Chapter 7 Marriage, Rank, and Seasonal Migration: Fractality in Social Structure Every society expresses and deals with competing principles. As in many other societies, potentially competing principles in nomad society present themselves most clearly in practice, namely, in concrete instances involving real kinship relations, marriage, and politics. Consequently, we need to turn to actual cases and practices to see most clearly principles that are in place and how competing claims, including rank and leadership, are negotiated or resolved. For the nomad clan the principles of rank and equality are often in conflict. Siblings and generations, for example, are ranked by seniority among the Aydınlı. Families and spouses are in many specific situations considered equals. These principles may also conflict: What if, for example, one marries someone (normally a tie of generational equality) who is of a different genealogical or chronological generation? Resolutions of conflicting organizational positions and principles can work effortlessly (like same-generation marriage rules) or can produce slight or notable ruptures of orderly relations. Factions and factionalism are a product of the latter kind. They can be viewed in nomad society as a breakdown in normal conflict resolution. From another perspective, however, factionalism, like competition, invokes the emergence of a set of rules or guides for disputants that provide flexible forms of conflict resolution. Indeed, it is the flexibility in the levels and manipulable boundary conditions that often result in the formation of factions and raise questions of where potential mediators are located within them, and that offer means of understanding their success as institutional practice. This chapter, then, has something of the quality of a detective story. Of necessity, it was constructed through analysis of the network data, which called for clarification of ethnographic questions from Johansen at various points, and then formulation of hypotheses that led to further analyses. Finally, out of the network analyses came some definitive so- 222 Chapter 7 cial patterns that the ethnographer might not have seen or thought, a priori, to include in a general ethnography. Here network analysis enriches the task of ethnographic study. Each of these more elaborate findings from different stages of analysis related to the problems under study were passed back to the ethnographer for commentary and integration with her knowledge of the case study. Equality and Rank Competition for Social Rank and Feuding In principle all nomad families are equal. Nomads, however, can become rich within one generation if they have good luck with their herds, slaughter rarely as was the common practice, and obtain some additional income by cattle trading, transportation with their camels or, in the last decades, tractor driving. They may become influential at the same time by having many sons and marriage links, which help them to form a numerous and powerful block within the clan. This type of strategy for herd increase or exceptional economic efforts produces an unstable balance in the standing of individuals and as between families. It is a reason for the never-ending competition of the patriarchs and their lineages, already discussed.1 In this connection it could be also shown how the modern propagation of birth control and monogamic marriage may be rendered less meaningful in societies like the nomads, where family size and ties also have considerable importance at this level. As mentioned in the previous chapter, the most important factor contributing to social rank is the backing by a large group of related families. Early in her fieldwork Johansen was aware of the central role of kinship in clan politics. She noticed the patriarchs’ permanently competing for a high position for their families in the social ranking whenever members of different joint families met, watched by and silently backed by the younger males. The boys and young men imitated the selfpraising and ambitious talks of their fathers when coming together in their special circles. The women accepted this practice of their men’s rhetorical capacities and occasionally showed their pride in being members of a high-ranking family whom they represented by keeping their tents orderly, serving guests with the best a family could afford, and showing full obedience to their men as long as members of other joint families were present. During Johansen’s first two stays shootings and ensuing blood- Marriage, Rank, and Migration 223 revenge sequences sometimes occurred. She observed four feuds up to 1964; in 1970, feuding had not altogether ceased, but there were no feuds by 1982. Traditionally, the more distant the relation between the parties, the more the feuding increased in severity and the larger the group mobilized. She never saw any feuding within extended families or lineages. Between the families of different lineages there were rare cases of feuding, especially in matters of abductions or elopements of girls. In the case of such feuding only the younger male relatives most closely related to the two extended families in conflict were involved. Feuds between two clans or a clan and a village were important up to about 1980 (cf. Johansen 1995). The entire group of unmarried young men sometimes carried out feuds with other clans or their lineages. Although feuds within the clan involved only the closest male relatives, it was of advantage in this situation to have at one’s disposal a number of venerable elder men with unusually extensive relations to other joint families and a large group of proud young people. Nobody would show aggressiveness with any seriousness against the members of such joint families and their closest allies. To settle disputes within this group by mediation was correct behavior. In contrast, compliance in cases of offenses from nonrelatives was looked at as cowardice on the part of young men, and a shame for the family. This is the same general context in which newcomers to the clan such as #8 and #9 had been eager to organize intermarriages as soon as possible. Barth’s Model of Nomad Dynamics Barth (1953:72) explains the dynamics of southern Kurdish tribal social structure, akin to that of the Aydınlı nomads, as a result of the lack of coercive means for leadership in an egalitarian nomadic society, as diagrammed in Figure 7.1. Leaders of small tight kinship groups such as lineage segments are present, but they derive their leadership from informal support. The tight kinship groups remain small through fission due to the limited and dispersed resources of nomadism, similar to our argument in Chapter 4. These groups are tightly linked through a preference for close marriage within the agnatic group, such as with a FBD lineage-mate. With no coercive means of leadership, feud is the main expression of power. Feuding reinforces the preference for FBD marriage, which cycles back in a positive feedback loop to reinforce lineage solidarity. Figure 7.1 expands on Barth’s diagram, including selected additional elements. His model also stresses that the combination of Chapter 7 224 power relations subject to manipulation, combined with political units defined by agnatic links, reinforce the political emphasis on FBD marriage from more than a preference to a highly frequent form of marriage. Figure 7.1: Barth’s Nomad Dynamics (family and lineage level) Agnatic links define political units Patrilocal residence Male husbandry of small animals, camels Close, tight kin group Fission Leader has/needs no coercive means Small Feud the main expression of power Pfd. marriages are FBD politically, close agnates Broader integration is weak Competitive power relations subject to manipulation Broader integration through marriages Non-agnate marriages link broadly Negotiations and Choice at Marriage Bride Payments and the FBD The lineage, clan, and tribal level dynamics in the type of pastoral nomadism that the Aydınlı and Kurds (Barth 1953) have in common is applicable to understanding FBD marriage among the Aydınlı and is consistent with our hypotheses in Chapter 4. When Johansen began her work with the nomads on their genealogies she already knew that the right to marry the FBD marriage was widely distributed among the Islamic peoples of the Middle East, and she was not surprised to find this right and preference in southeastern Anatolia too. Indeed cousin marriages were quite common, but it did not always have to be first cousins Marriage, Rank, and Migration 225 and cousins from the side of the young man’s father. Some facts that emerge from the ethnography of Aydınlı nomads bear on theoretical arguments regarding FBD marriage. The initiative of Aydınlı marriage negotiations was always taken by the family of the groom—officially his father and his father’s brothers— though there usually had been unofficial contacts between the women of both families long before, especially in the cases of cousin-marriages. These were looked upon favorably, but the choice of a bride was subject to many coincidences, for example, who were the neighboring families in the recent summer and winter camps, and the feelings of the groom, of course. The negotiations dealt first of all with the kalın, the price the groom’s family had to pay (bride payments), and the ceyiz, the dowry of the bride.2 These negotiations stretched over weeks with ample bargaining for the price even if the two families agreed to the marriage in principle because it was a shame for the girl, and for her family, to be handed over without hesitation. In the case of exchange of two girls the procedure was easier because there was no bargaining about the price, although the girls were still equipped with dowries by their own families. Weddings were celebrated with great expenditures. The entire lineages of the bride and the groom and all neighbors were invited. Mostly the groom’s family paid the enormous costs of such a three-daywedding. Its splendor and the high rank of the guests showed the prestige of the family and together with the bride payments and wedding presents often has a value equivalent to two to four years of family income. Bride payments themselves, for example, if delivered in animals, might easily come to twenty goats, which when they multiply to 120 can provide a living, so that twenty can provide a significant portion of one’s income. The great efforts and expenditure connected with a marriage were one of the reasons that couples had to keep together after marriage. They were under stress to do so from the sides of both families, who had the will and means to enforce compliance to retain their respective bride payments and dowry. The family of the bride was also interested in the marriages lasting because the kalın was often already spent for the marriage of a brother of the bride and could not easily be restored if the young woman ran away from her new family without a very serious reason. Thus divorces were relatively rare in the earlier generations although Islam allows men to divorce their wives. After 1982, when bride payments fell more and more out of fashion, the number of divorces rose accordingly. Only a girl’s first wedding was celebrated in its most ex- 226 Chapter 7 pensive form. If a divorced wife or a widow married again, only the next relatives and/or neighbors were invited. An animal was sacrificed and eaten and a sura from the Qur’an recited and the woman thereafter secured her bed at the side of her new spouse. If the widow or divorcée came from another lineage there had to be paid only a considerably reduced kalın for a young woman and she received only small presents. The groom himself, not his father, had to pay the bride payments for a second marriage. Because this was money he had earned personally, a man had more independent choice in this case, and he could be more influenced by personal love, not by family or lineage decisions. Thus second wives came more often from outside the clan. In the case of elopement the wedding and bride payment was also made in reduced form.3 Females can have only one spouse and their marriages are always arranged with the only exception that of elopements. If a young man and his friends organized his first marriage in the form of elopement (with the cooperation of the girl, of course), his father had to pay the kalın even when he had originally spoken that he did not consent. The father as a Muslim was looked at as being responsible that his son was sexually satisfied, married at the proper age, and not drawn to sexual intercourse outside marriage, considered sinful. Moreover, members of a family stand together and thus the father might grumble about an elopement without his consent but publicly keep solidarity with his son. Normally sons did not talk with their fathers about their love, but the fathers obviously anticipated elopements, which usually followed an official request for the girl that got a negative answer. If a widow marries the younger brother of her late husband, no kalın is paid. If she marries a non relative of her late husband or remarries after divorce, or in a case of elopement, a considerably reduced kalın is paid. The amount depends on the negotiations of the father of the woman and in the case of a widow also with the first father-in-law. Although there are many special circumstances and negotiations, kalın payments may be ranked accordingly:4 (++) bride payment is high for a first or virgin wife, the father pays. (+) bride payment is medium for elopement, paid by the father, and for a second wife, paid by the groom. (-) bride payment is lower for a widow or divorcée, the groom pays. (0) no bride payment for a widow if she marries the (usually younger) brother of her late husband. Marriage, Rank, and Migration 227 The quantities and obligations for payment of the bride payments for a woman, noting that for a second wife the man is considerably older, are roughly ordered according to how the strength and cohesiveness of ties between the groups is enhanced, including potential enhancement by the fertility of the bride. The last item in the list does not create a new tie (the husband has died) and fits the ordering. The pictorial representation in Figure 7.2 is also meant to convey that relationships of different tiestrength might also give rise to different levels of cohesion in groups linked by marriage. At the center, the bride payments for first and virgin wives are highest, and they are paid by the father. The bride payments for elopements are substantial but lower and paid by the father. Those for widows or divorcées in some circumstances are much lower and paid by the groom if they are his second wife. The moral economy of tiestrength is consistent with Barth’s model of how FBD marriages increase solidarity within the lineage. Although Johansen did not do a quantitative study of the amounts of bride payments, we may note from Bates’s study (1973:62) that the amount of bride payment for FBD marriages in a neighboring Yörük clan are more often in the upper than the lower quartiles as compared with other cousin marriages (p=.02). This may well be the case among the Aydınlı. It may also be the case that because extended families can fission so quickly (given the small size of the household production unit) the reinforcement of solidarity even among the closest of agnatic kin is important, as with FBD marriage. Figure 7.2: Tie-Strength Spheres Reflected in Bride Payments Key: - + ++ as above, see text Because the mothers-in-law have to be together with the young women after marriage, while men were off pasturing, managing the herds or fulfilling other tasks in daytime (and in the hottest period even at night), they had a lively interest in arranging matches and they were the ones who made the proposals about the choice of a bride to their husbands and explored the feelings of their sons. Women were often inclined to take a girl of their family of descent into account. If there existed good experiences with one young woman and a firm relationship to her family of orientation, her mother-in-law and her mother were inclined to pro- 228 Chapter 7 pose the marriage of another daughter to the younger brother of the young husband, and the sisters would always welcome the common household. In such marriages the brothers were called by the term bacanak (sister-related, bacı=sister), a Turkish term for sons-in-law to the same father-in-law (see Stirling 1965:173). Exchange marriages were of special importance and were frequent within and between lineages. The larger context of exchange of women between different patri-extended families is important for understanding the puzzle of the FBD marriages that occur within them. Also important was the fact that exchange of daughters and nullification of payments was another means of getting a good partner for one’s son, and could be used if difficulties were met to defray the bride payments. Marriage Exchange To examine the link between lineages, inheritance, wealth, bride payment, and exchange it is useful to define the concept of wealth-assets as it applies in an ethnographic context. To be analytically useful for the study of exchange (Bell 2002:16-17), a wealth-asset must (a) possess a capacity to grow in value, number, or size, (b) generate a flow of consumption benefits to those holding the rights to the wealth-asset, (c) be scarce in the sense that marginal increases in its growth must have a positive valuation and not constitute a surplus for which there is motivation for disposal, and (d) “be exploitable over an indefinite time horizon by a multi-generation group, linked by inheritance rules, that holds rights to its accumulation over that horizon” (Bell), or by functional equivalents such as corporate shareholding. Nonmarket wealth is the basis for corporate lineage formation and wealth-assets typically circulate in different spheres of exchange from those of nonwealth items (consumables). This terminology can be used to address the question of whether transfers of property at marriage from the husband’s to the wife’s group should be termed bride wealth (which implies that what is transferred qualifies as wealth) or bride price (which might be used to imply that what is transferred are not wealth-assets but money or consumption goods). In many cases in which these kinds of transfers apply, however, the same item, such as animals, may be scarce and qualify as a wealth-asset at one time period but may be plentiful at another time period and qualify as a consumption good. Hence, we prefer to use the term bride payment, which may or may not involve a transfer of wealth-assets at marriage. Women, in a given situation during the transition period from maid- Marriage, Rank, and Migration 229 enhood to marriage, may or may not share the attributes of wealth-assets: (a) fecundity may be valued as a source of children and growth of a lineage, (b) if the lineage holds rights to children certain benefits may follow from their labor contributions (e.g., as shepherds) as a source of consumption goods to the lineage, (c) marginal increases in having more children may have a positive valuation, and (d) the lineage as a multigeneration group may hold and transfer certain rights to its women (e.g., the transfer of a wife of a deceased lineage member to a brother) and children (e.g., provision for inheritance by children and their continued affiliation with the lineage even after the death of parents). Not all but some of the circulation of wives in marriage, then, may involve women as wealth-assets to a lineage. This does not entail diminishing wives’ status because women are themselves lineage members in this situation; they accrue rights such as inheritance of lineage property, as in most Islamic societies; they may bring property to their marriage in the form of dowry; and they may be heavily involved themselves in arranging the marriages of other women. Given that a woman may constitute (and constitute herself, that is, as a positive attribute rather than a state to which she is subjected) as a wealth-asset, certain marriages may come to involve the exchange of wealth-assets. This may take two forms, as we have seen: (case 1) the transfer of a bride from A to B and counter transfer of a bride payment from B to A, where A and B are family or lineage subunits; and (case 2) the exchange of brides as between two lineages, two extended families, or two families within the same extended family, which in all cases involve the annulment of any need for bride payment. As for case 1, the existence of bride payments for FBD argues for exchange in these marriages as well, contra the arguments of Bourdieu (1972) reviewed in Chapter 4. Case 2 involves the highest sphere of exchange involving marriage transfers, the exchange of two brides between families, in which marriage payments are nullified. Reexamining Figure 7.2, we might code the exchange of virgin brides in this case within the diagram as (+++). Further, as each bride brings her dowry to the marriage, they enhance the status of each bride, constituting herself as a wealth-asset through her fecundity. She is also an independent holder and source of consumption goods for her new nuclear family through her labor and her property, including dowry. Marriage Choice and the Extended Family 230 Chapter 7 We do not agree with Cole (1984:182), who observed a favoring of cousins similar to that of the Aydınlı for matches of the Al-Murrah Bedouins and gave an ordered list of marriage preferences among which a family chooses a bride for a young man. Johansen‘s experience was that the choice was much more occasional and that such a list is again a result of the inclination of anthropologists to make streamlined models of the behavior of the group members under study, ones that make for acknowledgment by their colleagues but fall short of the factually evident intentions of the people involved. How, for example, did Cole determine his list of the preference order in types of marriage partners? Was he simply looking at the frequency of observed behavior, and inferring that the most frequent marriages observed are also the most preferred? Can we not get closer to the thoughts and actions of the Aydınlı than by such an oversimplified assumption? Part of the advantage of the methods presented here is to overcome these kinds of shortcomings, once so common in anthropology, and that have also dissuaded many contemporary researchers of the value of kinship studies altogether. What the nomads themselves told Johansen about preferential cousin marriages sounded much less formal and more common sensical than most anthropologists’ writings on the Middle East. They told her that one must try to get a bride for one’s son from a family near to one’s own family. It should be a girl whom one has observed for a long time, which provides the opportunity to know her character, to know that she will become a skilled, diligent and obedient new member of the family. Moreover, cousins already know the members of their new families and their habits, which may help them to accommodate easily to special family traditions. This view is understandable when taking into consideration that there existed the rule of patrilocal residence and that extended families lived together in a tent of 50-80 square meters. The women worked together the whole day. There was no escape from this close togetherness, and disputes might have easily happened if the women were not already comfortable with each other. As another means to avoid quarrels with the newcomer in a time of accommodation, the custom existed that a young wife did not speak to anybody with the sole exception of the other daughters-in-law and the daughters, who shared her hard fate, respectively, or soon would do so in another family, and their husbands, who had to purchase her speaking by a valuable present—golden jewelry or a peace of cattle—on the wedding night. The men and the mother-in-law never heard her voice if they did not buy her speaking, each of them, by Marriage, Rank, and Migration 231 a valuable present of the same kind as the husband’s, which usually happened soon after the birth of her first son. After having given birth to children, especially sons, however, they not only broke the silence and talked with their new relatives but also became more and more integrated and tended to abandon their ties to their lineage, clan, or tribe of origin, if different. Women thus came as strangers into their new families at marriage, unless they were already of the same lineage. Women from outside were over time, however, increasingly looked at as female members of the clan. The important men in the course of a woman’s life, who fought for the rights of the women during quarrels, shifted from father, to brothers, and finally to their sons. In contrast to Bates’s (1973:69) study of a neighboring Yörük clan, Johansen witnessed that it was her husband who punished his young wife for adultery, not her brother, but he obtained the consent of her brother. A woman’s affiliation was a sort of sliding into membership with normal marriage, not a question of a lifelong either/or, as many male ethnographers have experienced and as theoretical works on the Middle East often assume, for example, Keyser (1974:307) and Holy (1989: 122ff.), although that exclusiveness might hold for different lineage memberships within the clan. Over time a wife is incorporated more and more into the husband’s family. Family and clan identities often override those of lineage. Beginning with the 1970s the habit of the young women’s silence vanished. Elopements came to be more common. Cousins were rarely if ever forced to elope as the FBD right spilled over to a presumptive right to cousins generally; elopement became necessary for a young couple in love when the parents of the girl were not willing to agree to the marriage, expressing their disagreement often by the demand of unpayably high bride payments. Elopements were more often between nonrelated couples, even between nomads and farmers, who may first have met in the school (cf. Bates, who did fieldwork with a more sedentary group, 1973:73,86; 1974). Females can have only one spouse (unless their husband dies) and their marriages are always arranged with the exception of elopements only. Meeting a Potential Spouse in the Circle of Families, Enlarged through Close Ties of Reciprocal Exchange Closeness of Ties Aydınlı statements about marriage choice, such as “one must try to get a 232 Chapter 7 bride for one’s son from a family near to one’s own family,” bear further examination. Ethnographers often report that potential brides are encountered on the basis of close ties, such as those acquired through visiting patterns. Obviously not everyone can marry a close relative, a lineage mate, or a FB son or daughter. Further, as we showed in Chapter 4, even if every man were to manage to marry a FBD marriage, the repetition of such marriages generates diverse connections between spouses that are not all of the same kind (Figure 4.6). Bates (1973:61,63) emphasizes this point for the neighboring Yörük clan.5 Two FBD marriages among the children of sororal marriages with brothers, for example, generates MZD marriage, but the relationship between spouses takes a compound form. Compounds also happen when two FBD marriages are arranged for the sons of two brothers. This generates sister exchange. A compound marriage is one in which one marriage is generative of the other by the diversity generating properties illustrated in Figure 4.6. We can ask, for example: How many FBD marriages are also sister exchanges? How many sister exchanges are within the same lineage? How many MZD marriages are within the context of FBD marriages of their parents? How many MZD marriages are within the same lineage? Compound marriage patterns of these different types are also ones that the Aydınlı consider “comfortable” in terms of marrying close kin or reinforcing kinship and marriage ties in multiple ways. Closeness, in sum, must be considered a relative term, a kind of weighted preference rather than a specific marriage rule. This opens several lines of inquiry: How is closeness extended through network linkages and through visiting patterns? These two aspects of interaction patterns may be mutually reinforcing or to some extent cross-cutting, one extending the other. How Are Close Marriages Attained? One very general idea as to how “closeness” of association is attained is the network neighborhood model: in a network with this characteristic (1) close ties are formed through reciprocity, (2) chains of reciprocal ties are formed, and (3) some chains are extended through weak transitivity, others form clusters or transitive triples of close ties, and others lacking transitivity link the clusters.6 For example, lineage A gives a bride to B and B reciprocates a bride to A: the close tie between lineages establishes the first element of the model. The second comes about when a chain of close ties is formed, for example, A to B and B to C (both reciprocat- Marriage, Rank, and Migration 233 ed). Weak transitivity requires only a significantly increased probability that lineages A and C, compared to an average pair of lineages, will form a tie, whether or not it is reciprocated. Strong transitivity may also occur, in which A and C form a reciprocated tie. In societies where the network model applies to marriages, as Lévi-Strauss might have expressed the transitivity principle, affines of affines tend to become affines. Hypothesis 7.1: Aydınlı marriages are characterized by the network neighborhood model. This hypothesis also entails that the Aydınlı are one of those societies for which the local clustering density of marriages between lineages is greater when the links between groups involve reciprocated exchanges of brides.7 That is, if the ties of A with B and C are reciprocated (B to A, C to A), then B and C are more likely to have a marital alliance. Recall from the small world model that local clustering density is measured by the extent to which each pair of marital allies of a family or lineage, such as ego’s lineage with those of alter 1 and alter 2, are themselves allies. In the network neighborhood model local clustering density should correlate with the weak transitivity of reciprocated ties: where there exists an A-B-C chain of reciprocated ties, A and C are more likely to intermarry, either asymmetrically or reciprocally. Hypothesis 7.1 is not true of all societies, and differences across societies in the network structure of marriage alliances are of considerable interest and importance for differences in the organization of cooperative and competitive ties. In societies with dual matrimonial organization, for example, an affine of an affine is concatenated with ego into one of the opposing and intermarrying groups, so that marriage links between groups cannot be transitive. The first thing we look for in models of network neighborhood is the means by which marriage itself, like friendship or acquaintanceship, might come to define a close relation between families. Family relations, especially affinal ones, may be quite brittle, with potential disputes over formal requirements, such as the repayment of the bride payments if the wife deserts the husband, or if tensions occur between the bride and groom or between their families, which is usually the case with elopements. We have also seen, however, that exchange marriages, in which two families each provide brides for the other, involve less formality in the sense that the requirements for bride payment are bypassed because it is regarded that the exchange is equal. Certainly, this doubling of marriage ties, which is also much favored among Aydınlı nomads, would 234 Chapter 7 seem likely to entail that each family is aware of and friendly with the other, more likely to visit back and forth, and hence more likely to form close ties. The second possibility that we look for in a model of network neighborhood is whether and to what extent, when close ties exist (like symmetric marriage exchange between families), they are more likely to be transitive, as with a friend of a friend becoming a friend: members of the circle of a family’s “close ties” are also more likely to become familiar with one another, to encounter each other in the context of their common ally, and eventually to get to know one another, perhaps visit each other independently, and to foster the conditions (or purposefully arrange) for their children to marry. In a society such as the Aydınlı, in which young women are closely circumscribed by their kin, the weak transitivity of close (reciprocated) ties through marriage is a much needed route to enlarging the pool of potential spouses within the group, without counting on random encounters. Analysis 7: Local Clustering and Curvature in Network Neighborhoods The network neighborhood hypothesis (7.1) is consistent with the progress anthropologists have made in understanding the evolution of reciprocity in human sociality (Boyd and Richerson 1988, Boyd, Gintis, Bowles, and Richerson 2003). Contemporary views on the importance of social reciprocity are consistent with the hypotheses we put forward here. Our concern with the network structure of reciprocity leads us to two further measures of network structure that have to do directly with the transition and reciprocity of social ties. The clustering coefficient (local clustering) for a network (e.g., Watts and Strogatz 1998) measures the extent to which, for each node in a network, others tied to it are themselves connected. Granovetter’s (1973) famous “strength of weak tie” hypothesis distinguishes between strong (frequent, close) and weak (occasional, distant) and argues that local clustering and hence transitivity among those connected to ego tends to be high for strong ties, while weak ties add span to a network and are typically less transitive.8 These tendencies are complemented by the network neighborhood model above that emphasizes situations in which chains of strong ties need not be strictly transitive and can bridge local clusters. Marriage, Rank, and Migration 235 Reciprocal marriage exchange is our indicator of strong ties between families or lineages, and we hypothesize that nonreciprocated marriage alliances, although balanced by bride payment, are weaker ties that reciprocal ones. Ties of reciprocal marriages between Aydınlı groups should be higher in closeness and trust as compared to one-directional marriage links offset by exchange of bride payments (not to be confused with the ++, + and – distinctions in Figure 7.2).9 Curvature is a measure for each node in a network of the extent to which others tied to it reciprocally—or strongly—are themselves connected. It is a strong-tie clustering coefficient that measures local clustering and transitivity for strong ties among those so connected to ego. Curvature captures an important feature of a social network: reciprocity is a crucial indicator of mutual recognition, of the presence of agency in building alliances, and hence of the closeness of ties. For this reason the clustering of reciprocal ties is far more likely to capture the phenomena that lead to the formation of cliques or larger clusters of ties. Curvature builds on Granovetter’s hypothesis that strong interactive ties—ones that are both frequent and reciprocally reinforced—are likely to be transitive and to form cliques or clusters. Eckmann and Moses (2002) use the curvature measure (density of ties among those linked by reciprocity to each node in a network) to define the local curvature of the network at each node, and to study network topology: How are local curvatures clustered in larger neighborhoods within the network? If curvature is a better measure for detecting the emergence of larger clusters in a network then the distribution of nodes that are linked and that have high curvature will give a better sense of the continuous topology of the network. We employ the measure of curvature to study the topology of Aydınlı nomad marriage networks. Figure 7.3 gives a visual test of Hypothesis 7.1 about the cooccurrence of symmetric exchanges at the dyadic level and the concomitant higher density of marriage ties predicted in the network neighborhood of each Aydınlı family. For marriages between maximal lineages with five or more members (scaled according to presence/absence of marriages) the figure shows solid lines for reciprocal bride exchanges and dotted arrows for one-way giving of brides. Where marriages mutually reciprocate between lineages, the arrows do not reciprocate. Three locally transitive network neighborhoods of strong ties are evident but there are also nontransitive chains of strong ties. Mustan’s lineage (#2) is the point of overlap between all three of these neighborhood clusters. The node labels of Figure 7.3 show the ID number of the highest ancestor of each lineage followed by the lineage number and then the size Chapter 7 236 indicator for the lineage, for example, 409#8:20 for lineage #8, with twenty married males. The structure of intermarriages in the figure can thus be checked against those in Figures 2.2 through 2.5. The genealogy in Figure 2.2, for example, shows reciprocal marriages between lineages #4 and #6. Two segments of lineage #1 were included and #10 omitted inadvertently but these particular errors make little difference to the analysis. Figure 7.3: Reciprocal and Directed Bride-Giving between Lineages 14 #9:5 H 409 409 Deveci Ali Oksüz Yusuf (emigrated in early times) 98 #1:9 19 1927 Fa of Yusuf (1926) and Ismaıl (1381) Koca 1228 122 Mustafa Mehmet Koca bey 224 1437 193 0 Fa of Yusuf (436) and Abbas (Mustan’s lineage) 507 1149 Ismaıl (98) Kırbaşı Fa of Ecevitli Network Neighborhoods with Curvature The sizes of nodes in Figure 7.3 reflect the clustering coefficient (for strong and weak ties) for each node. For a given lineage, it is calculated as the proportion of pairs of marriage allies of that lineage (to which brides have been given) that are themselves marriage allies. The measure is zero for nodes that lack two outgoing arrows or edges (the white circles: lineages #8 and #9), and indeed these are the two most marginal lineages, #8 being the descendants of the immigrant shepherd who affiliated with #5, like the descendants of #9 who affiliated with the more important lineage #1. For nodes with two or more outgoing arrows or edg- Marriage, Rank, and Migration 237 es, the measure takes values between one (the black nodes) and zero. Node 858#3, for example, is a black node and has a neighborhood density of one because it has four marriage allies and all are allies of one another. Lineages #1, #2, #5, and #7 are all allied in a cluster of reciprocal exchanges of brides, but they also have weak alliances with other lineages, getting or giving wives but not reciprocating the exchange of wives with them, and their indices of local clustering density are intermediate because many of these weaker allies are not in turn allies of each other. Intermediate density is a characteristic of central nodes that have broad span but lower local clustering density in their network neighborhoods because their ties are spread so widely. Note, however, that in the core of the network linked by reciprocal marriage alliances, every one of the major lineages is only one or two steps away from the others in terms of strong and locally dense linkages. Figure 7.3, however, is still incomplete as a test of hypothesis 7.1 because it renders only whether there are marriages or exchanges but not how many. The hypothesis of network neighborhoods (7.1) is confirmed in the visual pattern of Figure 7.3 given that there are a considerable number of strong-tie chains (A to B to C) that are not transitive at the strong-tie level and that these chains are often transitive at the level of weak ties. Other factors are also relevant to differences in local clustering density at the level of individual lineages. To have high local clustering a lineage needs to be isolated, separate, or embedded in some way. This is true at one end of the continuum for lineage #3, dependent on close allies because of its small size and lacking the wealth or leadership status to reach out broadly in its marriage patterns. It is true at the other end of the continuum for the more remote sheepherding lineages (#6 and formerly #4). Among the remaining lineages, those in leadership positions or their strong allies tend to have the medium local (network neighborhood) clustering density that is typical of broad span in a network. Very marginal lineages heavily engaged in emigration (#7), having exited through migration (#9), or immigration, having arrived from a village origin with very weak ties to their major sponsors (#8), might be expected to have very low or minimum local clustering density as their marriage ties are likely to be haphazardly spread around the network. A summary of these characteristics is given in Table 7.1, which shows for each numbered lineage the clustering and curvature indices (the former in qualitative terms, the latter as a coefficient that varies from 0 to 100). The associated network patterns and attributes such as alliance and leadership patterns will be discussed later. Lineage #10 was inadvertently omitted in the figures and table although they intermarried Chapter 7 238 with lineages #2 and #7. Table 7.1 Local Clustering Density and Lineage Attributes Lineage #5 #4 #6 #9 #1 #1 #7 #8 #2 #3 Clustering and Curvature Medium 87 Maximum 100 Maximum 100 Medium** 0 Medium 100 Minimum 0 Medium 100 Minimum* 0 Medium 60 Maximum 100 Network Pattern (So.=Southern) Tribal wives Tribal wives Traditional Traditional Educated (So.) Emigrate (So.) Emigrate (So.) Emigrate (So.) Mediators Emigrate (So.) Attributes wealthy/influential early on wealthy-sheep/separate/influential marginal/became sheepherders political allies of influential #1 get wealthy allied to #2/influential marginal/exited through migration marginal/many emigrated earlier marginal most central/relinking/influential very small/dependent *Affiliated with lineage #5 at a time when it had decreased in wealth ** Affiliated with lineage #1 at a time when it was large and important The Lowland-Highland Continuum In many rural societies, there is a continuum from more remote subgroups to those more connected into the larger national networks in which connections and differences between subgroups vary along the continuum. This suggests a topological hypothesis about the overall shape of the marriage alliance networks. An analysis and scaling of proximities between lineages based on the statistical tendencies toward intermarriage would reveal more about the topology of the network. The following hypothesis suggests the form this topology should take given that the Aydınlı tend to migrate north to summer pastures and back to southern lowlands in multiple routes far off the highway. Hypothesis 7.2: The topology of Aydınlı nomad marriage networks tends to have one major dimension of variation, along a lowland-highland continuum. Figure 7.4 uses a different version of the data used to draw Figure 7.3, which used a Pajek automatic drawing with the option to minimize line length but without differentiating strength of lines. Here, instead the strength or weakness of ties is measured, by their departure from the frequencies of intermarriages expected by statistical independence. Those lines representing frequencies greater than expected by chance are shown by the heavier lines, with thickness proportional to the statistical Marriage, Rank, and Migration 239 significance of the excess over expected frequency.10 These tend to form a bowl-shaped scaling of nodes. The more-frequent-than-chance reciprocal links along the perimeter of the bowl tend to form chains that are transitive not over longer distances but within more compact linearly organized neighborhoods. Those links that occur less commonly than expected by chance, given the total distribution of marriages, are rendered as the lighter lines. The automated drawing option used to scale the figure uses weights on the edges, with the lighter dotted edges that represent negative weights spreading nodes apart and connecting the more distantly related nodes along the bowl-shaped structure. The outer perimeter of the bowl thus represents a one-dimensional topology of the marriage network, consistent with Hypothesis 7.2. Figure 7.4: Reciprocal and Directed Bride-Giving, Lineages Scaled 2248#5:96 1149 1930 507 :68 #9: 95 409 #8:20 5 1437 1228 98 #1: 19 1927 The linearity of the marriage distances between lineages in the scaled Figure 7.4, along the semicircular outer rim of the scaling that reflects marriage ties that occur more frequently than expected by chance, is consistent with Hypothesis 7.2 of a scaling continuum on endogamous marriages that reflects an ordering of the lineages from highland to lowland following the network patterns listed in Table 7.1. The lineages in that table are ordered identically to Figure 7.4: from #5 and #4 on the right of the semicircle, the lineages more frequently married to tribal wives (highland oriented), to the series #1-7-8-2-3 moving from center to far left of the semicircle, which are the lineages that more frequently emigrate to the southerly villages (listed as “So.” in Table 7.1) and are thus more lowland oriented. The study of exogamous marriages was not part of Johansen’s re- 240 Chapter 7 search plan. In her collection of census and genealogical data, however, she systematically distinguished nomads from villagers and townspeople and in cases in which nomads married outsiders she always wrote down the names of the tribes, villages, or towns that her informants mentioned. The places mentioned were not exact locations, such as hamlets or outskirts of a given village, but the names of villages nearby where a spouse had come from. This was a common pattern for locational identifiers because many of the people whom the nomads married came from the families of former nomads who had settled near villages or towns but not in the town or village center. Although not all informants listed where people were from or where they emigrated, the data that she did collect allow the testing, even if imperfectly, of the following hypothesis. Hypothesis 7.3: The scaling continuum of Aydınlı lineages in terms of endogamous marriage alliances (as in Figure 7.4) is also reflected in differential rates of marriage to external groups ordered from highland to lowland, with exogamous marriage to those living near more remote villages or from other tribes at one pole and exogamous marriage to those living near more central villages, towns, and cities at the other. If this hypothesis holds, we have three sources of congruent support for a highland-lowland orientation continuum among the lineages: (a) from Table 7.1, rates of emigration to the southerly lowlands versus marriage with highland tribal members, (b) the scaling pattern of Figure 7.4 that reflects rates of intermarriage between lineage members that are greater than expected by chance, and (c) spatial patterns of the distribution of spouses for exogamous marriages. Table 7.2 tests hypothesis 7.3 by tabulating lineage and geographic origin of the spouse for exogamous marriages. It correlates the scaling of interlineage marriages with the geographic origin of spouses from outside. Although systematic data collection on exogamous marriages was not part of Johansen’s research plan, forty-eight persons considered by informants to be villagers or townspeople residing outside the clan were not identified by naming a nearby village or town as place of origin. Johansen did not ask for locational data unless it was offered. Table 7.2 gives geographical location data for thirty-four persons for whom information was volunteered that White was able to use to identify geographical location.11 In toto, Johansen wrote down locational identifiers for fifty-four spouses but not for another forty-eight, so these data are about 53% complete. We consider that a sufficient sample for analysis. Although there might be some bias in the names written down, there is suf- Marriage, Rank, and Migration 241 ficient data for testing Hypothesis 7.3. Table 7.2: Exogamous Marriages in More Recent Times Ordered from Distant Lowlands (K) to Proximal Highlands (V) and other Tribes (W) Correlated with Scaling Order of Lineage According to Endogamous Marriages Spouse Ego’s From: Lineage KL Kurdish, Distant M Kırıkhan N Adana (Misis) O Osmaniye Town + P Imamoğlu Köy Q Kara Tepe Köy R Kadirli Town ++ S Kozan Town T Saimbeyli Town U Tufanbeyli Köy V by summer camp W Other Tribes #3 #2 Ecevitli 3 1 1 1 1 3 1 1 2 1 4* 3 1 1 22 #7 #1 #6 #4 1 #5 #10 9 1 1 1 1 1 2 1 2 1 1 4 4 1 9 + Kırmacılı Köy ++ Eyüp Köy * Eski Mantaş Köy (3), Orta Köy (1) There is a moderate correlation (tau-b=.27, gamma=.38) that is statistically significant at 1 in a thousand (p=.0006) by chance alone between the order of lineages from the scaling in Figure 7.4 (endogamous marriages rates), given in the column labels of Table 7.2, and the geographical ordering of places of origin of spouses in exogamous marriages in the rows of Table 7.2. Examining the frequencies in the cells of the table, we can see that the correlation is not linear but rather a constraint relationship: The more remotely oriented lineages (#4, #5, and #9) have only remote ties while the less remotely oriented lineages have exogamous marriages with a fuller range of outside villages or towns or hamlets on the peripheries of cities. Marriages with families living near towns or cities are mostly with families of former nomads who have settled in villages or on the urban peripheries, such as Misis as a village 30 km east of Adana, or Eyüp Köy (=village) as an outlier of Kadirli town. The ordering of places runs from distant (Kurdish, Kırıkhan) or lowlands (Adana, Osmaniye, Imamoğlu) to the northern highlands paralleling the 242 Chapter 7 routes to the winter pastures. It ends with marriages with other mountain nomad tribes. Table 7.3 repeats the cross-tabulation of Table 7.2 for each historical generation for which we have data on place-names. In the earliest generation (d) we have only marriages with other tribes, except for one marriage into the Kadirli area. In generations e, f, and gh (combining g with the one case in h) the exogamy-endogamy continua correlations are fairly uniform although with a slightly higher gamma in generation e (taub=.23, .19, .23, gamma=.45, .24, .33; these fail to reach significance because of the smaller sample size). The pattern in Table 7.2, then, is fairly constant across generations. Lineage #3 is the major exception to the constraint and correlation patterns in Tables 7.2 and 7.3, but many of its members resettled as villagers rather than intermarrying with them. If this lineage is dropped, the correlations drop slightly (tau-b=.21 and gamma=.33). Lineage #2 (Mustan’s), not surprisingly, has the highest number of exogamous marriages, over half of which are with other tribes, either women marrying in or men marrying out. They have a disproportionate number of marriages with other tribes. If both these lineages are dropped the correlation rises (tau-b=.46 and gamma=.68) and even with many fewer cases the significance of the correlation rises (p=.00004) because the correlation takes on a more linear relationship. Marriage, Rank, and Migration 243 Table 7.3: Endogamy-Exogamy Continuum Correlations by Historical Generations Count Exogamous Marriages d V3 LINSPO Lineage of Clan Member 1 #3 R W Total e V3 M S T W V3 K L N P Q R S T U V W Total g h V3 Total O Q R S W 4 #7 6 #1 7 #6 9 #5 1 8 #4 1 2 2 1 2 1 10 #10 3 Total 1 9 1 3 3 10 1 Total f 2 #2 1 6 1 1 2 2 4 1 1 1 2 15 8 4 2 4 1 19 2 1 1 1 1 1 1 1 2 3 3 1 1 2 1 1 5 2 1 1 16 5 1 4 3 34 1 3 1 1 2 7 11 1 3 22 1 1 1 1 1 1 2 1 1 1 9 3 16 2 1 1 1 1 1 4 5 1 3 11 2 1 2 2 Key: Symbols K through W are the place names defined in Table 7.2, running from the most from distant lowlands (K) to proximal highlands (V) and other tribes (W) Figure 7.5 uses the data of Table 7.2 to address hypothesis 7.3 in a visual schematic that shows the geographic locations of out-marriages where members of the lineages have intermarried or settled, as in Table 7.2. Those locations show how the external orientations of the lineages differ in the spatial distribution of exogamous marriages. Because the scaling of lineages from endogamous marriage patterns in Figure 7.4 shows a one-dimensional continuum that corresponds to a remoteness-integration continuum, the idea in Figure 7.5 is again to see whether there is a corre- 244 Chapter 7 sponding geographic difference in the pattern of exogamous marriages. We are looking for a correspondence or correlation between the topology of interlineage marriages within the clan and the geographic topology of exogamous marriages according to the tribal or village origin of the spouse. The map in Figure 7.5 is completely schematic rather than geographic in the placement of lineages. The placement of the lineages on the map is not based on geographic location but on the scaling order shown in Figure 7.4. The north-south alignment of lineages on the left of Figure 7.5 merely replicates the semicircular scaling in Figure 7.4 in order to show visually the evidence supporting Hypothesis 7.3. Recall that lineage #3 at the lower end of this continuum is one with many emigrants who settle near towns. At the upper end are the more tribally oriented sheepherders (#6, #4) along with lineage #5, which provided the early leadership of the clan and which continues to be wealthy in herds. Actual migration routes are not shown on the map. The lines show some of today’s larger paved roads and highways and simply provide a geographic orientation within which to do a rough test of hypothesis 7.3. The road from Adana northeast to Saimbeyli parallels multiple routes along which the Aydınlı migrate between their summer pastures in the north and their winter lowland areas around Adana and Osmaniye, the cities closer to the coast. The summer pastures of Beypinar and hamlet of Arnaçkavak are shown just above Saimbeyli, while some of the many winter pastures (actually vast areas scattered between villages) are shown by the gray circles in the Çukurova in the south, the farthest east being that of lineage #6. Lineages changed their winter pastures often, depending on where they were cheapest, so the winter pastures shown on the map are merely illustrative. The villages that Aydınlı have married and/or settled into are marked in black. Villages and cities are placed geographically in Figure 7.5, while the placement of lineages shows schematically which ones are the more northerly mountainous populations, including numerous nomad tribal groups (also not placed geographically). The alignment of lineages is done to match the order of the multidimensional scaling in the automatic drawing of Figure 7.4, which scales lineages by the closeness of their ties. The fit is very close between the two orderings, and reflects a polarity between nomads intermarried with villagers and those who orient to the more remote mountain populations of other nomads. Marriage, Rank, and Migration Figure 7.5: Lineage Alignment and Ties along the Highway (Migration Routes are also north-south). 