of reconciliation among biol 61:197–204. captive long-tailed macaques (Macaca fascicula- 62 Sapolsky R. 1988. Individual differences and ris). Am J Primatol 19:39–51. the stress response: studies of a wild primate. 24 Silk JB, Beehner JC, Bergman TJ, et al. Adv Exp Med Biol 245:399–411. 2010. Female chacma baboons form strong, 43 Castles DL, Whiten A, Aureli F. 1999. Social equitable, and enduring social bonds. Behav anxiety, relationships and self-directed behavior 63 Sapolsky R, Ray J. 1989. Styles of domi- Ecol Sociobiol 61:197–204. among wild female olive baboons. Anim Behav nance and their physiological correlates among wild baboons. Am J Primatol l8:1–13 25 Schulke€ O, Bhagavatula J, Vigilant L, et al. 58:1207–1215. 64 Virgin CE, Sapolsky R. 1997. Styles of male 2010. Social bonds enhance reproductive suc- 44 Sapolsky R. 1998. Why zebras don’t get social behavior and their endocrine correlates cess in male macaques. Curr Biol 20:2207– ulcers: an updated guide to stress, stress related nd among low-ranking baboons. Am J Primatol 42: 2210. diseases, and coping, 2 Ed. New York: WH 25–39. 26 Kulik L, Muniz L, Mundry R, et al. 2012. Freeman. 65 Beehner JC, Whitten PL. 2004. Modifica- Patterns of interventions and the effects of coa- 45 Cheney DL, Seyfarth RM. 2009. Stress and tions of a field method for faecal steroid analy- litions and sociality on male fitness. Mol Ecol coping mechanisms in female primates. Adv sis in baboons. Physiol Behav 82:269–277. 21:699–714. Stud Behav 39:1–35 66 Hodges JK, Heistermann M. 2003. Field 46 Cheney DL, Seyfarth RM, Silk JB. 1995. The 27 de la O C,Mevis L, Richter C, et al. 2012. endocrinology: monitoring hormonal changes role of grunts in reconciling opponents and Reconciliation in male stump-tailed macaques in free-ranging primates. In: Setchell JM, Curtis facilitating interactions among adult female (Macaca arctoides): intolerant males care for DJ, editors. Field and laboratory methods in baboons. Anim Behav 50:249–257. their social relationships. Ethology 119:39–51. primatology. Cambridge: Cambridge University 28 Michelleta J, Waller BM, Panggur MR, et al. 47 Cheney DL, Seyfarth RM. 1997. Reconcilia- Press. p 282–294. tory grunts by dominant female baboons influ- 2012. Social bonds affect anti-predator behav- 67 Anestis SF. 2010. Hormones and social ence victims’ behaviour. Anim Behav 54:409– iour in a tolerant species of macaque, Macaca behavior in primates. Evol Anthropol 19:66–78. nigra. Proc R Soc Lond B doi:10.1098/ 418. 68 Beehner JC, Bergman TJ, Cheney DL, et al. rspb20121470. 48 Silk JB, Kaldor E, Boyd R. 2000. Cheap talk 2005. The effect of new alpha males on female when interests conflict. Anim Behav 59:423– 29 Cameron EZ, Setsaas TH, Linklater WL. stress in free-ranging baboons. Anim Behav 69: 432. 2009. Social bonds between unrelated females 1211–1221. increase reproductive success in feral horses. 49 Henzi SP, Barrett L. 2007. Coexistence in 69 Engh AL, Beehner JC, Bergman TJ, et al. Proc Natl Acad Sci USA 106:13850–13853. female-bonded primate groups. Adv Study 2006. Female hierarchy instability, male immi- Behav 37:43–81. 30 Fraser ON, Schino G, Aureli F. 2008. Com- gration, and infanticide increase glucocorticoid ponents of relationship quality in chimpanzees. 50 Henzi SP, Lusseau D, Weingrill T, et al. levels in female chacma baboons. Anim Behav Ethology 114:834–843. 2009. Cyclicity in the structure of female 7:1227–1237. baboon social networks. Behav Ecol Sociobiol 31 Osborne JW, Costello AB. 2004.Sample size 70 Engh AL, Beehner JC, Bergman TJ, et al. 63:1015–1021. and subject to item ratio in principal compo- 2006. Behavioural and hormonal responses to nents analysis. Practical Assessment, Research 51 Mitani JC. 2009. Male chimpanzees form predation in female chacma baboons. Proc R & Evaluation, 9(11). Retrieved August 17, 2012 enduring and equitable social bonds. Anim Soc Lond B 273:707–712. from http://PAREonline.net/getvn.asp?v59&n5 Behav 77:633–640. 71 Insel TR. 2010. The challenge of translation 11. 52 Hemelrijk CK. 1990. Models of, and tests in social neuroscience: a review of oxytocin, 32 Henzi SP, Lycett JE, Weingrill T. 1997. for, reciprocity, unidirectionality and other vasopressin, and affiliative behavior. Neuron Cohort size and the allocation of social effort social interaction patterns at a group level. 65:733–940. by female mountain baboons. Anim Behav 54: Anim Behav 39:1013–1029. 72 Snowdon CT, Pieper BA, Boe CY, et al. 1235–1243. 53 Siegel S, Castellan NJ. 1988. Nonparametric 2010. Variation in oxytocin is related to varia- 33 Seyfarth RM. 1977. A model of social statistics for the behavioral sciences. New York: tion in affiliative behavior in monogamous, grooming among adult female monkeys. J McGraw Hill. pairbonded tamarins. Horm Behav 58:614–618. Theor Biol 65:671–698. 