9. Adoption on Rotuma
[Published in Adoption in Eastern Oceania, edited by
Vern Carroll. Honolulu: University Press of Hawaii, 1970]
In this chapter adoption is dealt with from the standpoint of the decisions people make regarding the allocation of children.  I shall therefore be concerned primarily with the conditions under which children are transferred to foster homes and to whom they are sent rather than the effects of adoption either on the children or on the organization of Rotuman society. An attempt is made to develop decision-making models  that will account for the distribution of adopted children in Rotuman households both on the island of Rotuma and in five urbanized communities in Fiji. 
The ethnographic approach used in this analysis is based on Goodenough's dictum that the goal of ethnographic description is "to give the reader a basis for learning to operate in terms of the culture described in somewhat the same manner that a grammar would provide him with a basis for learning to speak a language" (Goodenough 1951: 10). In applying this approach in a paper on Rotuman land tenure, I asserted that an adequate ethnography should provide sufficient data to permit the construction of decision-making models in which the culturally perceived alternatives were designated, the principles which are determinate for choosing between alternatives described, and the relationship between the factors specified (Howard 1963b:409). Although such data were not presented in that paper, ideally the distribution of actual choices under given conditions should be presented and the decision-making models evaluated according to their ability to account for the distribution. In other words, if we find that among a given group of people half choose alternative "A" and half alternative "B" when confronted with a choice, then our models should account for the conditions under which "A" and "B" are selected. If the model includes all relevant conditions, and if the conditions are properly ordered and weighted, then it follows that the model will account for 100 percent of the cases. That is, it will be perfectly retrodictive. In most cases of interest to social scientists this is an unobtainable ideal and we must content ourselves with models of lesser accuracy, but under any circumstances I think it is imperative that we specify just exactly how much variance our models do account for. Anthropologists have been particularly lax in doing so in the past, at least when compared with sociologists and psychologists. Perhaps this is because the concept of "culture" has led us to concentrate on the ideational rather than the behavioral order. Although I would not condemn this emphasis as being erroneous, it is limiting unless we are willing to relate people's ideas to their behavior. What should be stressed, however, is that "ideas" must include more than ideology, i.e., the statements our informants make about how people should behave; for if we limit ourselves to ideological conceptions, the cultural models we are likely to develop will be only weakly related to behavior in most instances. It is not enough to state, as some have done, that behavior tends toward an ideal, or that it approaches the ideal when all other things are "equal." This is an evasion of the issue; things rarely are equal. The point is that we know almost nothing at all if we do not know how closely behavior approximates the ideal and under what conditions ideal patterns occur.
My argument should not be construed as an attempt to replace the ideational emphasis inherent in the culture concept with a behavioristic one. Indeed, I view decision-making as an ideational process, in full accord with the definitions of culture utilized by such theorists as Goodenough (1963: 258-259) and Wallace (1961: 28). It is rather that I think we deceive ourselves too easily into thinking we have discovered the significant cultural principles unless we systematically test our cultural models against observable behavior. By examining the distribution of behavioral choices under varying circumstances, we may discover cultural principles that our informants have never articulated, either because they took them for granted or because they were not conscious of their influence. In this sense, the type of analysis I am advocating has its prototype in linguistic description. To a linguist, an informant's ideas about the structure of his language may provide some useful hypotheses, but unless they actually account for the distributions found in native speakers' spontaneous utterances (and this is never the case), the linguist must hypothesize principles of his own intuition until he develops a model that does the job adequately. Furthermore, if he bases his own attempts to speak the language only on the principles overtly recognized by his informants, many of his utterances will be judged incorrect. They will, in short, be in violation of the linguistic norms. Unfortunately, we are too often spared this kind of embarrassment when it comes to other forms of behavior by virtue of the fact that visiting anthropologists ordinarily are cast into marginal roles for which the norms are ill-defined. Nevertheless there is good reason to believe that in most societies the large majority of decisions are neither arrived at nor judged on the basis of ideological considerations alone.
One rationale anthropologists have sometimes used to evade the task of accounting for observed cases is that ideal rules are sometimes violated, and that the actual incidence of conformity is less important than the knowledge that deviations are recognized by our subjects as contrary cases, that the rules are honored in the breach. After all, it could be argued, this is precisely what linguists do when they ignore the "errors" that occur so frequently in everyday speech. Although I am in full agreement that our ethnographic accounts should be theories of "competence," this should not become an excuse for avoiding the responsibility of presenting the distribution of cases, for surely rates of deviation present us with some of the best means for understanding the nature of a sociocultural order. But more importantly, it is often the case that what is passed off as deviation by the researcher on the grounds that the behavior violates an expressed cultural norm is not regarded as a deviation by the people under investigation. For example, in Rotuma a person is expected to be generous to someone who is needy. There are instances where people violate this norm and are criticized for having done so, but there are also occasions where refusing generosity is not only permissible, it is called for. For instance, if the person making the request is being exploitative-if he has frequently failed to live up to his reciprocal obligations-responding generously is considered stupid. It is by being generous under such circumstances that a person exposes himself to criticism. The point is that by trying to account for all cases of non- generosity we are much more likely to arrive at an understanding of which are actual deviations and which are not.