245 Lineage #5, the highest ranking of the lineages in the early migration from Lineages boxes #1-#9 are not geographic Aydın, and having the #5 leadership to the clan in Sheep #4 Other Tribes the last half of the nineherders teenth century, would ap#6 pear from Figures 7.4 to #9 match a pattern expected of a lineage oriented to a Summer Pastures #1 more traditional goat herding and mountainous way of life. That is, they are at H#7W#9 H#7 the “conservative” pole of the tribal-urban or re#2A moteness-integration continuum in their endogamous marriages with #3 lineages at the remote end of the continuum and be- H#8W#2 cause all their exogamous marriages are with other tribes. Almost at the other exGulf of treme from #5, lineage #2 Iskenderun of Mustan and his brother Key: H&W Migration as matchmakers projects a Wife assimilated very different pattern of Settled in wife’s town adaptation, one of mediaSettled in husband’s town tion between villages and Husband assimilated towns, the clan and other Pastures: summer winter tribes, just as they played a 100 km central role in mediation and integration of the line- Supplementary Key: Turkish place names lack diacriticals H&W migration indicates that both spouses emigrated ages within the clan. H#7, W#9 are indicators for a husband of #7 and Wife of Lineage #9, in the upper middle of Figure 7.4, #9represents a third extreme: complete exit from nomadism through emigration. Nomad life clearly has a number of very different and viable poles of adaptation. The pattern of external ties reproduces the semicircle of marriage proximities in Figure 7.4 and makes clear that lineages #2-3-7-1 are linked to sites near larger villages, while #9-6-4-5 include two 246 Chapter 7 sheepherding lineages that range higher in the mountains (#4 traditionally; #6 more recently) and they are more closely intermarried with other nomad tribes. Local Clustering Density and Continuum Scaling Figures 7.3 and 7.4, based on the same data (interlineage marriages) look very different, as indeed they are. The latter scales lineages according to departures from statistical independence in the observed frequencies of marriage. It identified a highland-lowland continuum, confirmed by the pattern of exogamous marriages, and it reflects social proximities between lineages along the continuum. Figure 7.3 was based on raw marriage frequencies and simply looked at reciprocated ties and local clustering among 3-4 lineages at a time, that is, a complete network of reciprocal marriage exchanges. There is no correlation between the highland-lowland ordering and the clustering or curvature coefficients. If we renumber the lineages in Figure 7.3 according to the highlandlowland orientation, it would be apparent that the dense “local clusters” with complete networks of reciprocated marriages do not represent highland-lowland clusters of lineages, but cross-cut and bringing together the poles of the highland-lowland continuum. The clusters are 1-2-9, 2-7-8, and 2-4-6-9.12 The cross-cut between low-frequency clusters of reciprocated marriage alliances and the higher than expected frequency of ties among neighbors on the highland-lowland continuum gives some indication of how alliances patterns and spatially based interactions might cross-cut to form the strong-tie small world of Aydınlı nomads discussed in Chapter 5. The Cascade of Lineage Segments and Their Fractal Relationships One of the problems in network and kinship analysis is how to study networks with hierarchies or units that have complex overlapping or incomplete hierarchies. For a society with lineages, for example, maximal lineages are hardly the only appropriate social units for study but only units of convenience because they are the highest level nesting of constituent and more effective units. As we have noted, the minimal lineage and residential unit is a 3-generation patriline, the effective lineage where genealogical links are almost always definitely known and accurate is a 5-generation patriline, and the maximal lineage may be deeper, Marriage, Rank, and Migration 247 embracing many effective sublineages. If local actors refer to kin with common descent from easily remembered ancestors at fairly shallow generational depths, these will be sliding lineages that shift over time with new generations. These more effective units often split apart within a maximal lineage. Conversely, as members of such units become more distantly connected with the passage of generations, they may try to recreate internal cohesion through intermarriage. A common tendency noted in the Middle East (e.g., Peters 1991), above five generations, is to attribute common ancestry to a patrilineage when in fact it is only that various smaller lineages have so densely intermarried that they create a fictitious common ancestry. Hypothesis 7.4: A high degree of local clustering densities applies not just to lineages (Hypothesis 7.1) but fractally, to overlapping sublineages. To test this hypothesis we extend our approach from network neighborhoods linking discrete kinship units such as maximal lineages, as we did for Hypothesis 7.1, to overlapping subunits, we use Eckmann and Moses’s (2002) curvature method to study marriage networks among lineage segments.13 These segments overlap if they belong to the same maximal lineage, but they are distinct if they belong to different maximal lineages.14 We apply Eckmann and Moses’s methods of analysis once again here but this time to sublineages of depth five. This gives us a means of seeing the “fractal” replication of units and subunits and how sublineages are also knit together. Our analysis of maximal lineages has already rehearsed the concepts and measures that we need in order to do a more fractal analysis of how subunits are interrelated by reciprocal and transitive ties. Figure 7.6 shows, for marriages between sublineages five generations in depth—a unit where ancestors are likely to be remembered—a scaled distribution of reciprocal bride exchange as well as one-way giving of brides. The former is shown by the solid lines and the latter by dotted arrows. As before, the size and shading of nodes reflects the clustering density (transitivity) of local neighborhoods: the large black node has a density of one, the medium gray nodes have densities above .75, and the white nodes (mostly in the upper part of the picture) have densities below .75. Consistent with Hypothesis 7.4, these are all very high. Figure 7.6: Reciprocal and One-way Marriages among Sublineages: 248 Chapter 7 Lineage #3 gen 1 Lineage #2 gen 4 Lineage #2 gen 3 Lineage #3 gen 2 Lineage #5 gen 3 Lineage #2 (Mustan’s) gen 1,2,3 #5g2 #5g1 #7g2 #1g2 #7g1 #1g1 #1g3 #1g2 #6g1 #4g3 g2 g4 g1 Legend: Sublineages and labeled by the generation of their highest ancestor within the lineage, for example, #5g2 is a segment of lineage #5 with an ancestor in generation 2. Height in the vertical axis shows “fractal” departure from the more general alignments of marriages among lineages shown in Figure 7.4 that replicate for a majority of sublineages at the bottom Figure 7.6 is scaled in the same way as 7.4, using Pajek’s automatic drawing (see Glossary) that treats the difference between expected and actual marriage frequencies as a proximity weighting in the scaling. Marriage alliances that occur less frequently than expected by chance are shown as dotted lines and are weighted negatively, so they are expected to be longer and the solid lines shorter as a result of scaling. Along the bottom axis are listed the lineages and generational ancestors for each of the sublineages. For example, #5g2 is the son of #5g1 as sublineage head. Except for #5, which is at the opposite end of the scale, the order of lineages is exactly the same as in Figure 7.4. But instead of being in a linear scaling order, sublineages of #3, #2 and #5—from diametrically opposite ends of the continuum—have migrated up and out of the plane at the bottom because they have a wider span than the other sublineages, and for Mustan’s sublineages (#2, generations 1, 2, 3, circled in the figure), the clan integrators, this is especially so. Many of these “higher” lineages in this figure have lower local clustering density (i.e., shaded white) due to their greater span, which detracts from local clustering. Sublineages in #5 change position in the scaling because of their Marriage, Rank, and Migration 249 links with #2 and because their few links with the “vacated” lineage, #7, which were statistically insignificant when considered at the lineage level, are more significant when broken down into the temporal segments of Figure 7.6. We see in fact that in generations 1 and 2, lineage #5 was allied with #7, but by generation 3 is more closely linked with lineages #4 and #1, at the opposite ends of the scale, but this is also after the period in which the early arrivals of founding lineage are amalgamated into the clan, that is, a time of structural change. With the exceptions noted, the fact that nodes belonging to the same maximal lineages tend to be proximate in this figure, which is scaled by closeness of ties, is an indicator that sublineages that share a patrilineal ancestor tend to both reinforce their cohesion through marriage alliances among themselves and share patterns of alliances with other groups. Figure 7.6, then, begins to show some of the dynamics of shifting alliances. Eckmann and Moses’s method, applied here to the study of the segmentation over time of patrilineal groups into cascades of descending sublineages, has a major payoff in visualizing the dynamics of shifting alliances. What the vertical dimension represents in this figure is precisely the variability of shifting alliances over time, with lineages #2 and #3 lifted off the plane of stable alliances below. Recall that #3 constituted a poor and small lineage allied in its marriages and fortunes with #2, which started as a poor lineage but advanced in status and wealth due to its abilities in making marriage alliances. Lineage #5 was originally the highest in status and wealth, allied with the Cırıklı tribe and with lineages #2 and #7 in early times. Structural Properties of the Sublineage Marriage Network All of the curvature analyses thus far have shown that the network of lineage and sublineage marital alliances has an overall topology, and that topology is built out of local clustering and transitivity. This can be tested formally. Hypothesis 7.5: A topology of clustering and transitivity in the Aydınlı sublineage marriage alliance network will show up in a triad census or reciprocal and nonreciprocal marriage alliances. Hypothesis 7.6: The triad census of the Aydınlı sublineage marriage alliance network supports a topological model of a small world (clustering, transitivity, and short chains). A triad census is a statistical inventory of the different types of structural 250 Chapter 7 relationships among nodes in a network, taken three at a time, and classified by the patterns of directed and reciprocated arcs. The sixteen possible patterns of arcs are shown in Figure 7.7, each labeled by three numbers showing how many pairs of nodes have reciprocated, unreciprocated, and no arcs, and in some cases whether those arcs point down (D), up (U), or across (C). Figure 7.7: The Sixteen Possible Triads for a Triad Census Results of the triad census are given in Table 7.4. Numbers of the sixteen types of triads are listed on the left, ordered by the topological models with which they are consistent named on the right. The standard output for triads testing shown here (Batagelj and Mrvar 2001) organizes the triad types in blocks (here separated by spaces) corresponding to successive models with additive features. The triad types marked “Forbidden” may actually occur and have interesting structural properties but they fail to fit any of these models of balance, clusterability, or transitivity. In the three center columns are the numbers of observed triads (ni), the number expected from the null hypothesis of statistical independence Marriage, Rank, and Migration 251 (ei), and the ratio of (ni-ei)/ei. Enclosed with circles are five triads that are significantly greater than expected by chance. The results show that the data fit a model of clusterability and transitivity, with balance included as a type of clustering, plus a triad that is normally forbidden in this model, the chain of two reciprocal links, A to B to C, with no transitivity, either strong (A C reciprocity) or weak (A C directed edge). The fitted model includes clusterability and balance of reciprocal ties without ranked or hierarchical clusters but also intransitive chains of reciprocal ties. This is the pattern expected in a network where the strong or reciprocal ties conform to the definition of a small-world in which ties are clustered by also connected so as to reduce the average distance between nodes. The high frequency of the last triad in Table 7.4, the intransitive chain of reciprocal ties (a forbidden tried in the pure clustering model but which occurs here more frequently than expected by chance) supports Hypothesis 7.6. Because a small world has clusterability and short average distance, it cannot have perfect transitivity with more multiple clusters because they would then be cliques with no links to other cliques. Hence, the data fit the small world model, consistent with our hypotheses from Chapter 5.15 They also support hypothesis 7.5 and back up the findings on curvature. Figure 7.6 has yet another interpretation, which is the temporally “fractal” pattern of marriage networks through time. The stable part of the scaled pattern is the largely stable “floor” at the bottom; changes boiling up to higher and higher levels above this floor carry an interesting implication: if your allies change position, your position changes with them even though you do not change allies. This is analogous to the concept of structural equivalence in social networks, but it is here applied to social dynamics. The implication is that not everybody needs to change in the network for the configuration to go through radical transformations; changes are self-amplifying structurally speaking, and so the “upward” variance of potential change is not self-dampening. Here, we see this only at the local level to be especially strong for lineages #2 and #3. These are lineages that were poor, were early adopters of migration, and became heavily intermarried into the villages; most members of #3, who are highest in the vertical axis of Figure 7.6, had emigrated by 1900 and did not sign the 1933 land agreement (see Figure 7.5). Those in #2, who were more diverse and structurally endogamous within the clan and whose segments are distributed along the fractal-change vertical axis of Figure 7.6, clove to the clan with little emigration. Chapter 7 252 Table 7.4: Results of the Triad Census among Sublineages Type Number of Expected Ratio triads (ni) (ei) (ni-ei)/ei Model 3 - 102 16 - 300 146 128 55.06 29.01 1.65 3.41 Balance (reciprocal ties) Balance (reciprocal ties) 1 - 003 73 14.60 4.00 Clusterability (no ties) 4 - 021D 5 - 021U 9 - 030T 12 - 120D 13 - 120U 26 46 26 43 57 55.06 55.06 123.47 69.22 69.22 -0.53 -0.16 -0.79 -0.38 -0.18 Ranked Clusters Ranked Clusters Ranked Clusters Ranked Clusters Ranked Clusters 2 - 012 140 98.22 0.43 Transitivity (one tie) 14 - 120C 45 15 - 210 175 138.44 155.22 -0.67 0.13 Hierarchical Clusters Hierarchical Clusters 6 - 021C 69 7 - 111D 103 8 - 111U 103 10 - 030C 15 11 - 201 135 110.12 123.47 123.47 41.16 69.22 -0.37 -0.17 -0.17 -0.64 0.95 Forbidden Forbidden Forbidden Forbidden Two-chain (reciprocal ties) Chi-Square: 1012.1780*** A Small-World Network of Strong Ties: Local Clustering and Short Distances among Reciprocating Sublineages Granovetter (1973), in the most cited of all network studies, hypothesized that it is the weak ties that serve as the bridges between clusters in social networks, with strong tending to close inward and form cliques. One of the open questions in network research is the relationship between small worlds and strong versus weak ties, such as reciprocated marriage exchange between Aydınlı lineages (our “strong” ties) and the unreciprocated marriages (“weak”). Yet, the data in Table 7.4, consistent with Hypothesis 7.6, show strong ties as the “forbidden” bridges between cliques in the clustering model (and weak tie directed paths that Marriage, Rank, and Migration 253 are not transitive are infrequent, for example). Still, Figure 7.4 and Figure 7.6 give the impression of a network integrated overall by the generalized exchange created by its weaker ties, which especially is a possibility because weak-tie integration would involve a one-directional flow of brides as against bride payments. Hypothesis 7.7: Strong ties (that is, reciprocated marriage alliances) act as the primary bridges reducing network distances in the links between sublineages. Figure 7.8 tests this hypothesis both visually and quantitatively by separating the strong- and weak-tie networks, scaled in a form equivalent to Figure 7.3. At the bottom of the figure, the graph of distances between nodes that the two separate networks create shows that pairs of lineages linked by reciprocal marriages are twice as common (132 versus 63) as the directed marriage alliances, and they link every pair of lineages by no more than a two-step path. The weaker ties of generalized exchange are redundant in that they span longer distances in the network, while the reciprocal ties, although clustered around network neighborhoods, also integrate the lineages and lineage segments in very short distances. Thus, Hypothesis 7.7 is supported. The strong ties, which we have defined so as to maximize the probability that they indicate relationships of trust and close familiarity, constitute a “small world.” But they also facilitate the weak-tie network that involves a one-directional flow of brides as against bride payments, and that too constitutes an integrated network, and one that is even more cohesive that the strong-tie network if we use as the measure of cohesion the minimum number of nodes whose removal is needed to break each of these networks into large disconnected segments. In Figure 7.8, as before, the size of the nodes indicates local clustering density. The solid triangles tend to be within localized parts of the network, so that strong ties are consistent with restricted exchange. But, as noted, strong ties also concatenate to produce low distances between different sublineages. Each sublineage is again numbered by its highest ancestor and the number of its men. Nodes 432:15 and 433:17, for example, are overlapping sublineages of size 15 and 17 that, along with 436:13 and 438:26, share membership at a deeper level in a common lineage (#4). The nodes also cluster by lineage. Figure 7.8: Reciprocal and Directed Bride-Giving among Sublineages 140 120 100 80 Chapter 7 254 Hypothesis 7.8: It is the strong and not the weak ties that create local clustering densities. Figure 7.9 contrasts weak and strong ties to test this hypothesis. Referent sublineages having two other sublineages as marriage allies are classified as before by whether their links to the allies involve zero, one, or two symmetric marriage exchanges (the horizontal axis), and the bars show the extent to which local densities obtain, that is, whether the two allies are themselves linked by a marriage alliance (the leftmost bar graph) or by symmetric marriage alliance (the rightmost bar graph). In each case the number of three-lineage configurations in the x-axis remains the same, as shown by the line graph. Figure 7.9: Marriage Alliances of a Referent Lineage with Two Other Lineages and the Local Clustering Density (transitivity) of Triples 0.9 0.8 0.7 0.6 0.5 1600 1400 Strong tie 1200 1000 800 0.4 0.3 0.2 0.1 0 600 400 200 0 0 1 2 Reciprocities for Two Marital Allies 0.9 0.8 0.7 0.6 0.5 1600 1400 1200 1000 800 0.4 0.3 0.2 0.1 0 600 400 200 0 0 Any Tie Triples 1 Reciprocities for Two Marital Allies 2 Reciprocal Tie Triples Because the leftmost bar graph in Figure 7.9 shows how many ally pairs have a weak tie (directional marriage alliance) and the rightmost graph Marriage, Rank, and Migration 255 shows the same data for ally pairs with strong ties (reciprocal marriage alliance), we can see that when the ally pairs are both themselves reciprocal alliances, there is a major jump to a high level of local clustering density (i.e., transitivity), and over 80% of that density is also obtained by a reciprocal alliance. The data in Figure 7.9, then, strongly support Hypothesis 7.8 and Granovetter’s hypothesis that the strong ties do the clustering in this social network. Weak ties do not have one of the crucial properties of small worlds because they do not form clusters. It is the strong ties that constitute a “small world problem” for the nomads, notably, that close relations (trust, reciprocal exchange) are highly clustered, but they provide a “small world solution” because they are sufficiently intransitive not to create mutually exclusive clusters and through their intransitivities, spanning different groups, to reduce the average strong-tie distances in the network.16 Hypothesis 7.9: The weak-tie network of directed marriage exchanges, which constitute exchanges in the flow of bride payments in which exchange for brides, is facilitated by the existence of strong-tie connectivities throughout the network, and, more specifically, tends to have a fractal self-similarity not with the local clustering but with the more global cohesion created by strong ties in the network. This hypothesis is a bit abstract, but it conveys the idea that the weak ties do not mimic strong ties in their clustering, but rather in their cohesiveness, as defined elsewhere. Note that lineages with many ties will necessarily have lower local clustering but contribute to overall cohesion. Figure 7.10 shows a relatively constant correlation between local clustering density for a given lineage and the number of marital alliance partners for that lineage. As the number of partners goes up, the local clustering density goes down. A self-scaling or fractal quality of the segmentary kinship networks is that the slopes of these two correlated distributions are the same, although reciprocal bride exchanges have a higher local clustering density. Hypothesis 7.9 is supported in that nodes with many links contribute to global cohesion both in terms of weak and strong tie networks, and the measure of local density for weak ties behaves like that of strong ties. This supports the idea that the weaker marital alliances are in some sense an “extension” of ties through the stronger relations of closeness and trust implied by reciprocal linkages, and they have important similarities as well as differences. Chapter 7 256 Figure 7.10: Bride Exchanges between Lineages and Transitivity of Triples, Correlated with Number of Alliances with Other Lineages 0.9 Local Density, no bride exchange 0.8 0.7 Local Density, bride exchange 1 0.6 Linear (Local Density, bride exchange) Linear (Local Density, no bride exchange) 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 60 70 Number of Alliances with other Lineages The social networks of kinship and marriage, then, tend to define the avenues by which people connect and extend their circles of familiarity leading to new marriages. More detailed modeling of how such social worlds are constructed might take the succession of generations into account, and show how at any one point in time, the existing “network neighborhoods” for each family, extended to include potential spouses one or two steps away through chains of close family connections, would set the context for the choice of a specific spouse. More detailed ethnographic vignettes in this context might substantiate the crucial role of webs of female kin in facilitating marriage choices. Having located potential brides or grooms within these family circles, however, the next problem for network analysis is how the relative generations of potential spouses figure in to actual marriage choices. We explore that issue first before getting back to issues raised by mapping of social structure from marriage frequencies, their structure, and their implications for group dynamics. Equality in Generation between Husband and Wife Having mapped out the network neighborhoods that create the social space in which young women can become well known to a family and be taken as wives by Aydınlı men or by the arrangements made by mediating female kin, we now explore the appropriateness of a particular spouse, whether man or woman. The fact that marriages between partners of different generations hardly ever happened had to do with the Marriage, Rank, and Migration 257 meaning of sexuality. Although Islam does not declare sexuality as sinful so long as it happens between spouses, among the Aydınlı it is looked at as something unclean and shameful. This ambivalence contains pride about one’s fertility on one side and extreme bashfulness on the other side. After Johansen had already been his “elder sister” for many years, for instance, the younger son of her family gave her a detailed report of his wedding with his wife, with whom he had been passionately in love. Recounting the point at which he had to display modesty and was pushed by his younger male relatives into the tent where the veiled, maiden bride waited, he abruptly broke off by saying “Orada ne bok yedim - biliyorsun” (=What a shit I have eaten there—you know). Problems of sexuality are not discussed, except as jokes between people of the same sex and same generation. This goes so far that young women visiting their parents’ homes showed their new babies to their mothers and sisters but were ashamed to show them to their fathers because they were testimonies of their having had sexual intercourse. Young men too are ashamed to tell their fathers that they are in love with a girl to have their fathers negotiate to get her as a bride. They turn to their mothers or mother’s sisters, who talk to the men of their generation. Kressel (1986:177) observed similar behavior among the Bedouins. Same-generation marriage prescriptions, however, are also a resolution to the potential conflict of principles that would arise in a first marriage between off-generation or markedly age-discrepant bride and groom. Husband and wife are taken as having generational equality in first marriages within the clan, which is a marker of the equality between their respective families. The Aydınlı are not a society in which wifegiving versus wife-taking confers differential status of the two families. This stands in marked contrast to another very different hypothesis about FBD marriage in the Middle East (e.g., Kressel 1986), namely, that the giving of women to another lineage typically lowers the lineage’s rank relative to the other lineage and that in this context parallel cousin marriage provided a means for maintaining status: Its acknowledged utility is the enhancement of social esteem. The custom is anchored in the notion that the sexes are unequal in intercourse; the male is cast in the role of conqueror or humiliator, while the female is the passive receiver. Because the parties to intercourse represent their groups of origin, daughters of lower-status kin groups are matched with sons of superior ones (hypergamy) or, ideally, to equal ones (isogamy). Grooms who are paternal cousins are the most equal. Therefore, they are the most likely choice for evasion of humiliation, insinuated by giving the daughter in marriage to a nonagnate. (Kressel 1986:178) 258 Chapter 7 The status equality of intraclan marriage among Aydınlı nomads is not the result of FBD marriage but present before the fact of marriage. One aspect of status equality between wife-takers and wife-givers is the age equality between spouses. The following hypothesis looks deceptively simple, but it masks a number of interesting questions. Hypothesis 7.10: Marriages, consistent with the Qur’an, are mostly of the same generation in the kinship network. Ambiguities of Individual and Marriage “Generation” In assessing generation, there is a problem. If two individuals are multiply related with different generational levels between them and not just one but several senior relatives, which of their relative generational level assessments do we use to test hypothesis 7.10? Individuals per se do not occupy an unambiguous generational position in a kinship network. Consider, for example, children of a couple who are not of the same generation, for example, where a man marries the daughter of his cousin. The man and wife are of different generations, but what of the children? With multiple marriages the problem may worsen. If a man’s first wife is his FBD, for example, his second wife his FBSD, and his third wife his FBSSD: To what generation do his children belong? They are all “next” relative to him, generation below, but the second batch is two below through the mother, and the third three below. Even if we assign the “deepest” generation, we still have siblings (the children) who are not of the same generation. To summarize, it is always possible to compute generations by the “deepest” generation rule, such that one is always of lower generation than either parent and, by this criterion, a husband and wife may belong to the same or different generations. None of this is a problem if birth dates are available, in which case we can evaluate Hypothesis 7.10 using the relative ages of husband and wife rather than generations. In the present case, as with many ethnographic studies, birth dates are not recorded. Johansen could only estimate approximate generational cohorts [a through i]. By taking these cohorts to be thirty years apart, there is a good separation between “generations” but sometimes parents and children will belong to the same generational cohort. Hence, there are two different principles at work, one of distinct cohorts and the other of distinct generations for parents and children. This is the ambiguity of assessing individual-level Marriage, Rank, and Migration 259 “generation” in the absence of data on birth dates. We may also think of marriages as having generations, and of a person’s marriages as being of possibly different generations. In the example above, this resolves the ambiguity: the first marriage is samegeneration, the second off by one, and the third off by two. Generations of marriages can be reliably and consistently assigned for kinship networks simply given the proviso that one’s parent’s marriage must be of higher generation than one’s own, inclusive of multiple marriages. This can be done consistently over an entire network, of any size, and lends a consistent interpretability to the concept of generation, especially because we can fit each marriage into the overall ordering of marital generations so as to be as close as possible to one’s parents. The algorithm for determining generational depth in a p-graph does exactly this. First, we find the longest ancestor-descendant chain in the whole network. That is the maximum number of generations that we need for all the descendants of that ancestor and ancestors of that descendant, each of whom can be placed uniquely in a generation. Any one of these descendants may have a lesser chain to another ancestor, and any one of these ancestors may have a lesser-length chain to another descendant. When these are added in, generations may be unambiguously assigned. When relinking appears, the marriage generations always fit within the constraints of previously assigned generations. There are no inconsistencies in assigning generations this way in that there is a minimum number of generations needed that depends on the structure of ascendant/descendant chains. With low structural endogamy or nearly exclusive same-generation marriages, the generations needed will correspond closely to parent/child birth intervals. But if there is considerable structural endogamy and many off-generation marriages, more marriage generations will be needed than in strict biological generations, in order to accommodate fractional differences, for example, between the average of men’s intergenerational birth times and those of women. In either case, while the assignment of generations is not strictly unique, it is always possible to do a consistent assignment of generations for any minimal number of marriage generations needed. For large populations, this assignment algorithm is sufficiently complex in many cases that it is not easily done by hand. That is the purpose of having computer programs in the Pajek and Pgraph packages that do work of assigning generations. The computer algorithm assignment, once done, can always be compared with actual birth and marriage dates, or with what is known about individual births and marriages in terms of general historical periods or cohorts. And, of course, it always best to do such a comparison, 260 Chapter 7 as a cross-check for errors in either recorded or remembered dates for errors in the assignment of parents. Individual and Marriage Generation Looked at as Mutually Informative Given the problems of ambiguity in assessing generations, one of the first things we did (Chapter 2) was to compute generations of marriages by computer. All but one of the marriages were positioned by computer in generations that were directly under a parent and over a child. We then compare these generations to those that Johansen had assigned to individuals. We looked at gross discrepancies to ascertain that we had not miscoded any of the data. All of that was completed before the analysis that we now present. The two systems of coding generation are mutually informative, which allows us to compensate for the lack of birth dates by mutual calibration of the time periods which they indicate. Generational depth as determined by the marriage algorithm correlate with the historical generations (Tables 6.1 and 7.3) assigned by Johansen to individuals, but with some differences in the phasing of the time periods that they indicate, relative to one another. As shown in Table 6.1(1-2), for the computergenerated generations 3-6 (for marriages), the intergenerational age difference in chronological time correspond to 30- or 40-years similar to the 30-year birth cohorts used by Johansen for individuals. After that, because of the accumulation of skewed-generation marriages with genealogies of increasing depth, the time gap between computer-generated generations shifts to roughly 15-year intervals, using Johansen’s historically dated cohorts as the clock. As we shall see below, these later periods also involve more irregularities in terms of different genealogical generations of husband and wife, which will usually shorten the chronological time lag between computed generations. Analysis 8: Same-Generation Marriage and the Qur’an Because network analysis of genealogies, even without precise birthdates, can be used to model the succession of generations and the relative standing in terms of generations, it can also evaluate whether marriages tend to be within generations or to involve “generational jumps” such as marriages between uncles and nieces, or older men with Marriage, Rank, and Migration 261 generationally younger women. Are Johansen’s ethnographic impressions that marriages are normally contracted among those of the same generation (Hypothesis 7.10) confirmed, consistent with the Qur’an? First, we examine the evidence from her coding of generations. Table 7.5 tallies the relative generations of husbands and wives for 410 of the 413 marriages in the genealogies. Each marriage classified has the husband’s generation listed at the left for each row, and the wife’s generation [a-h] listed along the top for the columns. The numbers of same-generation marriages (in bold) are 76% of the total. A nextgeneration wife accounts for 12% of the marriages, and a previousgeneration wife accounts for 8% of the marriages, leaving 4% with greater generational discrepancies. If we compare the rates of samegeneration marriages in the early generations [a-d] with those in the later generations [e-h], the rates of same-generation marriages are almost identical (74% versus 76%). Table 7.5: Relative Generations of Individual Marriages Using Johansen’s Classification of Husbands and Wives Generation of Generation of (N=410 married couples) the Husband the Wife Freq of Marriages Notes a b c d e f g h Total Same Generation a 3 1 0 0 0 0 0 0 4 3 75% a-b 86% b 0 9 1 0 0 0 0 0 10 9 90% Same: 76% c 0 0 13 1 1 0 0 0 15 13 87% -2 13 3% d 0 0 4 32 9 3 0 0 48 32 67% -1 49 12% e 0 0 1 9 79 21 9 0 119 79 66% +1 32 8% f 0 0 0 1 9 118 14 0 142 118 83% +2 5 1% g 0 0 0 0 2 9 55 2 68 55 81% Off: 24% h 0 0 0 0 0 1 1 2 4 2 50% Total 410 311 76% The Problem with Counting Generations for Individuals Another way to study consistency of generations in marriage is through marital relinking, and especially through consanguineal marriages. Table 7.6 shows the relative generations of husband and wife observed in 178 consanguineal marriages. Each type of blood marriage (with common ancestors up to six generations back) is classified by distance to the common ancestor for the wife [1-6] and for the husband [1-6]. Genera- 262 Chapter 7 tion is reckoned by distance from a common ancestor as computed by the Par-Calc Pgraph program. Here we see a much lower rate of samegeneration marriage, not 76% but 64%, which offers less support for the hypothesis. We asked ourselves: Which of these two results is a better test of the hypothesis, especially because Johansen’s classification of generations might suffer from the defect of tending to treat husband’s and wife’s generations the same. We regarded her classification only as a provisional one. But is the computer reckoning of generations any better, with its tendency to foreshorten the later generations in chronological time? In Table 7.6, we get a much better sense of how the generational rules might actually work: The closer the relation, the less discrepancy in generations. For marriage with first cousins compared to that with nieces and aunts, for example, there are no exceptions, perfectly consistent with the Qur’an. Exclusive of these cases (the smallest concentric box in the table), the exception rate for second cousins versus children of first cousins is 20%. When we go one more generation out, the exception rate is 27%; further, it is 42%, and then 48%; there are additional offgeneration errors in marriages to women of earlier generations (1% with aunt, a parent’s cousin, and so forth). Contrary to the ethnographers generalization in Hypothesis 7.10, this norm does not generalize beyond the avoidance of marriage with nieces and aunts: the errors increase to a majority of cases with more distant relationships. Overall, marriages with women in a lower generation account for 24% of the cases (116 marriage types), and the percentages of marriage types with a wife from the parents generation is 12% (N=57). Even in the marriages with distant relatives, however, there is no generational discrepancy of more than one generation, such as we had in Table 7.5. Examining these data more closely, it turns out that Hypothesis 7.10 holds only for intralineage marriages and, even then, not for lineage #2, the lineage of Koca Mustan and his brother, the master-relinkers. Of the 173 off-generation marriages, 95.3% are marriages outside the lineage. There are only eight intralineage off-generation marriages, and seven of these are within lineage #2. They all involve women who are grandchildren, great-grandchildren, or (mostly) great-great-grandchildren of Koca Mustan or his brother, and three of these involve exchange marriages or sororal marriages with brothers. More than likely, the greater generational depths here are ignored in favor of same-age considerations. This noted, then, for all but lineage #2, same-generation consanguineal marriages are avoided within lineages but not between lineages. Of the latter there are 108 different kinds of off-generation marriages, and only a Marriage, Rank, and Migration 263 dozen or so having more than three occurrences (FZSD having the highest frequency of seven). But we must remember that many of these 108 kinds of off-generation marriages involve the same couple with multiple kinds of collaterality. When we count the actual number of distinct couples involved, there are only thirty-nine extra-lineage off-generation marriages and eight intralineage. This, however, is still a significant number of exceptions (39/178=22%) to the hypothesis. Table 7.6: Generational Depth in Blood Marriages distance to common ancestor: (N=178 blood married-couples) on on the wife’s side hu’s 1 2 3 4 5 6 Side Freq Freq Freq Freq Freq Freq Total Notes 1 0 0 0 0 0 0 0 2 0 73 19 0 0 0 92 Same 302 64% 3 0 1* 78 45 0 0 124 - 1 116 4 0 0 24 85 39 0 148 +1 57 5 0 0 0 18 52 13 83 Off: 173 36% 6 0 0 0 0 14 14 28 Totals 0 74 121 148 105 27 475 * FFBD marriage (H1929-W341) Table 7.6 and the computer calculation of generations, thus supplement the impressions of the ethnographer, both in general pattern in detail. As distance to a common ancestor increases for marriages between blood kin, the off-generational discrepancy, or proportion of marriages where husband and wife are not of the same generation, increases as well. Further, off-generation marriage is consistently biased toward a lower number of generations for men than for women, which is consistent with men marrying women who are somewhat younger than themselves. The rules of the Qur’an are strictly adhered to only as stated in the Qur’an, for marriages between very close relatives, and not generalized further. Close examination of the data also showed that generalization of the same-generation rules held only within lineages, and even then, not for the lineage of Mustan where there is a great deal of internal marital relinking. Why is same-generation marriage important and why is the practice of same-generation marriage instantiated in this way, consistent with intralineage generational ranks for all but one lineage? This is a question we shall address later after further study of the relation between lineage and marital cohesion (Hypothesis 6.1) and a closer examination of prin- 264 Chapter 7 ciples of rank. Hypothesis 7.11: Given the possibilities of disruptive feuding within fraternal interest groups such as agnatic lineages with coresidential sublineage segments, and especially lineages with a right to marry lineage mates, same-generation marriage is an important functional element that reduces the competition between generations over wives. This is a provisional and incomplete hypothesis, as we expect it to be complemented by findings later on. Rank, Seniority, and Lineage In the opinion of the nomads, sexual relations between a man from the elder generation and a woman more than about twelve years younger than himself would spoil the order of seniority. Because sexuality was looked at as something for jokes and shameful doings of equals, a man lost his dignity when he took a wife from the younger generation, who had to respect him by strict obedience and hand-kissing and who should never joke with him. Because elder brothers had to be respected in a similar way, although not so strictly, levirate marriages were usually arranged between the widow and the younger brother of her deceased husband. The statement of the ethnographer that we have just read, however, requires a qualification that we have learned from the computer analysis of generations, namely, that this concern with generational levels is one that is behaviorally operative within the lineage, which would have been the context in which Johansen would have made these observations. She was affiliated with lineage #4, one of the ordinary lineages (all but that of Koca Mustan, lineage #2, for whom the exceptions are mostly 4-5 generations deep and more likely to accord with same age rather than same generation) that observed such rules. A different set of rules, perhaps not so easily observed by an ethnographer, appears to have operated in sexual and marriage behaviors between lineages. In Turkish culture generally (not just the nomads) there exist strict seniority rules. There are quite different words for elder and younger brothers and sisters. Especially between the generations there rules great respect, expressed by the gesture of devoted hand-kissing. Principles of respect are also set out in the Qur’an (sura 4,23-25): A man is forbidden to marry women from one generation older or younger than his own, Marriage, Rank, and Migration 265 specifically his aunts, stepmothers or stepdaughters, his mother-in-law, and the daughters of his brothers or sisters. In contrast, relatives of the same generation, cousins, especially the FBD, are preferential marriage partners among many Arabic groups. The Distance between Norms and Behavior The problems of evaluating the same-generation rule are an excellent example of the difficulties in establishing “norms” from ethnographic observations. These are difficulties posed in the task of the ethnographer to which network analysis brings something to bear. First, there is the problem of recording in a meaningful way what the ethnographer observes. Observing “generations” is problematic, and assigning or counting generations is difficult even with complete genealogical data and careful analysis. Second, there is the effect of variability. In many of the lineages in the clan, marriages are between relatives of the same generation but in Koca Mustan’s lineage (#2), off-generation marriages were frequent. These off-generation marriages were not salient, however, because they occurred only after the passage of 4-5 generations, and in such cases, relative age at marriage may outweigh considerations of generation. Third, as this example demonstrates, there is the influence of the context in which the ethnographer worked: in this case, residing with a particular lineage. Within that context, intra- and interlineage marriages were between relatives of the same generation, as was the case for most of the lineages. Had Johansen lived with members of lineage #2, however, in which off-generation marriages were tolerated, she might have reported a different pattern than the normative one. Network analysis is an aid in such cases such as this in accurately distinguishing norms and exceptions. Difference between Marriage Frequencies and Preferences The disparity between the frequency of observed behaviors, such as those pertaining to different types of marriage, and actual preferences— other things (the constraints of context) being equal—has been one of the great obstacles to the development of valid theories in the social sciences. Nowhere is that more obvious than in the intellectual division between the practitioners of alliance theory and its detractors. The practitioners are currently most numerous in France, where anthropologists 266 Chapter 7 had understood the strengths and weaknesses of Lévi-Strauss’s arguments, but more importantly, were convinced by the argument for the priority of norms over behaviors. Most of the social and cultural anthropologists in other countries, especially English-speaking ones, after an initial period of interest in structural models as representations of underlying forms of kinship (a critique of which was given in Chapter 4), did not remain convinced. Our data and methods allow us not only to test some of the relationships between frequencies of different marriage types and the contexts in which they occur, and to explore changes through time, but also to infer preferences and changes in preferences from the conjunction of these two types of variables: contexts and actual choices. We are able to control for the different rates at which cousins become available for marriage as a function of demographic change. Without such controls, temporal comparisons may be problematic. Whether there are changes in preferences or frequencies of FBD marriage over time is an issue of some importance where we can compare the ethnographer’s impressions with the computer analysis. Analysis 9: Fractality in a Marriage System Analysis 9 continues with fractality and issues of complexity in social organization that are evaluated by examining the entire distribution of different forms of marriage within a genealogical network. The reader may want to skip to the summary of this chapter. A summary in the complexity theory appears in White and Houseman (2002). Fractality is a pattern in which different parts or series, at different scales of magnification, are self-similar. In the pattern of competition between units in segmentary patrilineages there is a fractal process at work: competition occurs at all levels, and it occurs most frequently at the lower levels, scaling upward to larger units between which overt conflicts are less frequent but likely to be more severe. That small conflicts might spill over into fights between the lineages to clans, localities, tribes, and regions, for example, is partly due to the hierarchical ways these units are nested. If we compare the frequencies of two types of conflict that are constant multiples (e.g., double) in severity, the signature of a power-law pattern is that the two will also differ in that the frequency of the more severe type is always a constraint fraction of that of the less severe. If this ratio is “scale-free” it makes no difference whether we begin with the smaller conflicts, those of the middle range, or the Marriage, Rank, and Migration 267 larger. Power-law scale invariance is one of the characteristics of fractal growth and patterning produced by multiples: splitting of tendrils, growth of leaves, rich getting richer, segmentary conflicts in which the events of greater severity being the higher level conflicts that are more rare the larger their scale, and so forth.17 This is not, of course, a theory of all conflicts, but of those that occur within a segmentary structure. The processes that underlie such fractal invariance stem from the fact that a tiny event—a small conflict, a tiny seed germ of a tree—doesn’t know how big it will get, but it has a potential for escalation in a continuous network of accidents and interactive opportunities.18 You don’t get in one jump from A to B if they are not proximate, but if they are proximate, having gotten to A magnifies the probability of passing to B. In Chapter 4, we showed how a shift of focus from lineages to production units, taking us out of the confining assumptions of many kinship theories, let us see how—as well as when and where—FBD or “‘Arab” marriage operates in the context of the smaller patrilocal residential units that are also the production units in many of the societies influenced by Islamized Arabs in which FBD marriage is encountered. Competition and Fractality: Fractality Defined and Tested The diversity of marriages associated with FBD marriage, we have argued, works with and against the fractal gradient of segmentation (“I against my brother, my brother and I against my cousin, my cousin and I . . .”). On the competitive side, as we have seen in the previous chapter, there is selection at a distance against “those with whom one neglects to intermarry,” the reverse side of which is that those groups that do intermarry are potentially increasing the cooperativity and exchange relations on which they depend for survival. Several lineages in our genealogies (Table 7.1 :#1, #7, #8) fail to intermarry within the clan at a certain point, and tend outmigrate to villages. On the cooperative side, frequency of intermarriage scales with a topological distance in which cohesive clusters are continually expanding through the transitivity of intermediated relations (an ally of an ally becoming known, then familiar, and possibly then a new ally). Our argument about marriages associated with FBD preferences is that they constitute neither a single marriage rule nor a preference for a particular type of marriage, but a gradient of preferences and aversions across a great diversity of marriage types. That is, if we compute the frequencies of every type of consanguineal marriage, and plot these fre- 268 Chapter 7 quencies in a graph, ordering them from the types with high frequencies (such as FBD) to those to the lowest frequencies, we should see a gradient that has the characteristic fractal or scale-free distribution: neither flat nor linear, nor exponential decay, but a power-law distribution that is linear in a log-log graph of two variables, population frequency and the associated number of types at that frequency. This distribution, which we observe in Aydınlı nomad marriage data, expresses a scalefree organization of diversity of marriage types, consistent with questions in Chapter 4 about whether FBD marriage might be associated with strategies for diversifying types of marriage practices. Aydınlı nomads, then, practice a variety of types of marriages and the closer marriages, on whatever scale of distance or closeness one chooses, are the more frequent, following a constant gradient of dispersal. Hypothesis 7.12: The diversity of types of consanguineal marriage among Aydınlı nomads fits the small world model with navigability discussed in Chapter 5. Figure 7.11 shows the distribution of the percentages of marriages with cousins (Table 5.1: selective rate) at different genealogical distances. It shows a preference gradient for closer cousins, consistent with the navigability aspect of Hypotheses 5.1 and 7.6. Using a measure of kinship distances through a parent (P), child (C), or sibling (~), distances are three for first cousins (P~C), five for second cousins (PP~CC), seven for third cousins (PPP~CCC), and nine or more for fourth (PPPP~CCCC) and more distant cousins. The probability of marrying a cousin, as indexed by the percentages of available cousins married, declines inversely to the distance to the power 1.6, a power-law relationship very close to the optimum for navigability (Kleinberg 2000a) in a network lattice constructed on a constricted plane. Because we already know that the network is clustered and average distances are low, this supports hypothesis that the nomad clan network is a navigable small world, and suggests that its dimensionality is somewhere between 1 and 2, which is what we see in scaling images in Figures 7.3 through 7.6. Comparing the fit of r2=.975 for a power-law distribution, shown by the solid line in Figure 7.11, to that of r2=.912 for an exponential distribution (dotted line), the power law is the better fit. Figure 7.11: Cousin Marriage Probabilities Fit to a Small World Navigability Power Law of 1/distanceα with a Scaling Coefficient of α=1.6 Marriage, Rank, and Migration 0.12 269 percent married 0.1 0.08 Power (percent married) Expon. (percent married) 0.06 0.04 0.02 0 3 5 7 9 Figure 7.12a breaks out the preference gradient for all types of consanguineal marriages, regardless of distance. Here the x axis is a fixed number of spouses, from 1 to 32, who are married according to one of the 234 types of blood kinship within the range of fifth cousins (7 generations to a common ancestor). The y axis is the number of types of marriage with exactly that number of related spouses. Thus, we can read off this graph that there are 156 of the 234 types of marriage for which there is only one documented marriage. This number drops to 36 of the 234 types for which there are two marriages. If the graph were exponential it would keep dropping by a constant fraction, such as from 156 to 36 to 10 to ~2 to zero. Instead, the graph follows a power law, and drops from 156 to 36 to ~18 to ~10 to ~5 etcetera, if we follow the power law. Like all power laws, this distribution has a long tail of a few items with much higher frequencies than would occur if the type of marriage partner were randomly chosen. The extreme outlier in this breakdown of frequencies by type is FBD marriage (32 in raw frequency), nearly twice that of the next most frequent marriage type. The inverse power law of the distribution has an exponent of 2, an inverse square power law. If y is the number of types with frequency x between 1 and 32, then the power law equation is y=156/x2. The fit between this theoretical curve and the data is r2=.83 with an estimated slope of 1.97. Power laws of this sort, between marriage frequency of a marriage type and the number of types with this frequency, are strongly suggestive of kinship networks that operate as self-organizing systems. But in the present case, when we examine the types of marriage with the highest frequencies (FBD, MBD, FZD) they also seem to follow an order that reflects some kind of kinship distance reckoned from a perpective of patrilines. Chapter 7 270 180 160 140 120 100 80 60 40 Number of Types Figure 7.12a: Power Law Fractality of Marriage Frequencies Frequency 0 + 156/x^2 Frequency FFZSD FFBSD:10-11 FZD:14 MBD:18 FBD:32 Figure 20 7.12b repeats Figure 7.12a, this time taking the log of the x and y axes for the plot, and fitting a straight line to the distribution. 0 0 5 10 15 20 25 Figure 7.12b: Power Law Fractality of Marriage Frequencies - Loglog plot for Figure 7.1a with fitted line, slope ~ 2 Marriage, Rank, and Migration 271 Figure 7.13 shows the outcome of graphing the frequencies of all the 234 types of consanguineal marriages up to seventh cousins ordered by rank (as in a Zipfian distribution, hence the step-like appearance of the graph) with logged frequencies on the y axis and the logged number of observations for this type and frequency on the x axis. The distribution of raw frequencies is linear in the log-log graph and thus a power-law or fractal type of distribution, fitting our overall observation (and Hypothesis 7.9) about a fractal marriage pattern. Fractal marriage patterns function rather like Granovetter’s (1973) strong and weak ties, which have complementary strengths at complementary distances. The stronger and more frequent ties (of many fewer types) work at closer distances, in this case concentrically oriented toward close and patrilineal relatives, while the weaker ties of each type are individually less frequent but work as an ensemble in a distributed manner over longer distances. The fractal distribution of a strong/weak tie pattern of this sort, unlike the way that marriage preferences or rules are usually formulated, is continuously scaled rather than a simple dichotomy of types of marriage. 1000 Figure 7.13: The Fractality of Consanguineal Marriage Frequencies 1000 10 10 Percent Raw Freq. 100 100 Number of Possible Spouses Raw Freq % Raw / Possible Possible Spouses Power (Raw Freq) Expon. (Possible Spouses) Log. (% Raw / Possible) Log. (Possible Spouses) 1 1 11 Rank Order 1-234 by Frequency 10 100 1000 10 100 1000 Other types of distributions are also graphed in Figure 7.13. The upper curve for frequencies of types of possible spouses (all those available in a given category) shows an exponential decay or logarithmic distribution 272 Chapter 7 (here FBD is the most frequently available type of relative, MBD the next). The curve for percentage married of each type of those available is also a logarithmic distribution (again FBD is the highest percentage, MBD the next), unlike Figure 7.11, which was a direct test of Hypothesis 7.9 and the navigable small world Hypothesis 5.1. The logarithmic shape is due to the fact that there are many fewer types of consanguines at each kinship distance as we move closer to ego (four types of first cousins) but also, in a limited network, as we move to very distant relations that thin out if there are few apical ancestors, many of the vast number of combinatorial possibilities do not occur, and the closer relationships have already used up many of the relatives in the network. Only the raw frequencies fit the power-law distribution that is characteristic of fractality, and this is as it should be. In the following section, we show that power-law distributions for the Aydınlı do not apply to affinal relinking. The Fractality of Two-Family Relinking (nonpreferential) Figure 7.14 shows the analysis of all the various types of two-family marital relinkings in the nomad network. Marriage between two brothers and two sisters and sister exchange are the simplest examples of twofamily relinkings. There are over a thousand more elaborate varieties of two-family relinkings. Here the types they are rank ordered by their frequency of occurrence, as done for consanguineal marriages in Figure 7.13. Unlike the consanguineal marriages, however, that follow the power-law distribution characteristic of self-organizing systems, the twofamily relinkings follow an exponential distribution that is a characteristic signature of more-or-less random distributions. At the top of the upper left curve of this distribution are the more common types of twofamily relinkings, including the two simplest examples given about (e.g., sister exchange). They deviate somewhat from the exponential curve that fits the distribution generally, and they represent preferred types. There is no preference gradient over all types, however, as with the consanguineal marriages. Marriage, Rank, and Migration 273 Figure 7.14: The Fractality of Two-Family Relinking Frequencies 10000 turk relink2 redu Power (turk relink2 redu) Log. (turk relink2 redu) Frequency of Relinkings 1000 100 . 