54 Silk JB, Alberts SC, Altmann J, et al. 2012. 73 Seltzer LJ, Ziegler TE. 2007. Non-invasive 34 van Hooff J, Wensing JAB. (1987). Domi- Stability of partner choice among female measurement of small peptides in the common nance and its behavioral measures in a captive baboons. Anim Behav 83:1511–1518. marmoset (Callithrix jacchus): a radiolabeled wolf pack. In: Frank HW, editor. Man and wolf. 55 Cheney DL, Silk JB, Seyfarth RM. 2012. Evi- clearance study and endogenous excretion Dordrecht, Netherlands: Junk Publishers. p dence for intra-sexual selection in wild female under varying social conditions. Horm Behav 219–252. baboons. Anim Behav 84:21–27. 51:436–442. 35 Vervaecke H, de Vries H, van Elsacker L. 56 Cheney DL. 1992. Intragroup cohesion and 74 Crockford C, Wittig R, Ziegler TE, et al. 1999. Experimental evaluation of the consis- intergroup hostility: the relation between 2013. Urinary oxytocin and social bonding in tency of competitive ability and agonistic domi- grooming distributions and intergroup compe- related and unrelated chimpanzees. Proc R Soc nance in different social contexts in captive tition among female primates. Behav Ecol 3: Lond B 280 bonobos. Behaviour 136:423–442. 334–345. 75 Silk JB, Alberts SC, Altmann J. 2003. Social 36 Silk JB, Cheney DL, Seyfarth RM. 1999. The 57 Di Bitetti MS. 2000. The distribution of bonds of female baboons enhance infant sur- structure of social relationships among female grooming among female primates: testing vival. Science 302:1331–1334. savannah baboons in Moremi Reserve, hypotheses with the Shannon-Wiener diversity 76 Altmann J. 1980. Baboon mothers and Botswana. Behaviour 136:679–703. index. Behaviour 137:1517–1540. infants. Cambridge: Harvard University Press. ARTICLE A Practical Guide to the Study of Social Relationships 225

77 Silk JB, Rendall D, Cheney DL, et al. 2003. lying mechanisms. Perspect Biol Med 46:S39– 90 Brent LJN, Semple S, Dubuc C, et al. 2011 Natal attraction in adult female baboons (Papio S52. Social capital and physiological stress levels in cynocephalus ursinus) in the Moremi Reserve, free-ranging rhesus macaques. Physiol Behav Botswana. Ethology 109:627–644. 84 Bernardo J. 1996. Maternal effects in animal 102:76–83. ecology. Am Zool 36:83–105. 78 Silk JB, Beehner JC, Bergman TJ, et al. 2009. 91 Fre`re CH, Krutzen M, Mann J, et al. 2010. The benefits of social capital: social bonds 85 Sanchez MM. 2006. The impact of early Social and genetic interactions drive fitness among female baboons enhance offspring sur- adverse care on HPA axis development: non- variation in a free-living dolphin population. vival. Proc R Soc Lond B 276:3099–3104. human primate models. Horm Behav 50:623– Proc Natl Acad Sci USA 107:19949–19954. 631. 79 Silk JB, Beehner JC, Berman TJ, et al. 2010. 92 Weidt SE, Hofmann SE, Konig€ B. 2008. Not Strong and consistent social bonds enhance the 86 Onyango PO, Gesquiere LR, Wango EO, only mate choice matters: fitness consequences longevity of female baboons. Curr Biol 20: et al. 2008. Persistence of maternal effects in of social partner choice in female house mice. 1359–1361. baboons: mother’s dominance rank at son’s Anim Behav 75:801–808. conception predicts stress hormone levels in 93 Yee JR, Cavigelli SA, Delgado B, et al. 2008. 80 Mitani JC. 2009. Cooperation and competi- sub-adult males. Horm Behav 54:319–324. Reciprocal affiliation among adolescent rats tion in chimpanzees: current understanding and during a mild group stressor predicts mam- future challenges. Evol Anthropol 18:215–227. 87 Weinstock M. 2008. The long-term behav- mary tumors and lifespan. Psychosom Med 70: 81 Duffy KG, Wrangham RW, Silk JB. 2007. ioural consequences of prenatal stress. Neuro- 1050–1059. sci Biobehav Rev 32:1073–1086. Male chimpanzees exchange political support 94 Barocas A, Ilany A, Koren L, et al. 2011. for mating opportunities. Curr Biol 21:R586– 88 Seyfarth RM, Silk JB, Cheney DL. 2012. Variance in centrality within rock hyrax social 587. Variation in personality and fitness in wild networks predicts adult longevity. PLoS ONE 6: 82 King AJ, Clark FE, Cowlishaw G. 2012. The female baboons. Proc Natl Acad Sci USA 109: e22375. dining etiquette of desert baboons: the roles of 16980–16985. social bonds, kinship, and dominance in co- 89 Seuer C, Jacobs A, Amblard F, et al. 2011. feeding networks. Am J Primatol 73:768–774. How can social network analysis improve the 83 Cacioppo JT, Hawkley LC. 2003. Social iso- study of primate behavior? Am J Primatol 73: VC lation and health, with an emphasis on under- 703–719. 2013 Wiley Periodicals, Inc.