Not all behavioral decisions are subject to normative evaluation. Some choices that confront people involve several equally respectable possibilities, such as choice of occupation in our society. Under such circumstances individual motivation is a significant determining factor, and in order to account for such behavior we need to take into account psychological variables as well as cultural principles. I refer to these as strategic decisions. Decisions that allow no alternatives to what is agreed upon as being correct I refer to as normative decisions. One's "choice" under these latter conditions is between doing the right thing and doing something deviant.
The distinction between these two types of decision is of considerable significance. Normative decisions should be subject to formal analysis if all the relevant data are available. The logic behind this statement is that if our subjects approach 100 percent agreement as to whether or not an act of behavior is appropriate, they are presumably using a definitive set of principles in making their judgments and we should be able to specify them. Linguistic behavior comes closest to requiring total normative agreement, and it is therefore not surprising that formal analysis has focused on this area thus far. It is my assertion that the next likely area to explore is the area of distinct alternative choices, such as residence choice, property transfers, and the selection of leaders.
Models to account for strategic decisions are complicated by the necessity for measuring motivation or differential values. Thus, to account for occupational choice in the United States we would have to focus attention on the relative weights individuals give to a variety of considerations plus the strength of their motivation to achieve differential goals. In other words, whereas normative decisions can be accounted for entirely by reference to cultural rules, strategic decisions require consideration of both cultural principles (e.g., shared recognition of goal states and strategies for obtaining them) and psychological principles (e.g., strength of drive, habitual response patterns, etc.). This does not mean that models accounting for strategic decisions are necessarily less accurate, although ordinarily one expects this to be the case. As we learn to measure motivation and relative values more accurately we improve our predictive powers accordingly. Also, in some cases the weight of certain considerations is sufficiently strong to skew the distribution of strategic decisions decisively in one direction, rendering them highly predictable. On the other hand, genuine deviations from cultural rules may occur frequently enough in some areas to lower substantially the predictive value of models designed to account for normative decisions. Both strategic and normative decisions are involved in our analysis of fosterage among the Rotumans. For example, the decision a widower makes as to whether he should keep his children with him or send them to a foster home, is a strategic one. Either alternative is acceptable and his choice depends upon motivational as well as practical considerations. Once he has decided to send them to a foster home, however, his choice is governed by definite rules as to where is appropriate. It is, therefore, a normative decision that is predictable on the basis of a formal model.
The conceptual basis of Rotuman kinship is inherent in the term kainaga. In its broadest sense the word kainaga means 'kind', 'sort', 'variety', 'species', 'class' (Churchward 1940:235); in other words, 'belonging to the same category'. It can be used to describe people of the same nationality, or in a more limited sense to designate persons 'of the same blood', i.e., consanguinity. In this latter sense, a person's kainaga constitutes his personal kindred, a grouping which becomes functionally operative during life-crisis ceremonies and when an individual becomes critically ill. The term is used also in a still more restricted sense, indicating common descent from an ancestor who has resided at, and held rights in, a given house-site, or fuag ri. Every house-site is named,and a person usually describes his affiliation by such a statement as, "I am a member of the halafa kainaga." Associated with each house-site are sections of garden land, presumably those over which the ancestor held rights, and to claim membership in a given kainaga is to claim rights in these lands. Ideally, the person who lives on the house-site is the individual who has been granted control of the land by the kainaga members. He or she is known as the pure 'one who decides' and is obliged to grant usufruct privileges to any member of the group. Unless he is unavailable or judged incompetent, the senior member of the kainaga is supposed to be selected as pure. Seniority is based upon age, sibling position, and one's parents' sibling positions (cf. Howard 1963b, 1964).
Economically, the most significant social unit in Rotuma is the individual household. The average number of persons per household was 6.9 in 1960 although many households were considerably larger. Most consisted of a nuclear family, with various bilateral extensions on either the husband's or the wife's side. Social relations within the household are generally warm and cordial, and even the respect barriers traditionally prescribed between affines are apt to disappear in time. The men in the household do most of the farming and cut the copra, which is the main source of income. They also do some of the fishing, usually with lantern and spear, but sometimes with hook and line in the deep sea beyond the reef. The women weave mats, which are used domestically and for ceremonial exchange, and do most of the daylight reef fishing with nets. They also take major responsibility for taking care of and disciplining the children. The division of labor is not rigid, however, and quite often all the members of a household work as a team at a particular task. In general the status of women is very high and the 'good' husband does everything possible to make his wife happy.