10 Rank order of Relinkings by frequency 1 1 10 100 1000 10000 Courtesy of Michael Houseman, using software of Laurent Barry Power-Law Theory for Preferential Attachment to Degree and Preferential Attachment to Relinking (Ring Cohesion) Preferential Attachment to Degree In general terms, the theory of preferential attachment by degree in social networks entails a theory of competition between individuals or nodes over popularity, network resources, and a dynamic of “rich get richer.” This is made explicit in writers like Barabási (2002). The network theory given in the probability models of micro-macro linkages discussed in Chapter 1 provides a general explantion of the selforganizing feedback loops implicit in such attachments. The discussion of self-organizing feedback loops in social networks, however, focused on competition over how nodes in a network are differentially attached, and how micro behaviors that lead to differential attachment lead to degree distributions (frequencies of nodes with different numbers of links) in which a topology of centralizing hubs emerges in the network. How 274 Chapter 7 do the findings on the Aydınlı accord with the theory of network typology and dynamics outlined therein? Lack of Preferential Attachment to Degree in the Kinship Domain Kinship networks cannot by definition have network growth by preferential attachment to degree since children cannot choose to attach themselves preferentially to parents. And in spite of the parental preference for more children and the greater tendency of large sibling sets to remain nomadic, the extended tail of frequency of parents ranked by their numbers of children fits an exponential distribution (r2=.905) rather than a power law (r2=.659). Similarly, the number of descendants of ancestors is also exponential rather than power law. Perhaps the only kinship networks in human history that had a scale-free extended tail of frequency of parents ranked by their numbers of children have been in societies where male leaders were able to amass virtually unlimited numbers of wives, concubines, and children. If preferential attachment by degree does not apply to Aydınlı kinship, what can we learn from our results on the preferential attachment of marriages to a specific gradient of what we might call closer kin? Do Hypotheses 1.1 to 1.4 of Chapter 1, which are stated in terms of scalefree networks, apply to preferential attachment of marriages to closer kin? These preferential attachments are not to specific nodes, but specific types of nodes, that is, types of potential marriage partners. Preferential Attachment to Relinking Different types of marriage between those who are already kinshiprelated reinforce those existing links to create greater social cohesion. The possibilities for different kinds of relinking in kinship networks thus entails a different logic, which we might call preferential attachment by cohesion. Attachments of this sort have been documented by one of the present authors in studies of other types of networks (Powell et al. 2004, Brudner and White 1997). Relinking attachments that form power law distributions index preference gradients for different forms of cohesion. These might include repeating an existing type of tie between two lineages, reinforcing a pattern of visiting, creating an exchange relationship of greater trust between extended families, and others kinds of cohesive linkage. These forms are multivocal, and no single one can be taken as a preference “rule”; rather, there are differential gradients in different situations. The aggregate behaviors of individuals over these options form Marriage, Rank, and Migration 275 an aggregate preference gradient, aggregating individual and family preferences. Logics of cohesion through cooperation are too often ignored in network studies, partly because good network indices of cohesion have been lacking. The logic of relinking attachments extends to the negative case where relinkings do not form power laws and do not index preference gradients. Here we would expect to find that rankings of the frequencies of different choices will form distributions in which declining frequencies along the ranking will show the rapid exponential decay expected of a more random assortment of choices, rather than the slow power-law decay of a preference gradient. For Aydınlı this negative case applies to affinal relinkings. The logic of preferential attachment of marriages to closer kin, or more generally, preference gradients for different types of kin, is a result of how different forms of cohesion are constructed through cooperative and competitive processes. Constructing forms of cohesion through cooperation necessarily entails a social competition with distant others who are excluded from closer cohesion. Social competition of this sort is very different from the individual or node-based competition over preferential attachment by degree. In some societies, as for example, those with competition for class rank and heightened cooperation within social classes, such constructions often occur for affinal relinkings. In other societies like the Aydınlı they occur for marriages that relink consanguines; and in still others both, neither, or other preference gradients appear. We can be reasonably sure that in the case of the Aydınlı, given how our findings fit amongst these contrasting types of societies, we are dealing with a preference gradient for consanguines but not for affinal relinking, which have a more random exponential distribution. What we have learned in Chapter 1 about networks, especially from explanatory micro-macro principles, is how the feedback between local behavior and global properties of networks alters the context of local behavior to reshape social organization and structure. In the case of preferential attachments to relinking, the local behavior is that of who marries who and the global behavior is the resulting shape of the marriage frequency distribution, which, in turn, reshapes the context and frequencies of local behavior, so the feedback loop is complete. Our next question is how to model this loop. Ring Cohesion Theory The theory of network topology and dynamics discussed in Chapter 1 276 Chapter 7 bears some close analogies to our preferential attachment to relinking. To formulate a general theory of preferential attachments to relinking that is analogous to those for preferential attachments to degree, we need to identify what it is that relinking contributes to a network, and how feedback operates in shaping the power-law coefficients associated with attachments to prreferential relinking. All power-law phenomena tend to scale between 1 and 3, and for similar reasons. We are not looking, as physicists do, however, for a single universality class defined by a power coefficient that might be used to “explain” kinship organization in a scale-free sense over an extended range of magnitudes. We are looking for preference gradients indexed by the extended tails of power-law distributions, in this case for different types of relinking marriages. In our p-graph notation, relinking marriages are identified with closed cycles. In the Aydınlı case the preferential attachments are those brought about by consanguineal marriages. In the general case we call these cycles “rings” in order to give a name to a theory that concerns the analysis and significance of the frequency distributions of different kinds of cycles. Ring cohesion theory begins with the recognition that each independent ring or cycle creates a new link that adds cohesion to a social network. Ring cohesion theory also provides a method for dealing with what constitutes an independent cycle and why and how it is that only independent cycles add to a measure of cohesion. The graph theoretic definition of a set of cycles that are considered independent is that each cycle contains at least one edge that is not contained in any of the others. Each independent cycle then contributes one unit of additional ring cohesion. The theorem for the number of such cycles is that if n is the number of nodes and k the numbers of edges, the independent cycles number n – k + 1 in a connected graph and n – k + d if a graph contains d disconnected components. This theorem forms one leg for a theory of ring cohesion. A second leg for ring cohesion theory is the premise that shorter cycles contribute more cohesion than longer cycles (White and Harary 2001). The third combines these two findings. Because we can compute the number r = n – k + d of independent cycles in a p-graph, then finding the first set of smallest cycles of length h so that h > r gives us a means of finding a set of cycles that is sufficient to identify a superset that includes a set of independent cycles. The fourth leg of the theory is that all the cycles of a graph can be generated as products of intersections of independent cycles or their recursive products. (The intersection operator adds two subgraphs but removes any edges that they had in common). By this means we generate all of the cohesive cycles in the Marriage, Rank, and Migration 277 graph. In the general case, it is only a set of cycles that are independent that should be submitted for distributional analysis, so that they constitute a statistically independent set.19 The smallest such independent sets can also be tested for consanguineal and affinal relinkings separately, given the importance of this distinction. If only one of these sets has preference gradients indicated by a power-law distribution of marriage types, those types will be regarded as independent, and nonindependent types will be identified from the other set. The formula for the maximum number of independent rings or cycles allows us to calculate how much of the variation in marriage behavior is explained by the preferential set with the power-law gradient. A further step in the analysis would attempt to identify what kinds of principles would correctly predict the ranking ordering of the cycles in frequency, controlling for the marginal constraints on the types of edges present in the network. This would entail the type of Feynman situation studies that we discuss in Chapter 8. This more general approach of ring cohesion methodology was an outgrowth of our analyses of the Aydınlı data. The advantage of ring cohesion theory and its methodology is that it allows statistically appropriate samples to be tested for sets of smallest independent cycles. Ring cohesion theory currently occupies the status of a framework for investigation, and probabilistic models of network processes that involve how marriage or other types of cycles generate different types of cohesion in different types of networks requires the level of investigation that we reviewed in Chapter 1 for preferential attachments by degree. Here, we can only sketch in this particular case how such processes operate. What ring cohesion does as a substantive process of cycle formation within a network is to reinforce some ties in the graph more than others in terms of what they contribute to cohesion. In the kinship case, for example, a FBD marriage reinforces close ties along the male links in a shallow patriextended family. If all marriages were endogamous to a patrilineage or segment, the reinforcement would augment cohesion within these units, which exist mutually exclusive of one another. Marriages that involve female links would augment lines of cohesion running between lineages. These reinforcement patterns would affect what types of cohesive clusters form within the clan, and the lines of potential segmentation, the networks of trust and strong ties, and other topological features of the network. These macrostructures would then alter the context of behavior differentially for different members and groups within the clan, and the kinds of feedbacks we have described for network processes in general would need to be studied over time. An extension of the 278 Chapter 7 present study would be needed to study these feedbacks dynamically and structurally in relation to historical processes through time. Simulation studies would help in understanding these processes. Ring cohesion theory, then, opens a frontier of new research that is largely untapped, but that emphasizes social cooperation and social competition rather than the pure competition among individuals that is emphasized in the study of preferential attachment by degree (Barabási 2002). It is not just individuals but cohesive groups that complete as well as collaborate. We have shown that consanguineal marriage frequencies for the Aydınlı have power-law distributions that indicate a preference gradient. These marriages qualify as the independent cycles that generate cohesion in the Aydınlı genealogical network. The frequencies of different types of affinal relinking do not have power-law distributions that indicate a preference gradient. If we take those marriages shown by this means to be statistically preferential as independent cycles that generate cohesion in the Aydınlı genealogical network, we must also conclude from the formula for the maximum number of independent cycles that many of the affinal relinking cycles are the byproducts of consanguineal marriages. These marriages are therefore disqualified as cycles that contribute independently to constructing genealogical cohesion. When two brothers arrange FBD marriages their sons by an exchange of daughters, for example, the two independent FBD marriages generate a nonindependent affinal relinking of sister exchange. Nonindependent affinal relinking ramifies through the genealogy as a result of many different pairs of consanguineal marriages that have some of the same people in common. Fractally Segmented Lineages as Self-Organizing Systems: Navigable Small Worlds in the Kinship Domain The power-law distributions of consanguineal marriage frequencies for the Aydınlı exhibit the correct fit to the parameters of a navigable small world. The fit is between a power coefficient of 1.8 and the navigability model. This value corresponds roughly to a squeezed two-dimensional geographical configuration of nomadism that is reflected in how lineages organize their marriages. A small world that is navigable is one in which a specific ego can locate a specific alter with certain characteristics in relatively few steps, but it is also one that can mobilize its zones or segments at various distances from ego, and mobilize segments at levels up to and including Marriage, Rank, and Migration 279 nearly the entire network. That is precisely the property of segmented lineage systems. Figure 7.15 illustrates some of the properties of a segmented lineage in which each father produces three sons, for eight generations. The number of sons might vary but the self-similarity of the branches constitutes a fractal structure. The increase in the cumulative number of men in the lineage follows an exponential curve, shown in the graph. So do the cumulative numbers of potential sublineages branching from any of the men in this lineage, over all generations. The distribution does not fit a power law, as shown by the deviation of a best-fit power curve in Figure 7.15 to the exact curve that is exponential. These distributions remain exponential so long as the number of sons is a poison distribution around an average number of sons greater than one. Figure 7.15: Fractal Expansion for a Patrilineage with a Branching Number 3 – in each of k generations, starting with a single ancestor at level 0, each new descendant has three sons; the resulting cumulative distribution is exponential k 1 2 3 4 5 6 7 8 Fk=3k 3 9 27 81 243 729 2187 6561 Cumulative 3 7000 12 6000 39 5000 120 4000 3000 363 2000 1092 1000 3279 0 0 9840 Weak tie Exponential Power 1 2 3 4 5 6 7 8 9 Like Figure 7.14, the distributions of numbers in fractal lineages represent the negative case of absence of a power law. The power laws discovered for the Aydınlı, however, in Figures 7.11 through 7.13 relate not to the size of lineage segments but how they are tied together by marital relinkings that make the network of alliances between fractal sublineages cohesive. What makes power laws significant? The power-law shape of distributions is a characteristic signature of self-organizing systems (Barabási 2002:74-76) and suggests that the process generating the extreme tails of large or frequent events in the distribution is the same as that for the small ones.20 The possibility of the same process underlying the power laws of different networks is one of the themes of Barabási (2002) but it is by no means established for social networks. We have also seen from Analysis 7 that strong ties between subline- 280 Chapter 7 ages cluster to define for each sublineage its degree of local curvature of reciprocal endogamy within its immediate zone of strong marriage allies. The local clusters and continuities in the curvatures of proximal clusters of nodes define an important aspect of the topology of the kinship network. Fractally Segmented Lineages of Self-Organizing Systems In closing this chapter, we recall Hypotheses 5.1 to 5.3: Hypothesis 5.1: The Aydınlı genealogical network, which has both living persons and ancestors who are important as links between them, has the structural properties of a navigable CSW (complex small world). Hypothesis 5.2: The specific characteristics that constitute a kin-based CSW for the Aydınlı are: (1) the basis of trust is relatively near genealogical relationship and the rules of behavior between near relatives, (2) some close relatives will be found at larger spatial distances, (3) marriage frequency correlates with genealogical distance, (4) lineages provide one of the larger identities needed for navigability, (5) reciprocal marriages between lineages that extend the bounds of trust through female or affinal linkages (alliance groups occurring in specific historical periods) provide a second cross-cutting larger identity which is a minimum needed for SW navigability. Hypothesis 5.3: The five characteristics of the kin-based CSW network of the Aydınlı are not unique but are common to Middle Eastern societies influenced by that form of Arabic social organization that has segmented patrilineages and a right to marry FBD. Drawing on these while reflecting on the importance of same-generation marriages, consistent with the Qur’an, but mostly instantiated for within rather than between lineage marriages, we can now reformulate the more functionalist Hypothesis 7.11: Hypothesis 7.13: Given the possibilities of disruptive feuding within agnatic lineages with a right to marry a lineage mate, and with co-residential sublineage segments, in which marriage functions to create cohesion within as well as between lineages, samegeneration marriage is an important functional element that reduces the competition between generations over wives. Here the prediction is that disruptive feuding within agnatic lineages is Marriage, Rank, and Migration 281 minimized or absent, partly because such feuding would be of the most disruptive sort within a segmentary lineage society. In addition, the same-generation marriage pattern helps to minimize friction by eliminating competition for wives between men of different generations. The hypothesis, while inferential, is supported, or at least not contradicted, by the fact that Johansen observed no disruptive feuding within agnatic lineages. This reformulation, linking endogamous marriage to issues of social solidarity, might also help to understand—and again, this is highly inferential—why the lineage of Mustan is the major exception among lineages to the same-generation marriage rule. His Ecevitli A-B lineage, as we saw in Chapter 6, was one of the major sources of the marital relinking that established the clan as a cohesive unit. It might appear that the successful strategy of this lineage in establishing social cohesion might be accomplished at the cost of overriding the rules of samegeneration marriage as established by the Qur’an. At a more general level, however, going back to Hypothesis 5.2, we can also see why a same-generation marriage rule would be highly functional in the segmented lineage systems of the post-Qur’an Arabized world. Summary The term segmentary lineage derives from Evans-Pritchard’s (1940, 1953:26) study of the cattle-keeping Nuer, who lacked a centralized government, had exogamous lineage segments that corresponded to spatial divisions, and engaged in predatory expansion against other tribes, a feature elaborated by Sahlins (1961). Many of the features that these and other authors specify for this concept are not widely found in the Middle East, and they make this concept unsuitable for broader comparative analysis.21 As noted by Hart (2001: abstract; see also Munson Jr. 1989, Holy 1979): The concept of tribe has long been contentious in North African history and anthropology and the idealized concept of segmentary lineage theory is inadequate. Segmentation, however, is a useful concept as it corresponds to tribal self image and cohesion although it undermines chronology. We employ the more neutral term, embedded or segmented lineage, for “a descent group in which minimal lineages are encompassed as segments of minor lineages, minor lineages as segments of major lineages, and so on” (Anthropology Explorer glossary). Embedded lineages are characteristic of many Bedouin (e.g., Peters 1991), Arab (e.g., Cunnison 282 Chapter 7 1966), and Arabized societies of the Middle East that are often dependent on various types of husbandry and on long-distance trade by camel caravans. Few studies of segmented lineage structures have looked at how the cohesion between sublineages is generated by marital alliances; much less how such a structure forms a navigable small world model of reciprocal alliances. The comparative topology of segmented lineage social organization is an area open for study, and findings are likely to be linked to the properties of self-organizing systems. For the Aydınlı nomads, a structure of alignments among lineage segments emerges out of the interplay between nomadism and sedentism, egalitarianism and rank, competition and alliance, and it builds up from a local topology of transitivity of alliances. To see that topology clearly we have to think carefully about how to measure network dynamics at the local level. Analysis 7 uses a technique developed by Eckmann and Moses (2002). The method “fractalizes” the cascade of all the possible sublineages as they segment in time by considering all sublineages of depth five, and the network of relations between them. Because we want to focus on marriage alliances that are intentionally constructed, we maximize that possibility by considering only those marriage exchanges between five-generation sublineages that are reciprocal and in which a woman from A marries into B and a woman from B into A. When applied to the Aydınlı, because marriages are almost always in the same generation, these will almost always be true exchange marriages. The ethnographic context of exchange marriages is that the bride payments are cancelled, implying a high degree of coordination, trust, and informality between the families. It is not that these marriages constitute the majority of all marriages, far from it: the majority of all marriages involve bride payments. But as shown in Figure 7.9, the pairs of the twenty-one sublineages that are linked by at least one reciprocal exchange are more than twice as common as the pairs that have only directed exchanges of brides (e.g., one or more women from A marry into B, but not the reverse). With the much greater number of reciprocal over directed links, the average distance between two sublineages in the network of reciprocal exchanges is also much shorter than for the directed links, peaking at distance one. Marriage alliances show a high tendency toward transitivity of reciprocal but not of directed exchanges involving bride payment, and a vastly greater ratio of triples that are not connected at all than expected by chance, consistent with the idea of topological clustering and separation (Table 7.4 triad census).22 The topology is not strictly clustered into cliques but has more chains of reciprocal links than Marriage, Rank, and Migration 283 expected by chance. When we look at the global topology, it is indeed that of a long sheet of closely entwined paths (Figure 7.4, as in Figure 7.5), with a great deal of local transitivity among close segments. What does this topology represent socially? It conforms to our hypothesis that the sublineages that are indeed intimately related with one another engage in reciprocal exchange, and the informal visiting that occurs between these pairs creates a chain of three sublineages, say A, B, C, in which if A-B are close and B-C are close, then there is a strong likelihood that A and C will hear intimate information about and encounter one another’s members via the intermediation of B. This is the model of local clusterability (transitivity) of Eckmann and Moses, and it produces surprisingly good results: The global topology of the alliance structure is a long series of multiply connecting paths of reciprocal intimacies. One pole of the series begins with the sublineages most involved in marital alliances with villages. The other pole of the series ends with the more intensely nomadic sublineages, such as those engaged in sheepherding in the higher mountain ranges, who are more involved with each other and with other nomadic tribes. Figure 7.5 showed how this topological continuum can be visualized on a map of the region in which the links to villages are superimposed. The internal topology of marital alliances within the clan matches the external topology of outside alliances. This topology, then, is one that describes the opportunity space for marriages, including arranged marriages as well as meetings between the potential bride and groom, in a structured sociogeographic space. Orthogonal to the geographic social space is the temporal genealogical space in which the preference gradients are for same-generation marriages. Close visiting relations are between members of proximal sublineages who are also contemporaries in time; they give a dynamic to the topology because the range of visiting patterns can be extended through intermediation and transitivity. This topology, then, creates gradients of possibility and constraint on visiting and getting to know potential spouses, both for oneself and one’s children. The “fractalizing” method of analysis, however, which works well to discover the social topology from the analysis of network patterns, also bore another dividend: we could interpret change fractally in Figure 7.6, by the bubbling up of certain sublineages in the scaling pattern (which has the characteristic tubular shape at the floor of the diagram) into a dimension (on the vertical axis of the diagram) in which certain cascades of sublineages are shifting their position relative to that of their maximal lineage as a whole. Interestingly, variance among the norms and behav- 284 Chapter 7 iors is also probably fractalized within such a space. For most lineages, for example, same-generational marriage is an absolute rule, but not for any of the sublineages descended from Mustan or his brother. These master relinkers, as it were, seem to virtually ignore the norm of samegeneration marriages within their lineage once a depth of 4-5 generations is reached, so long as the ages of the couple are relatively close. The findings of this chapter connect with the model of fractal and complex small world networks constructed in Chapter 5. In that chapter, Hypotheses 5.1.1-5.1.3 and explicit models for small worlds were developed for an “‘Arab”-type fractal marriage network that integrated the levels of segmented patrilineages. That analysis showed how if such kinship networks, like those of the Aydınlı nomads, are to provide a basis for exchange, they must also have a topology that is navigable. The local curvature of clusters of reciprocal and transitive marriage alliances between lineage segments that we have discovered in this chapter provide a topology of localized and extensible relations of trust that are the vehicles for navigability. They give to the larger networks of villagers, nomadic tribes, and Aydınlı clans—within which our specific clan is embedded—a specifically fractal quality of a complex small world, one in which paths of mediated trustworthy connections are, on average, short, and denser clusters exist which help not only to construct cohesive and solidary groups that can compete for survival with others, but to make navigability for suitable exchange partners possible within the larger small world. The navigability characteristic needed for network navigability was confirmed in Analysis 9, Figure 7.11. Figure 7.12 confirmed a powerlaw distribution for consanguineal marriage frequencies, consistent with a kinship network with distributed agent-based self-organization. Figure 7.13 confirmed that the power law operated through raw frequencies, which constitute the demographic context of marriage behavior, and not through percentages distributions (Table 5.1). Affinal relinkings, however, were shown in Figure 7.14 to have an exponential rather than powerlaw distribution. The inference from these results is that consanguineal marriages represent a preference gradient scaled roughly with kinship distance which is potentially self-organizing with respect to clustering, formation of concentric cohesion, and navigability. Affinal relinking, in contrast, exhibits no such features, and patterns more randomly as indicated by the exponential distribution. Figure 7.15 reinforces the idea that lineage segments are basically fractal, as argued in Chapter 5, although a power-law distribution in sizes of sublineages could conceivably result from a preference gradient in which larger units have a greater success Marriage, Rank, and Migration 285 and retention rate for nomadic life, with smaller units having a higher probability of failure or emigration to villages. That hypothesis will be explored in the next chapter. Further Reading Granovetter (1973) and Eckmann and Moses (2002) provide some of the methodology that is central to the analyses of which chapter, with Barabási (2002) providing an approach to network topology that contrasts markedly with the models used here. Barth (1953) provides a study of sheep and goat pastoralism for the Basseri in Iran, a case somewhat comparable to that of the Aydınlı nomads. Bates (1973) provides a closely comparable case for a nomad group neighboring the Aydınlı and of the same ethnic, migratory, and tribal origin. There is a wealth of ethnographic detail in Bates’s monograph that can be used to compare with and flesh out some of the probable workings on Aydınlı society. Notes 1. An example of rivalries is evident from the 1957 funeral service of patriarch Fındıklı Abbas (840, lineage #2) shown in the Frontispiece, for example, where some of those in competition with the brother of the deceased who became tanıdık kişi that same year, including some of the Kırbaşı lineage (#4), did not attend. 2. On dowry for the neighboring Kayseri Province Yörük see Bates (1973: 77 f.n. 7). 3. For bride payments of neighboring Kayseri Province Yörük see Bates (1973: 76). The monetary equivalents of the payments for the two groups are roughly comparable and ordinarily very high for arranged marriages but for elopements (“kidnapped” brides) are adjusted to roughly one-quarter of the amounts for arranged marriages. 4. Bates (1973: 65) states for the neighboring Yörük nomad group that bride payments are very high for virgins taken as second wives, and while such marriages are almost nonoccurrent among the Aydınlı nomads we take the discouragement of such marriages to also entail high payment demands by reluctant fathers. He also reports “the brideprice will be high for kidnap marriages but not for divorcées or widows.” Squaring that with his statement in the previous footnote, that means that bride payments for arranged marriages are very high, and accords with our ordering of bride payments. 5. For example, “the likelihood that FZD’s . . . and MZD’s will be close agnatic kinsmen is high, the exact rate being determined by the incidence of FBD mar- 286 Chapter 7 riages in previous generations.” (Bates 1973:61) 6. A graph is transitive when for any subset of three nodes {A, B, C}, a pair of (directed) lines from A to B and B to C entails one from A to C. An example of transitivity is the idea that a close friend of a close friend is likely to become a friend. 7. “Restricted exchanges” is the term used by Lévi-Strauss for reciprocal ties between lineages. 8. Thus where strong ties link alter A to ego and ego to B, A and B are likely to be linked by a strong tie whereas if the ties from A to ego to B are weak, transitivity is less likely. 9. This contrast between the structure of strong and weak ties in marriage networks is analogous to Granovetter’s distinction in European societies, although the content of the two types of ties is very different. 10. To scale these differences quantitatively, the network is drawn with using Pajek’s automatic drawing option set to Values of lines/Similarities. The similarity value of a line between two nodes is measured by the chi-squared formula for the excess of actual marriage frequencies over expected values. 11. There were eighteen other village or town names given that White could not locate: six from Satır Köy, four from Otlu Köy, two from Mağara [“cave,” also an older name for Tufanbeyli], and one each from Arabica, Kara Köy (which might be Karabuçak, near Konya), Tarpan, Susamlı, Ilmasit [an older name for Antalya], and Buzaklı. 12. Lineages 1 and 2 (#2 and #3 respectively) at one end of the spatial continuum are thus brought together with 6-7-8-9 (#1, #6, #4, #5) at the other end. 13. Eckmann and Moses (2002) emphasize that mutual recognition, as indicated by reciprocated ties in social networks, is crucial to understanding how network neighborhoods are structured. Their approach is particularly useful in that they do not take networks in raw form for the analysis of reciprocity, but they take into account hierarchical links, like those within a person’s web pages, or those between members of the same minimal patrilineage, to find the reciprocal links between units that reside, so to speak, at a different address. Their approach has a useful analog for kinship networks in considering reciprocal marriage links between different parts of hierarchical structures, such as sublineages that have a meaningful social identity but are nested within larger lineages segments and within maximal linkages having more distant connections through remote ancestors. Their hypothesis is that cohesive content and thematics reside in the combination of reciprocal links between distinct entities capable of mutual recognition with locally dense network neighborhoods constructed out of these meaningful reciprocal links. They found that clustering as measured by the local density of neighborhoods surrounding single nodes, that is, the density of ties among those connected to a given node or ego, to be highly correlated with ties to ego that are reciprocal, and with the degree of overlap of ego neighborhoods to form larger clusters of higher densities. They found these patterns to be a common property of “social” networks that replicated for the World Wide Web, Marriage, Rank, and Migration 287 neuronal networks in the brain, gene regulation networks, and protein interactions. They call this pattern one of curvature, a term borrowed from topology. We find this pattern of curvature to replicate in our network of marriage among lineages. We found a well-defined structure of curvatures in overlapping and relatively dense network neighborhoods, as diagrammed in the bowl-like scaling of Figure 5.3, where reciprocal links tend to be along the circumference of the bowl and denser network neighborhoods tend to form around them. Our Figure 5.4 is an attempt to test whether the network neighborhoods themselves have some meaningful content to them, as indeed they do. 14. It may be easily seen how to extend this approach to bilateral kindreds that are also overlapping. 15. Further testing for a small world model could be done using statistical modeling of social networks, as, for example, the exponential random graph models discussed by Skvoretz (2002). The results from Table 5.2 are so clear-cut and consistent with other results, however, as not to require further testing. 16. Our finding that weak ties do not do the spanning in this network––or rather, they do, but they are redundant and, being more random that strong ties, also lack small world searchability—is consistent with findings from Dodds, Muhamad and Watts’s (2003) S-W experiment on 67,000 E-mail users and eighteen targets in thirteen countries, who found that successful chains depended on finding intermediate links through identities shared with the target, and did not depend on weaker ties. 17. A good example of a power law in the natural sciences is earthquakes with half the energy being four times more frequent over an extremely large range of magnitudes. 18. Similarly for slippage in the earth’s crust, which might start out as a tiny event below the surface and escalate through proximity along an existing series of faults and cracks to a massive scale. 19 .Counting cycles also entails minate isomorphisms among cycles so that they are counted only once. 20. The scaling of frequency of earthquakes in relation to their energetic intensity, for example, is a power-law distribution. For a physicist, this implies that there is no typical scale for earthquakes and suggests that the physical mechanism for large earthquakes is the same as that for small ones. 21. As we have seen, Korotayev (2000), for example, finds FBD and segmented lineages associated with Arabization under the conquests of the early Caliphates in the seventh and eighth centuries. 22. There is no tendency for ranked clusters, a finding that is consistent with egalitarianism. Chapter 8 Demographic Choice and Constraint: Historical Structure and Change Demographic Opportunity and Constraint as Influences on Social Cohesion, Resilience, and Change Lévi-Strauss (1969) shows how patterns of generalized (multiway) and restricted (pairwise) exchange, in terms of the circulation of goods, information, and people, entail different modes of stratification and patterns of global versus local cohesion. How women circulate, as wives and mothers, job holders or migrants, and in patterns of sociability, helps to define some of the key structural parameters in a society. Among the Aydınlı nomads, for example, bride payments circulate in the orbits of exchange while wives move in the opposite direction. Understanding the relationships activated when a bride takes a man to wed is a crucial part of understanding the system of social exchange by which the society is constituted. It is within this framework that anthropological theory has attached such importance to marriages between kin, the forms these marriages take, such as the different types of cousin marriages, and the implications of these forms for social exchange and cohesion. FBD marriage, as foreshadowed in Chapter 4, is problematic for anthropological theories of social exchange. Most theories of exchange formulated to aid in understanding systematic differences among kinship systems deal with the cross-cousin marriages, MBD and FZD. For societies that have a rule of unilineal descent or a gender-specific rule of residence (with husband’s family, wife’s family, etc.), these are marriages between different kinship groups. FBD marriages seem to contradict the very premises of anthropological exchange theories (Bourdieu 1972) because they are almost invariably associated with patrilineages and often with patrilocality, and are thus within kinship groups. This has posed one of the central theoretical questions that have framed our analysis of Aydınlı nomad kinship networks and of FBD marriage in general. 290 Chapter 8 In Chapters 2, 4, and 7, we argued that the making of inferences about marriage rules and the study of marriage choice requires a deeper understanding, as elsewhere in social life, of the constraints of demographic factors on the opportunities for making choices. Here we analyze three topics dealing with demographic opportunities and constraints: 1. Co-Selection Bias among Siblings 2. Parallel and Cross-Cousin Marriage 3. Parallel Cousin Demography Analysis 10: Co-Selection Bias among Siblings Demographic fluctuations, such as family size, provide opportunities for making selective choices, such as whether to emigrate, emigrate with one’s siblings, or to stay at home because of the additional support that a large set of siblings provides. Kinship links can be a source of social cohesion, as with residential clusters of kin, or a potential for dispersal, as with siblings who emigrate either jointly or independently. For Aydınlı nomads, for whom male lineages are crucially important, we decided on a demographic analysis of whether the size of the sibling set influenced people’s decisions to leave the nomad clan or to stay with a nomad residential group. For males, that residential group is typically the patrilocal extended family. Hypothesis 8.1 (Co-Selection Bias among Brothers): Because cooperation among co-resident siblings and lineage mates is crucial to the local production units of a pastoral economy, we expect that males with more brothers would be more likely to continue to reside (patrilocally) within the clan, while those with fewer brothers would be more likely to emigrate. The analysis of this and related hypotheses was not conducted by using any of our usual network software (Pajek or Pgraph), but required a more specific spreadsheet analysis of the size of sibling groups and an analysis of how size varies with other variables. The auxiliary analytic routines (in the Fortran programming language) that were used may become available with the development of new software packages. Robust health and the value of children to the extended family point to high fertility of women and population growth in the nomad clan, yet limitations on resources force outmigration to villages. How is the potential for cohesive integration in the clan, the lineage segments, and the ex- Demographic Choice and Constraint 291 tended families affected by the demography of sibling groups? In Table 6.4 we examined the question: Who stays and who leaves? We now examine more closely: Who stays together or emigrates together? Figures 8.1 and 8.2 are the results of testing hypothesis 8.1. It turns out that the more brothers in a sibling group the more likely they are to remain nomadic and not to migrate to villages. This is a crucial factor in nomad demography, supporting the lineage system itself. The relationship between number of brothers and migration, however, is curvilinear, as shown in Figure 8.1, for brothers of the same father. Here the frequencies on the y axis are number of sets of brothers, not number of brothers. The ratio of stayers to leavers is simply computed as numbers of stayers divided by leavers but then multiplied by ten to conform to the scale for numbers of stayers and leavers. The ratio of stayers to leavers rises as the number of brothers rises from one to four, then dips between five and seven, and rises sharply for eight brothers, as are sometimes produced by two or more wives. The biggest exception to the generalization that supports Hypothesis 8.1, causing much of the dip in the curve of ratios, are six brothers, five of whom emigrated to the village of their mother in one of the recent generations. Figure 8.1: Stayer-Leaver Brothers by Same Father 350 300 leavers 250 stayers 200 ratio*10 150 100 50 0 1 2 3 4 5 6 7 8 Num ber of Brothers Another part of the explanation for the two inflections in Figure 8.1 is the aggregation of full with half-brothers. Figure 8.2 shows the same data graphed for full siblings and, given the small numbers of larger families, averaged in the upper categories of numbers of brothers. The num- 292 Chapter 8 bers of brothers now shows a single mode, and the ratio of stayers to leavers is more of a curve rising with the number of full brothers. The two graphs taken together support the idea that full brothers are more likely to be cohesive than sets of brothers with two different mothers. What also matters statistically is that these larger sets of brothers remain nomadic and are less likely to migrate to town. Figure 8.2: Stayer-Leaver Brothers by Same Parents 120 100 80 r=.47 60 leavers 40 stayers ratio*10 20 0 1 2 3 4 5 6 7 Number of Brothers of Same Parents 8 Our demographic data cannot provide much more detail than this because of the problem of missing or incomplete data. The blank areas in Table 8.1 indicate where data are missing on sibling sets for the early generations. If we take only those after generations d (born after 1875), where data are more complete, the Pearson’s correlation for brothers of the same mother increases from .47 (over all generations) to .55 and accounts for 30% of the variance in migration ratio. For brothers of the same father the correlation increases from .47 in generations starting in a-c to .53 in generations f and below. For the two generations for which data on completed birth cohorts are accurate, generations f (born >1935) and g (born >1965), the average number of brothers decreased from 3.3 (n=75) to 2.8 (n=86; p<.001). It is thus likely that from generation g forward, although cohesion is still higher among larger sets of brothers, the absolute size of these sets is also shrinking. Table 8.2 also shows more men in the >1965 generation Demographic Choice and Constraint 293 emigrating, but also more whose fathers were from the clan returning from villages. For women, there is no correlation (r = -.07, p = .68) between number of sisters and likelihood of migration, although there might be a tendency for more women to stay if they have more brothers. Table 8.1: Number of Males by Number of Brothers by Generation a b c d e f g h i Totals 1 2 0 2 4 8 16 28 16 2 78 2 2 0 8 7 14 26 44 19 0 120 3 0 6 6 3 26 53 44 8 0 146 4 0 4 4 16 20 58 35 0 0 137 5 0 0 0 0 20 15 5 10 0 50 6 0 0 0 6 7 22 36 0 0 71 7 8 9 0 0 0 0 0 0 Missing data0 0 0 0 0 here0 0 0 0 13 24 9 15 15 3 2 0 0 0 0 0 30 39 12 10 0 0 0 0 0 10 8 0 0 18 Totals 4 10 20 36 95 246 233 55 2 701 Table 8.2: Male Demographics by Generation ab Men: % Emigrating Men: % Returning Men: Herders Immigrating Men: Tribals Immigrating Men: Villagers Immigrating Men: % with wives unknown Men: % Single 2 2 40 >1845 c >75 d 4 >05 e 9 >35 f 12 >65 g 15 3 >80 hi 9 12 53 69 91 2 1 30 6 Marriage Choice and Constraint Spatial and Network Constraints in Finding a Spouse For the Aydınlı nomads, relevant factors in the choice of spouses include spatial distributions and routes of communication, such as the nomad migration routes that influence the demography of emigration, network neighborhoods and the structure of meeting places, presence of different kinds of relatives in those neighborhoods, and the generational levels of those relatives. These factors were examined in analyses 7 through 9 in Chapter 7. They operate as contingencies that must be factored in before we can factor out the extent of actual preferences. Demographic factors need not be overly complex if at each step we integrate our knowledge of the structure of networks and interactions, as we have done for the 294 Chapter 8 nomads in Chapter 7. From that analysis of network neighborhoods and the structure of meeting places there emerged a coherent pattern of interaction among lineages and between them and the sedentary villages that lie parallel to their migratory routes. Variability in the surrounding environment as it differentially impacts local groups is evident in the network topology of the clan, and may be taken as a general principle of nomad social organization. We have also seen how emigration rates correlated with position in the interlineage marriage topology of network neighborhoods, and how the segmentation of lineages, their merging by intermarriage, and changing environments impact on their alliance patterns. Changes in alliance patterns (and the fact that lineages are not fixed entities but groups that shift membership as new generations mature and receding ancestries are forgotten) impact back on the network topology, in some cases changing the structure of the clan. Reading the data of Figure 7.6 from a complexity perspective, for example, changes in alliance patterns among the lineage segments whose members change through time show a temporal fractality as changes in clan structure bubble up through time at varying magnitudes, some of which entail major structural changes. All that is to say that our study of marriage alliances by means of network analysis has shown fruitful results up to this point, although we have not yet addressed the issues of Chapter 4 and the anthropological theories of lineage structure or marriage alliance as social exchange. We now address some of the theoretical questions about “forms of marriage” that we raised in Chapter 4. Demographic Constraints in Choice of a Spouse In Chapters 2, 5, and 7, we argued that anthropological progress in the field of kinship studies has been impeded by lack of systematic analyses of how the types of kin available for marriage are contingently affected by demography, even in how to define marriage rules or preferences (Table 5.1), much less how to study changing rules or preferences. This became evident with Hammel’s (1976) publication of the principles that govern how the rates of FZD and MBD marriages are systematically distorted in favor of the latter when and to the extent to which there are differences in the age or other status characteristics of husband and wife. Given the focus on cousin marriages in anthropological exchange theories, Hammel’s principle seemed to invalidate much of the previous research. What if a disproportionately frequent type of cousin marriage, for example, could be attributed not to preferential rules or choice, but to Demographic Choice and Constraint 295 demographic contingency, in which random encounters with relatives nearby or near in age or of the right status were already biased in the direction of the higher observed frequencies? The kind of thinking that we see in Hammel’s analysis, about spatial and demographic constraints on behavior, is what underlies the kinds of methods and analyses we use in the following analysis. Only by factoring out such constraints can we begin to talk about preferential rules or choice. Analysis 11: Cousin Marriage Demography In contrast to counting the frequency of each of four types of cousin marriages and comparing the percentages of each, as a method for inferring marriage preferences, having our data on the kinship network allows us to compute the percentages of different types of consanguineal marriages relative to the availability of relatives of each type (White and Jorion 1992). This is hardly possible by manual enumeration, and offers a more realistic baseline given Hammel’s principle (Chapter 5) and the potentially massive effect of demographic factors. The Par-Calc program (Pgraph software: White and Skyhorse 1999) performs this type of analysis. The program examines every possible consanguineal relation between men and women, and counts the frequency of each type, such as FBD marriage, for each ego, in each generation, and then norms these frequencies as percentages of those available. White and Jorion (1992) describe in detail the analytic techniques used, which we will not repeat here. The program computes how many men (overall, or by generation) have at least one relative of each type. The percentages of actual over possible marriages tell us how many of those men having a FBD, for example, actually married a FBD, and similarly for other relatives. The analysis includes blood relatives of each marriage type where a couple may have common ancestors up to seven generations back. This includes first, second, and more distant cousins, as well as relatives of different generations. Johansen concluded from her pre-computer survey of her genealogical scroll that FBD marriage was decreasing in numbers through time. Taking FBD marriage as an index of the viability of the system of patrilineages, she concluded that the lineage system was in decline, and that the traditional kinship system of the Aydınlı nomads was breaking up. The frequency observation was correct, but the conclusion is not so simple. Figure 8.3 classifies the frequency and percent of FBD marriages by 296 Chapter 8 historical generations, as well as the percentage of FBD marriages out of all cousin marriages. The table shows each of the three major measures in Table 5.1, with raw frequencies on the left, percents in the middle, and relative numbers on the right. The results confirm Johansen’s earlier impression for the raw frequencies. The percentage of men who married an available FBD (center graph) as a percentage of all cousin marriages (right graph), however, are at variance with the raw frequencies. FBD marriages are rising slightly in relative rate compared to marriage with other cousins, and stable or rising in the interactive rate of percent of men marrying those available. The low absolute numbers of cousin marriages in the generation born after (>) 1874, especially for cousins involving female links, are undoubtedly due to missing data on females in early generations; hence, this generation will be ignored in other figures, and the high relative numbers of FBD in this early generation should be ignored. The low absolute number for FBD marriages in this generation is probably due to lineage segmentation because earlier segments of a patrilineage are likely to have split off or perhaps died out, so reports here are also incomplete. Thus, there is considerable discrepancy between the measures. If absolute and relative numbers for FBD marriage give the impression that FBD marriage is falling out of favor, this is not confirmed for FBD marriage as a percent of those available. In general, the correlation between raw and relative rates of cousin marriages and the selective rates for percentage married of those available (as in Figure 8.3) are r = .36 and r = .28, accounting for only 13% and 8% common variance. Demographic Choice and Constraint 297 Figure 8.3: Comparison of Changes in Cousin Marriage Rates using Absolute, Percent Married of Those Available, and Relative Numbers Absolute numbers 50 45 % of Available Relative numbers as a % of cousins >18745 40 35 >1905 30 >1935 25 20 >1965 15 10 5 0 FBD MZD FZD MBD %FBD %MZD %FZD %MBD fbd mzd fzd mbd Legend: The actual frequency of each type of cousin marriage (on the left) is divided by the number of men who have a relative of this type (not shown) to give a percentage (center graphs) that may be used for comparative purposes and are not subject to distortion by demographic variables such as number of siblings that affect the relative percentages (graphs on the right) of cousin marriages by type. What we need to explain to understand the evident changes in Figure 8.3 are the following: a. Raw frequency declines for all four types of cousin marriages in the last generation. This is a likely result of higher rates of emigration, so that the size of sibling sets is smaller and the likelihood of finding cousins to marry is lower. b. At the same time, MZD (uterine line) marriages increase in the percent married of those available. This might be the expected result of greater visiting with relatives who have emigrated to villages. c. FZD marriage rises gradually in percent married of those available. Note that FZ is an agnatic relative, and while FZD is a child of a lineage mate she belongs to her father’s lineage. Changes for some of the more remote or nonagnatic cousin marriages should follow some of the same principles as those closer in. Hence, we want to examine those data, on more remote cousin marriages, to verify that we have correctly identified consistent patterns of change. First, however, we examine the more extensive patterns of patrilineal endogamy beyond FBD proper. 298 Chapter 8 FBD and the Agnatic line As Figure 8.4 shows, the selective rate of marriage with agnatic lineage mates is also on the rise over recent generations, even as the numbers of agnatic relatives available to marry is on the decline. If we compare rates of marriage with patrilateral relatives of various types, as in Figure 8.5, we see a rise in percent of marriage with patrilineage mates (the first two bar graphs in the upper figure). These rises contrast with the low rates overall for marriages with other patrilateral relatives (the last four bar graphs in the lower figure). Data for early generations are not given because the absolute numbers are too close to zero to compute percentage rates. Demographic Choice and Constraint 299 Figure 8.4: Changes in Marriage Rates within Agnatic Lines 30 90 80 70 60 50 40 30 20 10 0 25 20 15 10 5 0 >1905 >1935 Patrilineage mates married as % of Numbers available (25-80) >1965 Figure 8.5: Comparison of Changes in Patrilateral Marriage Rates using Absolute Numbers and Percent Married of those Available 50 40 30 20 Absolute Numbers >1905 >1935 >1965 10 0 FFBSD FFFBSSD FFZSD FFZDD FMBDD FMZSD The raw frequencies of marriages with designated types of relatives are shown by bar graphs for successive thirty-year periods from 1905 to 1965. Below are the percentages of marriages with the same designated types of relatives, percentaged over the total number of each type of relative available to marry. 50 40 30 % of Available >1905 >1935 >1965 20 10 0 %FFBSD %FFFBSSD %FFZSD %FFZDD %FMBDD %FMZSD 300 Chapter 8 The following hypotheses are generally supported by these data: Hypothesis 8.2.1: The agnatic principle in marriage is not diminishing in importance. Hypothesis 8.2.2: There is a general preference for closer patrilateral parallel cousins than for more distant ones. Hypothesis 8.2.3: The tendency of FBD marriage to decline in absolute numbers in the twentieth century is a result of demographic factors. These might include greater numbers of nomads shifting to village life, or a decline in the number of siblings due to a demographic shift. Such factors would account for a decline in the numbers of FB and FBD relatives, if the cohort size for parallel cousins were reduced, say, by lower fertility or outmigration. FBD Marriage: Index of Tradition or Generator of Diversity? Frequency of FBD marriages is a weak index of agnatic lineages in the context of nomadism in the Middle East. Rights to FBD marriage may exist even in instances where people choose not to exercise the right (Berrenberg 2003, Bell 2002). When massively impacted by the demography of the turn to settled life, for example, our selective measure of percentage of marriage with those available may be a better indicator of preference than raw or relative frequencies (Table 5.1). Within the context of lineage endogamy, many other types of cousin marriage occur as well. Intralineage MBD marriages, as diagrammed in Figure 4.6 for example, occur within lineages #1 (2 cases), #2 (2 cases), and #5 (1); lineages #2 and #3 have reciprocal MBD marriages; and almost all lineages have at least one MBD marriage (wife-taking for #1-23-4-7-8 and wife-giving from #1-2-3-4-5-6). MMBDD and MMMBDDD occur selectively within lineages (#1 and #2) and where MBD marriages are already present, and in one case creating a cycle of wife-giving from #2 to #7 to #5 (through MMBDD and MMMBDDD) to #2 (the latter with concurrent cases of MBD). MBD combines with dense intermarriage between as well as within lineages to form a marriage system of generalized exchange. FZD marriage also does not occur as a privileged type, nor is it restricted to merely local or reciprocal marriage exchanges between lineages, but it is part of the pattern of generalized exchange. FBD marriage, for all its importance, is only part of a diversity of marriage types and strategies in a system of shifting competitive and cooperative alliances. Demographic Choice and Constraint 301 It would seem that there is abundant empirical evidence for the hypothesis of Chapter 4, that FBD marriage is associated with diversity in types of marriage. MBD Marriage: The Effects of Spatiality and Demography Because the selective percentages of FBD, FFBSD and FFFBSSD marriage decrease with kinship distance (17%, 8%, 3%, respectively, over all periods), for example, it is evident that the expressed preference for FBD marriage is a function of both common lineage and close patrilateral ties. For MMMBDDD, MMBDD, and MBD marriages (with rates of 14%, 10%, and 10%), however, the selection for distant marriages is stronger than for the closer ones. These marriages (see Figure 8.6 for MBD, and MMBDD) start from a zero baseline for those born in the nineteenth century, rise dramatically in the twentieth century, and fall again in the most recent generation. Hammel’s (1976) principle that inequality of age at marriage increases the number MBD versus FZD relatives of an appropriate age for marriage does not provide an explanation for the drop in selective MBD marriage in generation g because age differences in marriage were minimal in both the earlier and the later time periods. No systematic differences of status or age in marriage were apparent to Johansen. Figure 8.6: Changes in Matri-Cross and Similar Cousin Marriages 25 >1905 20 >1935 15 >1965 10 5 0 MBD MMBDD MMBSD MFBSD %MBD %MMBDD %MMBSD %MFBSD MZD and the Uterine line Emigration of women significantly exceeds that of men, as shown in Figure 8.7. This seems to be especially true in later generations but— given the underreporting of females in early generations—this is likely to have been true earlier as well. Patrilocal residence and higher rates of female exogamy and emigration would imply a lower availability of 302 Chapter 8 MZD relatives as potential spouses, even in the early periods. We know, for example, that women often marry into other nomad tribes, but men rarely if ever do so; hence, spatial mobility is greater for women than men (see Figures 6.1 and 6.2 for effects on agnatic and uterine lines). The same is true for marriage of women into settled villages. What is evident in recent times, however, is heavy migration of men as well, and with new forms of mobility through trucks, buses, and cars and more visiting of relatives in villages. Looked at in terms of an identity or descent principle, the rates of MZD and MMZDD uterine marriages rise in two most recent generations (rates of 6%, 9%, and 25%), as shown in Figure 8.8, even while the numbers of these relatives available for marriage are falling sharply. In traditional pasturage, girls’ movements are restricted, and they have little daily contact with relatives outside their extended family (which would include the FBS). The head of the family and married sons and children are a separate unit of residence and production, and girls work in or nearby their family’s tent. Unmarried boys would also have little contact with their MZD, who would reside in another place and belong to a family affiliated with another lineage. MZ was a nonetheless a preferred confidant of young people, including boys, so this might easily lead to a preference for MZ contacts in visiting patterns under the right circumstances and hence lead to MZD marriages. Figure 8.7: Changes in Percent Emigration of Men and Women, from Generations 3 (1875) to 7 (1990) 30 1875 1900 1930 1960 1990 25 To village-Men 20 To villageWomen To other tribeMen To other tribeWomen 15 10 5 0 3 4 5 6 7 Demographic Choice and Constraint 303 Figure 8.8: Changes in Marriage Rates within Uterine Lines 30 70 25 60 20 50 15 Matrilineage mates married as % of 10 Numbers available (12-62) 40 30 20 5 10 0 0 >1905 >1935 >1965 Historically, the increases in selective rates of MZD and MMZDD marriages among Aydınlı nomads correspond to increasing sedentary contact in the sense of Aydınlı settling near villages in winter, in which they may rent land and have access to grazing from the village common lands. The pressure to do so has increased along with increased population density in southeastern Turkey generally and the enclosing of lands available for pasture. When residing on the edge of villages, nomad families from different lineages come to reside on adjacent plots of land and families are more likely to interact. Hence, Johansen’s hypothesis concerning social change in MZD marriage practices is this: Hypothesis 8.3: The increase in MZD selective marriage rates is a result of greater sedentary contact in which visiting patterns entail a greater likelihood of meeting relatives linked through females. From her perspective as an ethnographer, Johansen showed that, consistent with the statistical analysis, increased sedentary contact has had the effect of changing the visiting patterns among nomad families. Analysis 12: FBD and MZD Demography Compared Theory and Measurement: FBD and MZD Demographics In comparing theories about marriage and types of marriage, appropriate means of measurement are needed that are sensitive to how the frequencies of different behaviors are conditioned by the demographics and em- 304 Chapter 8 pirical relations existent within the network in which the behaviors are found. As an illustration of how measurement issues might affect interpretations of marriage frequencies, and hence theoretical conclusions, we will compare some of our findings on types of cousin marriage with those of Barry (2000). Barry’s hypothesis is that parallel cousin marriages, both of the MZD and FBD varieties, are affected by behavioral and psychological tendencies to avoid or not to avoid relatives on the basis of concepts of identity based on shared substance transmitted in the uterine or agnatic lines, respectively. Given the favored position of MZ as a confidant of unmarried boys among Aydınlı nomads, the rise in visitation between female relatives and the percent of marriages with MZD among those available in recent generations, and the decline in absolute numbers of FBD marriage, neither Hypothesis 8.3 nor Barry’s hypothesis is implausible. This use of relative rates for different types of cousin marriages, as explained in Table 5.1, however, leaves much to be desired. Both MZD and FBD marriages are on the decline in absolute terms, but on the rise in selective percentages. As for Aydınlı nomad concepts of identity and shared substance, “bone” (Turkish =kemik, the male contribution through the visible entry of semen) is thought to be heritable and to give rise to morphological similarities between lines of fathers and sons, while “flesh” is thought to be the female contribution to the child in the womb and in later nurturance through breastfeeding. Thus, the Aydınlı do not have a concept of resemblances between mother and children due to the maternal contribution of inherited “flesh.” These resemblances are recognized, but thought to come from nurture, not inheritance. Presumably, Barry could assimilate the idea of substance transmitted through nurture, but this would deviate from the source of his theory. While the “seed” that is nurtured is thought to come from the father, the concept of ovulation and a hereditary line of resemblance through the mother are unrecognized. Hence, while female lines are of fundamental importance for the Aydınlı, the importance of these links does not correlate with a concept of a uterine line. To examine the evidence for Barry’s theory, we reproduce in Figure 8.9 his regression results. The inverse correlation (r 2=.72) between FBD and MZD marriages, as measured by raw frequencies, is evident from the nineteen data points labeled by numbers in square brackets, each of which represents for a one society the relative frequencies of cousin marriages. The regression line through these cases has a steep negative slope. The other two (flat) lines are the regressions lines of FBD with Demographic Choice and Constraint 305 Fréquences du marriage avec les autres cousins MBD (r2=.0357) and with FZD (r2=.064) marriage frequencies. Figure 8.9: Bipolar Continuum between FBD and MZD marriages Fréquences du marriage avec FBD pour 100 marriages de cousins We have added to Figure 8.9 (Barry’s figure) a dashed and dashed ellipses around the sedentary and nomadic groups, respectively. The negative regression line through the cases within these ellipses is due to a contrast between sedentary (upper left) and nomadic (lower right) societies. FBD marriage is much higher and MZD marriage much lower for the nomad than for the settled groups. The Aydınlı nomads fit Barry’s negative correlation regression line for FBD and MZD marriage frequencies: Their frequencies relative to all cousin marriages over all time periods are 45% and 18%, respectively, at the midpoint in Barry’s distribution. Barry’s theory sensitizes us to the importance of enunciating principles of social identity. We have argued throughout this book that uterine ties are important for establishing social cohesion in the nomadic clan. He gives us an insight as to how the importance of uterine ties as a lateral extension of kinship identity may be linked to the importance of vertical ties in agnatic descent groups. MZD-FBD Inverse Correlation: Residence as a Third Factor There is a simpler explanation of Barry’s finding of a strong negative correlation, which is unusual in comparative studies, between FZD and MZD marriage frequencies for societies where FZD marriage is strong to 306 Chapter 8 moderate, that is, constituting over 25% of all cousin marriages. The two ellipses that we have drawn on his graph, one for settled populations (the dashed ellipse) and the other for nomadic groups (solid ellipse), provide a different interpretation of his findings. Our explanation for the negative correlation is that the nomadic societies (within in the solid oval within the figure) are more likely to be patrilocal so that, while men tend to stay in their local group after marriage, women are more likely to marry out and emigrate from the population. This reduces the number of women who stand in relation to a male ego as MZD, as opposed to those related as FBD. Hence, the stronger the patrilocality, the more FBD will outnumber MZD marriages numerically. Conversely, the weaker the patrilocality, as in settled populations (those within the dashed oval of the figure), the more equal will be the numbers FBD and MZD marriages. That is what we see in Figure 8.9. Hypothesis 8.4: FBD marriages tend to increase in frequency as MZD marriages decrease, but relative to a third factor that influences them both: Greater patrilocality and lesser visiting of female relatives who have emigrated from the home community augment the likelihood of FZD marriage and diminishes that of marriage with MZD. We have already seen evidence for Aydınlı nomads that MZD marriage percentages (not absolute numbers) increase as the nomads visit in recent generations with their settled relatives (Hypothesis 8.3), although FBD marriage percentages (not absolute numbers) increase as well, with patrilocality remaining the rule in the nomadic population. Table 8.3 tests a generalized Hypothesis 8.4, that MZD marriages tend to increase in frequency as FBD marriages decrease among the Aydınlı nomads, without further examination of the underlying causation. By Barry’s method of percentaging the frequencies, a change in this direction is weakly indicated in successive time periods. An inverse trend is not evident by the method of measuring relative rates (Table 5.1) Table 8.3: Parallel Cousin Marriage through Time by (a) Relative Rates (relative to all cousin marriages) and (b) Selective Rates (relative to available cousins of each type) (a) Barry’s % method (b) Our selective rate method FBD MZD FBD MZD generation c 50% 0% born 1846-1875 generation d 50% 12% 28% 7% born 1876-1905 generation e 37% 11% 25% 6% born 1906-1935 Demographic Choice and Constraint 307 generation f 39% 13% 33% 10% born 1936-1965 generation g 43% 29% 30% 20% born 1966-1995 A straightforward reading of these results using Barry’s interpretation would conclude that uterine identity (MZD avoidance) has weakened for those born 1966-1995. As MZD marriage increases in frequency, FBD marriage should fall, according to his regression line. But discounting the early generation bias inflating FBD counts, there is no tendency toward lower percentages of FBD marriages (agnatic descent principle) by either method. The trend in Table 8.3, in both Barry’s and our measure of percentages married from those available, is toward more equal rates of cousin marriages (this, even if low in absolute terms, would indicate for Barry a shift away from “Arab” type marriages toward his “complex” type of marriage system). Other data that need to be examined are the rates of parallel second cousin marriages: Barry’s theory would predict that the agnatic and uterine types will also vary inversely. As shown in Table 8.4, using both Barry’s and our methods, second cousin FFBSD marriage increases for those born 1966-1995, but MMZDD marriage remains roughly constant by Barry’s measures and increases using our method, contra Barry’s argument. As we might expect from the indigenous theory of kinship and inheritance, his concept of uterine versus agnatic descent lines does not work for the Aydınlı nomads. For all the importance of female links as the source of nurturance for children and for the Aydınlı tracing of alliances, the concept of motherhood is not one that emphasizes transmission of hereditary substances, and hence not one that would lead to the “identity” within a uterine line, nor to proscriptions against marriage within that line. Table 8.4: Parallel Second Cousin Marriage through Time by (a) Relative Rates (relative to all cousin marriages) and (b) Selective Rates (relative to available cousins of each type) (a) Barry’s % method* (b) Our selective rate method** FFBSD MMZDD FFBSD MMZDD generation c 0% 0% born 1846-1875 generation d 0% 0% born 1876-1905 generation e 25% 18% 7% 5% born 1906-1935 generation f 27% 9% 12% 9% born 1936-1965 generation g 67% 16% 50% 50% born 1966-1995 * contrasted with FFSSD and MMBDD Findings from Controlled Simulation 308 Chapter 8 Controlled or Feynman simulation (White 1999), as noted in Chapter 5, holds constant the demographics of a kinship network, generation by generation, and reassigns marriage choices randomly within each generation, subject to the actual rules of marriage prohibitions of a population. The technique used here is to hold the male links in the network constant, plus the parents of females, and to reassign married females to married males randomly within each generation, within the constraints of a realistic marriage prohibition. This is then repeated in the randomized structure, holding constant this time the female links and reassigning the marriages of the males. This holds constant the distribution of nuclear families and the gender composition of their offspring in each generation. Comparisons of actual marriage choices to controlled simulation outcomes is the most powerful tool presently available to examine the evidence for marriage preferences as against the background of random choice given demographic and proscriptive constraints on marriages. Given our findings thus far, we reject Barry’s theory in favor of Hypothesis 8.4, and argue that: Hypothesis 8.5: Cousin marriage preferences for the Turkish nomads are not extended through lineage principles. That is, the only parallel cousin marriages that occur with a frequency greater than expected by chance, in comparison to a controlled simulation of clan marriages (with siblings proscribed), are those of first cousins, not those of more distant cousins. The results of testing Hypothesis 8.5 are shown in Table 8.5, using comparisons between actual and simulated frequencies of parallel cousin marriages. For first cousins, but not more distant cousins, the rates of parallel cousin marriage are much higher than expected by chance (p<.001) for both FBD and MZD. For second through fourth cousins, the parallel cousin marriage frequencies are within the range of those expected by chance. This finding would be expected from and consistent with Turkish nomad beliefs concerning nonhereditary contributions of “flesh” in mother-child relationships, but consistent with the ethnographer’s observation that the preference for FBD marriage is a matter of close relationships within the extended family, and is not a matter of lineages. The use of FBD marriage as a strategy to reinforce the agnatic line does not exceed the level that might be expected of a variety of widely dispersed or distributed marriages. Beyond the first degree, cousins are Demographic Choice and Constraint 309 neither strongly avoided nor preferred either for agnatic or uterine kinship lines. The exception is the MMZDD, for which Johansen’s Hypothesis 8.3 rather than the lineage principle might provide an explanation, but there is also a slight and equally significant tendency (weaker in terms of actual/expected ratios; more significant because of the greater number of male links in the population) for FFBDD marriage. Table 8.5: Comparison of Actual with Simulated Parallel Cousin Marriage Frequencies Patrilateral parallel Ratios Matrilateral parallel Ratios marriage marriage F(F) . . . B(S) . . . D M(M) . . . Z(D) . . . D Cousin Actual SimulatP Actual SimulatM Marriages data ed data data ed data First 3.8% 4:1 5.4% 0.7% 8:1 16% p<.001 8/149 1/149 p<.001 31/187 7/186 Second 7.0% 4.8% 1.5:1 6% 0% 3:0 11/157 7/146 p=.14 3/50 0/45 p=.15 Third 3% 3.2% n.s. 0% 0% n.s. 4/126 3/94 0/19 0/16 Fourth 0% 1.7% n.s. 0% 0% n.s. 1/103 1/59 0/15 0/5 Marriage and Social Change The social pressure that preservation of nomadism exerts on reproduction is evident in marriage customs. In spite of schooling, many fathers pressure their sons to marry before they have to do their two years of military services, at the age of eighteen or nineteen years, with the purpose of possibly getting grandchildren from them even in the case their sons lost their lives as soldiers and, when serving in a town, preventing “sinful” sexual experiences with prostitutes. The wish to delay military service was the reason why births of boys were often announced to the administration not before the child reached an age of two to three years. The importance of marriage to nomadic life is reflected in the norms applying to marriage as a normal state for both men and women. Girls were very rarely unmarried and until about 1980 it was looked at as ridiculous and blameworthy if a healthy girl did not marry. Likewise, all men were married. A sixty year-old man in a village about 30km away who had never married was looked at as a sensation. The usual age for 310 Chapter 8 marriages was fifteen to twenty for girls and seventeen to twenty-five for young men. In 1982, after it had become a habit that all boys and many girls to receive at least five forms of schooling, the age of marriage rose by two or three years. Formal schooling, as in most traditional cultures, is a major transforming factor in the lifeways of the clan. Stayers and Leavers Revisited Barth (1964) noted for the Basseri goat- and sheepherders of Iran that fertility is high, and the pressure of the nomadic way of life on the nomads is so severe that great numbers of families and individuals are sloughed off through migration to villages and towns. In most cases these are the poor and unsuccessful, but in some cases they are those who have invested wealth in land in and near villages or towns, and leave nomadism for landlordism. For Aydınlı nomads, similar pressures and processes are operative. Our study, in addition, has shown that the larger sibling sets (particularly those of brothers) are the most likely to be stayers. In Chapter 6 (Table 6.4) we found that marital cohesion through relinking was a strong correlate of remaining nomadic, within the clan, rather than emigrating. FBD marriage contributes to both sides of the equation, as both a form of relinking and a means of cohesively relinking the families of brothers who have taken up or are likely to take up separate residence after their father dies. Summary The sloughing off of population among the nomads occurs more frequently in the sibling sets with fewer brothers, and less frequently the greater the number of brothers (Figures 8.1 and 8.2). Because nomads typically have a healthy environment in which women’s activity promotes high fertility, our findings in exploring Hypothesis 8.1 provide needed detail about the process of sloughing off population also reported by other ethnographers (e.g., Barth 1953) of Middle Eastern sheep and goat pastoralists. This greater cohesion of larger sets of brothers, especially full siblings, and the lack of a similar pattern for sisters, implies a demographic situation in which shallow patrilineages are quite large, and those men who do not migrate to villages then have many living males who are close lineage males. Over time, this pattern will also tend to develop three-generation lineage segments that are deeply embedded in Demographic Choice and Constraint 311 large and deep segmented patrilineages. The smaller patrilines, given this pattern, are both more likely to migrate out as well as to die out given their smaller size. Figure 8.10 supplements Barth’s diagram in Figure 7.1 by showing some of the operative pressures, factors, and processes affecting size of lineages, co-selection for residence with the clan, marital relinking versus emigration, and flexible ranking. These outcome variables are shown on the right side of the diagram, with arrows to or between them to indicate probable influences. Two of the many factors that affect them are the quality of the environment (upper right, influencing high fertility) and the pastoral division of labor (lower right, enhancing value of cohesion and belief in equality). Other influences can be read from the diagram. Figure 8.10: The Effects of Environment, Fertility, and Sibship Size on Stayers’ Sublineage Sizes Good Environment High Fertility Many Brothers High Population Pressure Pastoral Division of Labor Enhances Value of Cohesion Sloughing off Population Belief in Same Age Equality Marriage Large Segmented Lineages Co-Selection for In-Group Residence Emigration versus Marital Relinking Flexible Ranking, given Emigration Every new marriage is one that knits together various components of the functioning society of the nomadic clan. We have seen great diversity in the ways that marriage as a coherent and cohesive force operates: within the extended family, within lineages, between lineages, between tribes, and establishing patterns of visiting and exchange with villages and towns. Structural endogamy is a key factor in continued residence in the nomadic group but also provides coherence between the complex bodies of skills and knowledge possessed by clan members, both men and women. Marriage preferences for FBD continue to be expressed by rural families in the present era but the raw frequencies of FBD marriage plummet in the most recent period, along with those of other types of cousin mar- 312 Chapter 8 riage. Is this an index of the decline of the lineage principle? The network calculation of young women available in these categories also drops with the increases in emigration (Figure 8.7). It may be demographic constraints, not a change in the kinds of spouses that are selected, that account for the decline in frequencies. What is evident from the demographic data is that the size of sibling cohorts is shrinking. The answer to our question about effects of demography needs further close evaluation, but one of the key questions is whether there is still solidarity among large sibling sets, especially whether larger sets of brothers tend to remain nomadic. To get a preliminary answer, we can compare rates of emigration of males who were sons of women in two recent generations (e and f), the pre- and post-World War II cohorts for women with completed births. For the prewar mothers, the husbands’ rates of emigration (an index of the family’s moving) were 15% for those with five children or less, and 13% with six or more. For the postwar mothers, the rates rose to 23% for the smaller families but dip to only 4% for the larger. This is a much greater rate of retention of nomadism among the larger completed families of the more recent generation. It seems likely, however, that a demographic shift to smaller family sizes is underway. Whether this will affect nomadism as a viable way of life in this region remains to be seen, but we must wait for completed fertility to make a new assessment for the most recent generations. A further demographic analysis, which must be the subject of a separate study, might well show that as migration rates have increased in recent years there is an even greater effort on the part of parents and siblings to keep intact the larger families that seem to be essential both to nomadism and to patrilineages. The answer to the question as to the continued viability of patrilineage segments is still unfinished, but there is no strong reason from our evidence that its viability is diminished as an element of nomadism. The self-selection process fits a general pattern observed for camel nomads and small animal pastoralists in which members of smaller units are more likely to emigrate (Barth 1953), pushed out by competition over limited resources whose productive use depends on effective sublineage-size production units that are also defensive and raiding units (Bell 2002). New methods of analyzing marriage rates among kin open new questions. Cousin marriages on the mother’s side, for example, remain significant for MBD and increase for MZD if we take as our measure the percentage of men with cousins available in these categories that make such marriages. The reason for some of these changes seems to be the combination of increased migration and the visiting of the mother’s kin in the Demographic Choice and Constraint 313 villages. Lineage membership, either agnatic or uterine, does not seem to be a primary factor in the favoring of parallel cousin marriages. The more preferential parallel cousin marriages are with first cousins; those with third cousins occur at the same rate as simulated random marriages under existing demographic constraints (Feynman simulation). The marriage patterns in general are in line with the complexity distributions introduced in the previous chapter. The general hypotheses advanced in Chapter 4, namely, that FBD is a marriage type associated with a preferential gradient for diversity in marriage types, is supported. Further Reading Ethnographic examples of the network study of marriage, rank, and leadership are found in White and Schweizer (1998) and Houseman and White (1998b). Barth (1953), Korotayev (2000), and Bell (2002) provide an overview of the issues surrounding FBD marriage. The property of fractality as a component of self-organizing systems is reviewed in Solé and Goodwin (2000). Chapter 9 Decentralized Leadership and Network Cohesion Decentralized Leadership and the Aydınlı Case A well-analyzed social network is one of the rare contexts in which it is possible to observe political processes closely enough across generations to pose questions about the nature of emergent leadership and the pathways by which it is produced and reproduced. In centralized political systems, which occur in both state and nonstate societies, recruitment into leadership positions is hemmed in by highly institutionalized processes. In state systems, political institutions, such as parties and formal bureaucratic governmental institutions, play the role of defining leadership positions and shaping the careers and opportunities of potential leaders. In many pre-state societies, the institutional forms of relatively centralized political institutions, such as rules of hereditary succession, play a similar role in institutionalized recruitment. Not so obvious are the career pathways of emergent leaders. To observe recruitment and influence in its more protean form, decentralized and relatively autonomous political contexts are thus of special interest. These include societies with emergent leadership where leaders are neither routinely appointed nor elected nor selected according to overt prescribed criteria, and those in which leadership is not hierarchical, that is, in which the leader does not command but is regarded as a leading councilor among peers. Weber termed emergent leaders “charismatic” to call attention to the unique personality of the leader. It is the social process by which leaders emerge that is of more interest to us, regardless of whether they are charismatic in the Weberian sense or not. In an extremely wide variety of contexts in which the ethos of the group is egalitarian, the more protean processes of leadership occur (Boehm 1993, 1997, 1999), and leaders emerge by informal processes. Such systems are pervasive in many parts of the world, including inter- 316 Chapter 9 stitial groups in state societies. Leadership processes in relatively new nation-states, and emergent leadership is characteristic of many pre-state societies that operate with a decentralized political process. What we may learn from the Aydınlı nomad case concerns the network processes of this more protean form. We will identify candidates, their positions in networks, and outcomes, and determine whether similar kinds of outcomes, given network location of actors, are repeated over generations of emergent leaders. Aydınlı Leadership and Network Cohesion One of the ideas we want to test is whether potential candidates become qualified for leadership, in addition to their personal attributes, by the extent of their cohesive embedding in a social network. The Aydınlı nomads are an interesting case in point. There is no rule of succession to the leadership position of tanıdık kişi (see Table 9.2 and 9.7). For all the importance of kinship networks among the Aydınlı, their system of leadership is not based on hereditary succession. They depend for social and political support on cohesive embedding in social networks. Only in the earliest generations of the late nineteenth century, at a time when lineage #5 had already arrived from the west and was clearly the dominant group, did succession go from father to son (228 to 343). Thereafter, typically, each major lineage had its contenders for tanıdık kişi, and succession often went to a contender in a lineage in competition with that of the current tanıdık kişi. Tribal “two-party systems” of interlineage competition in the Middle East are discussed by Gellner (1981:186 & 190) and Yalçin-Heckmann (1991:115). The acceptance of a leader became obvious by the men’s frequent visits in his tent. A tanıdık kişi has to own a four-pole tent, not only the usual three-pole tent. Thus his selâmlık (=men’s and guestroom within the tent) was doubly larger than normal. A patriarch’s pretensions to play a leading role in the clan were shown, in the cases where it was not already necessary by reason of the extraordinary largeness of his family, by his having the women of his family weave such a tent. Some of the patriarchs had special reception tents. But it would be seen as ridiculous to own such rooms, if honorable guests did not regularly crowd them. A decision was usually known by consensus long before a new tanıdık kişi finally emerged, as evidenced by attendance as the tents. Such discussions, in which respected men of each lineage were seated in Decentralized Leadership and Network Cohesion 317 prominent positions, did not prevent other candidates from offering themselves or competing for recognition. Even before such a leader might begin to make what were widely considered mistakes of judgment, however, criticism of a tanıdık kişi would often be heard from families who thought they had a better candidate for leadership, and there was no shortage of such discussions. Kozan Mehmet (32), for example, the tanıdık kişi from 1930 to 1957, was ruined by the criticism of his egotism when he choose a new village location and then choose the best site for himself, after which others moved elsewhere. The tanıdık kişi in the time from 1957 to 1982 was Fındıklı Hacı (=Hazelnut Hacı— Hacı=“Pilgrim” was not a title but an official first name) from the Ecevitli A-B lineage (#2; 818). In the generation before him Kozan Mehmet (=Cauldron-Mehmet—but Kozan is the name of a town too) from the Dolaşıklı lineage (#1; 32) was the tanıdık kişi, but he became sedentary before Johansen’s 1957 stay with the nomads. His younger cousin Hacı Molla (=Hacı, the Religious Student or Pilgrim; 99) tried to become the tanıdık kişi after him and until the age of about eighty (he is now over 100 years old) always muttered against Fındıklı Hacı’s decisions. He almost never went to the latter’s tent and was the center of all those who complained about Fındıklı Hacı’s egoism. His second marriage with a Ecevitli A woman (674) was a failure: The woman he was given turned out to be mentally ill. Kırbaşı’s lineage, which had developed close ties to Kozan Mehmet’s family by agreeing to the marriage of a sister of the ruling patriarchs to him, also spoke of the Hazelnuts often as “swaggerers” and did not fully accept them. Nevertheless, Fındıklı Hacı as the second and most able son of the pious Fındıklı Ali (=Hazelnut Ali; 784, the second son of Mustan) and—so far as Johansen could elicit—the most powerful competitor of Kozan Mehmet, was accepted as tanıdık kişi, not Hacı Molla. After Fındıklı Hacı’s illness and death, the role of most influential personality shifted to Mustafa (597), nicknamed “Dede” (=Grandfather), from Johansen’s extended family of the Kırbaşı lineage (#4)—the group somewhat opposed to the Ecevitli leadership. She observed in 1982, in the period of his leadership, that “Dede” received no less than five visitors daily, often more; some of them patiently waiting for him for hours, served by his two wives in the guest room when he was away, while the women and children used the other room of the tent (see Johansen 1965). Fındıklı Hacı (818) successfully received nearly every day a number of male guests and decided, in the above-mentioned way, affairs such as the settling of disputes between the families, if this could not be ar- 318 Chapter 9 ranged by the families themselves, over issues such as the use of wells, the confining of herds with epidemic diseases to a special part of the summer pasture, or the restitution of bride payment, when a young woman had fled from her husband’s tent back to her parents. Fındıklı Hacı organized matters such as the elections of the official mayor of the village they pretended to live in—in the first period himself, and in 1963 putting up as his representative a young man from an allied lineage who was dependant on him. He also negotiated winter pasture for his joint family and some less influential families with village elders in the Çukurova, the transport of poor people to the physician in the next town, and the engagement of lawyers in lawsuits of common interest for the group. Such lawsuits he watched personally and whenever there was an occasion he appeared as a witness for the sake of his group. He had no means to enforce the carrying through of the decisions made in his tent, but they were looked at by most of the clan members as public opinion against which it was not popular to act. Because the tanıdık kişiler were members of the most numerous lineages and not only family solidarity but the standing together of all relatives is a well-known necessity in societies where governmental institutions are far away (cf. Barth 1953:70), they could always rely on a large group of followers. Thus, the many marriage relinkings shaped a basis for their power and their power was the reason that many families were interested in establishing new relinkings with their families. To try to document the importance of this type of diffuse sociopolitical support, we will measure cohesive embedding in networks in a way that has not been employed before in the social science literature on leadership and social or political support. Hence, this study speaks to issues of how to understand complex and decentralized political systems. For a variety of reasons, and because social science has had little detailed ethnographic and network data available for study of decentralized and relatively autonomous political systems, a test of the links between leadership emergence and cohesive embedding among the Aydınlı nomads may be of general interest. This study may thus potentially contribute to a much broader literature on latent sources of social and political support. This type of diffuse leadership also has considerable importance for complex societies, which is not so evident given that more formally institutionalized process are typically considered sufficient to explain the recruitment of leaders. As suggested earlier, examination of Aydınlı leadership—being relatively marginal to Turkish national political institutions—is all the more important because it is Decentralized Leadership and Network Cohesion 319 situated outside the context of parties and formal governmental institutions, and outside the contexts of hereditary or elective leadership that are important in pre-state societies. Network Cohesion Created by Marriage as a Predictor of Emergent Leadership: Hypotheses and Measures In the previous chapter, our problem of measuring cohesion was relatively simple: we were concerned only with identification of the clustering and transitivity of reciprocal exchanges between lineage segments. There, we looked at local cohesion or clustering of local lineage segments in terms of the extent to which each such unit had a set of like units in its immediate network neighborhood (i.e., consisting of those to which it is directly linked) that were in turn directly linked with one another so as to form broader clusters, and at the still larger societal level how the local clusters link to form a coherent structure. Here, we are interested in cohesion on a scale broader than network clustering. Aydınlı nomad leaders, the tanıdık kişi, according to informant reports, are not simply proposed by lineages, or selected on the basis of their reciprocal alliances, or even on the basis of the size of their clusters of ties. The ethnographer’s report was that they emerge as candidates for clan leadership through their support from the broadest possible networks of family, distant relatives, and allies through marriage from throughout the clan. Second, it is not the pure number of persons who are close in the network of a tanıdık kişi, but the personal weight of the clan members who are on his side. Third, in any dispute there are always some among the patriarchs, for example, who stand aside and do not want to interfere. These perceptions led us to examine broader cohesive embedding and network centralities as potential predictors of emergent leadership. Cohesive embedding is the extent to which an individual or couple possesses a high level (for the moment read: density, but be prepared for other concepts to follow) of cohesive linkage with as broad a group as possible within the larger group. This conception of cohesion, shortly to be defined more precisely, differs from centrality in a network (Freeman 1977, 1979), which is often measured for individual nodes as the raw number of links a person has (degree centrality) or, at the global level, by having in the aggregate the shortest possible paths of reachability (closeness centrality) to every other member of the group. Both are too 320 Chapter 9 narrow to capture cohesion, and they ignore how cohesive subsets are themselves organized in the larger network. Cohesion also differs from two other concepts of centrality at the global level: betweenness centrality, which is the extent to which a person mediates the paths that connect other pairs in the network, and the recursive centrality of links to others that are central. Both of these concepts weight the strategic placement of being between groups as much as being at the cohesive center of a group, and they do not capture the dimension of cohesion that we want to explore. Cohesive embedding in a social network—the extent to which someone or some couple is at a high level of cohesion that is shared by as large a subset of others as possible within the total group—sounds like a difficult network concept to measure (see Friedkin 1998) as an alternative predictor of emergent leaders to the standard measures of centrality. It is a measure of social cohesion recently proposed by White and Harary (2001) and one which they and others (Moody and White 2003) have found to be predictive of many of the theoretical consequences that should follow from the differentially cohesive placement of individuals in a network. Friedkin (1993) finds similar results for processes of interpersonal influence. In the present case of candidates for emergent leadership, we expect that those with higher levels of network cohesion in a potentially largescale social group (in this case the larger clan) will also have higher levels of participation with others in joint activities, and enjoy higher levels of broad-based support. This leads to testable hypotheses as to the consequences of varying levels of cohesion in kinship networks. Hypothesis 9.1.1: Differences in the extent to which individuals or couples are cohesively linked within the kinship network of the broad social group of the nomad clan are predictive of broad-based social and political support for potential leaders, and hence of the emergence of particular individuals as tanıdık kişi, the emergent leaders of the clan. Hypothesis 9.1.2: Measures of cohesion will be better predictors of tanıdık kişi emergent leadership than measures of degree, closeness, betweenness, or eigen centrality (see Glossary). We describe how cohesion is measured, as distinct from centrality, and then test these hypothesis using network measures of cohesion, centrality, and other variables that may affect support for potential leaders. Decentralized Leadership and Network Cohesion 321 Measuring Network Cohesion Created by Marriage As social cohesion is one of the most subtle concepts in social science, at least in terms of measurement, its logical rendering is dependent upon insights into the fundamentals of network structures. The following introduction to the logic of cohesion is not only theoretical, nor merely of descriptive value in the present instance, but also provides the basis for the measurement of cohesion that can be used, in appropriate form, for any network. It thus provides the basis for testing hypotheses about both the antecedents and the consequences of cohesion, including its connection to leadership selection. The logic of embeddedness of actors in specific positions of the network structure is discussed here to demonstrate the basis for the network measurement of cohesion used subsequently in this chapter. It provides also a means of understanding how the existence of universal scaling parameters (k-connectivities) for cohesion can be used for measuring the extent to which networks may be constructed so as to be relatively invulnerable to structural ruptures through disconnection by removal of nodes, or to disruption of flows by the closure of paths. The understanding of these parameters of network cohesion gives us a potential for deep insight into the logics of cohesion that are in play in social groups. The Logical Construction of Social Cohesion The level of embeddedness of a person in a social network (White and Harary 2001; Moody and White 2003) is defined by considering the most cohesive group to which that individual belongs. This is identified from network data using as a criterion for level of cohesion the minimum number of people who must be removed from the group to disconnect any of its members. The highest level of cohesion attainable for a group is when every pair of its n members is connected in a clique, with n-1 as its level of cohesion: here even after removal of all but one member no pair can be disconnected. A group within a network is said to be k-cohesive (a cohesive group, at some level k) when any subset of fewer than k members can be removed without disconnecting the when no removal of fewer than k of its members disconnects its remaining members. The variable k is the scaling parameter for level of cohesion. The cohesive embeddedness of a member of a series of cohesive groups is the highest level of kcohesion of any of these groups. 322 Chapter 9 remaining members. One way to find a set of people who might belong to a common kcohesive group in a network is to prune away all those who have fewer than k direct connections to others, then do the same for the resulting group, and to repeat this pruning until every remaining member is directly linked to at least k others in the resultant group. This “outer limit” is in technical terms the k-core of a network, in which every person is connected to at least k others in the k-core. A k-core, however, is not necessarily k-cohesive because it may have segments that can be broken off by removal of fewer than k of its members. That second kind of pruning is the basis for algorithms, some recently developed (see Moody and White 2003 for an example), for identifying k-cohesive groups. One of the fundamental insights of graph theory (Harary 1969) is the theorem of connectivity level1 that shows that if a pair of nodes in a network can be disconnected by removal of k but no fewer intervening nodes, then there are precisely k completely distinct paths that connect them. Conversely, if there are only k completely distinct paths that connect the pair, then only k nodes are needed to disconnect them. The theorem generalizes to groups whose boundaries, robustness to disconnection, and internal redundancy or strength of connection are precisely defined by the pairwise parameter k, as follows. A k-component of a graph G is deThe pairwise cohesion or pairfined as a maximal set of nodes that wise connectivity k of two nodes cannot be internally disconnected by in a network is the minimum removal of fewer than k nodes in the number of nodes whose removal set. The theorem of connectivity level will disconnect them, and also the proves that a k-component will have number of completely distinct paths between them. at least k completely distinct paths be- The k-components of a network tween every pair of its nodes. It also are the largest possible sets that shows that a maximal subgraph S of G cannot be disconnected by removwith at least k completely distinct al of fewer than k nodes, and paths in S between every pair of its where all pair of members in each set have pairwise connectivity k. nodes is a k-component of G. The theorem thus shows the strict equiva- The k-components within a netlence of the dual aspects of the pa- work form cohesive blocks that rameter k: strength of ties through k may be hierarchically nested in node-independent paths and invulner- parallel series, from 1 to k, each like a Chinese box. ability to k-disconnection amount to exactly the same thing. The twin aspects of the way in which cohesion is constructed in so- Decentralized Leadership and Network Cohesion 323 cial networks, as shown by the theorem of connectivity level, parallel what people everywhere seem to understand and operate with intuitively when it comes to social groups, namely, recognition of the importance of the redundancies of paths that make social groups cohesive and of the fact that these redundancies create potentially greater levels of invulnerability to disruption if one or more members of the group were to be absent or withdraw. These two axes constitute a dual framework, one dealing with structure (effects of node removal on connectedness of the whole) and the other with traversal (properties of paths connecting parts of the network), that the theorem of connectivity level shows to be unitary for a single cohesion parameter k of connectivity and path redundancy. Reliance on Dependent Nodes for Cohesion in Kinship Networks That relinking marriages produce cohesion in terms of creating or increasing the number of multiple paths by which people are connected within a kinship network is one of the recurrent themes in this book, undergirding our network analyses. The theorem of connectivity level shows that when each is taken to its largest limit the following two sets of couples are identical: (1) those that are all connected by multiple paths in the network and (2) those relatives and affines that cannot be disconnected by removal of single nodes in the network. This unit of cohesion is a bicomponent of the kinship network, as discussed earlier. An important distinction must be made, however, to clarify how we measure the cohesion created by marriage in a kinship network. Note that in a nuclear family, the marriage between father and mother and the birth of their children creates the 1-connectedness of a unique configuration of relationships between each pair. Only if brother and sister married would relatedness in more than one way create an additional level of cohesion: in this case, the siblings would now also be spouses, the parents would also be in-laws to their children, and so forth. The way we define cohesion in terms of multiple Recall that a p-graph provides a paths or biconnectedness in the pmeans of studying cohesion in graph that we use to represent kinship terms of distinct multiple paths networks is consistent with the obserof connectedness in kinship networks that is consistent with the vation that it is only relinking marordinary concept of how people riages that create new multiple paths. acquire multiple kinship relations The maximal connectivity level in a pwith one another. graph, however, is only two. In a p-graph bicomponent, which has been made more cohesive by re- 324 Chapter 9 linking marriages, every member is related in more than one way. Figure 9.1 is a p-graph that illustrates the limits of cohesion in kinship networks: it has an order of generations, from top to bottom, as nobody is ever their own ancestor, and dashed lines represent female parent/child bonds, while solid lines represent male bonds. The upper four nodes are couples involved in a FBD marriage, which creates bicomponent cohesion among them (i.e., they are multiply related by virtue of this marriage). In the lower two marriages we have another FBD as well as a BDD marriage.2 The skewedness of this last marriage is not important, as other examples could be given without skewedness: what is important for our purposes is that each of these last two marriages increases the pairwise connectivity among each of the four top couples, while the bottom two nodes have only bicomponent cohesion. These two lower nodes, however, mediate the additional pairwise connectivity of the upper nodes. It is characteristic of kinship networks that ancestral nodes and descendant marriages mediate the extra pairwise connectivity beyond bicomponents. Figure 9.1: p-graph with pairwise connectivity beyond a bicomponent Exocohesive 3-block (4 nodes) FBD ( embedded within) Bicomponent of six nodes FBD BDD No matter how we array such configurations of cohesion-throughrelinking, at the group level of cohesion, biconnectivity forms a limit against which kinship networks must contend. In a genealogical network, there must always be in any p-graph bicomponent of that network at least one couple who lacks children (assuming that parentage cannot cycle back on itself) and (in order to be in the bicomponent) who has links to other parental couples in the group, which can be none other and no more that the parents of the husband and the parents of the wife. Thus it is impossible to construct a p-graph of genealogical links that is cohesive through marital relinking alone—the type of cohesion examined here— beyond the level of the bicomponent. Decentralized Leadership and Network Cohesion 325 Direct versus Mediated Forms of Cohesion To accommodate this insight on the limits on cohesion that stem from the intrinsic structure of marriage and parenthood in kinship networks, and the mediation of pairwise connectivity by ancestral nodes and descendant marriages, we describe one of the ways that these limits may be overcome in measuring cohesion: White and Harary (2001) give us the A k-block is a unit of means to define a k-block of a graph G as a potential cohesion that maximal subgraph of at least k+1 nodes that has as many members as has at least k node-independent paths in G be- possible, but at least tween every pair of its nodes. Thus, a k- k+1 members each pair of which is connected component of a graph G will be a k-block in G by at least k completely because a k-component is a maximal subgraph distinct paths. It is not of G that has within it at least k node- yet a k-component beindependent paths between every pair of its cause some paths may nodes and must have at least k+1 nodes. A k- go outside the unit. block, however, is not necessarily a kcomponent. Figure 9.2 illustrates, for a graph with six nodes and eight edges, an example of a k-block that is not a k-component, for k=3. The whole graph is a bicomponent (outer oval), with 2-cohesion (every pair of nodes has two or more completely distinct paths between them). Each pair of the four nodes in the inner oval, however, has three completely distinct paths between them. Some of these paths are with the nodes that lie outside the inner oval, so this 3-block of four nodes is not a 3component. It is cohesively embedded within a bicomponent, but has extra pairwise connectivity beyond that needed for a bicomponent. Figure 9.2: A k-block That Is Not a k-component for k=3 Exocohesive 3-block (4 nodes) (embedded within) Bicomponent of six nodes This is the kind of cohesion that we find, beyond bicomponents, in kinship networks. The graph in Figure 9.2 is a copy of that in Figure 9.1, 326 Chapter 9 but ignoring the difference between male and female edges and the order of generations. These are details that—at the most general level of analysis—make no difference in measuring cohesion. The k-block, defined above, takes us one additional step toward identifying the additional mediated cohesion in kinship networks that goes beyond the level of the bicomponent.3 That is, in addition to 2-connected blocks, there is additional cohesion that may occur pairwise between individuals. To reiterate, cohesive embeddedness in kinship networks may exceed the parametric level two only at level of specific pairs of individuals, or in larger sets of such pairs. Sets of pairs may also resemble in important ways a k-cohesive group or k-component. That is, there may be many completely distinct paths that link nodes A and B in a kinship network, and B and C, and even all the pairs in a large set of individuals, although these may never form an actual k-component where all these multiple paths connect nodes within the set. This observation leads us to define an exocohesive set of nodes in a network as follows: A k-block B of a graph G is exocohesive when there are not, within B, at least k nodeindependent paths between every pair of its nodes. The nodes in the inner oval of Figure 9.2, for example, constitute an exocohesive 3-block. Figure 9.2 is also an example of a k-block that is not a k-component. An exocohesive k-block is one in which not all pairs of members are All pairs of the four nodes within connected by at least k completely the circle have three node- distinct paths. It is not a k-component independent paths connecting one because some of its paths go outside another, but one of these paths the unit. We may say that such a kpasses outside the 3-block to a node block has extraconnectivity k. This that does not form part of the 3- concept is perfectly suited for the study of cohesion in kinship netblock itself because it has only two works. node-independent paths connecting to the others. A k-block defines a cohesive group in terms of a minimal number of links or node-independent paths between every pair of its nodes, and is of sufficient size to be a k-component, but an exocohesive k-block may draw upon independent paths and hence “dependent” nodes, like the lower node in Figure 9.1, that are not themselves sufficiently connected to be joined to the k-block, but which contribute to its cohesion. This concept is perfectly suited for the study of cohesion in kinship networks, which will show different structural levels of exocohesion. Thus the fundamental insight about cohesion in kinship networks, that grow out of parent-child connections, is that pairwise connectivity Decentralized Leadership and Network Cohesion 327 can grow beyond the cohesion of bicomponents to a level of k > 2, where k is the parameter of cohesion, only by additional connecting paths to what might be termed “dependent” members of a group, that is, members who do not share membership with the cohesive pair in a common kcohesive group or k-component. In Aydınlı nomad kinship networks, we look for blocks of k or more couples in which every pair of which has k or more completely distinct paths that connect them, but where some of those paths pass through “dependent” nodes that are outside the block, such as children, grandchildren, or ancestors. We use the term exocohesive groups for such blocks, within a kinship context, because they represent the only available type of higher-order pairwise that is available in kinship p-graphs as defined by genealogical links. That is, kinship networks may have 2components but not k-components where k > 2. It is very common in kinship networks, and indeed in networks of all sorts (Moody and White 2003), that cohesive and exocohesive blocks form embedded hierarchies. That is, there is typically a nice buildup in kinship networks of embedded k-connected pairs of higher order that can be well modeled by hierarchical clustering. If in a kinship network we compute a matrix of the number of node-independent paths between every couple, then the maximal submatrices that are all filled with values at some level k are the exocohesive blocks (White and Harary 2001, White and Newman 2001) where the node-independent paths may (and indeed in the case of k > 2 for p-graphs, must) pass outside the block.4 Exocohesive hierarchies are exactly what we find in Aydınlı nomad kinship networks, and hierarchical clustering of these matrices is an appropriate method. Couples who are deeply embedded in these hierarchies can be thought of as a set of nested exocohesive cores of the community, at different levels of intensity. The deeper the nesting of the block, the more the kinship links among block members involve distinct multiple connections between them, and the fewer people they have nested with them at that level.5 Hence, we come away from this discourse on method with a sense of the topology of social networks, and of kinship networks in particular, in which there are different elevations, similar to the topological contours of terrain, but here consisting of well-defined and measurable levels of social cohesion. The reader may well understand by now that it is precisely the development of a new set of tools for charting this terrain and for relating its social contours of cohesion to other aspects of social dynamics, such as emergent leadership and social change, that is an addi- 328 Chapter 9 tional motivation for undertaking this work, for it is now possible to measure such complex aspects of social topology, and to apply this topological mapping to the Aydınlı. Hence, the testing of the hypotheses of this chapter may merit the heaviness of this introduction, which bears the weight of defining a methodology for the network analysis of social cohesion in kinship. We turn next to the actual network components out of which these contours of cohesion are built. Analysis 13: Hierarchical Embedding of Cohesion The analysis of cohesion in networks needs a method for displaying and locating how exocohesive subsets are hierarchically embedded at different levels of cohesion. Hierarchical clustering analysis (HCA) is a method for analyzing square matrices of similarity coefficients, or measures such as numbers of node-independent paths, in order to show subsets for which all elements or nodes have a minimum number, or level of similarity. Such subsets may form one or more hierarchies at successively higher levels of similarities or cohesion. Hypothesis 9.2: The nomad clan is integrated not by mutually exclusive exocohesive clusters with some overlay but by a single hierarchical order of successively embedded exocohesive blocks. Concentric Rings of Decreasing Pairwise Connectivity When pairwise connectivity for the couples in the bicomponent of the kinship graph is analyzed by HCA,6 as shown in Figure 9.3, a hierarchical embedding of exocohesive groups becomes evident. Hierarchical containment is exactly what is expected from pairwise connectivity (White and Harary 2001), but what is evident is that there is a single such hierarchy and not two or more semi-independent hierarchies, which would be the case with multiple clusters of similar items. Decentralized Leadership and Network Cohesion 329 Figure 9.3: Hierarchical Clustering of Pairwise Connectivity Values VIII VI II I III IV V VII Support for this hypothesis is evident in the hierarchical clustering results in Figure 9.3. The single peak that rises out of the pyramidal structure to the left in the HCA diagram out of a wider base on the right shows small sets of actors with high pairwise connectivity values from the many node-independent paths between them—as with two ancestors with many intermarried or common descendants. As we move from these sets toward the right of the figure, with successively larger blocks labeled I-VIII in the column to the right, the general pattern is that these smaller sets of nodes are successively contained within larger sets of nodes with deceasing minimum pairwise connectivity. These results are consistent with the results of the previous analysis, namely, that exocohesion disperses and diminishes as total group size grows over time (counting ancestors) and greater cohesion adheres to ancestors farther back.7 These results are further examined in the following figures and tables that show the results of multivariate analyses. The exocohesive groups labeled by roman numerals I-VIII at the base of Figure 9.3, for example, are ordered by similarities in cohesion scores of individual nodes in a principal components analysis of the matrix.8 The groups with the lowest roman numerals (starting with I and II) have the highest correlations among them. Table 9.1 tabulates the number of couples for each of the groups I-VIII by the number of lines (nodal degree) that connect them to their parental couples and children. It shows that the exocohesiveness 330 Chapter 9 structure correlates almost perfectly (tau-b=.99) with nodal degree. This is an expected result because nodes with pairwise connectivity k with another node must have at least degree k. The upper groups (V and VII) are smaller than k+1 where k is their minimum degree, which is an indicator that they cannot be k-components, but only k-blocks. DEG * REGRP9 Crosstabulation Count Table 9.1: Crosstab of Nodal Degree of Couples by Cohesion Group VIII DEG Total 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 14.00 2.00 143 4 VI 3.00 8 II 4.00 REGRP9 III 6.00 2 I 6 5.00 5 33 3 IV 7.00 V VII 1 8.00 9.00 7 22 16 1 9 1 11 4 3 143 36 3 1 1 Symmetric Measures 1 1 Asymp. 22 17 10 11 a 7 b7 Value Std. Error Approx. T Kendall's tau-b .994 .002 21.547 3 Total 143 33 25 16 10 12 4 6 1 1 1 1 Approx. 253 Sig. .000 These results support Hypothesis 9.2 by showing a series of levels of Ordinal by Ordinal N of Valid Cases exocohesive subgroups within the clan nested 253 at successively higher leva. Not assuming the null hypothesis. els of connectivity. The more exocohesive subgroups are not differentithe asymptotic standard error assuming the null hypothesis. ated into opposingb. Using factions, but form a single core. It is along this exocohesiveness axis that we expect the central leadership (individual tanıdık kişi) to emerge. Leadership and Levels of Exocohesion Once we reached this point in our analysis of leadership and cohesion, we were in a position to reformulate our thinking about how the overall cohesion structure of the clan plays into emergent leadership in each generation. In considering leaders and the exocohesive groups that support them, it is common to think of differentiated exocohesive clusters around each leader, with support groups tending to be mutually exclusive. The structure of cohesion in the clan is radically different from this because it consists of a single hierarchy of exocohesively embedded groups. If our initial hypothesis 9.1.1 had been correct in affirming that level of cohesion predicts emergent leadership, then a hierarchical embedding of cohesion in the clan would place all the emergent leaders at the top of this hierarchy. Instead, they are distributed across different levels of the Decentralized Leadership and Network Cohesion 331 hierarchy, (if instead of looking at Table 9.1. we look ahead to a figure based on the same data, the location of leaders is as shown by the solid dots at the top of Figure 9.9). Nonetheless, our hypothesis was partially successful because the leaders all scaled at exocohesive levels 3 or more, and not simply at level 2. Having discovered the exocohesive hierarchy, we thought that, if leaders were selected in each generation from a family that had attained a considerable level of wealth but whose lineage had not been previously selected for leadership, then there might be the following association between leadership and exocohesive levels. Hypothesis 9.3: Positions of clan leadership are associated with differential exocohesiveness levels within the clan bicomponent. For this hypothesis to work, however, given our findings about wealth gained in recent generations in a lineage that has not yet led the clan as a predictor of leadership, there would also have to be some historical process that ties level of cohesion in each generation into the wealthgeneration process. This would add to our serendipitous discoveries about the entrepreneurial fathers and well-connected mothers of the emergent leaders and to the constituencies that they might represent. The Erosion of Exocohesion over Time A partial solution to the puzzle of how cohesion is linked to leadership became evident when we plotted Figure 9.4 to show how, for each exocohesive group associated with the tanıdık kişi for a given period, the level of that group’s cohesion changes over time. Cohesion might be expected to be lower in early generations because of memory loss concerning genealogies, but we can allow for that by correcting our estimates to give probable greater exocohesive levels for earlier generations than we extrapolate given missing data. This is shown by the dotted lines in the graph for the earliest two leaders from lineage #5. We then see a coherent pattern: The k-blocks in which the tanıdık kişi are embedded, extrapolating for early generations with missing data, seem to decrease in exocohesion with time. The finding of an erosion of exocohesion over time is theoretically significant for our understanding of kinship networks, and may be understood as follows. In a society with large sibling sets, the ancestral couples who spawned many offspring who stayed and relinked by marriage within the group (the clan in this case) will have more pairwise connectivity, in proportion to their offspring, than those who are relinking in the lowest generations. Further, if the size of sibling groups stay- 332 Chapter 9 ing within the group is shrinking due to emigration or a changing demography, then the most exocohesive groups in each generation will be ordered in time from those with highest pairwise connectivity in the earlier generations (if or when data are complete) and lower levels of exocohesiveness in the later generations. This is a possible explanation for the pattern we see for the Aydınlı nomads in Figure 9.4; namely, the loss of cohesion dating from the early twentieth century (when our data start to be complete) represents a real historical decline in cohesion along with a growth of inclusive group size, that is, as founders have more descents. There is also some growth in absolute group size as well. The numbers on the graph next to the lineage numbers of the leaders, such as #4:7 (14), refer to the sizes of the cohesion groups at the level in which the leader is embedded, in this case a cohesion level of 8 (as read from the vertical axis of the graph) with 7 couples at that level (and 14 at that level or above). While cohesion falls, these numbers rise. That part of the story follows from the definition of hierarchically embedded kblocs (or HCA structures of exocohesive pairwise connectivity), in which as levels of cohesion rise the cumulative size of all groups with that level of cohesion or higher necessarily shrinks. Level of Exocohesion Figure 9.4: Erosion of Cohesion Level of tanıdık kişi Groups over Time 10 (#5) 9 (#5) (inferences given memory loss) 8 #4:7 (14) 6 5 4 #5:11 (25) #1:11 (25) #5:17 (42) #2:22 (64) 3 1850 1875 1900 1925 1950 1975 #4:36 (100) Growth of Size as a Dilution of the Exocohesive Groups Decentralized Leadership and Network Cohesion 333 Increases in group size (inclusive of ancestors) partly explain the erosion of exocohesion that we observe over time. Table 9.2 shows the inverse correlation between the pairwise connectivity level of a group (exocohesiveness) and its size. Group sizes, including higher levels of exocohesion, are given in parentheses. Lineage numbers as in Figure 9.4 and generation and dates of leadership of the tanıdık kişi are also listed in the table. The entries in the table are ordered by the dates of leadership of each of the leaders. 334 Chapter 9 Table 9.2: Exocohesive Groups by Dates of Leaders, Size and Pairwise connectivity ExocoSize of Pairwise Lineage of Gen Date Cohesive connectanıdık kişi hesive Groups tivity Groups I 17(42) #5 (228) c ca.1850-75 >5* III 11(25) #5 (343) d ca.1875-99 >6* VII 7 2 No t.k.*** [20th C.***] 9+ V 7(14) #4 (517) d ca.1900-30 8 III 11(25) #1 (32) d ca.1930-57 6 IV 11(25) 1 (99)** d** 1957- ** 7 II 22(64) #2 (818) e 1957-82 4 VI 36(100) #4 (597) f 1982-94 3 VIII 143 None 2 * Cohesion probably higher because these are early generations with missing data ** Failed attempt *** No tanıdık kişi: foremost member Fındıklı Ali 784, S of 716 Mustan, and F of 818 Tanıdık Kişi and cohesion Table 9.3 shows, in support of Hypothesis 9.3, the correlation between exocohesive blocks at different cohesion levels and the existence of a tanıdık kişi leader. The correlation is nearly perfect. The one exception is cohesion group VII, with the highest degree of cohesion and whose leading member was Fındıklı Ali (784, lineage #2), son of clan founder Mustan (716) and father of Fındıklı Hacı (818), the tanıdık kişi from 1957-1982 This venerable and pious man was olne of the well known spiritual leaders of the clan and not unlike a tanıdık kişi, although group political decisions might not have been undertaken in his tent. Other than that, at each cohesion level there is one or more tanıdık kişi (or someone who attempted to be a tanıdık kişi) and each tanıdık kişi is in one of the groups that are distinguished by their cohesion levels. Table 9.3: Membership of tanıdık kişi Couples by Cohesion Groups Structured: Unstructured: p=.003 In one of exocoheIn noncohesive sive groups I-VII group VIII 6 groups, groups with a tanıdık 89 couples kişiler or aspirant \groups lacking a tanıdık 1 group w. religious 1 marginal group, Decentralized Leadership and Network Cohesion kişiler leader, 22 couples 335 143 couples The historical process that we think accounts for these well-structured results is this. The early generations are remarkably cohesive, as is known from the relinking marriages in the generation of Mustan’s children, including Fındıklı Hacı (the exceptional case to the correlation noted in Table 9.3). Out of this cohesive network-building by the founder generations of the clan comes a system of generalized exchange in which bride payments circulate in one direction in long paths or cycles among families while brides circulate in the other direction. The social embedding and potential for economic cooperation created by marital relinking facilitates entrepreneurial men to attain wealth, multiple wives, married sons and daughters-in-law, and to expand their tent size. The cooccurrence of these factors acts as an attractor for those in their kinship web to gravitate to their tent for political discussions. Given these factors, emergent leaders and their fathers will tend to be in the most cohesive segment of the clan in their generation. The same processes tend to occur in succeeding generations, except that the gravitation is toward a newly emergent wealthy family from a lineage that has not served in office previously. Their inclusion in the successive circles of leadership is at a lower level of cohesion in part because cohesion decreases on average over generations and in part because the former leading lineage, associated with higher cohesion, is now avoided as a meeting place for new leaders. This family brings in, and the new leader represents, a broader set of additional families that are linked overall at lower cohesive levels. This account is too complex to test directly by regression analysis, but it is consistent with the previous regression analyses and with our hierarchical analysis of cohesion as it varies with leadership through time. With this historical model in mind as to how kinship cohesion relates to emergent leadership for the clan, several types of further analysis of cohesion are merited before we shift to the personal and family attributes as a statistical explanation for the emergence of leaders from among the potential candidates (Analysis 16). One is to examine more closely the structure of cohesion itself in relation to avoidance behavior associated with prior leadership (Analysis 14). Another is to examine how groups of nodes at different exocohesive levels are distributed in the kinship networks of the clan (Analysis 15). In our final analysis (17) we pull all our results together into an overall perspective on emergent leadership. 336 Chapter 9 Analysis 14: Exclusion Principles: Cohesion versus Adhesion The exclusion principle of competition by the avoidance or withdrawal of social ties can operate in many ways, and it can have many different effects. When a person or group withdraws some of its ties within a clique this leaves those who are still attached in a more central position where they are now an intermediary to those for whom ties were withdrawn. Withdrawing a tie that provides an intermediate bridge for others outside the clique, on the other hand, may leave them more isolated and more distant from others in the network; hence, with lower centralities. Withdrawal of ties is thus a means of manipulating the centrality of one’s or another’s group. Here we explore whether there is evidence that Aydınlı nomad marriage strategies, in a context of cooperation in large zones of high curvature and competition with groups whose zones are remote, include the avoidance of marriages that give centrality to others. Avoidance of marriages that give centrality to others might have the effect of dispersing one’s relinking throughout the network, but in some circumstances can lead to segmentation in the network. There remains an important element that offers a different dimension for understanding how cohesion and support operate, one that is especially relevant in the decentralized political context of Aydınlı nomad marriage networks. The idea is that exocohesive groups and support networks may be both polycentric and dispersed, interpenetrating in ways that go unrecognized in conventional approaches to thinking about cohesion. What we learn from the decentralized case, in which this kind of structure is more evident and supported by ethnographic observations, may make it easier to also identify such structures in centralized social systems, once we are attuned to conceptualize and look for them in diverse settings. By way of background, the ethnographic data we have provided thus far makes it clear that Aydınlı nomads, like many tribes in the Middle East, are deeply egalitarian, a value that makes nomadic life attractive not only in terms of setting and lifeway but also in getting out from under the boot of central authority. The amassing of wealth by the tanıdık kişi from lineage #1 leading to his retirement to town with many of his kin as a newly wealthy landowner, for example, was much commented upon and often resented by other nomads. The unsuccessful bid for leadership by his cousin from #1 was also a frequent source of gossip. The Decentralized Leadership and Network Cohesion 337 nomads distrust the concentration of wealth and power. They do not state a preference for rotation of leadership, nor state this as a principle, nor do they do so voluntarily; yet, this is the pattern that recurs instead of succession within the same lineage or group, which would be hotly contested. This attitude toward concentration of wealth is also ambivalent in that as individuals and families, each also wants to succeed economically, and leaders emerge in part out of respect for their economic success.9 Distrust or ambivalence about concentration of wealth or influence might also be a common attitude toward those who might capitalize on more central positions of influence within the kinship network. Mustan, his brother, and their children, while they may not have been not rich or powerful, were clearly the major sources of network integration within the clan. Yet, no member of Mustan’s lineage Ecevitli A emerged as leader until the grandchild generation after Fındıklı Hacı’s father had achieved economic success 10 and neither did they retain leadership in the next generation in spite of their formidable combination of wealth and social influence because all his sons obtained a higher education and gave up nomadism. Some of the sons gained great influence but in another social context. Thus, from the ethnographic perspective, we may lay out the following regularities of competition and cooperation that apply to Aydınlı nomad behavior, ones that can be seen to generate by preference or outcome polycentric, dispersed, interpenetrating forms of cohesion: Everyone wants to reinforce alliances within their group because the production units are small residential groups, and lineages are subject to segmentation and hence weakly bonded, but also: Everyone wants to disperse their alliances broadly so as to have unanimous the support of the clan as a whole and all of its subgroups. Another way of putting this is that no one wants to make enemies (and no Aydınlı mentioned such a practice) by allowing schisms in the exocohesive structure of the clan, with the result, of course, being a network of generalized exchange. This leads to the following hypothesis: Hypothesis 9.4: Members of more central groups will tend to avoid the creation of alliances such that the centrality or dominance of others is enhanced. To thus conceive of a resultant social cohesion that is polycentric, dis- 338 Chapter 9 persed, and interpenetrating is to break out of the mental template that sees cohesive groups as based on boundaries between densely connected centers whose members are mutually exclusive sets. The common alternative, when density of ties is used to define such groups, is to conceive of groups that shade into another, in which some people are members of multiple groups. But if we shift to thinking of groups in terms of overlapping sets, as with the idea of overlapping cliques, the problem of overlap is one that is complex. Still, both sets of ideas about cohesion— mutually exclusive groups and overlapping groups—are intuitively and often formally based on the ideas of distance and density. What is lacking in these conceptions is the idea of cohesion at a distance, namely, that one group can reach into another, by multiple paths, even if the distances are considerable (see Friedkin 1993). This is the multiconnectivity or multiple-paths conception of cohesion (White and Harary 2001; Moody and White 2003) that we have employed throughout this book to identify the boundaries of exocohesive groups and measure the degree and structure of cohesion in various subgroups. This conception, based on the graph-theoretic concepts of connectivity, runs counter to a common preconception that social interaction has only proximal effects, and that indirect effects quickly decay as we move from direct effects (distance 1) to effects along paths of distance 2 or 3, beyond which indirect effects, in the social networks literature, are mostly thought to be minimal.11 There are, however, two different graph-theoretic conceptions of connectivity that we can put to good use in testing hypotheses. The one we have been working with up to this point has been that of nodeconnectivity, which has provided our measures of social cohesion. We now introduce a second notion that is based on a slightly different measure, that of edge-connectivity, which White and Harary (2001) call the measure of social adhesion. Adhesion versus Cohesion (Maxflow versus Node connectivity) The notion of social adhesion is that when many paths converge on a central node, such as an urban center or junction in a transport network, the paths that pass through this node still provide independent routes of traversal, ones that are not node-independent, but edge-independent. Thus, if all transport routes between Los Angeles and Tijuana passed through a certain checkpoint at the border in San Diego, one could measure the flow capacity of transport from Los Angeles-Tijuana as the sum of capacities on all these different routes, but the measure would Decentralized Leadership and Network Cohesion 339 fail to take into account the potential bottleneck effects of the mediating node. The commonly used measure of flow capacity in networks is, in fact, the Ford-Fulkerson algorithmic computation of the sum of capacities on edge-independent paths (in UCINET and Pajek this is called the maxflow computation). Edge-independent paths differ from nodeindependent paths in the following way. Intermediate nodes on nodeindependent paths between u and v must all be distinct. The number k of node-independent paths between nodes u and v is a measure of cohesion because u and v cannot be separated without removing k of the nodes that connect them. Edge-independent paths between nodes u and v in a graph, however, may pass through a common node intermediate to u and v, so long as all the edges of the paths are The cohesion of two nodes in a distinct. Hence, the number k of edgenetwork is the lowest number of independent paths between nodes u nodes needed to disconnect and v is not a measure of cohesion. It them and equals the number of is referred to, instead, as a measure of node-independent paths between adhesion. High levels of adhesion bethem. Their adhesion is the lowest number of edges needed tween two nodes in a network, like the to disconnect them and equals Tijuana bottleneck problem, are no the number of edge-independent guarantee that there will not be a sinpaths between them. gle node whose removal from or blockage in the network would disconnect them. As regards computation, the UCINET program currently computes matrices for node- or point-connectivity (cohesion) as well as edge- or maxflow connectivity (adhesion). Pajek at present computes only maximum flow or edge-connectivity matrices.12 Zachary (1975) made use of the Ford-Fulkerson algorithm to compute maximum edge-independent flows (maxflow) between two rival leaders of a karate club during a period when the club was segmenting into two. The edges were those of friendship, weighted by the number of contexts in which pairs of individuals hung out together. The assumption was that each member of the club, during and after the potential split, would tend to adhere to the leader with whom they had more edge-independent path flow capacity. Maxflow in this context is an indicator of capacity for potential communication flow such as information, sentiment, and directives that in this case might influence decisions to support one or the other leader on the basis either of direct edges or indirect paths of 340 Chapter 9 friendship links. On this basis, he was able to accurately predict the factional split in terms of who went with which of the leaders. White and Harary (2001) showed that, using node-independent paths as a measure of cohesion within Zachary’s karate network, there is an equally good if not better prediction of the same outcome (factional division). It is a better prediction considering parsimony because it does not rely on weighted edges. Beyond that, however, it is difficult to evaluate which concept gives the better prediction in this case, however, because the measures were so highly correlated that they did not give different predictions. These two measures do not always give the same results because they measure very different structural properties. There are many studies that show excellent predictions of diverse outcomes from the measures of cohesion that we use, so on that count there is greater support for White and Harary’s hypothesis but few if any other studies that replicate Zachary’s result of predicting similar outcomes from adhesion rather than from cohesion when the two measures give different results. The fact that the measure of adhesion versus cohesion differs in The k-edge-components of a netprinciple, if not always in specific work are the largest possible sets each associated with a value of k, networks, however, gives us the where all members of each set have potential for testing Hypothesis 9.4 at least k completely edge-independabout strategies of network- ent paths to every other node in the building, which predicts that the set, and by the connectivity theorem Aydınlı nomads avoid, in their the set cannot be disconnected by marital relinking behavior, the cre- removal of fewer than k edges. The least or minimum edge-cut of a ation of alliances such that the cen- graph is the fewest number of edges trality or dominance of others is whose removal is needed to disconenhanced. We need the graph theo- nect it. retic definition shown to the right. Table 9.4 explains the basis for this test of our Hypothesis 9.4. The table simplifies the contrast between high versus low levels of cohesive (k-connectivity) and adhesive (k-edge-connectivity) connectivity, independently measured, within any given k-edge-component of an actual network. One cell of the table is empty because it is impossible for the node-connectivity of a k-edge-component to be greater than k, although it can be less than k. When the latter occurs—as in the cell of the table labeled by high adhesion and low cohesion—then there must be highadhesion members of two or more high-cohesion groups. Decentralized Leadership and Network Cohesion 341 Table 9.4: Adhesion versus Cohesion Adhesion: EdgeLow Connectivity High Cohesion: Node-Connectivity (Multiconnectivity) Low High Little or no cohesion [empty cell] Adhesive Mediators between Cohesive Groups Cohesive groups with no mediators Marriage Behavior That Avoids Enhancement of the Centrality of Others Aydınlı nomad marriage networks show no difference between cohesion based on node connectivity and adhesion based on edge connectivity, that is, their exocohesive groups have no high-betweenness mediators.13 The only way that this can be true is if discrepant situations of high adhesion and low cohesion are avoided. This result, then, is consistent with Hypothesis 9.4, namely, that nomad members of central or more exocohesive groups avoid alliances that make other groups more central. When one person or lineages starts to become excessively central as a mediator between other groups, those groups begin to ally among themselves and reduce the competing lineage’s centrality.14 Analysis 15: Distributed Cohesion in Kinship Networks Automated drawings that minimize line length have the capability of showing cohesive subgroups in networks, and the potential for showing how other phenomena, such as the positions of leaders, are structurally located with respect to how cohesion is distributed in a network. Figure 9.5 is an automatic drawing that shows the kinship links among couples in exocohesive groups I-VI, with nodes as couples: Solid arrows point to husband’s parents, and dotted arrows point to wife’s parents. The visual message of the graph, which shows up more clearly in the web version where groups are differently colored, is that nodes of the same group are clustered, but each exocohesive group also has members that are widely dispersed. The large shaded nodes indicate leaders, and they are often but not always located near the center of their cluster. This is the typical pattern of dispersed block-cohesion. In Figure 9.5 there are V=253 vertices or nodes, A=431 arcs, and an 342 Chapter 9 index of relinking of .74, which is a very high level of cohesion. The nodes are colored by group, and the larger nodes indicate tanıdık kişi group leaders. It is evident that while the cohesion groups overlap, leaders seem to come from opposing sides of this graph, which would account for the small second eigenvector (principal component) in the factor analysis of pairwise connectivity (Analysis 14). Note that there are two leaders, 2 and 4, at extracohesive level 6, group III, and they belong to lineages #5 and #1, respectively. That also explains why there are seven large superimposed nodes, coded by leader, in the graph rather than six. Figure 9.5: The Six Leadership and Exocohesive Kin Groups— Atemporal Cohesion (V=253, A=431, index of relinking=.74) Group I Cohesion 5 Leader 1 (# 5) III 6 Leaders 2 (# 5) 4 (# 1) V 8 Leader 3 (# 4) IV 7 Challenger 5 (# 1) II 4 Leader 5 (# 2) VI 3 Leader 6 (# 4) The color-coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads Hypothesis 9.5: Tanıdık kişi group leaders are differentiated by relatively long genealogical distances among themselves, without diminishing their overall cohesion in terms of multiple connections, which may vary in their distances. The feature shown by Figure 9.5, and others that we will see of this type, is the wide distribution of nodes of the same cohesion (color) group. Decentralized Leadership and Network Cohesion 343 What this signifies is that each tanıdık kişi factional leader builds many independent paths that connect to other couples who are distant in the kinship network, thereby achieving high exocohesiveness overall with other significant groupings of the clan. Figure 9.6 shows the same network as the previous drawing, with the addition of a third (vertical) dimension for generational time. There is an oscillation every two generations between a left side of factional leadership and a right side that is slightly more central in the kinship network. What we think this shows is that the tanıdık kişi, as factional leaders of different allied groups of lineages within the overall exocohesive structure of the clan, occupy positions of latent opposition within the kinship networks, linking through multiple connections to the supporters of others leaders (a structure of diffuse overlapping cohesion for each leader), but avoiding closer links with the core segments of opposing factions. Figure 9.6: The Six Leadership and Exocohesive Kin Groups— Group Cohesion Temporal Perspective I 5 Leader 1 (# 5) III 6 Leaders 2 (# 5) 4 (# 1) V 8 Leader 3 (# 4) IV 7 Challenger 5 (# 1) II 4 Leader 5 (# 2) VI 3 Leader 6 (# 4) The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads In Figure 9.7 the peripheral nodes (degree 2: ones that do not correlate with the exocohesiveness structure) of Figure 9.5 are deleted, and the figure is then rescaled. What this graph shows is that the high-cohesion leadership factions are themselves directly connected and cohesive among and across themselves. The son in the father-son pair of early leaders from lineage #5 is in the center; leadership then switches to more 344 Chapter 9 peripheral nodes for two generations, and then returns to the center for the last two generations, with a failed challenge from the nephew of the fourth leader (a peripheral) to the emergent leader in the fifth. Figure 9.7: The Six Leadership and Exocohesive Kin Groups Minus Peripherals—Atemporal Cohesion (A=110, V=161, relinking index =.55) Group I Cohesion 5 Leader 1 (# 5) III 6 Leaders 2 (# 5) 4 (# 1) V 8 Leader 3 (# 4) IV 7 Challenger 5 (# 1) II 4 Leader 5 (# 2) VI 3 Leader 6 (# 4) The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads In Figure 9.7, compared with Figure 9.5, it is more clear which is the more central cluster of groups, and that the outlying factions tend to have somewhat higher levels of cohesion. Perhaps the reason for this is that outlying groups need greater cohesion than do more central ones in order to gain leadership. Figure 9.8 shows the same data as in Figure 9.7, adding the generational time dimension. Again we see a temporal oscillation between a left side factional leadership and a right side that is slightly more central in the network. Decentralized Leadership and Network Cohesion 345 Figure 9.8: The Six Leadership and Exocohesive Kin Groups Minus Peripherals—Temporal Perspective (A=110, V=161, relinking index Group Cohesion =.55) I 5 Leader 1 (# 5) III 6 Leaders 2 (# 5) 4 (# 1) V 8 Leader 3 (# 4) IV 7 Challenger 5 (# 1) II 4 Leader 5 (# 2) VI 3 Leader 6 (# 4) The color-coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads In Figure 9.8 there are V=110 vertices or nodes, A=161 arcs, and an index of relinking of .55, which is a high level of cohesion, but not as high an index of relinking as in Figure 9.5 where the peripheral couples are present. The hierarchical clustering of maximum flow values for 110 couples in structured groups (minus peripherals) identified by factor analysis is almost identical to that of the bicomponent, except for a few irregularities. Removing the more marginal nodes of degree 2 makes the exocohesiveness structure more irregular, but it is clear that these marginal nodes add to and balance out the overall clan cohesiveness rather than adding to any one particular group. As before, the level of pairwise connectivity goes up to 13, which is the maximum number of edge-independent paths between any two pairs of couples. Hypothesis 9.6: Nomad cohesion is not organized in tightly clustered factions, each having short-distance connections, but in terms of distributed connectivities. We test this hypothesis by a factor analysis of the matrix of geodesic or shortest distances between nodes in the bicomponent of the network. Large factor loadings of the first few factors, provided they are greater than one, which indicates a lack of clustering for each successive com- 346 Chapter 9 ponent, indicate clustering. As shown by the results from UCINET15 in Table 9.5, the results show a lack of clustering in the distances. This is very different from the factor structure of pairwise connectivity, which shows a very high degree of clustering. Hypothesis 9.6, then, would explain the results in Table 9.5 in relation to our results on cohesion. Table 9.5: Factor Analysis of Geodesic Distances Factor Value Percent Cum% Ratio ------- -------- ------ ------ ------1: 65.468 26.9 26.9 1.493 2: 43.853 18.0 45.0 1.560 3: 28.108 11.6 56.6 1.373 4: 20.468 8.4 65.0 1.389 5: 14.739 6.1 71.0 1.095 6: 13.462 5.5 76.6 1.307 7: 10.304 4.2 80.8 1.329 8: 7.754 3.2 84.0 1.254 9: 6.182 2.5 86.6 1.349 10: 4.582 1.9 88.4 1.110 11: 4.129 1.7 90.1 1.490 To summarize the findings of this chapter in the form of an hypothesis to be investigated further in other chapters: Hypothesis 9.7: The rotation of leadership within the clan, while it must occur among men with the attributes and qualifications for emergent leadership, also depends on networks of kinship support in terms of distributed cohesion throughout the clan. Analysis 16: Network and Attribute Predictors of Leadership Network methods facilitate testing broader hypotheses than are usual in social sciences because structural properties as well as attributes can be used as independent variables. Six different attribute and network hypotheses about leadership have been hinted at thus far in our discussion of Aydınlı nomad politics. Any one or many of these might account for the relation between the leadership position of males at various historical periods and their position in the social network of genealogical ties relative to other factors, including the ties through marriage across the geographical and cultural landscape, and the movement of Aydınlı themselves or people to whom they are linked. We can fairly quickly eliminate the hypothesis that wife-giving ver- Decentralized Leadership and Network Cohesion 347 sus wife-taking confers changes in family or lineage status for the relatively egalitarian but competitive-status society of the nomad clan, in which wife-givers and wife-takers are considered as equals. Barring that, the following six hypotheses were formulated prior to developing quantitative variables that could be used for testing different models of leadership. Hypothesis 9.8.1: Status may just vary by numbers of supporters from one’s lineage. (Predictor: Size of husband’s or wife’s lineage.)16 Hypothesis 9.8.2: The broader the overall span of their links to others in the clan, the greater their support from others and the higher the support for their rank. (Predictor: Network closeness centrality.) Hypothesis 9.8.3: The more prominent males occupy positions of maximal betweenness (betweenness centrality) with respect to different segments of the clan, that is, their genealogical links put them in the best positions to mediate between various factions and thereby to gain support through conflict resolution. (Predictor: Network betweenness centrality.)17 Hypothesis 9.8.4: As between patriarchs of two contending groups, vying for leadership, the boundary in the network where removal of the minimum number of nodes would cleave into two opposing factions would predict the division into two clusters of supporters. (Predictor: Exocohesive embedding. This is a restatement of Hypothesis 9.7, but now stated in terms of pairwise connectivity (which applies to k-blocks). Hypothesis 9.8.5: The variables involved in the ranking of prominent men (and affecting the status of their lineages) involve not only the size of their group, the alliances of their family and lineage, and influence exerted through the kinship network, but also qualitative variables such as personal influence, character, and charisma. With many competing hypotheses it is best to use a multivariate model for testing hypotheses. Using linear regression, the first five variables predict only about 9% of the variance in leadership status (r=.31, s.e.=15), not such a poor result because there are only six leaders to predict. The hypotheses supported are better indicated by their tests of significance: size for husband’s lineage size times wife’s lineage size 348 Chapter 9 (p=.20: weak, but stronger than any other direct size effect); betweenness (p=.001), and exocohesive embedding (p=.002). Closeness and eigen centrality had no effect independent of these factors (p > .5), and the results for maximal pairwise connectivity are negative for outliers (negative effect for this factor squared, p=.003) but positive controlling for each of the previous variables (p=.019, taking as a measure the excess of pairwise connectivity above the exocohesive or group level cohesion). The group-level variables of size and cohesion work best when size is multiplied by cohesion and this interactive term predicts leadership (p=.076) more strongly than size alone, while cohesion and other variables continue to have the same effects (now the variance is 10% accounted for, with r=.315). The variance accounted for with this model is brought up to 22% if other attributes are added (see below): older sons, having 3-5 married sons, both husband and wife born in and stay in the clan and come from one of the four largest lineages, and father a tanıdık kişi. In the expanded model, betweenness centrality and exocohesive blocks continue to predict, and the square of pairwise connectivity to negatively predict, emergent leadership. Figure 9.9 shows how the variables used in testing these hypotheses work in tandem. In this log-log plot, the horizontal axis is an ordering of the 256 marriages in the bicomponent of the kinship network, which we used for testing hypotheses, on the basis of their exocohesive embedding and secondarily their betweenness centralities. These are the strongest predictors at the level of the group and the individual couple, respectively. The vertical axis are logged multiples of the network variables, shown in the lower part of the graph (scale is not important here), while the dependent variable of political leadership (labeled “Tanidik and wife”) is given for the six marriages shown as round dots at the top of the plot. They are positioned to show their order within the rank order of marriages by k-block exocohesiveness. All the leaders have a minimal exocohesiveness between the 5th and 55th percentile within the bicomponent, statistically significant at p=.0001, but other factors are also needed (as in the regression analysis) to predict their emergence as leaders because there are many couples with higher cohesion. As we have seen in Analyses 12 and Figure 9.4, however, a strong direct correlation between leadership and cohesion in the bicomponent for all time periods would be impossible because cohesion in that context decays across generations. The three asterisks at the top left of Figure 9.9 are the top three nodes on the scale of group cohesion. These include Mustan (the middle of the Decentralized Leadership and Network Cohesion 349 leftmost starred nodes at the top), whose high betweenness centrality is seen by the upturn in the profiles of betweenness centralities shown below, and two of his sons, one with high and the other with low centrality. These men were prominent in the socioemotional leadership of the clan: Mustan the great founder-relinker, one son (868) being the father of the clan healer and the other (Fındıklı Ali, 784) the father of the political leader and also a spiritual leader. From a lineage that may have been poor at that early time period, they themselves did not become political leaders, but their cohesion as a form of social capital was converted in the grandchild generation into a leadership position. Figure 9.9: Logged Plot of Variables Used in Predicting Leadership Percentiles 1% 5% 30% 50% 100 Betweenness Cohesion 10 Closeness MarrChildren Leader&Wi Tanıdık kişi Tandik&Wi and Wife 1 LEADER 1 10 100 1000 TANDIKI Figure 9.9 is drawn to show the fractal or power-law (linear in log-log) Power relationship between levels of cohesion and betweenness and their per(Cohesion) centage distributions in the population. The parallel Power heavy regression lines show the power-law slopes of cohesion (k-block membership, (MarrChildren) Power and numwhich begins at 2 and rises to 14 in the bicomponent sample) ber of married children (out degree of nodes as parents,(Closeness) a measure of social support for leadership from having multiple married children). A parallel lighter regression line shows the same power-law slope for betweenness centralities, which have considerable independent variability even while correlated with cohesion. Closeness has a nearly flat slope and is basically uncorrelated with cohesion. The fractal distributions of 350 Chapter 9 these variables are a likely indication that when we take samples of varying size from the population, within certain limits, we will observe the same form of such distributions. Pairwise connectivity and the degree (number of kinship links) of nodes in the kinship and marriage network is associated with tanıdık kişi leadership in different factions, as we have seen. The degree of nodes in a network is one measure of node centrality. More detailed modeling of cohesion as a predictor of leadership is needed within the six different time frames of the different leaders, but focusing on one leader at a time, for only six leaders, makes it difficult to assess statistical hypotheses when the dependent variable has only one case that varies in each time period. Reframing the time periods would provide an outcome variable once again of six leaders to be predicted, as we have done here, but perhaps that game is not worth the candle, especially because there is an added dimension of how leadership itself, its structure and dynamics, changes over time. In Chapter 5, with the curvature analysis, we were able to break the structure of lineages into temporally coherent fractal (and overlapping) pieces to get a larger set of units to work with to study the structure of connectedness and the dynamics of change, but for our six political leaders, a comparable methodology has not been invented. To go into more detail, rather than testing hypotheses quantitatively, what will have to suffice is a qualitative depiction of the position of leaders in the networks of each time period, and how leadership seems to be changing at various time periods as indicated by differences in leadership histories and positions in relation to the temporally specific networks. The following hypothesis was developed after the others: it is intuitively obvious although it does not require quantitative testing, so obvious that we did not think to develop data to document its relevance until after the previous section had been written. Hypothesis 9.9: In most periods of succession, a son of one of the wealthiest families in that period that has not already held leadership tends to emerge as leader. Because of the father’s wealth, that man can acquire several wives even before taking leadership, and the wealth, wives, and married sons (plus well-connected daughters-in-law) already presuppose that this man’s family will have acquired the large type of tent that is a sign of leadership status and of a size in which others can gather for discussions. The following narratives about fathers and sons (in bold: the emergent leaders) show this hypothesis to be accurate: Decentralized Leadership and Network Cohesion 351 #5. 1850. 224/228. The Koca bey (224) lineage (#5) was originally the highest in social ranking. Before the eastward migration, headman Veli Kahya (228c) had three sons by his wife from Antalya, but was rich enough, as the most influential man of the clan in his generation, to adopt into his lineage a smart but poor boy, younger than his own sons, to help care for his flocks. We do not know much about his father, Koca bey (224b), but the fact that he is remembered (along with his brother) speaks to his prominence and possible wealth. This does not contradict the hypothesis in spite of missing data. #5. 1875. 228/343. Veli Kahya’s (228) son Hasan bey (343d) became the next headman. His lineage segment was among the first to leave the Antalya region for the eastern pastures, in the 1870s. He left in order to spare his most gifted son Hafız Ali (345e) from a lengthy military service. He fits the pattern of having a wealthy father who, because he did not migrate to the east, had not been headman of that lineage segment but of the pre-migration group. #4. 1900. 514/517. The Kırbaşı oğulları have been fairly rich because the time of Ali (514b) of the c-generation, whose wife was daughter of an efendi (formerly gentleman or man of some influence in town, now simply Mr.). His son Erkek Mustafa (517c) was a tanıdık kişi and wealthy. This fits the hypothesis. #1. 1930. 31/32. Hacı Dolaşıklı (28b) himself and his brother and their sons began poorly, but the sons were successful at last and owned herds of at least medium size (about 300 animals). The eldest son of Hacı Dolaşıklı (28b), Hacı Mehmet (31c), who already owned large herds, and whose wife was heir to Ecevit Mehmet (whose family had no male heirs), had a very rich son, Kozan Mehmet (32d), who was tanıdık kişi for a certain time. He married three wives, which only the very richest can afford. He was a breeder of small cattle, and also cattle merchant and did other business. He became sedentary in a small town thereafter, which created distance between him and the clan. (As Eberhard has shown, nomads who become wealthy absentee landlords settle in or near towns, while poor nomads settle near villages.) This pattern of wealthy fathers also fits the hypothesis. #2. 1957. 784/818. The Ecevitli were poor until the c-generation. Only thereafter did they advance in richness and influence, as we have already written. Among the offspring of Mustan (716c, son of 659b) and his brother Hacı Ketir Mehmet (661) of Ecevitli lineages #2A-B, the most influential and rich in the d-generation was Fındıklı Ali 352 Chapter 9 (784d). His wife was daughter of a Molla (trained in a religious school) who would be above the average in wealth. Their position had influence on their son, Fındıklı Hacı (818e), as tanıdık kişi. Fındıklı Hacı’s herd was a good, large one, estimated in 1964 at 500 goats or more, but it included the animals of his nephews and obviously also two of his brothers. This fits the hypothesis. #4. 1982. 584/597. The eldest of the sons of Erkek Mustafa (above, #4), who reached old age, was Molla Ali (532), who was also the richest. He owned about 700 goats all together in 1957, but lost many because in this summer an epidemic of foot-and-mouth disease happened. He had two very able sons who managed to stay rich, nevertheless. The next son after Molla Ali, Isa (584d), was at the beginning of his independence a rich man too, and had married a woman from a well-to-do family of the Dazkırlı (#6). They had an elder son who was honest and of fine character, but not very intelligent, and his second son died after Isa had spent much money to save him and even more money to get the daughter-in-law, Ayşa (359), for his eldest son, who became lovesick for her and had to be cured too. Moreover, Isa had had some losses in the years before, especially from a failed second marriage that cost him dearly. Thus, he owned only about 140 goats, which fell to below 100 during the epidemic of 1957. This was a critical situation for the family, but then the youngest son Mustafa (597), called “Dede,” and later tanıdık kişi, had grown up sufficiently following his military services to take the reins of the family economy, and had also learned reading, arithmetic, and ways to make contacts in towns from Johansen, his lineage “sister” for whom he was also responsible. Within fifteen years he managed to make it the richest in the clan. His brother continued goat-breeding with his advice and he himself did many sorts of business very effectively and extremely diligently. He was twice married. This fits the hypothesis, with some ups and downs in the family fortunes. The pattern of a tanıdık kişi son succeeding a father who had recently become the richest man in the clan in a lineage other than the preceding tanıdık kişi is recurrent in these narratives. A secondary theme, one that creates conflict, is the consolidation of wealth by the son as tanıdık kişi, which allows him and his relatives to return to village life as owners of significant fixed property. This was the case with Kozan Mehmet from lineage #1. This finding, which we had not expected to emerge so clearly, sug- Decentralized Leadership and Network Cohesion 353 gests that our quantitative testing might well be redirected to the question: what are the network and other factors that dispose toward the emergence of an entrepreneurial father whose sons have the best chance of becoming tanıdık kişi? Having constructed all the necessary variables for use in testing hypotheses 8.4.1 through 8.4.6, we used multiple regression once more to try to predict the six fathers of the tanıdık kişi. Hypothesis 9.10: The fathers of tanıdık kişi can be predicted from the same type of network and other variables specified in hypotheses 9.7.1 through 9.7.5, as applied to the fathers rather than the son as a candidate for tanıdık kişi. Entrepreneurship and Parents’ Status as Predictors of Emergent Leadership We stumbled onto the investigation of leader’s parents by serendipity, and as a purely exploratory model simply used the same variables that predicted emergence of tanıdık kişi leaders, this time trying to predict, from the network characteristics and attributes of the parents, whether they became the parents of a leader. What we found, although there was more missing data, was that 11.3% of the variance was predicted from three variables: embedding in exocohesive blocks (p<.001), excess pairwise connectivity (p<.001), and less than one link to parents (p<.001). The last predictor in this list reflects something about the mothers of the tanıdık kişi, two of whom came from the Antalya homeland (the earliest cases from lineage #5), or from a place unknown (three cases). In these latter cases, judging from the mother’s father, the women were from prestigious families.18 What these results seem to imply is that Aydınlı nomad entrepreneurship and emergent leaderships is roughly equivalent to a very common pattern worldwide, namely, immigrant entrepreneurship. This often included high status from the sending community, the establishment of multiconnectivity in the new locale, often by joint or network migration (moving with others with already established ties), and the endowment of children with the resources from a “good family” on the mother’s side and from an entrepreneurial father’s achievements to allow them to attain political prominence. The features of Aydınlı nomad politics that are unusual are that the leader is not selected by an electoral or formal procedure but an informal one, and that leadership at the lineage level is based on a norm of turn- 354 Chapter 9 taking. The turn-taking is not automatic, but often disputed, with competing contenders. The following narrative illustrates how conflict and cohesion enters into this process. The case of a failed contender is also illustrative. After the time of Kozan-Mehmet’s (32, #1) leadership and his abdication to town, his younger patrilineal cousin Hacı Molla (99, #1) tried to become the tanıdık kişi after him. Nevertheless, Fındıklı Hacı (818) as the second and most able son of the pious Fındıklı Ali (784, the second son of Mustan) was accepted as tanıdık kişi, not Hacı Molla. The interesting structural factor in this completion was that the two were almost exactly matched on all of our network and attribute variables, except that the lineage of the winner was different from that of the previous incumbent (and somewhat larger than Hacı Molla’s, although when it came to the lineages of the wives, the latter’s was the larger). Perhaps this structural similarity was the reason why Hacı Molla continued to mutter, until the age of about eighty, against the decisions of Fındıklı Hacı and to complain of Fındıklı Hacı’s egoism. It is also noteworthy that this failed contender was the only hereditary candidate (as cousin of the former leader) pushing himself for tanıdık kişi status in recent times, which was perhaps a factor in his rejection, although reasons given had to do with his actions and poor judgments. His kinship and wealth qualifications were impressive, however: Not only was his cousin (32, #1) a tanıdık kişi, but his father (31) had also become rich at the same time as his cousin’s father, and his mother a FBD of his father and daughter of the lineagename founder, Ecevit (1, #1). Leadership, Marriage, and Social Change A final theme on which to close this chapter on the issue of leadership in the complex but decentralized society of the clan is the importance of women and marriage to political leadership. Although more than one marriage is not legally permitted in Turkey, Islam allows up to four wives at a time. To have more than one wife was looked at as a sign of power and the most influential men were often married to two, and occasionally to three wives, which required considerable wealth. As Table 9.6 shows, it is not just the number of wives that correlate with tanıdık kişi leadership rank but wives plus daughters-in-law. The minimum number of wives and daughters in law for leaders was four, excepting the present leader, “Dede,” whose sons had not yet married. In the next Decentralized Leadership and Network Cohesion 355 chapter we consider marriage as a source of cohesion in the network of kin-based support that is important in a decentralized political system. It is not only the wives and daughters-in-law of the leader who are important,19 but the marriages that make his kinship block cohesive that may be considered important. Other patterns that link leadership to marriage alliance are evident in Table 9.6. One set of patterns are those of repetition of alliances from one leader to the next: Each turn at leadership renews some of previous marriage alliances of the leader. In each turn except the last, there is a marriage within the lineage. There is also a discontinuity in the sequence, labeled as phases one and two in Table 9.6: The succession of #4 from #5 is accompanied by a marriage alliance with #5 (daughter-in-law), but this does not repeat. The leaders in Phase One (#4 and most likely, #5, for which data are missing) have alliances with other tribes, but not with villages. The pattern changes in Phase Two to marriage alliance with a village for the more recent leaders, starting with #1, and continuing through #2 and the last successor, #4. Another pattern is that lineage #4 repeats a leadership role, once in Phase One and again in Two, after a two-generation lag. Overall, there is also a pattern of the temporal succession of lineages occupying the leadership position mimicking the continuum from traditional to more village-oriented lineages that we saw in Figure 7.4, a continuum that resulted from scaling the similarities in marital ties among the lineages. The alliance structure among leaders, that is, mimics the alliance patterns among their lineages, and the traditional-to-village oriented scaling is mimicked by a secular trend toward more village-oriented leaders. 356 Chapter 9 Table 9.6 Leaders and their Wives and Daughters-in Law Lineage Leaders Wives’ lineages Da-in-law’s lineages Total Phase One: Traditional Pattern of Marital Alliances #5 228 Veli Kahya two: ? ? five: ? ? seven #5 343 Hasan bey one: #2 two: #3, ? four #4 517 Erkek Mustafa one: #4 four: #2,#5,#6,A,K*five Phase Two: Village (V)-Oriented Pattern of Marital Alliances #1 32 Kozan Mahmut three: #2 #5 V five: #1,#5,V #2 818 Fındıklı Hacı three: #2 #5 V two: #2,V #4 597 “Dede” two: #5 V zero (six sons)** * A,K are tribes eight five two ** None married as yet Analysis 17: Overall Cohesion, Lineage, and Leaders Network analysis makes it possible to formulate hypotheses about alliance patterns that do not restrict themselves to characterizing the social norms of entire societies but to analyzing variations on social structure that occur within societies or across various types of social boundaries. The structure of cohesion that we see in the graphic figures over all time periods, as analyzed in this chapter, is the cumulative result of relinking processes over successive time periods. Overall structure is the historical residue of past strategic action. Thus, we can return to our consideration of lineages, alliances, factions, and leadership (analyzed in Tables 9.2, 9.6, and 9.7). We want to understand how the overall pattern of cohesion relates to the strategies of lineages and patterns of social coalitions used to provide social support to those of their leaders who became known men (tanıdık kişiler) in different historical periods. Hypothesis 9.11: In temporal sequence, the lineage with the greatest distributive cohesion within the clan is the one in which a leader emerges (the causality here is one of recursion, or mutual influence, rather than unidirectional; it should be noted that “Dede” would be an exception for reasons to be explained). During the tenure of leadership, competitors emerge and the next lineage to emerge with the greatest distributive cohesion within the clan is the one that takes over clan leadership. As time goes on, the distributive cohesion of lineages of earlier leaders decays. In considering the cohesion of a lineage, the role of allies and marriage alliances needs also to be considered as contributing to cohesion of a Decentralized Leadership and Network Cohesion 357 leadership factional core that is larger than a single leader. The extent to which the cohesion of lineages is distributed across the clan or norm narrowly focused within small segments of the clan is gauged by the distribution of members of each clan within the energy scaling of the entire clan over all time periods, shown in Figure 9.10. Nodes are colored according to lineage in original version of Figure 9.10 (now found at the URL for the web site indicated), as in Table 9.7. Hence, all nodes of the same lineage are connected by male father/son links. This energy scaling of marriages for the whole clan is the same scaling used in Figure 9.9, for which subsets are shown for different periods. The difference is not in the location of nodes or marriages in these scalings, which is invariant, but that Figure 9.10 includes all the nodes that have lineage membership within the clan, for all time periods. Figure 9.10 is simplified by removing female links so that connectedness among agnatically related nodes can be identified. The scaling, however, reflects the invisible presence of female ties. If the female ties reinforced the tree-like structure of the male lineages the black-and-white graph would simply radiate out from the center (where the marital cohesion of Mustan’s grandchildren holds it together). Instead, we see many lines that cross over into other sectors of the graph, and the pattern is one of rather diffuse or distributed integration. The lineages do tend to cluster (this is only evident in the colored graph), but each lineage has its outliers who are found at longer distances from the lineage cluster. 358 Chapter 9 Figure 9.10: The Scaling of Marriage for Overall Cohesion (The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads; lines for individual males and females are readably numbered at the WWW site ) Table 9.7 presents data on leadership and alliances among the lineages, ordered by the time period in which they held the principal leadership position for the clan. The numbers of lineages and the coloring of nodes for each lineage in Figure 9.10 are given in the first column of Table 9.7. In the second column are given the periods of leadership for each lineage, as in Table 9.2. Column 3 summarizes the kind of cohesion held by the lineage, whether widely distributed or concentrated in certain locations in Figure 9.10. Lineages occupying correlated spaces within Figure 9.10 are, by virtue of the scaling algorithm, allied through marriage. Alliances of a lineage with a lineage appearing earlier or having a higher leadership position are summarized in column 4 (taking the data from Table 9.2. and Table 9.6). Relinking within lineages (Table 6.7) is summarized in column 5. Finally, the location of specific leaders in the cohesion structure of Figure 9.10 is given in column 6. Decentralized Leadership and Network Cohesion 359 Table 9.7: Summary of Data Relevant to Lineages, Their Exocohesive Integration in the Clan, and Emergent Leadership (Hypothesis 9.11, for the Cohesive Structure in Figure 9.10) Lineage in Figure 9.10 and colors* #5 Green #8 Lite Green #4 White #6 Lite Blue #1 Yellow Leadership Periods Table 9.2 1.1850-75 2.1875-99 client Position in Fig. 9.10 Cohesion Structure Quasidistributed*** upper center 3.1900-30 6.1980-95 (ally) Lower left Lower left Lower center 4.1930-57 Distributed central band lower left upper right like #1, but denser on left Fully distributed Thin central band #9 Orange #10 Purple #7 Red client client (ally) #2 Pink 5.1957-80 #3 Dark (ally) Green * Colored in web version ** Table 6.6 *** Displaced to left center Alliances Table 9.2, Table 9.7 Relinking within Lineage** Third Highest Position of Leader Fig. 9.10 1 Center 2 Center Fourth Highest 1 Center 2 Center Second Highest 1 Center right Highest 1 Center upper #4 #1/9/10 #5/8 #2 From the data in Table 9.7, it is evident that Hypothesis 9.11 holds up to the end of the 1980s (period 5), which ends with the leadership of Fındıklı Hacı (818, lineage #2), and does not hold for the leadership of “Dede.” Up to this point the most recent lineage to hold clan leadership (#2: 1957-1980), along with their marriage allies in lineage #3, has the most fully distributed cohesion in Figure 9.10 (URL image: the pink nodes, that is, are the most widely and uniformly distributed across the figure). We will examine shortly the pivotal role of Fındıklı Hacı in the social change and modernization that the clan experienced after the end of his tenure of leadership. The previous lineage to hold clan leadership (#1: 1930-1957), with their marriage allies in lineages 9,10, and 7, has the next broadest distribution (URL image: yellow nodes) across Figure 9.10, more focused on a central horizontal band (the client lineages have complementary distributions). Farther back in time, lineage #4 held the leading personality position in period 3 (1900-1930)—and again later in period 6 (after 1980)—and has one of the least distributed patterns of 360 Chapter 9 cohesion. It should also be noted that the leading personality position is not unequivocally transmitted. In the long and turbulent period 3, for example, it is not always clear as to the relative influence of Erkek Mustafa (517) and Fındıklı Ali (784). The first lineage to hold clan leadership (#5, URL image: green; 1850-1900) is somewhat dispersed but displaced to left center. According to Hypothesis 9.11, the most recent lineage (#4, the first to hold leadership in discontinuous time periods) ought to have a pattern of distributed cohesion. A similar logic may be operative, however, but now dependent on personal networks rather than cohesion through the lineage. The recent leader of lineage #4 is personally located at the center of Figure 9.10 (and is the nephew of the previous leader who also occupied a central position in terms of cohesion). In all cases, leaders are quite central in the cohesion network, but the leader of #4 in period 6 (597, “Dede”) is more central in terms of his personal network than either of the leaders in periods 4 and 5, although their lineages had more dispersed cohesion. The possibility of a change from a lineage-based distributed cohesion as a leadership base to a person-based centrality in the network of cohesion is another hint of changes in nomad social structure in recent decades (see Hypothesis 9.13). Hypothesis 9.12: To use the alliance theory terminology of LéviStrauss, it might be inferred from the preceding analysis that, historically, the lineages of clan leaders occupied positions of clan cohesion corresponding to generalized exchange (in our terms: distributed alliances). In contrast, lineages marginal to clan leadership, or after a period of time delay, have restricted exchanges (more focused and restricted marriage alliances). Hypothesis 9.13: The basis of leadership is changing, after the 1980s, from distributed cohesion in which a leader’s extended marital ties through lineage mates are of central importance, to cohesion based on personal networks that are more limited in their ability to integrate broad support within the clan. Ethnographically, there are two aspects of social change that lead up to and then occur in the transition of the 1980s. The first was the pivotal role of Fındıklı Hacı (818, lineage #2), the last of the “traditional” clan leaders, in modernization. The second was the pivotal role of “Dede,” the first of the “modern” leaders. The first supported the emergent role of the second in the social changes that occurred. Fındıklı Hacı was a traditional tanıdık kişi in terms of his selection, his kinship support, and Decentralized Leadership and Network Cohesion 361 his comportment, but his contacts outside the clan proved decisive in altering his perspectives. Fındıklı Hacı and his brothers were highly intelligent and respected men, but he himself was illiterate. The children of his first wife (389) were traditionally oriented clan members and lifelong nomads. One of his brothers (855) left the clan and resided in the small town of Saımbeylı, where one son (855) became a lawyer. His children by a second wife (894, Emine, who had no special education), advanced through formal education, in which Fındıklı Hacı had a special interest because he represented the clan in its negotiations in villages and towns. He had the oratorical style of a lawyer, and he sent his children to schools in town. One become a high-level district administrator and later a member of the Turkish parliament (Koçali, 826), another a doctor (829), and a third a director of a high school (830). Another brother (840) married a great-granddaughter (652, Nuriye, D of 851M of 652F) of a man (630) who was one of the few to rejoin the clan after retreating to the high mountains for many years to avoid the tax collectors, but who was a grandson of the lineage #1 founder (1926) nicknamed “Quarrelsome” Mustafa. The four sons of Nuriye (652) went on to become a policeman (846), a high school teacher (851), a primary school director (853), and a tractor driver (854). When her husband died, his brother Fındıklı Hacı took her in leviratic marriage as a co-wife to Emine. Of their children, one son (838) made a career in the police force and the daughter (639) became a high school teacher. When Fındıklı Hacı reached the age of retirement in 1980, he clearly understood that times had changed. All of his sons except the eldest were educated professionals. In the meantime, the Turkish authorities had required from the 1960s forward that the clan, having a small settlement in the mountain camp, elect a mayor, which was managed by Fındıklı Hacı by appointed someone to the task who would take orders from him. A local committee of young men was elected to help the mayor, one of whom was “Dede.” Fındıklı Hacı opposed the choice of “Dede,” the most dynamic of the young men to succeed him as tanıdık kişi, but, after his health worsened and he moved to town, he relented and “Dede,” who had then become mayor, also became his successor. “Dede,” from lineage #4, then, not only represented a new set of values but also a willingness to participate in a new organizational structure imposed on top of the traditional leadership position. “Dede” had started poor, worked extremely hard, became a cattle trader, came to own trucks for transport, and, by selling animals and buying property built up a respective position among the sedentary population, although he had never 362 Chapter 9 been to school; still, he had learned so much by himself that he earned a basic school diploma. We can examine in detail the structures of marital alliances that still operated to create kinship cohesion for “Dede” (597), but now in reduced form, if we return to the example of Figure 6.5. This was the figure we used to illustrate the existence of multiple large bicomponents (fragmented social cohesion). We used this figure to illustrate the way that the network ties in this period might have looked from the egocentric perspective of “Dede” as the clan leader of that period. The figure also illustrates how changeable or emergent are such elements of social structure: a marriage between a daughter of couple 44 in Figure 6.5 and a son of couple 1, for example, would radically reduce the kinship distances in this network, and would integrate the two bicomponents. Yet the fact that this has not occurred given ample opportunity, that “Dede’s” lineage is not widely integrated in terms of distributed cohesion, and that it is only his personal network that places him at the center clan cohesion, gives convergent empirical support for Hypothesis 9.13, of a fundamental change of the basis of political leadership. Still, there were other factors behind the changes that the succession of “Dede” represented in 1980. Traditionally, prior to the 1930s, the poorest and the richest had turned to sedentary life: the richest by selling commons pasture land as if it was their own and investing in village lands to become landlords, and the poorest because they were forced to make a living as field hands. Up to this time, however, usually there were no land ownership documents. Beginning in the 1930s the Turkish administration started to provide and administer land documents as requirements for sale. In the following periods it became increasingly recognized that economic success for those who would become wealthy or move upward to better positions depended either on education—and emigration—or on the type work ethic that “Dede” represented in the transition to a different basis for leadership. Summary Emergent leadership for Aydınlı nomads is part of a multigenerational process. At stage one, the pattern is that an entrepreneurial father and a mother from a well-off family (with the social cohesion needed to have access to the nomad exchange networks) may emerge as one of the richest families in their generation from a lineage either not having had a re- Decentralized Leadership and Network Cohesion 363 cent turn in the rotation of leadership among competing and cooperating lineage segments, or where the father was a headman in a different location, prior to migration, as was the case with the second of the leading personalities of the clan. If they have the local attributes of cohesive marriage and family practices, including several married sons who retain their nomadism, one of these sons may have the character, economic success, and respect associated with achievement within the contemporary economic setting to become the tanıdık kişi, the important person who coordinates clan discussions and decision making. In a society that does not appoint or elect their leaders but lets them emerge through a consensus process according to how many others (eventually, all) beat a path to their tent, one of the requirements is that his family have one of the larger tents associated with large and wealthy families. At stage two, for the son to be successful, he must also have the attributes of cohesive marriage and family practices, and partly because of his father’s wealth he may also been able to afford several wives, which are an asset in producing more married sons, endowing daughters with bride payments for further marriage alliances, and in providing the labor needed to host large groups in a large tent for discussions. The attributes and the network variables of kinship cohesion fit together in a tight package associated with a process that is not ascribed, or even rigidly predictable (as we see from the low r-squared of attributes in regression analysis, although these have significant effects that fit this description). We see the process as an intricate pattern of emergence and synchronization, in which behavioral practices and goal-driven choices interact with societal values and the judgment and respect of others. The process is fundamentally a social one, an elaborate social choreography, deeply embedded within the needs and drives for economic success, but attuned to the political nuances of alliance-making and social support. Cohesive social practice seems an apt label for this kind of emergent process. Let us not forget, however, that it is the personality that turns the scale for leadership. Here we come to the limits of our network data, where we have no adequate measures of differing personalities. Marriage alliances and structural endogamy, however, are also crucially important for Aydınlı nomad clan leadership and equally so for lineages and sublineages at various levels. Leaders and their sons tend to have one wife who relinks with other segments within the same lineage, another who relinks with the lineage of the previous leader, another who repeats the alliances of the former leader, and so forth. In each case there is continuity from one period to the next, but also evidence ruptures or 364 Chapter 9 change to a more village-oriented alliance pattern in the 1960s and later, such as the importance of formal schooling. With this, and questions about continuity and change, we introduce in the next chapter our examination of political leaderships. Further Reading Aydınlı nomad emergent leadership fits the pattern of emulation of successful role-= models or prestige-biased cultural transmission described by anthropologist Joseph Henrich (2002) and colleagues in a series of studies carried out cross-culturally. Boehm (1993, 1997, and 1999) has extensively studied egalitarian leadership and its effects on human evolution in a series of prize-winning studies. Friedkin (1993) shows, for example, in one case study, that social cohesion (measured by an access measure that incorporates number of communication paths, length of paths, and strengths of constituent ties along paths) is the primary determinant of issue-related communication. The latter turn is the primary predictor of issue-related influence, controlling for rewards, coercion, authority, and expertise that have influence through the elementary structure of differential power. Bourgeois and Friedkin (2001) argue from the evidence of a case study that while interpersonal ties (including measures of cohesion) foster social solidarity, the expected negative effects of social distance may be salient in certain core-periphery structures. However, they are not ubiquitous implications of social differentiation but properties of particular forms of social organization. Friedkin (1998) gives less attention to social cohesion as a basis of interpersonal influence in his empirical analyses than he does structural similarity and centrality, but he includes an excellent discussion of social cohesion. Notes 1. Categorical connectivity or the category of connectivity (Friedkin 1998:164) refers to four categories of reachability in a directed graph: unilateral (one or the other of each pair can reach the other on a directed path), weak (every node can reach any other through a semipath), strong (every node can reach any other through a directed path), and disconnected (there is some node that cannot reach another through a semipath). In contrast, the connectivity level k as we use it here corresponds to the type of connectivity in Menger’s Theorem (Harary 1969: Ch. 5), for example, the number of nodes whose removal is needed to Decentralized Leadership and Network Cohesion 365 disconnect a graph, and the minimum number of node-independent paths between pairs of nodes (White and Harary 2001). 2. Marriage with a grand niece (BDD) is not found among the Aydınlı but is only a hypothetical example in the illustration. 3. Note once again that although treating the p-graph as an ordinary graph in this way provides a means of identifying distinct multiple paths of connectedness, and of measuring levels of cohesion, there cannot be a k-component in a pgraph that is more than a bicomponent, that is, having a level of group cohesion greater than k=2. This is because, at the group level, no set of couples may be connected to all other couples in the set by paths within the set that begin with three or more parental edges. To illustrate this more concretely, imagine that we tried to increase the cohesion of the p-graph in Figure 9.1 by adding an edge from the lower FBD marriage to the upper ancestral node. This would violate the principles of the genealogical ordering of generations. The upper node cannot be a child of this couple because it is an ancestral node. Nor can it be an ancestor for either member of this couple because both sets of parents for the couple have already been identified. Thus, if there were a kinship network for which the underlying graph was a 3-component, then someone would have to be their own ancestor, and this cannot be the case in genealogical relations because parents are always in a preceding generation. 4. A 1-block is necessarily a component, and a 2-block is a bicomponent because it can have no nodes outside the block that add extraconnectivity. Only kblocks for k >2 are distinct from k-components. 5. It is known mathematically that no two k-components can have k nodes in common, but it has not been proven that two exocohesive k-blocks cannot have k nodes in common. It is obvious, however, that every k-block is embedded in a k1 block, just as every k-component is embedded in a k-1 component. 6. Bicomponent analysis, pairwise connectivity, and PCA are available in UCINET. 7. When we scale exocohesive blocks that have a minimum size-of-group requirement (equal to the level of embedding plus one) the embedding structure is somewhat flattened (data available from D. White. A Fortran program (maxflow.for) is available from the first author that identifies the highest kblocks in UCInet output matrices for point or edge- (maxflow) connectivity. 8. Because the first component of a principal components analysis (PCA, available in UCINET) corresponds to marginal sums of the cohesion matrix, the values on the second eigenvector sort the nodes as to their similarities in level of cohesion. 9. Whether wealth that is achieved, such as by the father of a potential tanıdık kişi, is more respected than wealth that is hereditary and consolidated, in the sense that the latter is a threat to the delicate balance of equality among nomad lineages and families, is not at all clear from the experience of the ethnographer. 10. Fındıklı Hacı was not very rich as a young man and never became the rich- 366 Chapter 9 est, but his father was a highly respected personality too and a fairly rich man in the c-generation. 11. Recall from Chapter 6 that while genealogical distances among individual Aydınlı can be quite large, distances between lineages at a higher order are small. Higher-order linkages is one way to overcome long linkage-distance between individuals, but because cohesive groups are often the basis of higherorder groups, it is well worth repeating some of the relevant quotes about how cohesive groups overcome distance in the general case: [W]hat this bias [against long-distance ties among individuals having demonstrable effects] in preconceptions of social cohesion omits are the two fundamental properties of the redundancies created by multiple independent pathways and multiple-node cut sets. First, independent pathways are convergent in their indirect effects, even at a distance. Independent paths between every pair of nodes in a cohesive block defined by connectivity level k (which necessarily equals the minimum number of such paths) may more than compensate for the decay of effects of cohesive interaction along long paths. Studies of large-scale social diffusion, for example, typically rest upon and demonstrate the fact that long paths matter. What connectivity provides in terms of transmission effects within the internal networks of cohesive groups is the possibility for repetition along multiple independent pathways of rumor, information, material item, and influence transmission. Second, multiple independent pathways (equi-numerous to minimum cuts) necessarily imply stronger bonding between pairs of nodes, regardless of distance decay. It is k times as hard to break apart a network tying nodes together by k nodeindependent pathways than it is to break apart a single chain that connects them. (White and Harary 2001:351) The effects of multiple bonding and redundancy or repetition along convergent independent pathways are crucial in the formation of social coherence, social norms, sanctions and solidarities, and the emergence of socially or culturally homogeneous groups, and thus should be of focal interest to the study of social cohesion, including cohesion on a very large scale. (White and Harary 2001:355). 12. To reiterate, the number of edge-independent paths linking actors is the number of paths that are independent in terms of their edges and corresponds to the number of edges that must be removed to disconnect a pair of nodes. It is always equal to or greater than the number of node-independent paths linking actors. These computations are usually made for relatively small matrices (<400) because computation can be quite lengthy. White and Harary (2001) explain why “adhesion” is a better label than “cohesion” for the maxflow measure. Two nodes may require any number of edges to be removed before they are disconnected, and yet be disconnected by removal of a single node. Hence, high adhesion does not entail high cohesion (where the number of node-independent paths Decentralized Leadership and Network Cohesion 367 >> 1). 13. That is, when we compute the pairwise matrices of adhesion and that of cohesion, they are virtually identical. As in the high-high cell of Table 9.4, the Aydınlı nomads have no adhesive mediators between exocohesive groups, mediators that would attain added influence in view of their betweenness centrality. 14. One might ask: Why do we not test the Zachary’s hypothesis that minimum edge-cuts in Aydınlı nomad social networks are a predictor of leadership? They are, indeed, but only in the sense that the edge-cut predictions are virtually identical to the node-cut (cohesion) predictions. Our answer is that the empirical equivalence between the two measures (which is not foreordained because they are defined to be partially independent) is itself an indication that the Aydınlı themselves express in their behavior a preference for cohesive strategies and avoid adhesive ones. By doing so they also avoid giving greater betweenness centrality to other nodes, clusters of nodes, and lineage or support groups. Accordingly, the betweenness centrality index of the bicomponent of the kinship network, as a percentage of the possible maximum for this many nodes, stands at 27%, which is not particular centralized given that certain ancestors such as Mustan have had so many descendants. 15. The UCINET options are Network/Cohesion/Distance and Tools/MDS/ Metric. 16. We know that leading known-persons are always members of the larger lineages, but not necessarily the largest. There is a complex relation between lineage size and succession to leadership, as was shown in Figure 6.10. In the earliest period, for example, lineage #5 is apparently not the largest lineage, at least from our retrospective data, but holds leadership for two generations. An alternative possibility suggested by accounts of some of the old men is that lineage #5 was more numerous than the genealogical accounts would imply because many of the early collaterals of the lineage segmented from the clan and left no descendants. Lineage size in genealogical reconstructions is not necessarily accurate but must be supplemented by oral historical accounts. 17. This hypothesis gains support from the role of Mustan’s lineage as mediators. In a separate analysis of betweenness centralities for all the nodes, organized by lineages, Mustan’s (#2) lineage and its members are by far the most central in terms of relinking within the clan, but they started as a poor lineage, and they only attained the leadership role very late in the succession. According to Figure 6.10, however, they had become the largest lineage by 1900 when the tenure lineage #5 ended. Again, however, it may be that their central role in integrating through social cohesion (marital relinking) those who choose to stay rather than migrate only makes it appear retrospectively, through a stayer-bias, that they were most numerous a century ago. 18. One was of a prestigious origin (515, #4: judged from the father’s title, efendi, a man of influence), another heir to property (10, #1), a third the daughter of a Molla (1413, #2) and thus probably from a well-off family, as was also the 368 Chapter 9 mother (445, #4) of the sixth tanıdık kişi. 19. A young wife is described in relation to her father-in-law, as for example, Molla Ali’s daughter-in-law (Molla Ali’nin gelini) for whom the term used means Molla Ali’s young woman. Chapter 10 Graphic Approaches to Nomad Solidarity: The Endoconical Clan Informants gave a history of specific clan founders and the genealogical data on the marriage network on the clan and its members’ links to villages and the other tribes. From this information, we analyzed the structure of relinkings and structural endogamy so as to ascertain some of the principles of the formation of the clan. Can we illuminate these astoundingly complex linkages with graphic representation and then use analysis of graphs to help evaluate further social aspects of their construction? Structural analysis of the network as a whole is a task of formidable complexity. The Pgraph package used at the time that Figures 2.2 to 2.5 were made (circa1997) lacked the requisite scaling algorithms for such a task. Development of the Pajek program (Batagelj and Mrvar 1998) for large network analysis made possible an automatic drawing of the entire structure of relinked marriages. (As noted, Pajek reads both the genealogical GEDCOM file made by the Ego2Cpl program from the data in Appendix 1, whose format is given in Table 2.1, and the NET file also made by Ego2Cpl). Pajek was instructed to scale the network of parent/child relations among couples in a three-dimensional representation, so as to produce Figure 10.1.1 Here, generational time is the vertical or z-axis while the generally conical shape of the graph results from a Pajek automatic drawing (energy scaling)2 in the x-y planes that project out toward the viewer. The scaling shows how the early generations were closely knit through marriage. Over succeeding generations, some couples at the center remain closely knit while others diverge along the lower slopes of the cone into sectors that are locally but less centrally knit within the entire clan. One gets the immediate impression of a “conical” clan and of the occasional dying-out of descent lines. 370 Chapter 10 Figure 10.1: 3-D Graphic of the Entire Nomad Genealogy 1860s 1890s 1920s 1950s 1980s The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads The Endoconical Clan In this chapter we explore the idea of endoconicality in the sense of clan members coming from a common root or roots (Analysis 4), having some form of endogamous cohesion that keeps the clan together (Analyses 3 and 5), and a flexible principle of ranking among stayers that accommodates the high rates of emigration. The black-and-white shadings of nodes (originally colors) in Figure 10.1 indicate generations, roughly Johansen’s [a] through [h], but generations are rescaled by Pajek to the fewest levels needed for parents to be prior to children. A Pajek partition that computes generational layers Graphic Approaches to Nomad Solidarity 371 was used as the z axis in the graph, which here is the vertical time dimension. In 3-D renderings the nodes are smaller as they recede in the xy horizontal plane orthogonal to the time dimension. The x-y coordinates were computed by Pajek’s automatic drawing that minimizes length of lines. Couples that are closely relinked are pulled to the x-y center of the graph along the vertical axis. The couples at the outside of the cone are those less centrally connected.3 A partial structure consisting of all the relinkings through marriage is extracted in Figure 10.2 from the complete genealogical network of Figure 10.1. The shading (colors) of nodes remains the same as in Figure 10.1, and members of the earliest generations with no relinkings (and only partial data) are removed. Figure 10.2: 3-D Graphic—Relinking Marriages among Nomad Kin Koca Bey #5 Abbas Kirbasi Ecevitli #4 #3 #2 1860s 1890s 1920s 1950s 1980s The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads The automatic drawings in Figures 10.1 and 10.2 pull to the center of the graph those who have closely intermarried and push apart those who 372 Chapter 10 have not. This is ideal for displaying the general morphology of marital relinking given our questions about tribal structure, alliance, and rank. Figure 10.1 presents the results of the embedding for the clan as a whole—the entire genealogy—while 10.2 selects out only those couples who are connected to this structure by two or more relationships. These marriages are structurally endogamous (White 1997, Brudner and White 1997) in that every pair of these couples is maritally relinked. Of the total of 414 marriages and 557 single individuals, for a total of 956 vertices, there were 243 relinked marriages that are graphed in Figure 10.2. Marital relinking among the nomads, shown in Figure 10.2, is quite dense and occurs around a set of people in lines constituting a distinct center of the graph that focuses on two of the Karahacılı ancestors (#2 Abbas and #5 Koca bey), whose son and grandson are the first of our leading known-persons. These and two other Karahacılı lines, from the Ecevitli C village ancestor #3 and the Kırbaşı ancestor #4 furnish the main central starting points of the central line in the first few generations. Karahacılı line #4, descended from Koca bey, joins the central line in the fourth and fifth generations. No other apical ancestors appear to be central to the early generations. The father of Yusuf and Ismail of “Dolaşıklı” lineage (#1) is the only early ancestor in the first six generations who begins at a distance from the central lines. Because he does not himself relink to others, but relinking in his line starts only at the grandchildren’s generation, he is not shown in Figure 10.2. Recalling Analysis 4, the genealogical network reveals in its general morphology something of interest as to how dense intermarriage can lead to the perception of being “all from one root.” While there are multiple apical ancestors, their descendants become so closely fused through blood marriages and relinking marriages from the central confluence of this graph in the third generation, it is easy to see how “from one root” might be justified. There are very few other apical ancestors from lines that are not already linked after the third generation. From the central stock, a central core of descendants remain tightly intermarried in the center of the graph, but other lines of descendants relink more and more remotely to the main lines. There appears to be a certain point in the tribe’s history at which a “common root” is formed by dense intermarriages. This funnel of relinking density probably marks the emergence at a certain place and time of the clan itself. The automatic drawing, seen as a structure over generations, captures much that is significant about marital relinking in the structure of tribal formation, as seen through the genealogies. Multiple marriage affinities Graphic Approaches to Nomad Solidarity 373 hold the clan together as do blood lines, and intermarriage leads to common bilateral bloodlines. The structure gradually fans out as relinking becomes more and more divergent. This might correspond to a process whereby the more central members are more likely to remain nomadic, while the more divergent members from the central line might be more likely to convert to sedentism. The structure of marital relinking also helps to see how people from the outside are absorbed, through multiple marriage alliances, into the core group. Indeed, one of the early founders of the nomad clan (7, father of 1230 Deli Abdurrahman) as well as the three adopted shepherds (founders of lineages #9,10,11) seem to have “disappeared” from Figure 10.2 as distinctive lines outside of the central cone of the genealogy. All that has happened is that due to their relinkings they have been assimilated into and obscured within the main part of the structure, due to their proximity to affines. To what extent are the marital relinkings that pull the couples in the center of the graph marriages among blood kin as opposed to affines?4 Chapters 7 and 8 suggest the importance of preferential blood marriages while Analysis 7 suggests that marriage exchanges are central for constructing cohesion. The question remains unresolved. Analysis 18: Age Ranking and the Endoconical Clan Network analysis can address questions of network topologies of kinship in quite precise ways. A clan, for example, might have a concentric marriage structure if those at the center are the most closely relinked and those at the edges are less so. A clan might be “conical” if the descent lines of a unilineal or ambilineal descent group that has an apical ancestor and status ranking such that the line of highest successive ranks can be drawn at the center, and successive peripheries can be scale concentrically toward the sides. Figures 10.1 and 10.2 give an image of the nomad clan as distinctly “concentric” in structure in that what holds members of the clan closer together in the automatic drawing are the gradients in intensity of cohesion that come from marital relinking. But they are conical in this limited sense only in the compactness of nodes at the top, and the growing number of nodes and distances moving down. Whether the Aydınlı nomad clan is also conical in the sense of a ranking system can be answered only by a network analysis of how the attributes and relations of individuals (e.g., the section on “Rank” in Chapter 5) are related to the network structure. 374 Chapter 10 Fustel de Coulange (1864) and Gifford (1929) were among the first to describe the ranking principles that define a conical or core-periphery ranking structure of the respective clan systems of Greco-Roman gens and Tongan ha’a patrilineages. In the Tongan case, primogeniture creates the center-periphery ranking. Mason (1954) found ranking by primogeniture in Marshallese matrilineages. Leach (1954) found an inverse ranking, based on ultimogeniture, among the Kachin (gumsa). Kirchoff (1955), describing the general principle of ranked conical clans, noted that descent might be traced unilineally through either males or females, or ambilineally through either men or women. Other principles of rank may enter in, such as sibling rankings by gender, or differences among affines. In the Tongan case, for example, sisters outrank brothers, husbands outrank their wife’s brothers, and wife-takers outrank wife-givers. In the Kachin (gumsa) case, wife-givers are superior to wife-takers. Marriage alliance is often implicated in the dynamics of the ranking systems that define the core-periphery structure of conical clans. All of the cases of conical clans discussed by Kirchoff, and later reviewed by Hage and Harary (1996), also allow endogamy, so that Kirchoff’s concept of the conical clan stands in contrast to the exogamous clan. Does the Aydınlı case represent a different type of conical clan, one with a conception of a “common root” in such a deeply endogamous set of ancestors that the root appears unitary, but in which there are degrees of membership that depend on how deeply one is intermarried into the clan, and relationships are reckoned bilaterally even though a single lineage principle may be operative? If so, might there be a principle of rank that applies to individuals, ranking siblings by age, for example, such that higher ranked individuals also tend to be at the highest levels of endogamy created by marital relinking within the clan? We might call this an endoconical clan not just in the sense of endogamy but of endogamic intensity as an important basis of differential identification, membership, and leadership. This conception would constitute a coherent type of clan system with its own dynamic (marital relinking), with leadership and other forms of participation in the life of the clan varying according to a looser conception of how individuals are ranked. This endoconical type of clan, in our conception, is one that, compared to Kirchoff’s conical clan, has greater gender equality because of the importance of bilateral relationships and a lesser degree of rank inequality because the principle of rank is more emergent than prescribed. This is the idea we explore here. The Aydınlı clan system approximates what we called above an en- Graphic Approaches to Nomad Solidarity 375 doconical clan, marked by intense endogamy, lineages, lineage segmentation, and the tendency to attribute descent from a common ancestor or “root” to those with whom they have relationships of affinity. 5 The concept of an endoconical clan is not so far from that of Kirchoff’s conical clan, in which endogamy is an important element, in contrast to the exogamous clan. As opposed to strict status ranking, however, the Aydınlı are fiercely egalitarian in their political leadership, egalitarian in their distribution of inheritance, oriented toward same-generation marriage rather than age-skewed marriages, and recognize no status inequality between wife-giving and wife-taking. Aydınlı nomads do recognize, however, elder/younger sibling distinctions in greetings and other behaviors, and, like all Turkish speakers, distinguish the elder brother by a distinct kinship term (ağa, as distinct from kardaş, yB). Given the importance of age rankings, a question remains: do elder brother/younger brother distinctions carry over to the ranking of lineage segments? Hage and Harary (1996: Chapters 4 and 5) identified the system of recounting genealogies known in computer science as depth first search (DFS) as indexical of Kirchoff’s (1955) conical clans with strict age- or status-ranked descent systems, such as ultimogeniture for gumsa (Kachin) or primogeniture for Tongan patrilineages. A strict DFS ranking is exemplified on the left of Figure 10.3. Here, the recitation of ancestors begins with the highest ancestor (1), takes next the highest ranked offspring (2), and goes down through each of their highest ranked offspring (3,4) as far as possible (4) before going up as little as possible and then down again as far as possible, repeatedly (5; then 6-7; then 8-9). In this way each branch is exhausted, until the whole descent tree is traversed. A sibling-set depth first search, shown to the right in the figure, imposes the recounting of sibling sets within the order of a DFS, that is, each time a lower node is taken the siblings of that node are recounted next before going on to the descendants of the initial node. This method is used by the bilateral Tory Islanders (Fox 1978) for recounting deep genealogies, who may represent another variant of endoconical clans. Starting from the highest ancestor (1), the age-ranked children are listed from eldest to youngest (2-6), then the most prominent sibling is selected (2) for moving down to the next generation so as to list his children (79), then down another generation by the same principle, arriving at the lowest generation in one branch (10) before going back as little as possible to the level of lesser ranked descendants (8, then 9, 11-12, and 13) who were not previously recounted, always down again as far as possi- 376 Chapter 10 ble, continuing until the entire set of known offspring of the lineage (ending with 14-15 and 16-17) is recounted. Figure 10.3: Depth First Search (DFS: leftmost) and Aydınlı Siblingset DFS Reckoning (right) 1 2 1 10 11 16 3 5 6 12 15 4 7 8 13 9 14 17 2 3 4 7 8 9 10 11 12 5 6 14 15 16 17 13 Tribal peoples Islamized by the Arabic conquests that began in the seventh and eight centuries often acquired Arabic customs that included FBD marriage practices and familiarity with deep genealogical accounting. Such Arab tribes, by all reports, recite genealogies orally only back from the present (Paul Dresch, personal communication). Strict DFS is not the system of recounting deep genealogies from memory in the Middle East, nor is sibling-set DFS. Aydınlı nomads are not so hierarchical in recounting genealogies as to use top-down DFS-type systems. Like other tribes, they start from the bottom and work up to F, FF, FFF, and eventually to the oldest known relatives. Genealogical recounting has a certain rhetorical style and pattern, but as it is done in the context of a group of men, who join in to recount their relations to the early founders of the clan and the intermediate generations. They remember and comment upon the relative age of siblings (especially brothers) when discussing genealogy. In asking them to give a fuller recounting of lineages, Johansen found they could easily trace the sibling-set variant of DFS recounting, as shown on the right of Figure 10.3 and as found among the Tory Islanders. Even in recounting genealogies upwards, however, they would give the siblings in the descent line, with other men coming in later to recount how their descent line connects with one of those siblings.6 Age-ranking within sibling sets, then, is an important principle in the reciting of Aydınlı nomad genealogies. Further, when Aydınlı men recount genealogies, women are usually recalled only as ranked members Graphic Approaches to Nomad Solidarity 377 of sibling sets, when first recounted, and as wives, but not as among the vertical elements in the depth-first reckoning. Men’s recitals often left out the tracing of a woman’s children, and Johansen would have to do separate interviews with both women and men to complete the female lines. Blood feud is given by another ethnographer of the Yörük Black-Tent peoples (which includes the Aydınlı nomads), as one of the reasons why nomad tribes tend to avoid the telling of deep genealogies from the top down: A sharp limit to the scope of genealogies is seen in informant efforts to recite their patrilineal ancestry. A point is reached below which every male can place most of his patrilineal relatives, but above which even elderly males of a higher generation cannot pass. This is not a feature unique to Yörük segmented lineages: Irons notes a strikingly similar pattern of “limited recall” among the Yomut Turkmen (1969b:58-65). His explanation is that the Yomut obscure their genealogies above a certain point in order to avoid the risk of being included in vengeance killings or feuds which through remote linkages of agnatic kinship might be held to define them as objects of revenge. Vengeance among the Yörük [including, we note, the Aydınlı] is an obligation which falls to close patri-kin of the victim of murder, but without clearly defined outer limits of either involvement or responsibility. (Bates 1973:47)7 Aydınlı nomads also fit the pattern described by Bates (1973:50-51) for his Yörük group, in which the relative age of brothers . . . is not merely a linguistic phenomenon; it has considerable importance in interpersonal relations among siblings. What is relevant here with respect to segmentation is that the eldest of the brothers is held to be senior to all younger, irrespective of wealth, in situations of formal etiquette; he serves as spokesman when brothers act in concert. After the father’s death he is obliged, more than the father in his lifetime, to provide for his single brothers, and to assist them in time of trouble . . . marriage takes place in order of birth, which again sets the order of household fissioning to form new ones as younger sons marry and bring their brides into the tent. This, of course, gives older brothers in any generation an earlier start in the production of progeny to further their name. . . . However, just as the point of segmentation does not depend entirely on genealogical depth, neither does the relative seniority of brothers escape the impact of political and residential fact in determining which of several will provide the name under which the group passes. Given our conjecture of a loosely ranked endoconical clan system for the Aydınlı, we wanted to test whether the network data supported the idea 378 Chapter 10 that first or early born sons, as listed in the genealogies, bearing in mind that some of them will migrate or will fail economically, tended to be more important actors in the clan, both at the level of cohesive relinking and of political succession. Hypothesis 10.1: First and second sons are more important in marital relinking within the clan than are younger sons. We state the hypothesis this way because first born sometimes emigrate, leaving the second son who remains behind the one with highest agerank,8 but also because our numbers of first and second sons (400 marriages) and second and younger sons (408 marriages) are roughly equal, which is useful for network comparisons. To test the hypothesis we compared relinking in the first network to that in the second network to see if there is a difference in the extent to which each forms a cohesive bicomponent of the clan in its own right. Recall that bicomponents in a p-graph are the largest units of structural endogamy. As Figure 10.4 shows, there is a massive difference: the network of marriages among the first and second sons (400 marriages: 374 edges; 277 male and 97 female) forms a bicomponent of 314 nodes, while the network of marriages among the younger sons (408 marriages: 296 edges; 230 male and 66 female) forms a bicomponent only two-thirds that size (226). Johansen (1999) also supports Hypothesis 10.1, for which we may give a brief summary. The principle of seniority is not an absolute rule and in practice the smartest will be the most respected. Neither Fındıklı Hacı nor “Dede,” for example, were the eldest sons. Nomad societies, especially the more egalitarian, have to be flexible if they want to survive as a group. 500 400 Number males females bicomponent 300 200 100 0 2 thru 9 sons 1&2 Graphic Approaches to Nomad Solidarity 379 Figure 10.4: Comparison of Bicomponents (Structural Endogamy) Formed by Birth Order Sets To guard against the possibility that the older sons’ network was biased toward early generations, in which some brothers are missing, we also took the network of marriages in generations 3-6, computed the distribution of sons’ birth orders (which is uniform across these generations), computed the bicomponent again, and, within that, computed the new distribution of sons’ birth orders. The results are shown in Figure 10.5. Overall there are 34% first sons. There are many more first sons within the bicomponent (41%) than outside the bicomponent (29%), a significant confirmation (p=.002) of the hypothesis, especially because first sons are in the minority in the overall network. Figure 10.5: Comparison of Birth Orders within Bicomponents 100 80 60 In Bico Outside 40 20 0 1 2 3 4 5 6 7 8 9 Hypothesis 10.2: In the traditional system of leadership, emergent leaders and their fathers are likely to be first or second sons (or to segment to form a new lineage in a different clan). This hypothesis applies only to the first five tanıdık kişiler and not the last because we have argued that “Dede” represented a break from the traditional pattern of leadership. Excluding the father of the first, about which we know nothing but the name, this leaves nine tanıdık kişiler and their fathers, of which eight are first or second sons (p=.07 in the direction of the hypothesis), with the exceptional case being one of lineage segmentation (i.e., no older brother within the clan itself). Taking this into account, all nine fit the hypothesis (p=.01 that all nine will be first or second sons given the total distribution of birth orders). Six of our 380 Chapter 10 eight cases of first or second sons are second sons, which also deviates from the demographic expectation that most will be first sons (p=.07). One first son (785, Fındıklı Mustafa) was married with children at the time that his younger brother was tanıdık kişiler, consistent with the lack of a strict rule of primogeniture. What we find, then, is that first and second sons do play the social roles that might be expected of those with higher rank: Compared to what is expected demographically in relation to younger brothers, they have a greater frequency of involvement in the cohesive networks and in the leadership roles of the clan. Older brothers are only primus inter pares, and inherit equally with other brothers, although they have greater obligations to support them after their father dies. They are assured bride payments for attaining wives, however, and if they attain more wives than their younger brothers this alone might explain how and why they play greater roles in kinship cohesion as well as leadership. Where there are only two brothers, for example, the first born is significantly more likely to take a second or third wife (p=.02). Further, none of the twentyfour brothers in birth orders 5 to 10 attained more than one wife (p=.0002). Thus number of wives, associated with birth order, contributes to the prediction that early-born children will have higher exocohesive relinking (more wives; more children) and a better chance (because of the wives) at emergent leadership. Our evidence is consistent with the idea that the endoconical clan is an emergent structure among the Aydınlı, not based on a strict rule of succession such as primogeniture, but on an age-ranking tendency for older sons to have a higher place as primus inter pares. Competitively, older brothers simply have early-arrival advantages in the division of tent space, herds, bride payments, in securing wives. The endoconical clan, as we have described it, forms part of a continuum with Kirchoff’s conical clan, the former based on an emergent primus inter pares ranking, and the latter on a more ascriptive and ineqalitarian ranking system. Kirchoff’s conical clan can be contrasted with similar systems having looser and more emergent ranking systems, as we could discern for the Aydınlı, as shown in Table 10.1. Here we summarize the ways in which the endoconical clan is intermediate in terms of stratification between his conical clan (rigid stratification) and his exogamous clan (strictly egalitarian), although there ought to be a greater range of types on that continuum. In terms of descent principle, however, it contrasts with both of Kirchoff’s types, of unilineal or ambilineal clans. The Aydınlı nomads have a bilateral clan, but one that contains patrilineages which may fuse Graphic Approaches to Nomad Solidarity 381 or segment, a common pattern among the tribes of the Middle East. Genealogical amnesia beyond 5-6 generations is common in Middle Eastern tribes and greater genealogical depths are mostly recorded only in written form. In the Tory Island variant of the endoconical clan lineages are lacking and descent is bilateral, genealogies are deeper, and genealogies are recited from the top down, apparently in relation to the inheritance of land. We do not see this classification as a discrete set of types but suggestive of dimensions of variation in types of conical clans. Table 10.1: Defining Features of Clan Types Conical Clan Unilineal or Ambilineal Endogamy allowed Strict Ranking Stratification Genealogies may be recited from the top Deep political Genealogies Endoconical Clan Bilateral clan, lineages may fuse or segment Endogamy allowed Emergent Ranking Primus inter pares Genealogies recited only upward Genealogical amnesia beyond 5-6 generations Exogamous Clan Unilineal or Ambilineal Lineage Exogamy No Ranking Egalitarian Ancestor may be putative or traced Genealogical depth varies Analysis 19: Levels of Relinking Graph theory and graphic approaches are useful not only to visualize but also to help us to model, measure, and explain social phenomena. In the present case, we have made considerable use of concepts from graph theory to formulate testable hypotheses about social structure. Relinking, structural endogamy, p-graph, bicomponent, exocohesive blocks, centralities, controlled simulation, and ranked search trees are all concepts derived from graph and network theory that have helped us to do so. As another test of this approach, we can ask whether graph theoretic concepts contribute to an explanation of the question: Why do Aydınlı nomads have so high a level of marital relinking? We begin to examine further properties of the nomad kinship network by considering the lengths of paths between various relatives, paths in which the lack of branching relations weakens the cohesion between the endnodes. A chain is a path that contains no branches. We will use the length of maximal chains as a graph theoretic concept to help define the weaker kinship ties in a p-graph. 382 Chapter 10 Maximal Chains A maximal chain of a graph G is a path of maximal length between endnodes u and v such that no node between u and v has degree greater than 2. That is, a maximal chain between u and v is a path that has no branches, yet is as long as possible. There exists a unique set of longest chains for each length d in any graph. The maximal chain length of a graph G is the length d of the longest maximal chain in G. The maximal chain length of a graph G is a measure of the compactness of G. If the maximal chain length d is 1 for a graph G then G is a clique. Graph (a) of Figure 10.6, in contrast, is minimally compact. As shown in graph a of the figure, if every relationship from ego in a kinship network were composed of maximal chains with chain length d > 1 (e.g., for ego u: chain u-w-v), then ego would be a sociocentric “star” in which we could label every relative by a unique path by which they are linked, as shown in graph b of the figure. Star-like kinship networks such as this occur when there are few siblings to create branching paths. Ego’s relationships also weaken with longer chain distances, especially when they are not compensated by the cohesiveness of multiple independent paths created by relinking marriages. Figure 10.6: Egocentered Graph (a) and Labeled Reduction b v v' where x = u'v' = uw+wv (its label) w u graph a u' graph b In many populations where genealogies are collected, many of the paths between nodes are maximal chains of moderate to very long length. Relinking may occur so as to define large-scale structurally endogamous groups, but the shortest distance between spouses prior to a relinking through marriage is high. This is not the case for the Aydınlı nomads. In societies where distances between relinked spouses are very high, however, the distance may be so great that relationships are not recognizable even to those who are relinked. Hence, to find the denser cores of re- Graphic Approaches to Nomad Solidarity 383 linked endogamous groups, we may need to remove from the genealogical graph those maximal chains of some length d or greater that defines more compact and recognizable relationships. If such reduction—of removing longer maximal chains—is done, and bicomponents are computed on the network of either shorter maximal chains or of branching linkages, these bicomponents will have a higher relinking density. Chain-Reduced Graph A chain-reduced graph Gd of a graph G is one in which every maximal chain of length d or greater is eliminated. Thus, if we define a cutoff d and remove all longest chains of length > d for a graph G, the result is the graph Gd. Hypothesis 10.3: The kinship network of Aydınlı clan members is not only highly cohesive, but compactly cohesive: It has very short maximal chains (e.g., no longer than two), and the removal of longer maximal chains has little effect on the boundaries of structural endogamy or on the relinking index. Figure 10.7 gives an inventory of maximal chains with various vertical orientations corresponding to possible p-graph segments (subgraphs) in which there are no branchings. The various types of kinship links between u and v in these graphs are (a) GreatGreatGreatGrandparent (b) GrChSpGrPa (c) Grandparent’s Sibling’s Child (d) Grandchild’s Spouse’s Sibling’s Child, (e) GreatGreatGrandparent (f) GrChSpPa (g) Parent’s Sibling’s Child (h) Great grandparent, (i) ChSpPa, and (j) Sibling’s Child. None of the length 5 chains in fact occur in the Aydınlı data, and none of those of length 4 except for two cases of chain g. As for chains of length 3, the only type which occurs empirically is that of chain i, involving new marriages for which there are no children, which will cease to be a maximal chain when children are born. Hence, there are no stable maximal chains greater than length 2, consistent with Hypothesis 10.3. Chain h does not occur in the data, and even the grandparental/grandchild segment of this type of length 2 does not occur. Effectively, this finding entails that people do not occur or stay in the nomad clan if they lack siblings (chain h, and shorter versions thereof). This provides an explanation for the high degree of relinking in the clan. The absence of maximal chains with no branchings implies a high degree of relinking. The lack of nonbranching maximal chains generates high relinking density. 384 Chapter 10 Demographic Change in Stayer Bias for Larger Families The demographic regime under which this kind of very dense relinking is possible is closely connected with families that produce many children and where having siblings is very common among adults. Our shortchains finding, as it reflects on sibling links, suggests the following: Hypothesis 10.4: With adult siblings as a source of a coresidential and cooperative group, individuals are less likely to emigrate. Figure 10.7: Maximal Chains (nonbranching) of Length 5, 4, and 3 u v u u v v u chain a chain b v chain c Length 5 Maximal Chains chain d u u e1 v v chain e nonoccurrent u v e2 u chain f chain g nonoccurrent nearly nonoccurrent (2 cases) Length 4 Maximal Chains u v e1 e2 chain i v chain j v chain h u Graphic Approaches to Nomad Solidarity nonoccurrent occurs Length 3 Maximal Chains 385 nonoccurrent Figure 10.8.1 shows that the closer we move to the present (generations f and gh), where the selection bias of leavers is minimized, the more being in a small sized sibling set (1-2) disposes to emigration, while being in a larger one disposes to remaining nomadic (p=.02). Figure 10.8.2 shows, for the latest p-graph generation, that having fewer brothers is correlated with a higher rate of emigration (p<.001). Figure 10.8.1: Small Families Dispose to Emigration (generations a to h) 100 percentages 80 60 40 Stayers Migrate Small Large Small Large Small 0 Large 20 a-e a-e f f gh gh Figure 10.8.2: Small Families Dispose to Emigration (absolute numbers in latest generation as computed by p-graph; tau b=.32) 100 80 60 Stay 40 Migrate 20 0 3-8 Brothers 1-2 Brothers 386 Chapter 10 If the clan passes through a demographic transition where sizes of sibling groups are much reduced, our prediction would be that current basis of economic cooperation among the nomads would be greatly diminished. Analysis 20: Time Slice Graphs of Network Change Successive historical periods can be analyzed in longitudinal perspective and displayed in time slices that help to visualize their dynamics. Dating nodes by their historical generations allows us to display time-series graphs using the Pajek program. In Figure 10.9, graphs for four historical periods each having four adjacent generations are displayed graphically. They are left purposefully very fuzzy because our only interest here is not in the detail but in the size and cohesiveness or, in this case, the density of certain clusters compared to white patches that indicate separations between cohesive segments. The marital relinkings in the last of these periods (spanning those born between 1860 and somewhat after 1960) have little missing data, but they are clustered in a way that suggests that the cohesiveness of the clan as a whole may be breaking up. In the color version of these graphs (not shown), the nodes colored blue are those couples known to have emigrated. Hypothesis 10.5: There is a breakup of cohesion in the generation born after 1960 (Table 10.2 Period 6, Figure 10.9 graph 4). Table 10.2, in which for our seven historical periods four sets of up to four adjacent generations are considered provides a test of Hypothesis 10.5. Figure 10.9 shows the four periods from periods 3-6, shown in bold in Table 10.2, for which data are mostly complete in informants having memories of the four generations corresponding to that period (except completed marriages in the latest generation of period 6). Given the memories of ancestors in each period, changes in levels of cohesion can be compared, with some special cautions as to period 6. The smaller number of relatives remembered as we go back toward period 3 may not accurately reflect the complete demography of that period, because if early segments of the clan migrated out, there may have been elements of the population at earlier times who left no descendants to remember them. Still, the fact that the percentage of females recalled in each of these periods is roughly the same is a good indicator that there is little memory bias in the recall of these generations (because such bias favors Graphic Approaches to Nomad Solidarity 387 remembering males). Within the four periods, the percentage of female links is relatively constant, a good sign for data quality in that there is no memory decay for female ancestors going back in time through these four periods (this is not true for periods 1 and 2). Relinking and bicomponent percentages increase in the first three of these periods, but, consistent with Hypothesis 10.5, the last period in which data are relatively complete (the 6th) shows a drop-off not in the percentage of nodes in large (connected) components, but in the largest bicomponent, and the index of relinking. Table 10.2: Analysis of Kinship Cohesion in Bicomponents for Successive Historical Periods of Two-Four generations Period, # of nodes/ arcs 1* 10 / 5 Length of % Nodes in period & Fem-- Large Comdates ales ponents 2-17851835 2* 34 / 29 3-17851860 3 75 / 82 4-17851885 4 148/172 4-18101910 5 326/401 4-18351935 6* 686/800 4-18601960 7** 876/917 4-18851985 8** 812/685 3-19101985+ Nodes in Bicomponents Index of Relinking 20% 3(30%),3,3 0 n.a. 8 24% 19(56%),7,6 0 0 34% 57(76%),14 18(24%), 5,5,4 52(35%), 5,4,4 149(46%) ,4 220(32%) ,4,4 165(18%), 4,4,4,4,4 25(3%),20 ,4,4,4,4 23% 6 35% 140(95%) 42% 305(94%) 42% 668(97%),3 44% 44% 810(92%),21, 8,7 368(45%),24, 17, 16,15,12, 11,10 . . . 7 28% 5 35% 4 22% 3 10% 2 2% 2 * Cohort not complete (partial data). ** Extensive emigration (after WW II), circa 50 couples, but some completed marriages are lacking in the latest generation. Table 10.3 repeats the analysis for periods 3-6, removing those couples known to have left the clan and to have migrated to villages. The percentages of nodes in the bicomponent and index of relinking by period change very little with the removal of migrants, although the percentage 388 Chapter 10 of females drops (see Table 8.2 for relevant demographic changes). Table 10.4 again repeats the analysis, this time for periods 3-8 and the total population, with Feynman-simulated random marriages (avoiding marriages between siblings) in each generation. This allows us to test whether the drop in cohesion in the sixth period is real, or possibly due to females in the last generation who were so young at the time of last fieldwork that they had not yet taken spouses. We used the controlled simulation method described in Chapters 5 and 8 (Analysis 11). Without comparisons of actual data to a controlled simulation, the testing of Hypothesis 10.5 would require further fieldwork to determine if there is evidence of a breakup of the social organization of the clan in terms of social cohesion in period 6 (and later, as will be explored below). Figure 10.9: Continuity, Migration, and Fragmentation in Four History Time Periods (those marked in bold in Table 10.2) (The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads) Table 10.3: Analysis of Kinship Cohesion (with no migrants, if different) in Bicomponents for Successive Historical Periods of two- Graphic Approaches to Nomad Solidarity 389 four generations Period, # of nodes/ arcs Length of % Nodes in period & Fem-- Large Comdates ales ponents Nodes in Bicomponents 3 4 140/163 5 276/334 6 534/616 Index of Relinking 23% 4-18101910 4-18351935 4-18601960* 41% 133(95%) 47% 255(92%) 46% 515(96%) 49(35%),5, 4,4 126(47%), 4 177(33%), 4,4 28% 5 35% 4 22% 3 Comparing the actual Table 10.2 and simulated Table 10.4 results for period 6, the number of nodes, edges, females, and rate of relinking are all held relatively constant by the simulation method, and the comparison shows that the simulated random marriage results do not differ for either the size of the largest connected component or the largest bicomponent. Hence, Hypothesis 10.5 is rejected. There is in the actual data of 10.1.1 a significant drop in the rate of relinking, hence more people marrying outside, but controlling for the demographic increase of the number of single males in our genealogies after World War II (Tables 8.1 and 8.2), there is no decline in cohesion relative to a random Feynmanmodel of relinking, given the number of marriages, just more male links to nodes that do not represent marriages. The simulation, in fact, shows slightly higher rates of relinking in the two earlier periods. This is what would result from random marriages as compared to actual marriages in which there is greater self-selection among larger sibling sets and among stayers rather than emigrants. 390 Chapter 10 Table 10.4: Analysis of Kinship Cohesion in Simulated Bicomponents for Successive Historical Periods with Comparable Data as in 10.1.1 and 10.1.2 (all results here are Feynman-simulated data) Period, # of nodes/ arcs Length of % Nodes in period & Fem-- Large Comdates ales ponents 3 76/83 3-17851885 4-18101910 4-18351935 4-18601960 34% 4-18851985 44% 4 148/169 5 326/399 6 686/793 7** 876/916 * Nodes in Bicomponents Index of Relinking 71(95%) 31(41%) 21% 34% 138(48%) 60(41%) 26% 5 41% 304(94%) 146(45%), 4 225(33%), 4 35% 4 42% 657(96%) 6 22% 3 828(95%),8 193(22%), 10% 2 *810(92%),21, *165(18%), *10% 2 8,7 4,4,4,4,43 8** 812/682 3-1910 44% 496(61%),46,1 42(5%) 1% 1 * 1985+ 7, 15,12,12. . . *368(45%),24, *25(3%),20, *2% 2 17,16,15,12,11 4,4,4,4 * Reliability results given for a second simulation. Size of simulated large component has large error bound for period 8. ** Cohort not complete (partial data). Recent Breaks in the Cohesive Relinking of the Clan The visual results in Figure 10.9 (fourth panel, period 6) suggest factional lines of cleavage within the clan that would not have been evident to the ethnographer even in the field situation because cleavages are not always evident nor verbalized by informants. The graph of the fourth panel shows a possible cleavage between two sides of the clan, in terms of sparse connections between them. If we examine this cleavage in period 7, however, we still find thirteen marriages between the two segments. Because there are now fewer than four generations, the blood relations that connect the two segments include second cousins and closer relatives. We can examine this potential cleavage in more detail for time period 8, in which the older generation from period 7 (born after 1910) has died off, and the most recent generation of marriages has only begun. The blood relations that connect the two segments now include, because great grandparents have been omitted, only first cousins and closer rela- Graphic Approaches to Nomad Solidarity 391 tives. Hypothesis 10.6: There is a breakup of cohesion in the generation born after 1985 (Period 8). Figure 10.10 shows the four generations 5-8 in period 8, colored in the left panel blue, red, green, and just a few yellow, with some colored light blue in the right panel to show emigrants. Only two marriages (one each from the two middle generations) create strong blood or affinal ties connecting the two segments of the clan in this latest period. The 812 nodes of Figure 10.10 break up into disconnected components of sizes 368, 53, 24, 17, 16, and many smaller components, including 100 disconnected nodes. Closer examination shows that those in the leftmost pie-shaped wedge of this graph tend to be highly disconnected and most of those in this part of the clan (nearly 50%) have emigrated (see Hypothesis 6.1.1). For those clusters with stronger kinship ties, the largest connected component of 368 nodes, whose spatial dispersion represented the relatively closeness of their kinship cohesion in previous generations, spans more than half the nodes outside the wedge of emigrants. New marriages in the next generations (beyond the last period of fieldwork) will undoubtedly restore a good deal of the cohesive relinking among these families. The lower wedge of about seventy-five marriages, however, is presently connected only by two marriages to the upper right half where the majority of the clan is located in the graph. The middle (green) generation of children in this lower wedge are all unmarried daughters, who at this late date in the clan’s history have the greatest likelihood of leaving the clan for village life and a higher level of schooling. All the unmarried male children of the clan are located with the groups in the upper right of the figure. At this point we are at the edge of the contemporary marriage behavior of the clan, and it is the future marriage behavior of the latest generation that will determine whether the clan will fragment or remain cohesive. We cannot evaluate the outcome until after Johansen’s next period of fieldwork. We can, however, examine periods 7 and 8, those with still incomplete data on completed marriage histories of the currently younger generation, in comparison to the controlled simulation data for these generations. 392 Chapter 10 Figure 10.10: Kinship Disconnection and Migration in the Most Recent (green, yellow) Generations of Period 7 The color coded figure is found at http://eclectic.ss.uci.edu/images/TurkishNomads Table 10.5 lays out a summary, based on the results in Table 10.2 through 10.1.3, of comparisons for periods 3-8, including both actual and simulated results. The new variable that appears in this table is the index of relinking within the largest bicomponent or bicomponent in each period. In period 8 the size of the actual bicomponent goes down to 59% of that which would have resulted from random marriages. In periods 5 and 6, those with the best data, the sizes of bicomponents in the actual and simulated data are equal. Hypothesis 10.6—decline of cohesion in the most recent periods—is supported by the comparisons in Table 10.5. A drop-off begins in period 7 and continues in period 8. This appears to be a trend for declining marital cohesion, or marital fragmentation within the clan. If this trend continues, and less cohesive fragments of the clan choose to emigrate en bloc, as seen in period 6, the viability of the clan as an entity may come into jeopardy. This would raise questions of whether it will become too small to survive, its members will join with another pastoral group, and so forth. Graphic Approaches to Nomad Solidarity 393 Table 10.5: Simulation Analysis of Kinship Cohesion (Bicomponents) in Successive Historical Periods of Three-Four Generations Period and # of nodes/arcs Length of period and dates 3 75 / 83 3-17851885 4-18101910 4-18351935 4-18601960 4 147/169 5 326/399 6*686/793 Largest Component Ratio of Actual/ Random 57/71 = 80% 140/138 = 101% 305/304 = 100% 668/657 = 102% Largest Bicomponents Ratio of Actual/ Index of Random Relinking 18/ 31 = 58% 6 22% 52/ 60 = 87% 5 67% 149/146 =102% 4 74% 220/225 = 97% 3 76% 7** 876/916 4-1885810/828 = 165/193 = 85% 2 71% 1985 98% 8**812/682 3-1910368/496 = 25/ 42 = 59% 1 75%, 1985+ 74% 100% * First period to include extensive emigration (after WW II), circa 50 couples, but some completed marriages are lacking in the latest generation. ** Cohort not complete (partial data). While the cohesive core (bicomponent) of the clan is shrinking and fragmenting in the latest generation, the final column in Table 10.5 also shows that relinking remains high (over 70%—as in all periods 5-8) within the bicomponent. In period eight, there are two bicomponents (one of size 25, the other of size 20), each with over 75% rate of relinking. Hence, there is a fragmented core of members of the clan who are closely intermarrying, the larger involving lineages 1-4-5-6-9 and the smaller lineages 2-3-4-7, with lineage #4, in which “Dede” (597) has held clan leadership since 1980, holding the two together in a single component, the upper right half of Figure 10.10. With further marriages, these two bicomponents will clearly be knit together as the cohesive core of the clan in the future. The problems of clan cohesion and fragmentation have been discussed in relation to leadership in previous chapters. Figure 10.10 shows once again how two sets of couples in each of the smaller bicomponents in period 6 have left to settle in villages, and only one woman from a village has married into these bicomponents. If the emigrants are removed 394 Chapter 10 from these graphs, their departure fragments the smaller bicomponent of which “Dede,” the latest clan leader (597), is now a member. His bicomponent, however, will remain cohesive even with these departures. Summary Hage and Harary (1983, 1991, 1996) were the first to go beyond the limited applications of the network approach in anthropology in the 1960s and to implement a continuous series of projects of ethnographic reanalysis, in their case carried out on a comparative scale in Oceania that benefits from representation as networks of relations and from graph theoretic analysis. Our analysis of Aydınlı nomad genealogical networks has benefited greatly from their example, and it is one of the few analyses both of complete community-level networks over time and of the ethnographic data and background that puts the networks into context.9 The analysis in this chapter uses visualization of the cohesive shapes of networks over time employing contemporary 3-D graphic rendering. The nomad clan has a particularly strong conical pattern of cohesive topology, commensurate with the idea of a “single root” that often goes with the concept of the root ancestor of an ordinary lineage or ambilineage, but which, in this case, is associated with the density of relinking in a bilateral clan, in which there is also a core ancestor, Mustan, who has 80% of the subsequent generations of clan members as their descendants. This visualization prompted further analysis into the principles of ranking (Analysis 18), and whether Aydınlı nomads, and possibly other societies, had a collective structure to the clan that was comparable in some ways to the hereditary rankings of the “conical clan” (conical in view of its strictly core-periphery rankings) and conceptualized by Kirchoff (1955). And indeed, consistent with the loose age rankings that give the older siblings a privileged position in the family, we found that older brothers were much more frequently involved in the structurally endogamous networks of the clan than their numbers alone would predict. We shall say more about this in our next and concluding chapter. Besides our concepts and measures of cohesion, which come directly from graph theory (Harary 1969, White and Harary 2001), another purely graphic conception that Analysis 19 explored posed the opposite question: How do we define and visualize the relative sparseness of some of the segments of a network? Our graph-theoretic rendering of “thin” kin- Graphic Approaches to Nomad Solidarity 395 ship structures, in Figure 10.7, consisting of chains of relations with no branchings, gave a first inkling, by virtue of their absence, of just how important were “thick” kinship embeddings in Aydınlı nomad kinship networks. From there he proceeded to test the hypothesis that “thin” families tended to emigrate and “thick” ones to remain nomadic (Figure 10.8.1 and 10.8.2). Finally, the investigation returned to cohesion itself, this time through a scaling of cohesion in successive time periods (Analysis 20, intended to provide moving images of changing structures of cohesion), and a retake on how cohesion related to leadership over time (Analysis 17). The analysis confirmed the importance of cohesion in integrating the clan in the first analysis, and in the leader’s integration across the clan in the second; but, in both cases noting how clan wide cohesion begins to break up in the latest period of leadership, commensurate with ethnographic observations of a different and more town-oriented style of leadership. Further Reading The use of graph theory in ethnography is extensively discussed in Hage and Harary (1983, 1991, 1996) who in their seminar work on through ethnographic problems in Oceania have found graphic theoretic insights and analyses to be extraordinarily useful and revealing in each of the major topics of ethnography: exchange, marriage, politics and stratification, kinship systems, terminologies, ritual and myth, and more. Their discussion of Kirchoff’s model of the conical clan is the best introduction to that topic but also serves as the point of discovery of one of the underlying principles of systems of hereditary ranking. Freeman (2000) discusses the use of visualization and provides a gallery of on-line illustrations. The ethnographic contributions to Schweizer and White (1998) provide a wide variety of cases in which ethnography is combined with network and other kinds of analyses of decision making and exchange. Hèran (1995) and White and Jorion (1996), once again, may be revisited as an inventory of diverse principles, well beyond those in this book, for the analysis of kinship networks. Notes 1. As described in White, Batagelj, and Mrvar (1999), the procedure for computing generational depth uses Pajek options Partitions/Depth/Genealogical, 396 Chapter 10 which creates the partition, then Draw/Draw Partitions to activate the graphics page, then Layers/Type of Layout/3D, then Layers/In z direction, and finally Layout/Energy/Starting Positions/Given z. This initializes the drawing to 3-D and makes generational layers the z-coordinate. Layout Energy/FruchtermanRein-gold/2-D then scales the drawing in the x-y plane, and use of the x, y, and z keys rotates the image. 2. The description of Figure 6.9 gives the Pajek options for 3-D drawings. 3. The 3-D image in Figure 10.1 was exported for printing in 2-D to a Windows buffer by the PrintScrn key whereby it could be inserted into a word processor document or graphics editor. Scalable vector graphics (SVG) output can be made by Pajek for viewing and printing from the web. Complex ethnographic data in renderings such as these can also be viewed interactively in 3-D on the World Wide Web to explore the global structure of the network and its constituent details. A 3-D representation can be exported from Pajek into a molecular chemistry format for viewing in the Chemscape Chime Web browser (http://www-uk-midi.com/chemscape/chime/chime.html). It can also be exported from Pajek for viewing using virtual reality (VRML 2.0) formats that can be rendering by the VRML viewer in Internet Explorer or internet browser plug-ins like Cosmo Player (http://cosmo.sgi.com/cgi-bin/download. cgi/index.html). In the VRML browsers viewers, the viewer can click the nodes to see the names of couples and their attributes, rotate, zoom, or walk through the 3-D image. 4. Closeness to the central z-axis of the graph can become a new spreadsheet variable for analysis, although we skip this analysis because automatic drawings are not a particularly good scaling procedure for this purpose. 5. We have shown that an objective tracing of common ancestors is often possible through bilateral lines of descent rather than strictly patrilineal ones. One conceptual leap is to infer that one’s affines are probably related also by blood given generations of intermarriage. Another is to infer that because patrilines often segment and then intermarry, those who intermarry between the lineages of today may well be linked by patrilineal links in the remote past that have segmented into distinct lineages. Both of these conceptual leaps seem to occur with some frequency in Middle Eastern societies with segmented lineages. 6. The rightmost model of genealogical recounting in Figure 10.3 is not the way that Aydınlı conceive of lineage rankings, or rankings within the clan. It does, however, reflect the way that Johansen organized her genealogies, and could recount them when asked by the patriarchs, much to their delight. Johansen made ample use of the sibling set DFS principle in Aydınlı rankings. The spatial arrangement that she used to record the genealogies in her scroll, and the system she used later to number individuals (see Appendix 1), in which women will sometimes be next to (and have numbers next to) their husbands, and sometimes next to their siblings, corresponds to either the full or partial DFS principle shown in the figure. While the DFS recounting used in very hierarchical conical clan systems is not directly employed in Aydınlı men’s mode of recounting ge- Graphic Approaches to Nomad Solidarity 397 nealogies, it is not a principle with which they are unfamiliar. 7. Bates (1973:47) goes on to say: “Thus, Iron’s suggestion might have validity here in that genealogical amnesia likewise serves this function. . . . Another positive aspect of this is that flexibility in descent group restructuring is enhanced, a feature that will be shown to be well suited to highly variable camp groups.” This might be true for the Aydınlı but Johansen has no direct confirmation. 8. Michael Fischer notes that for Pakistani Arabs, when the eldest brother comes back after emigrating he cannot necessarily reclaim leadership, which is often a source of dispute. Rank is here partly a concept of priority, but largely one worked out through actual leadership. A returning elder brother may buy his way back into a leadership position if he has both the resources to do so and the requisite moral as well as behavioral authority. Our “Hınalı” Mustafa (=“Quarrelsome” Mustafa, 630), who returned from town as an adult, may have been such a man 9. Among the precursors are Brudner and White (1997) and Houseman and White (1998a). 399 Chapter 11 Overview and Conclusions Long-term field research has changed the face of anthropology. It has held up both change and persistence to be regular features of human society and has revealed the complexity of both. . . . It has brought new research questions to our agenda and pioneered new methodologies. It has acknowledged that such involvement over time transforms anthropologists, and so has anticipated the whole debate over the role of anthropologists and the impact of anthropology on the societies being studied. In short, it has changed the nature of the field. — Royce and Kemper 2002: xvi Chronicling Cultures: Long-Term Field Research in Anthropology Overview Chapter 1 asked why network analysis is needed in ethnography to develop an understanding of social processes. We distinguished between phenomena that are easily observable and require little adjustment to normal ethnographic practices and emergent phenomena that are not easily observable and that have important consequences for understanding social dynamics and historical change. We argued the case of cohesively emergent groups and the need for network analysis to identify the boundaries of such groups and to test for their consequents. We drew upon Firth’s contribution in distinguishing social structure and social organization as partly filling the gap between structure and behavior in developing the analytic concepts of anthropology. We concluded that there still existed an analytic gap in our ability to account for social process, and the types of emergent phenomena that we have studied in this book—such as structural cohesion—can be conceptualized so as to fill that gap between social structure and social organization and provide a link to the theory of complex phenomena that emerges through interaction. The research results of our successive chapters have amply supported the productivity of this approach. Our goal has been not to present a full ethnography based on years of 400 Chapter 11 fieldwork with the Aydınlı nomad clan but to present sufficient ethnographic data to place in an ethnographic context what we learn from a specific, extended genealogical network analysis, and to evaluate what we have learned. Placed in a processual framework for analysis, this includes our concern with the processes of ethnogenesis, transformation, and dissolution of a community or society. It also includes how we identify the various levels of cohesive social groupings that provide a principled theoretical approach to understanding segmentary and cohesive dynamics in societies in which social relations are highly mediated by multiple kinship ties. It is this very multiplicity of interactive levels and variables that provides a conceptual foundation for the study of social cohesion through the formal definitions and analysis of marital relinking (Chapters 2 and 5), structural endogamy (analyzed in Chapter 6), and changes in bicomponent or exocohesive structure (Chapters 9 and 10) over time. We show how to use the study of changing practices to investigate emergent or changing rules, groups, and norms. The nomad clan is a particularly relevant case for this analysis because kinship and networks of marriages alliances (Chapter 7) underlie many of the facets of its social and political organization. Figures 2.2 to 2.5, introduced in Chapter 2, build on our identification of the clan founders to form a detailed computer-drawn genealogy that locates all the relinking marriages in the clan by their lineage. Those figures have been annotated to show links between lineages through the wives. The genealogies corroborate what older informants had said about the early histories of the clan, as also confirmed by independent historical sources. Our argument that the relinking marriages knit together the entire nomad clan as an alliance network is supported by evidence in subsequent chapters—Table 6.4 and tests of Hypotheses 6.1-6.5 and of successive refinements of Hypothesis 9.1, 9.3, 9.7, 9.8, 9.11—that central leadership positions within the clan are established by the strength of their position within the relinking network (Chapter 9). This case study has also been fruitful because societies with “Arab” type FBD marriage rights have been problematic in studies of the Middle East and have posed theoretical challenges both to theories of lineage structure and to theories of exchange and marriage alliance. Chapter 4 introduced the theoretical problems of FBD and segmentary lineages, Chapter 7 led to the discovery of some new principles, consistent with complexity theory, for studying and understanding segmentary lineages, and Chapter 10 developed a model of a flexibly ranked clan system with segmentary endogamy (“endoconical”) that may help to raise new prob- Overview and Conclusions 401 lems about Middle Eastern social organization. Chapter 8 showed how we might study processes of social change more profitably with data from long-term fieldwork. One of the fundamental advantages of a computerized social analysis of genealogical data is that the structure of the entire genealogical network can be examined both in toto and in its respective parts. One of our first uses of the genealogy was to locate the leading known-persons whose influential personalities provided informal leadership to the clan during succeeding generations. All of these men, it turns out, had relinking marriages and so could be located on the reduced genealogy of relinking marriages (Figures 2.2-2.5). The first part of this concluding chapter will discuss the advantages of long-term field research and the problems engendered by the enormous amount of data that may result from such fieldwork. We will indicate how these problems can be solved at least partly by network analysis. We also summarize some of the ways the deepening of the ethnographic results have been achieved by the study of networks in relation to other data and evaluate the methodological and theoretical approach of network analysis. Dynamics: The Long-Term Findings Long-Term Ethnography versus Longitudinal Analysis Long-term field studies (Royce and Kemper 2002, Foster, Scudder, Colson and Kemper 1979) are carried out by many anthropologists, if we take as our criterion systematic returns to a field site to collect data. They usually involve rethinking how data that were initially collected in one time frame begin to look very different, even with a lapse of one to two years of fieldwork, with a multiple time perspective on social change. Longitudinal research takes a further step, and it is not synonymous with but rather builds upon long-term field research (Johansen and White 2002:81). It entails the capacity to track samples of cases through time, and to link families from one time period to those in the next. This places our understanding of people’s lives within the framework of genealogical and often migratory succession, in which the new generations that follow the old are not simply a new set of autochthonous individuals, as a Western perspective might have it. People’s embedding in a cultural mesh of kinship and culture is rooted in concrete networks, and 402 Chapter 11 these are themselves deserving of study if we are to understand the strategies of cooperation and competition—and the results of interactive encounters and contexts upon their opportunity spaces—as people seek to strengthen the place of their families, groups, and ways of living within larger societal and changing global frameworks. While we might achieve an understanding of people in a given context in terms of the choices they are likely to make in that context, when they move out of it we know far less about them. For that, we need to follow migrants and movements, only some of which we have been able to do here. Long-term field research has often been associated with the rise of new methodologies. Many long-term field research projects attempt to collect what has been termed minimum core data, including “detailed maps, censuses, and regular surveys to update the situation of households and their members” (Royce and Kemper 2002:xx). These in turn produce enormous data sets. This allows for the periodization of the data into time spans that can be compared and contrasted to facilitate analysis of change as well as of stability in networks. Such data bases can also be compared with various informant reports at different time periods. The potential richness of the database and the very specific long-term field experience of the researcher in capturing or at least discussing aspects of the data in great detail can also, in turn, stimulate new questions, give rise to puzzles, and ultimately new methodological practice and measures. Our argument, as elsewhere (Johansen and White 2002), is that most anthropologists who conduct long-term research, as a necessary mode of knowing the people with whom they work and the contexts of their lives, also collect systematic genealogies and personal histories. Such collections of qualitative data can provide the basis for systematic network research, and the reverse. These data, including biographies and genealogies, are in fact a central part of the record of a particular social group’s history. The need for such an approach is consistent with the emphasis in anthropology on understanding the history of social groups in their relationships to nation-states, streams of migration, colonialism, and globalization, and with the histories of ethnic groups and their interrelations. Lévi-Strauss (1949:#-#) argued that the kinship networks were so impossibly complex that the only way human beings could conceive of them was through normative abstraction. Norms, rules, and roles were what people carried in their heads and how they organized their behavior, so goes the argument, and the analyst was in no better position than people themselves to understand in detail how such abstractions related Overview and Conclusions 403 to the complexity of networks. Today, far from being intractable, data on kinship and social networks constitute a viable analytical object—an entire network being a veritable new object—for the study of marriage practices and how behavioral contexts are situated in terms of networks. The emergent forms of social organization that result from social practices in their concrete form—who marries whom, the formation of families, who stays or migrates, and so forth—give us keys to understanding, through network analysis, a whole bundle of social processes. Kinship networks constitute a central element in social organization. The ability to analyze kinship via longitudinal network analysis benefits significantly from longterm field study. The longitudinal element of tracking individuals over time, when coupled with tracking a genealogical network, and, often, connections to land or property, allows us to study processes crucial to understanding change and transformation, or conversely, stability through time. Ethnogenesis of the Clan: Understanding Dynamic Complexity In tracing the process of clan amalgamation we have seen the principles of adaptive radiation of nomadic tribes, the fusion, differentiation, and manipulation of tribal identities, the segmentation of lineages and their amalgamation into clans, the attachment to villages or return again to nomadic life. How are these divergent pathways stabilized to reproduce a socially and spatially organized clan? Aydınlı nomads do not fit easily into the standard concept of an endogamous clan, technically called a deme: “communities revealing a marked tendency toward local endogamy but not segmented into [exogamous] barrios” (Murdock 1967:48). The concept of clan as putative descent group is somewhat misleading. The nomad clan is as much based upon affinity—or, to be precise, structural endogamy or marital relinking—as upon putative descent. Putative descent, here as elsewhere, is often the normative expression for structural endogamy as registered in local discourse. Hence, the anthropological concept for this type of clan has been embraced under an emic rubric of “from a single root” or a putative ancestor, a definition that may have local salience but is not a defining characteristic unless we also take marital relinking into account. What we have seen for the Aydınlı nomads, with structural endogamy in the context of a clan that lacks exogamous segments, is that dense cy- 404 Chapter 11 cles of intermarriage produce in subsequent generations an increasingly greater density of common ancestry or ancestries (Hypothesis 6.3). If the clan is an entity with a single leadership or corporate existence, as we have seen, the genealogical morphology is likely to take a conical form, with one or a very few densely intermarried ancestors as its discursive “single root” (Figures 6.10 and 10.1). Simulations of population dynamics by B. Derrida et al. (1999) clarify some of the fundamental emergent properties of kinship networks. Endogamous populations with random variations in number of offspring, normally develop “common single ancestors” that are statistically dominant over other candidates (following an exponential decay distribution) for a majority of a population at intermediate genealogical depths (5-9 generations) and modest population sizes (two to twenty thousand). Beyond the fact that the marital practice of endogamy itself helps to create “root ancestors” in a population that keeps track of genealogies, the role of salient ancestors in helping to bridge the lineages by brokering marriage is particularly interesting in terms of clan identification with a “common root.” There is no literal single founder of the clan, but the clan cohesion created by Koca Mustan (716) in generation 3 of lineage #2 gives the clan its ancestral unification. Examining the genealogical data on the number of current descendants left by each founder (Analysis 4) yielded Mustan as a close approximation to a “single root” of the clan. Hence, informants’ statements of a single root of the clan, when distinguished from the concept of a single founder, are basically correct. Mustan played the role of a relinking broker in establishing clan cohesiveness. His ability to create marital relinking through his children’s cohort (along with their allies in lineages #1 and #3) provided the coalescence of the clan as it migrated to and amalgamated in its new pastures to the east. By so doing, Mustan and his wife became an ancestral root for a great majority of the clan. Their respective parents are two of the identified founders (#2, #3), and their daughters link through their marriages to lineages #1 (98, the founder son of Hacı Dolaşıklı, 28), #3 (1169), #5 (343, grandson of founder Koca bey), #7 (1230), and #1 (630—a grandson of the founder). Mustan’s lineage became the largest and the only one that intermarried with each of the other main lineages in the clan. The decisive alliance and relinking of lineages #1-#2-#3-#5#6 takes place among the children of Mustan’s children’s generation and the next (adding lineage #4).1 Among the types of marital relinking occurring in this epoch of clan consolidation, the exchange of sisters and recurrent wife-giving from one lineage to another were key to the con- Overview and Conclusions 405 solidation of clan cohesion and subsequent ancestral “rootedness.” This is an ethnogenesis where endogamy was present at the start and grew with the growth of the clan.2 Clan Cohesion, Segmentation, Expulsion, and Transformation Computer analysis of the properties of genealogical networks helps to identify the boundaries of cohesive groups in a society, and to measure the degree and structure of cohesion in various subgroups. The frequency of marriage ties among lineages is one aspect of subgroup structure that generates overall cohesion. Clan cohesion involves a continual balancing in each generation of processes of segmentation and amalgamation. As we saw in the Chapter 4 background to understanding variations in FBD marriage practices and the analyses in Chapters 5-8, there are strong pressures in both directions. The smallest lineage segments (3-generation patrilineages) are the effective production units that compete for survival, the fertility of nomads in healthy conditions is high, and excess population is sloughed off, with those units unable to meet their production needs from herding being more likely to emigrate to villages (Bates 1973: Ch. 7, Barth 1964, Bell 2002). There is also, as we saw in the demographics of Chapters 6 and 8, a small but significant differential probability for those with fewer agnates (sons and brothers) to emigrate to the villages. Because minimal lineage segments are the units that must be sloughed off, segmentation of patrilineal groups over time into cascades of descending sublineages is a continual process. Allies within the larger segments and maximal lineages are important in the competitive process, as are marital allies in other segments: access to pastures and allocation of herding rights depends critically upon negotiations carried out at the level of such larger segments and, at the level of the clan itself, on decisions taken at the meetings in the large tent of the tanıdık kişi. Segmentation is illustrated by the accounts of a segment of lineage #1 leaving the clan, continual fissioning of families as couples or individuals leave for settled village life or marry into another nomad tribe, and the tendency for some of the lineages to dwindle in size or die out. Analysis 5 shows a further segmentary principle, namely, that in the competition for leadership between the larger lineages there was a slight tendency to reinforce claims to leadership and perhaps to heighten one’s own 406 Chapter 11 lineage ranking by not giving daughters to other lineages and by greaterthan-chance endogamy within the larger lineages (Hypothesis 6.5). In the competition for survival one of the means by which lineage segments at various levels are selected against, besides economic competition, is by the social exclusion of others not being willing to ally through marriage. Hence, those who are excluded and who settle in villages (and we have seen several examples) are also more easily culled from the genealogies by selective processes. Relinking behavior is thus a kinship practice that instantiates cohesion as the opposing force to exclusion and segmentation. Analysis 16 (Table 9.6) shows that the frequency of relinking by leaders and their offspring is much greater than expected by chance. This is compensated on the integrative side, however, by near-random distribution of marriage between the larger lineages and by the tendency of smaller lineages to ally more frequently with larger lineages within which they become in a sense amalgamated. Further analysis of intermarriage among relinked tribes (Table 6.5) shows something of a larger structurally endogamous network (Hypothesis 6.2). The marriage of women into the clan from other tribes, including more socially distant tribes not relinked by marriage (Analysis 3), provided a source of recirculating personnel familiar with nomadic ways of life. The clan amalgamated members not only of one dominant tribal identity but also occasional members of outside tribes, mostly village shepherds adopted by a nomad sublineage. The formative links among clan members prior to the eastward migration incorporated a series of new lineages through (or perhaps just continued existing patterns of) marital relinking (Figure 6.7). In our evaluation of Barth’s hypothesis (Chapter 6), both fissive and integrative elements are seen to operate at the level of competitive leadership: FBD marriage reinforces close family support but keeps the core kinship support group tight (Figure 7.1). The need for highly capable leadership under nomadism serves to stress competition and personal characteristics. Wider network recognition—as well as core factional support—is broadly integrative but without a coercive basis of social control. This can also lead to factionalism and feuding (sometimes with struggles between tanıdık kişi as their focus), which a good Hacı (older and pious pilgrims to Mecca) or at other times an established tanıdık kişi as a peacemaker can help to mediate through lengthy negotiations in which everyone involved gets an opportunity to express their frustrations. While the raw frequency of FBD or patrilateral parallel cousin Overview and Conclusions 407 marriage declines in prevalence over generations, its selective percentage with available relatives in this category does not decline (Hypothesis 8.2.3, Figure 8.3). This is one of the most significant differences of our computer analysis from the conventional forms of frequency analysis of cousin marriages. For the period 1875-1965, the period with adequate statistical data, the percentage of kin in the FBD category who are taken in marriage actually rises from 25-28% in the earlier period to 30-33% in the later. The drop in absolute numbers, then, is because there is a greater proportion of persons resettling in villages and perhaps because of smaller families and sibling group sizes (Hammel’s Principle of Demographic Bias). This drop is compensated for by a marked rise in the percentage rates of patrilateral parallel second cousins, which are also lineage mates. The much higher numbers of male linking relatives in consanguineal marriages (table on request: a standard output of the Ego2Cpl analysis) also supports the pattern of favored marriages within the paternal line. Enlarging the Concepts Concerning Clans We explored the question of whether the Aydınlı clan system constituted a variety of what Kirchoff (1955) and others have called a conical clan, but which lacks exogamy as a salient characteristic. The conical clan systems described in the ethnographic literature (Hage and Harary 1996), often associated with politically hierarchical societies, have coreperiphery structures clearly demarcated within genealogical rankings that are ordered from top to bottom by relatively strict ranking systems of primogeniture or ultimogeniture, often with wife-taker versus wifegiver inequalities playing the role of a more dynamical status-altering element. Instead of Kirchoff’s conical clan, we developed in Chapter 10 the idea of another type of clan which we named endoconical: one with a ranking dynamic based on marital relinking in which leadership and other forms of participation in the life of the clan co-vary with a looser conception of age or status ranking. The gradients of endogamic cohesion within the group are elements of loose relinking (along with relative age among siblings, number of spouses and affines, etc.) that follow from the cohesive behavioral practice of an individual’s marital relinking. Compared to Kirchoff’s conical clan, the endoconical clan has greater gender equality, given that it is based on bilateral descent and a lesser degree of 408 Chapter 11 inequality because rank is a partly achieved characteristic rather than a strictly prescribed one. Decline: Demographic Changes through Outmigration and Continued Viability of Nomad Social Forms Tribal labels, while recognized, have less political significance in the twentieth century than they did up through the nineteenth. Political positions that had once been occupied by organized tribes ceased to be recognized by the Turkish after World War II (Chapter 4). Clan and lineage, however, have continued as important organizing principles, although only among nomads. Still common is the preference for FBD marriage by young men as well as their fathers, who still tend to arrange marriages. In addition to expressed norms, however, our study has emphasized social practices in marriage choices. From the decline in prevalence of FBD marriages one would conclude that the agnatic lineage is in decline. We consider it a mistake, however, to take changes in the prevalence of FBD marriage as an index of the viability of a lineage system in a Near Eastern society. Having a computer-based method of analysis makes available a more precise analysis of behavioral choices in marriage, that is, by selective baseline analysis of available relatives of different types, the numbers of which will change under different demographic regimes. There are several advantages of this type of measurement, which we use to examine the rates of different types of blood marriages and affinal relinkings. Nomads are adapting to changing demography and outmigration pressure, partly as a result of population growth in Turkey generally and within the clan itself. Declines in prevalence of FBD marriage in the late twentieth century are partly a result of fewer co-resident father’s brothers’ sons and daughters, whether from increased migration or smaller family size. The selective percentage rate used in our computer analysis of marriage choices has a great advantage in that it controls for the significant demographic factors confounding the raw frequency or prevalence rates of marriage choices. Because the selective percentage rates of FBD marriages do not decline (normed on the number of FBD relatives available for marriage), it follows that the strength of the preference for FBD marriage may not have declined either. From our results one would not infer that the traditional lineage system was in decline (Hypothesis 8.2.1, Overview and Conclusions 409 8.2.3). Indeed it is as if to compensate for the loss in absolute numbers of available FBD relatives that the selective rate of marriage with patrilateral second cousins has risen in the last two generations (Table 8.4). Far from demise, the nomad clan repopulates sedentary villages with its offspring as the core population grows for its given region, but it has managed to keep a relatively stable population in the ecological niche to which nomadism is adapted and to keep its social institutions viable. Barth (1964) makes similar observations for the Basseri goat- and sheepherders of Iran. One of the fundamental elements in the continued viability of nomadic clan is the health and associated fecundity of the lifestyle: almost all of its members (except the very poorest, who may be eventual emigrants) have many children and bring them to maturity given the quality of their diet—with many dairy products, including yogurt in summer and cheese in winter, considerable protein, the collection of plants, mushrooms, peppermint for teas, and the easy availability of mulberries, grapes, and raisins by purchase. There are few miscarriages and relatively little difference in the survival of babies among different families, all being born in tents and generally receiving two years of maternal care and milk before the next child is born. Women bear children between 16 and 36, and 10 children per woman are not atypical, 8 being considered normal. Even the considerable walking (but not overwork) that women do as part of the nomadic way of life is good for averting stillbirths during pregnancy and keeping mothers healthy. There are quantitative indications, however, of changes in clan and lineage organization. A shift from an agnatic lineage-based pattern of cohesion supporting political leadership (Hypothesis 10.6) is underway in the latest political period (post-1985: Table 10.2) and evidence for a breakup in clan cohesion (Table 10.5), although further fieldwork would be required to see if patterns of relinking will continue to replenish a cohesive core of the clan or continue to allow further segmentation to occur in the cohesion structure (Figures 10.9 and 10.10). The evidence on leadership patterns and social cohesion point to changes that begin in the 1980s. This corresponds to the period of leadership under Mustafa (597 “Dede”, lineage #4) when the basis of political support was also seen to change. “Dede” himself was shown to lack membership in a cohesive giant bicomponent but instead was embedded only within one of two factionalized bicomponents (see Figure 6.5). There is a triangulation of results on our inferences about social change in the basis of leadership support in that “Dede” also lacked a basis of 410 Chapter 11 wider support throughout the clan through the distributed cohesion of his lineage (Analysis 17). The 1980s was also the period in which feuding ended, suggesting again social structural changes, from quasi-corporate factional groupings such as lineages to the personal basis of leadership support that we saw in Dede’s leadership (Hypothesis 9.13). “Dede” was a transitional figure and was also highly respected by the outside world, called by the sedentary population of the neighborhood also by the special term of “patron.” He was at the same time official mayor of the village located at the edge of the summer pastures, head of a transport union in the town of Kozan, and an owner of a citrus tree plantation. The Anthropology of Relations Norms and Behavior: What People Do and What People Say Computer analysis of the intricate patterns and levels of contextual embedding in which individuals make their choices is capable of making a precise evaluation of the relation between expected norms and actual social behavior in marriage choices as can be identified from genealogically recorded data individual by individual. Informants, for example, reported that spouses are generally chosen of the same age. Although age data are lacking, Analysis 8 examined the statistical tendencies toward or away from consanguineal marriage of the same generation and found strong evidence of generational symmetry (Table 7.5 and Table 7.6) commensurate with the stated norms and those of the Qur’an (Hypothesis 7.10). A closer look, however, showed that norm compliance was mainly within lineages, although one lineage constituted an exception even to this.3 A more general conclusion lies at the base of our FBD example, however, as regards norms and behavior, namely, that conventional analyses of marriage preferences are inadequate without properly normed percentage data for observed marriage choices (see, for example, Hypothesis 8.2.3). Given demographic constraints on relatives available for marriage, inferences from frequencies alone as to marriage preferences of rules are likely to be invalid. We have found common variance between prevalence and selective rates for different types of marriage across a range of societies (see, for example, Table 8.3). In this light the entire field of analysis of marriage systems may benefit from reanalysis with newer methodologies. Overview and Conclusions 411 The analysis of different types of marriage, marital relinking and exchange marriages can be based, as in the present case, on rates or comparisons that do control many of the confounding demographic variables that make solid conclusions from such analyses so difficult. Hence, this book may provide a more fundamental theoretical reorientation as to how we norm behaviors in a statistical sense before comparing them with verbal norms. Because marriage systems have been at the center of a great deal of work in anthropology, how we measure behavioral orientations makes a great difference to our results. As an example of this fundamental change in orientation we may contrast our approach with that of Barth (1953:72). His model of the processes that linked the resource base in nomadic societies to egalitarian leadership began at the level of small, tight, kin groups such as the lineage, conceived of feuding as the expression of political power and of local leadership as based on support through kinship alliance, including FBD marriage. We extended his argument to another level in our study to show that higher-order clan endogamy and relinking marriages are the basis for informal leadership, factionalism, and mediation in feuding. Barth, who spent only a short time in the field for his study, could not have perceived the complexity of network structure above the level of the local group. Structural Endogamy as Exocohesive Connectivity Reproduced in the Practice of Marital Relinking: Social Inclusion and Exclusion Using the concept of structural endogamy, the large-scale social boundaries of self-reproducing groups can be identified by means of computer analysis. Structural endogamy is the empirical phenomenon in which the intermarriages of an arbitrarily large set of individuals—arranged in couples—form a set of marriage cycles that are self-enfolding. Every couple is connected to every other by one or more circles of kinship and marriage relationships. The index of relinking used to evaluate these overlapping circles is an index of the social redundancies by which members of a community cohere through multiple paths of connectedness. The maximal boundary encompassing those couples included within these overlapping circles is the region in which endogamy is extended within the social group, a boundary that is self-defining or emergent through marriage behavior. 412 Chapter 11 What we demonstrated at a societal level commensurate with the nomad clan are fundamental principles of social inclusion and exclusion that operate jointly through the practice of marital relinking, which in its absence constitutes a basis of social exclusion. We showed empirical support (Analysis 2) for structural Hypothesis 6.1 that clan membership is based on relinking. Those couples who stay with the clan are relinked through marriage, while those who are not relinked are those who tend to decide for settled life (Hypothesis 6.1 and Table 6.4). Those who are not relinked are vulnerable to exclusion from the social life of the group, lacking a spouse with access to and knowledge of nomadic ways of life, and in consequence they are typically not taken into consideration in invitations to weddings, funerals, and so forth. What is striking about the concept of relinking is that large-scale but definitively bounded social groups may be formed that lack a unifying descent principle but are linked through affinity. Thus a principal argument for our use of p-graphs and computer analysis of marital relinking has been that the relinking marriages knit together the entire nomad clan as an alliance network. Not only are central positions within the clan established by relinking but the emergence of the clan itself and the varying cohesion of its subgroups is a function of the cohesion provided by marital relinking. This theory works particularly well for the egalitarian nomad clan, in which intensive cooperation is required at multiple levels, but cohesiveness is poised in opposition against continual segmentary and factional tendencies because there is no centrally binding authority to mediate disputes. Network analysis and the study of relinking as a dynamical basis of cohesion is potentially a better fit to seize on such social groups and their shifting social boundaries than are conventional structural models. Every marriage that creates a new relinking creates a locally cohesive subgroup. This can be in terms of either a consanguineal marriage, in which relinking adds a new connection between relatives who already have a common ancestor, or a relinking affinal marriage. The latter adds a new connection between already affinally connected relatives. Subgroup Analysis One of the most fascinating set of findings of the computer analysis results from applying new concepts about social cohesion (White and Harary 2001) to subgroups within the structurally endogamous boundaries Overview and Conclusions 413 of the clan. Analysis 15 gave the following findings: Cohesion between pairs of couples within the clan as measured by the number of independent paths between them is hierarchically structured in terms of a series of embedded units of increasing cohesion. Social units identified with different levels of cohesion created by relinking within the last 150 years are associated with either a particular tanıdık kişi (=known-person) or a failed aspirant to tanıdık kişi status. In one case such a unit is associated with a religious notable who is a son of a founder and father of a tanıdık kişi). All but one of the seven tanıdık kişiler (=known-persons) from four of the five main lineages4 and the one aspiring tanıdık kişi are within the N=110 couples who occupy different positions in these structured cohesion levels (p=.003). The salient characteristic of the cohesive units is that they are not clumped at close distances like cliques. Rather, each of them spans the other subgroups of the entire clan. They differ and are detectable via eigenvectors analysis, because they have a uniform number of independent paths between members, which also implies characteristic differences in density. The structure of these cohesive units is concentric, by differing intensities or characteristic number of independent paths between members. Over time, it appears that the characteristic levels of cohesion that distinguish each successive tanıdık kişi coming to occupy the principal role of leadership in the clan are decreasing. However, this is a function of the greater size of sibling sets as we go back in time and the fact that the larger sets have a slight tendency as well to remain nomads rather than to emigrate. The size of cohesive sets and the number of independent paths that give them cohesion are by definition in complementary distribution. A successful candidate for tanıdık kişi can have a small but intense following whose members span the clan, or a large and less intense following, or somewhere in between. In any case the members of the cohesive set must span the clan and cannot simply occupy a solidarity group with very close ties that exclude other subsets in the clan. Simultaneous and competing aspirants to the role of tanıdık kişi do appear, each associated with a cohesion group, sometimes at the same level. The limitation to a single occupant of the role is not by exclu- 414 Chapter 11 sion but by the fact that people will tend to throng to a single tent, constructed so as to hold sufficient occupants, that serves as the site for a discourse among the majority of men. While an aspirant can hold smaller meetings in another tent of sufficient size, if and when the number of attendees dwindles to very few, the aspirant will desist and sometimes deign to attend the meetings of his contender. The concentric structure is not perfectly unidimensional; rather, there are oscillating sides of the concentric structure that appear to compete and take turns in terms of emerging leadership. A static image of clan structure in terms of cohesion and leadership is shown in Figure 11.1. The eccentric circles represent levels of cohesion and the letters represent leadership positions (not in any particular historical order, but see Figure 9.4). They center around the highest cohesive core but also differentiate into opposing factions each of which tries to span through multiple independent connections the whole reach of the clan. Figure 11.1: Embedding of Successive Exocohesive Groups in Clan Structure and Support for Exocohesively Embedded Leadership e c a b d f (The nonsequentiality of lettering is intended to call attention to oppositional segments) All of these aspects of how cohesion is related to leadership were hypothesized and tested prior to our construction of attribute-level variables to test predictions about which of all the potential candidates for emergent leader (the single tanıdık kişi in each generation). Those tests (Analysis 16) showed a multigenerational process in which an entrepreneurial father and a mother from a well-off family may emerge as one of the richest families in their generation and their sons become candidates Overview and Conclusions 415 for tanıdık kişi provided that they are sufficiently numerous, retain nomadism, and practice relinking marriages. The son must have the character, economic success, and respect associated with achievement within the contemporary economic setting to become the tanıdık kişi and should be viewed as experienced and intelligent. Ideally he should be the eldest of the sons who have not emigrated. As Johansen (1999) has written, however, the principle of seniority could survive in the fragile nomadic societies only because it was employed in a flexible way. Only Kozan Mahmut was an eldest son but not Veli Kahya, Hasan bey, Erkek Mustafa, Fındıklı Hacı, or “Dede.” Ambitions for leadership were greatest in the families of former tanıdık kişi. The candidate must be able to equip one of the larger tents associated with large and wealthy families; he must be able to afford several wives, which are an asset in producing more married sons, endowing daughters with bride payments for further marriage alliances, and providing the labor needed to host large groups in a large tent for discussions. It was an outcome of competition but not a preference or stated rule that the tanıdık kişi come from a lineage not having had a recent turn in the rotation of leadership among competing lineages, as in the opposing segment structure envisaged in Figure 11.1. The attributes and the network variables of kinship cohesion fit together in a fundamentally social process that involves an intricate competitive and cooperative pattern of emergence and synchronization, in which behavioral practices and goal-driven choices interact with societal values and the judgment and respect of others. An apt label for this kind of emergent process is cohesive social practice, which includes the perceived personality and character critical for leadership. Leadership and the Political System A more dynamic account of how nomad politics operates would shift the focus from a static model such as shown in Figure 11.1 and the attribute models of Analysis 16 to a moving image that is suggestive of the processes that operate across generations. What follows is an attempt to depict a more dynamic model, diagrammatically, in Figure 11.2. Through successive generations emergent loci of leadership slowly rotate around the clan center as different segments of the clan are articulated to the center. Given a clan with a relatively stable center, each successive leadership faction after the obtaining eastern pastures after 1875 416 Chapter 11 is focused on a different lineage but also links both to the center of the clan and reaches out to other subgroups. Rotating around a central axis of clan cohesion, then, as in the figure, shifts of leadership over time sweep through different segments or factions of the clan to augment overall political cohesion of each successive lineage group and their closer allies with both the clan center and its peripheries. The operation of this political system is dependent upon multiconnectivity as the basis of a form of social cohesion that is distributed throughout the clan, linking most clan members both to the current but changing leadership and to the relatively more permanent central social figures of the clan. The triangular wedge in Figure 11.2 show by double dotted lines represents the lineage or faction of the current leader, which may include links to central clan figures by lineage brothers or sisters. A larger political faction of the leader is shown in the figure as the off-center oval that contains lineage members plus close allies. The arrows represent high inclusive multiconnectivity out of the larger leadership faction into other sectors of the clan. The sweeping arrow represents change of leadership from sector to sector and lineage to lineage over successive generations. Figure 11.2: Dynamics of Clan Reproduction in Terms of Cohesion, Social Embeddedness, and Leadership Clan allies center leadership faction Key: designates omnidirectional or diffuse cohesion through multiple pathways of connection throughout the clan A central concern in our search to understand emergent leadership has Overview and Conclusions 417 involved the measure and importance of different types of connectivity. There are, of course, two tendencies as concern leadership. A tanıdık kişi tries to keep leadership in his extended family but competing lineages try to support the striving of another able man to become tanıdık kişi as soon as the current incumbent shows inability or weakness. Hypotheses 9.1 through 9.11 speak to different aspects of these two entwined processes within this general model. The special role of multiconnectivity in distributed but interpenetrating cohesive groups is identified in Hypotheses 9.2, 9.3, 9.7, and 9.11. The theory we advanced about marital relinking in relation to leadership is that multiconnectivity is essential to large-scale social cohesion, which can include the basis for social formations not only of kinship groups and clans but of social class as well, as Brudner and White (1997) have noted. Where the multiple connectivities of a social group extend into central political positions, this theory can also provide an account of the association of political power in relation to cohesive social groups with boundaries delimited by the limits of multiconnectivity. In the present case political positions are themselves emergent out of the cohesiveness structure along with rules of turn-taking between social segments. There are still pieces of our analysis of emergent leadership in Chapter 9 that need to be put together. Figure 9.4 showed how it was likely, in spite of missing data from the early generations, that the cohesive embedding of leadership groups declined in succeeding generations as the clan grew in size and differentiated into looser networks. Cohesiveness of nodes in the structurally endogamous core declines over time both for the whole population (tau-b=-.302, p<.001) and for leaders. Our multiple regression analysis attempted to predict emergent leaders from both attributes and network variables such as cohesive embedding. There we found that when cohesive blocks are taken into account, network centralities (variables for degree, flow, betweenness, and closeness), which are often taken to predict leadership (Hypothesis 9.8.3), had little or no additional predictive value. But level of cohesion of leaders was only above average and, while statistically significant, was a poor predictor in terms of variance accounted for, especially after controlling for the attribute variables. Failure of global cohesion variables to predict emergent leaders more precisely may be due to missing data, failure to measure cohesion within the relevant generations rather than globally, or misspecification of our hypotheses. Still, the attribute variables that did predict emergent leadership were 418 Chapter 11 the very attributes that translate from the individual level—more brothers, more married sons, more wives, a particularly well-connected mother and father, and so forth—into a successful local practice of behavioral cohesiveness within the specific time horizon. Given these local attributes it is virtually certain that within their own generation each emergent leader would be a highly cohesive member of their cohort. This is an effective demonstration of our thesis about cohesive practice and it obviated the need for breaking the datasets into different time periods and measuring node cohesion in each segment, an approach that can suffer from misspecification of the generational boundary conditions that are associated with the position of leaders in terms of cohesiveness. Were we to use this approach it would have to be a fractal analysis such as we used for the study of marriage alliances between lineages and cascading lineage segments, namely, to create all possible generational time-units and measure the effects of cohesiveness for each unit that leaders are members of to see how such units differ. But because the leadership dependent variable consisted of only six cases it is probable that even that method would not be efficient in testing further hypotheses. In the analysis of time-segments of the dataset in Chapter 10, as shown in Figures 10.9, 10.10, and Tables 10.2-10.5, we also saw that the cohesiveness of support for the most recent leader appears to be unraveling with the change in style toward a more village- and town- oriented leadership, partly due to the increase in outmigration that makes a village orientation all the more important. Whether the social cohesiveness that was traditionally endogenous within the clan and adjacent tribal segments will now reconfigure to embrace a broader community that spans also the villages where nomads have settled is something that only future fieldwork can resolve. It is also possible that if more of the larger sibling sets remain nomadic, while the small ones are sloughed off, the original concentric cohesion pattern could reassert itself, but that again would require continuing the long-term research project into the future. Measuring Relinking Density and the Porousness of Structural Endogamy When the quantity of marriage circles reaches its maximum of 100% within a social group as measured by the index of relinking, the group is fully endogamous in the canonical sense of a caste: everyone marries within the group. When the relinking density is less than 100% there are Overview and Conclusions 419 still clear-cut boundaries of the largest structurally endogamous unit of a society but the group may be open to a greater or lesser extent to marriages with outsiders. In the case of the nomad clan, the index of relinking within the largest structurally endogamous unit is 75%, which is extremely high in comparison with other studies (see for example, Houseman and White 1998b). The largest structurally endogamous unit spans all of the lineages of the clan and includes all the ancestors and significant descendants who have contributed to the social reproduction of the clan. By this measure the nomad clan is enormously cohesive. When we take out the more peripheral couples of the structurally endogamous unit—those with the bare nodal degree of 2 necessary to have multiconnectivity—the index of endogamy falls to 57%. This is still high but does not imply that the remaining 43% will be married or linked to outsiders because some of these links are with peripheral couples. The porousness of structural endogamy is an evident characteristic not only for the Aydınlı but for all noncaste societies. Identification of Emergent Forms Some of the principles of nomad social organization that are now evident relate to our discussion of the polysemy of terms such as kabile (clan, lineage, sometimes tribe or aşiret) and aile (wife, also used for small families). The sliding terms of kabile and aile are not a result of a weaker sense of logical definitions among the Aydınlı in comparison to scientific logic as might have been thought; rather, they are seen as precise terms for emergent phenomena that themselves have sliding boundaries, such as the segmentation over time of patrilineal groups into cascades of descending sublineages. The network basis for the formation of shifting kabile and aile groupings involves principles that apply from the extended family up to the tribal level (Hypothesis 6.2). Flexible principles of marital relinkings provide a means for asserting group membership at the broadest level through structural endogamy, at middle levels through connectivity sets, and at the most intensive and local level through consanguineal, intralineage, and two-family relinking marriages. To reiterate some examples of these patterns, we found structural endogamy at the intertribal level (Hypothesis 6.2) and at the level of the clan, and relinking as the basis for cohesive subgroup formation at many different levels. Having discussed some examples of these principles of 420 Chapter 11 emergent cohesive groupings at higher and middle levels or social organization, the formation of cohesive units at the lower level remains to be reviewed. Intralineage and Blood Marriages as Exocohesive Relinking Analysis 11 of consanguineal marriages and historical change builds, as we have seen, on a unique feature of kinship analysis by computer. The latter gives the possibility of analyzing the occurrence of different kinds of marriage in terms not only of prevalence but also their selective rate (i.e., compared to the maximum possible number of such marriages) given the demographic structure of the population. Prevalence or raw frequencies (Table 5.1) are a poor guide to changing patterns and preferences. While there are only sixteen MBD marriages among the 414 recorded nomad marriages (a relative rate of 4% of all marriages), for example, the selective rate of MBD marriages is 16% (Figure 8.6: i.e., among men who have a recorded MBD available for marriage). Thus MBD is much more often selected as a mate than would appear from its raw frequency. MBD marriage is one of several types of cousin marriages that LéviStrauss (1969) examined as an object of study by examining their consequences for social integration. MBD marriage is a generator of “generalized” exchange or open-ended cycles of marriage. Hammel’s (1976) principle for the antecedents of differential frequencies of cousin marriages establishes that inequalities of age or status characteristics at marriage favor a significantly greater number of MBD cousins of an appropriate age for marriage than other cousins such as FZD. Among Aydınlı, MBD selective rates of marriage declined in the most recent generation to zero frequency. This cannot be explained by Hammel’s principle because, other things being equal, this would predict equal rates for all four types of cousin marriages. The method of calculating selective rates of marriage is also important because it compensates for gaps in the genealogical data due to memory loss, bias, and missing data. There is much more extensive memory data for males in the early generations, for example, than for females (Table 6.2, Figures 6.1 and 6.2). In relation to group boundaries and cohesive subgroups, the point is that each consanguineal marriage reinforces a specific social boundary within the kinship network. Because regions of higher cohesion may Overview and Conclusions 421 emerge within the network, each kind of consanguineal marriage helps, in proportion to its prevalence, to create certain systematic types of cohesive groups within the society. The idea here is closely related to the Axiom of Choice (Chap. 5, Section 2, on Deepening the Foundations for a Network Theory of Kinship) and the importance of prevalence for the consequences of marriage choices, while selective rates are useful in comparisons across time and between societies, and measures of choice are the most relevant to determining preferences. One way that reinforcement of kinship cohesion occurs within the clan is through marriage with blood relatives. The way in which this occurs, however, is through preferences for marriage with closer rather than more distant relatives (Hypotheses 8.2.2, 8.4, and Figure 7.11), ones who are also of the same generation (Hypothesis 7.10). Thus the emphasis is on extended family relinkings. With multiple types of marriage to relatives connected through females as well as through males, the effect of both is to densify relinkings within moderate-sized subsets of nodes but also, through overlap, to create broader relinking and cohesion in a distributed fashion throughout the clan (Hypothesis 6.1). FBD marriage, for example, reinforces lineage cohesion, as do second, third, or more distant cousin patrilateral parallel marriages (but see Hypothesis 8.5). While its prevalence fell dramatically in the twentieth century (Figure 8.3), its selective rate rose from about 25% in the late nineteenth century to 30-33% in the twentieth, and that of FFBSD marriage rose to 50%. Hence, what would seem by crude rates to be a custom in decline is discovered to be highly resilient in terms of selective rates into the contemporary era. The declining prevalence of FBD marriage is linked to smaller sibling sets via outmigration. Measures of choice in comparison to random marriages also show FBD to be highly preferential (Table 8.5). Although the decline in prevalence of FBD marriage is accompanied by an absolute and selective rise in FFBSD marriage (a greater dependence of intralineage marriage ties on more distant kinship connections), measures of choice in comparison to random marriages do not show FFBSD or FFFBSSD to be preferential (Hypothesis 8.5). Matrilateral parallel (first, second, third) cousin marriages (Tables 8.3 through 8.5), while lower in selective rate than its patrilateral counterpart, might be seen to reinforce cohesion within the uterine line. According to Barry (2000), the absence or diminution of such marriages might indicate higher identification with a female principle in which inheritance of a substantive identify leads to avoidance of marriage, except 422 Chapter 11 that among Aydınlı nomads there is no concept of inheritance of the female contribution (“flesh” or nurturance). Given the lack of a concept of a female “line” of identity, Barry’s argument would not explain why the selective rates of MZD and MMZDD marriages rise together (Figure 8.8: table on request). Johansen’s argument about changes in visiting with female relatives that accompany sedentization (Hypothesis 8.3) does provide an explanation. Her argument does not explain, however, why there would be statistical evidence for MZD as a preferential marriage (Table 8.5; the small numbers for MMZDD pointing in the direction of marriage preference are mute on this question because of lack of statistical significance). Hypothesis 8.2.2 would provide the answer: presented with the opportunity to interact with relatives (as argued, for example, in Johansen’s Hypothesis 8.3) the preference is for marriage with closer relatives rather than distant ones. Presumably this is because the prospective couple has a better chance to know more about one another’s character and personality the closer the relationship. The bulk of our evidence points to contemporary consanguineal marriages continuing to play a major role in reinforcing extended family cohesion. As lineages are thinned by outmigration the complexity and diversity in marriage choices takes a somewhat different shape. While parallel cousin marriages, which reinforce patrilineage principles, are increasing in terms of selective rates and preferential choice in the most recent generation, they are decreasing in prevalence. A comparison with simulation results confirms that cousin marriage preferences are not generalized through lineage principles or second or third cousins (Hypothesis 8.5) but they are instead restricted to first cousins. Among first cousin marriages, because only MBD marriage has declined up to the present, and FZD (associated by Hammel’s principle with same-age marriages and direct exchange between lineages in successive generations) has been stable throughout the twentieth century (table on request), the sum total of evidence about marriage exchanges and relinking via cousin marriages points to a trend toward reinforcing shallow agnatically extended families and shallow matrilines by parallel cousin marriage. Graphic Approaches Certain of our graphic images, such as Figure 6.1 and Figure 6.2 (patrilines and matrilines, respectively) are partial views of the nomad clan Overview and Conclusions 423 genealogy, organized so as to bring out only certain aspects of the network structure. More generally, the ability to use computerized genealogical data to draw accurate, well-organized, and informatively labeled genealogies, as in Figures 2.2 to 2.5, is of obvious use to ethnography in the presentation of data. It is also a means of visualization to provide both the ethnographer and the reader with a source of intuitions and insights about social structure. We have used the genealogies themselves, elicited from informants, to form a skeleton from which to give a detailed historical narration of the nomad clan’s formation, growth, and change up to the present. Nowadays, as a majority of those born into the clan turn to village life as adults, there remains a structurally endogamous core who continue to adapt the nomadic way of life to new challenges. In our Chapter 10 on Graphic Approaches we used automatic drawings (see Glossary) that minimize line length to show something of the overall structure of the clan. Figure 10.1 showed a three-dimensional graphic of the entire nomad genealogy in which the conical structure of the clan is made visual. Within it we can see both the fusion of the descendants of separate founding ancestors through intermarriage and the dense relinkings that occur within the core of the clan. Magnifying the relinked core of the nomad clan, Figure 10.2 showed a 3-D graphic of the relinking marriages among nomad kin. The density of the core, as we have seen, reaches the incredibly high index of relinking of 75%. Among the more striking images of this book, these figures may help the reader to grasp how the clan is organized as a conical structure. To understand the density of the nomad clan genealogies generally we also examined (Hypothesis 10.3) cases in which a given kinship link from an ego did not subsequently branch out to reach multiple alters but continued along a single path of successively more weakly or distantly linked relatives. What we found is that there are no paths of this sort (homeomorphic segments) greater than length 2 (length 2: node u links to v links to w with no branching at v) that we could consider stable, that is, other than ones involving a newly married couple. Focusing on the data in Figure 10.2 for the maritally relinked core of the clan, the computer analyses of cohesion (Tables 10.1 through 10.2, Figures 10.4 and 10.5) led to results that could be diagramed more abstractly in Figure 11.2. The figure shows clan structure in terms of hierarchical levels of cohesion, favoring the higher cohesion of couples whose marriage links back to highly cohesive ancestors and whose grown children have married so as to also become core members of the 424 Chapter 11 clan. These cohesive subsets give rise to positions of leadership and to a certain degree of competition between overlapping but opposing broad segments of the clan. Here leaders compete from relatively distant positions for overall cohesive backing of relatives distributed throughout the conical structure of the clan. Our ability to bring out a distributed structure of cohesive subsets within the clan, from which leaders draw their support, is novel within the sociology of small groups and subgroup cohesion. It is dependent on the computer’s ability to compute multiple independent paths of connection as well as to scale their structure, which turns out to be hierarchical. These results are very different from the identification of subgroups based on cliques as clusters of persons densely linked by direct ties. The salient feature of the nomad cohesive subgroups is that they are based on indirect ties, and they bring together, through the multiplicity of independent paths, people who may be quite distantly connected and who are dispersed throughout the nomad network. General Methodological Conclusions This concluding chapter has provided a summary of our specific findings on the Aydınlı nomads and has evaluated the degree to which computer analysis has enabled obtained a more precise representation of social structure and individual strategy as compared to usual fieldwork. A fuller ethnographic background (see Johansen 1965, 1994, 1995; Johansen and White 2001) is not summarized because the choice of facts is already limited to a minimum necessary to illustrate the introduction into the methods explained here. Instead, in a relatively short compass, we have tried to use computer methods to visualize data and test hypotheses, to explain the methods for others to use, and to demonstrate the advantages of such methods. The benefits of this approach include expanding our tools for representing basic ethnographic data, the precision of the statistical and structural analysis, and the usefulness of graph theoretical analysis and concrete visualization of more abstract properties of social structure. Such studies can serve as a formal basis for the comparison of ethnographic cases. The methods used gave rise to hypotheses and results many of which would be virtually impossible either to formulate or to evaluation using standard approaches (Analysis 11). Because our approach is based on graph theoretic concepts implemented for analysis via computer pro- Overview and Conclusions 425 grams, the formal concepts used in the analysis have been explicated and details have been given as to how to use the Pajek programs for largenetwork analysis that will allow the researcher to move through a variety of analyses of social network structure and dynamics. A fundamental contribution to network ethnography made by the approach taken in this book has been to surmount a twofold problem: 1) the separation between synchronic or structural and diachronic or historical study (discussed in the introduction to Schweizer and White 1998), and 2) the separation between stated norms and data on actual practices as can be collected from genealogical data individual by individual and then assembled over time. Basically, we have looked at networkembedded behaviors that form the basis of social structure as to whether they change over time or they remain constant. We were able to control for the changing demographic contexts of marriage behaviors and the social cohesion created by structural endogamy on a large scale and cohesive subgroups on a smaller scale. By doing so we have been able to identify the existence of preferences, avoidances, and social rules regarding marriage behaviors, and of specific historical turning points at which rules or social structures change. Because we can study the emergence of cohesive social subgroups within the larger network, we can examine how leadership emerges out of different aspects of social cohesion. The analysis showed how to use multiple network methods and measures for assessing convergent evidence for hypotheses about social structure, social strategies, leadership, and change in decision making processes.5 The measurement basis for many of the hypotheses in this book derives from the idea of a language of behavior put forward in the Preface and expanded in the technical vocabularies of Chapters 2, 5 and the Glossary. Our findings, derived from testing hypotheses concerning languages of behavior, support the idea that in certain contexts a language of behavior can be read and understood, especially if we utilize new forms of network analysis combined with an ethnographic understanding of indigenous knowledge systems. It is from actual choices or behaviors, especially as compared with the network background of possible choices as a system of constraints, that preferences and social rules can be inferred and cultural systems can be more accurately understood. 426 Chapter 11 Complexity Theory The relevance of complexity as a theoretical framework for kinship analysis constituted one of the last of our analyses. It emerged as one of the unexpected results to come out of our collaborative study. White had been using the language of complexity analysis informally in our descriptions but the predictive hypotheses about the power-law distributions (Chapter 7), as a signature of self-organizing systems came as a true surprise. This provides a new avenue of research that will not be followed up here but elsewhere. One of the most interesting conclusions that can be drawn concerns the type of distribution data that form powerlaw as opposed to exponential distributions. It is the raw behaviors that constitute the experienced social world that form the power-law distributions (Figures 7.11 to 7.14). The indicators of cognitive preferences such as the normed selective rates of different types do not form power-law distributions but are exponentially distributed. The implication is that behavior is not formed strictly in the mind by internal models and preference gradients as Lévi-Strauss and cognitive anthropology would have it but as Leach and Murdock have argued in raw frequencies of interactions in the social world itself. One does not exclude the other. The work undertaken by White and Houseman that informs this study is not improperly called by French reviewers the théorie de la pratique (praxis theory) in contrast but also complementary to cognitive structuralism and we continue to consider this to be an interesting avenue of investigation. Looking to the Future of Longitudinal Studies The particular results of this case study have been worth the effort to construct a systematic database, given the related capacity to formulate and test with it a variety of hypotheses against a rich body of ethnographic and potentially available network data. The hypotheses incorporate and develop network concepts appropriate to anthropological studies and enable us to test older structural formulations about kinship as well as newer ones. To provide guidance for comparable studies we have left the student of social structure with a vocabulary for analysis that is well grounded and equipped for deployment in other case studies.6 Comparisons between case studies will enlarge a new theoretical foundation for the network study of social organization and dynamics. Such study in Overview and Conclusions 427 many societies is extremely important for sociopolitical and historical research. We hope that this detailed example will stimulate others to undertake a deeper analysis of the societies or social groups that they study. Accordingly, we have detailed throughout our footnotes and in the appendix where to find (e.g., on the Internet) and how to use the data on which our analysis is based, where to find the analytic programs that we have used, and how to use the software for purposes of analysis. We hope the variables constructed in this process, according to the ethnographer’s and analyst’s intuitions, both prove to be of lasting value in that they can be revisited for the explorations of further ideas. 7 This is also true for other long-term field research projects (see examples in Royce and Kemper (2002) and of those of anthropologists dedicated to recording and preserving data and examining hypotheses through time as societies change). A dialogue in which the experience and understanding of a long-term ethnographer—formulating questions and giving descriptions, explanations, and narratives—is balanced against a research analyst’s proactive formulation of hypotheses is fairly rare. Such a dialogue can be very productive, however, especially in developing and testing hypotheses against a systematic database. This kind of dialogue in a collaborative model of anthropology may become increasingly important when researchers acknowledge that both continuity and change, similarity and difference in individual and group choices are potentially highly significant from one time period to another and that these in turn are the basic characterizations of the societies, groups, and individuals that we study and live among as anthropologists. Seriously considered, this acknowledgment about the complexity of cultural formations and social histories may provoke the rise of deeper and more various methods of analysis equal to the task. Longitudinal analysis of field data enables anthropology and social science generally to continue to deepen their contributions to understanding the richness and complexity of human society and experience, even as the effects of globalization continue to complexify the societies that we study. These, as we have seen, already pose a considerable challenge to comprehending the protean and often simple principles concerning social processes that can be gleaned from the wealth of empirical detail available from long term ethnographic studies. 428 Chapter 11 Further Reading The April/September 2000 special issue of l’Homme (154-155) on the theme of “Questions of Kinship (Parenté)” contains several extensive reviews of the approach taken in this book.8 Collard (2000:638,640, 646,651) reviews and identifies the work of White and Jorion (1992), Houseman and White (1998a, 1998b; including contributions to the volume by Godelier, Trautmann, and Tjon Sie Fat 1998), and Schweizer and White (1998)—using the p-graph approach—as one of the main contributions to la théorie de la pratique approach to kinship today. Jamard (2000:735-736) devotes a long exposition to the methodological and theoretical importance of this approach.9 Augustin (2000a), in the same volume, reviews the network approaches taken in Schweizer and White (eds. 1998). While the Pul Eliya and Amazonian analyses are concerned with a structural logic of dual organization entirely different from the structural themes of the present book, the extent to which the vocabulary of network analysis, adapted to the concerns of social structure, has entered the canon of social and structural anthropology in France. In the “Glossary of Kinship (Parenté),” Barry et al. 2000) devote entries to dividedness (724), sidedness (731), matrimonial network (730), and pratique matrimonial (729) as opposed to matrimonial norms. Hence, a network approach to kinship and matrimonial practices is firmly established within the French intellectual terrain of social anthropology, if not within the dominant paradigms of English-language paradigms (but see reviews of Schweizer and White (1998) by Dow (1999) and Gregory (2000)). Further, White’s (1997) definition for “structural endogamy” has entered the canon in several recent publications (see Augustin 2000b: 594). As Augustin notes therein: One finds such clusters [of endogamy in bilateral society] in abundance in the majority of European societies in the form of matrimonial enfolding among a set of persons linked in a manner more or less distant. . . . [The] structural endogamy discussed by Douglas White [1997, Brudner and White (1997)] is the matrimonial concomitant of this same phenomenon.10 The theme of structural endogamy, along with the concepts of structural (and regular) equivalence now commonly used in network analysis (White and Reitz 1983) is a prime focus of our study of Turkish nomads. Notes Overview and Conclusions 429 1. As told in the ancestral narratives, the grandchild-generation intermarriage with the Karahacılı lineages (Kırbaşı #4) was an alliance of amalgamation between the new lineage and the existing clan: “after becoming rich off” exchanges with #1 and #2 (at this point #3 was also allied with #2 as relinking wife-giver while #1 was allied by sister exchange), Koca Oğlan’s (#4) offspring agreed to intermarriage. In this generation that the other Karahacılı lineage #5 becomes relinked to both lineage #4 and the central lineage 3. Karahacılı Dazkırlı lineage #6 does not intermarry into the main lineages until three generations later. 2. There were obviously also many endogamic ties in Kurşun, but we do not know if a break in cohesion led to the migration of Mustan’s family, for example, or whether all of the founder lines were already cohesive back in Kurşun, given the loss of genealogical knowledge over intervenint generations. 3. In refinement of our earlier observations, the numerouus off-generation marriages of lineage #2 may have been due to considerable age differences between brothers in early periods and the long time period involved. 4. Lineages #1 Dolaşıklı, #2 Ecevitli A, #4 Kırbaşı oğulları, and #5 Koca bey oğulları, but not #3 Ecevitli B) 5. A rational choice framework is also closely incorporated within our approach. The network approach allows an analysis of individual roles, biographies, activities, and social choices as to cooperative, competitive, and selfinterested behaviors. We have seen how social choices and roles such as those of Mustan and early founders contributed to the cooperative establishment of a network of cohesive families that was critical to the ethnogenesis of the clan. We have shown the costs (in social and economic support) and the benefits of marital choices such as whether or not to choose a spouse who relinks with others in the clan, and how such choices correlate with decisions to stay or leave the nomadic lifeway (Hypothesis 4.4 and Hypothesis 4.4.1). We have shown how specific kinds of marital relinkings contribute to the formation or reinforcement of cohesion in larger or smaller social groupings, and, in the recent period, how the increasing frequency of emigration and consequent reduction in the size of sibling sets has created potential problems involving the shrinkages of lineages and fragmentation of social cohesion. As outlined by Schweizer and White (1998: Introduction) , the integration of a rational choice and actor-oriented perspective has been one of the goals of the network approach to ethnography. Building on the base of network analysis, a great deal of further work can be done in this direction. 6. Including in the simplest case documented collections of genealogical data, but hopefully also rich ethnographic field data. Genealogical data are commonly available as *.GED files that may be accessed using Pajek and are often built using commercial programs such as Family Origins or Family Tree Maker. 7. Equally, we would be thankful if others contributed suggestions for deepening our analyses. 8. We call attention once more to the support for development of this approach 430 Chapter 11 in France, through the auspices of Clemens Heller, Jean-Luc Lory, the Maison des Sciences de l’Homme, Maison Suger, and the Ministry of Research and Technology. 9. Jamard (2000:735-736) states, translated from the French: Methods and techniques [of kinship analysis] have strong implications on the theoretical side. For that reason, their use pertains to the reexamination of kinship nomenclatures. [In the Godelier et al., edited volume, 1998] One article vigorously distinguishes itself in the domain of precise procedures. In contrast with Tjon Sie Fat, who presents a meticulous algebraic treatment of purely terminological kinship, Michael Houseman and Douglas R. White, using a variety of computer tools, collaborate to show the emergent properties of a network of marriages that are effective through their dynamic aspect in the pratique—behavioral practices—of matrimonial alliances, where they find observed regularities that are not a simple effect of a terminological logic and rules of marriage. These constitute, at the level of practice, a sort of primary behavioral regularity [encodage], of a complex order. This is precisely demonstrated in that the two researchers, in the course of their analysis, are able to detect a structure of sidedness [structure à coté], or bipartite network where a pair of supersets of marriages, connected by agnatic and uterine decent links, operate so as to organize network configurations of marriage alliances across a range of societies in lowland Amazonia. The authors succeed in creating an empirical sociology of high quality that takes the first steps toward a conceptual and theoretical advance toward a sort of grounded theory (Glaser and Strauss 1967) based on facts established methodologically through carefully controlled working hypotheses [and precise analytic definitions]. 10. Translation from the French.