Children are greatly desired and are welcomed into every home. As a general rule infants and young children are a focal point within each household and are indulged by everyone, often to the extreme. Until they are able to walk they are continually in someone's arms. Demonstrations of affection continue in the form of fondling and caressing long after the child has begun to walk, and children tend to seek physical contact with those whose affection they desire. Indulgence also takes the form of gift giving. The 'good' parent is the one who grants his children's every reasonable request; if he can anticipate their desires, so much the better. Toys and other presents are given to youngsters whenever there are a few spare shillings, and it is not unusual for parents to sacrifice household effects, which an outsider might regard as necessities, to buy their child a toy. In most households this pattern of indulgence extends to eating behavior; children are fed first and are given the best food, while the elders content themselves with the leftovers. This association of affection with material giving is continually reinforced, and is reflected in the connotation of the word hanisi, the nearest Rotuman equivalent to our word 'love', which implies willingness to give tangibly.
In Fiji, land is not available to Rotumans, so they have to make a living as wage earners. Unmarried women beyond school age, and often married women as well, work if the opportunity is present. The household structure has not significantly changed, however, with nuclear families providing the stable core around which more mobile relatives congregate. Household size is only slightly less than in Rotuma, averaging 6.4 persons per unit, and my evidence suggests that despite their being a greater burden to wage earners than to farmers, children are welcomed as eagerly by Rotumans in Fiji as by their kinsmen on the home island.  To the Rotumans, in short, children are valued commodities. They are valued not for their economic potential, for they contribute little to household economies, but for their stimulus value as love objects, and as is true of all valued commodities, their allocation is governed by culturally explicit norms.
There is very little information on adoption in the scattered literature on Rotuma, but it is quite probable that in the aboriginal society it played a significant role in politics. Churchward lists two entries in his dictionary that designate politically motivated adoptions. One is ati, which he defines as 'to adopt and bring up a boy of chiefly rank with a view to his becoming chief of the town or district which adopts him' (Churchward 1940:177). The other is po gagaja, which is defined as a 'chief or person of chiefly rank procured from elsewhere, by marriage or otherwise, so as to give higher status to those acquiring him' (Churchward 1940: 286). Neither of these concepts is in current use. Politics in Rotuma nowadays are sufficiently unstable from generation to generation to make adoption virtually irrelevant (Howard 1963a, 1966). Its significance for land tenure has likewise been lost. Prior to cession to Great Britain,  persons would sometimes will all their land to an adopted person, usually an adult, in preference to willing it to their own children. They could justifiably do this when their own children had neglected them or had otherwise failed to fulfill their filial obligations. Under other circumstances, adopted children were specifically willed a share along with other children. But shortly after cession a regulation was passed explicitly prohibiting the transference of land rights to adopted children  (Eason 1951: 98-99). With these broader functions removed, adoption has become in Rotuma, as it is among Rotumans in Fiji, an exclusively domestic affair. That is, it focuses upon the care of the children involved and has no profound implications for groups larger than the household.
Adoption is ordinarily an informal matter among the Rotumans. I have encountered no cases of legal adoption, although litigation has been brought before the courts from time to time in which parents have attempted to reclaim children after long periods of fosterage. Transactions are usually between close relatives, and although there are instances in which it is understood that the transference is permanent, the duration of fosterage more often depends upon subsequent events than explicit agreement. Thus, there is no Rotuman term differentiating permanent from temporary fosterage. Both are covered by the word putu 'to care for', and includes among possible objects children, sick persons, or young animals separated from their mothers (Churchward 1940:291). Whereas requests by close relatives to putu children are generally unceremonious, if a distant relative or an unrelated person wishes to adopt a child, it is appropriate for him to accompany the request with a ceremonial gift of kava and a 'baked pig' koua, but this is likely to occur only when the prospective adopters have no close relatives from whom to request children. Such circumstances are rare; none occurred during the period of fieldwork.
The act of adoption is covered by the general verb re 'to do' and includes a wide range of sub-meanings including 'to make', 'to produce', 'to get', 'to provide', 'to give', and many others (Churchward 1940:297). Adopted children are referred to by the same term as 'grandchildren' ma'piga regardless of their genealogical relationship to their adopters. If already closely related, they are not likely to alter the terms with which they refer to other kinsmen,  but if permanently adopted by unrelated persons or distant relatives they ordinarily begin to use terms appropriate to their new position in their adopters' kinship network. The actual terms used may not always be consistent. For example, although an adopted child is called ma'piga 'grandchild', he is apt to call the genealogical children of his adopters by sibling terms rather than terms which designate members of the parental generation, particularly if they are near his age. Rotuman kinship terms can be qualified by two suffixes, -pu and -'aki. The former designates the primary kin type while the latter designates extensions of the term. For example, ego's children's children are referred to as ma'piag pu, while all other consanguineally related persons of the grandchild generation are referred to as ma'piag'aki. The latter term includes adopted children and the -'aki suffix is used to qualify any person who is a kinsman by adoption.
The categorization of adopted children with grandchildren is probably based historically on the custom of sending children to the grandparents for a period at the time of weaning, or approximately when one year old. This is still done today by some couples, depending upon circumstances. Visits by older children to grandparents for periods ranging from weeks to years is common even today. Finally, grandparents are the chief adopters, and as the data will show, other adopters are under most circumstances grandparent substitutes in a very real sense.
A custom that has particular relevance for adoption is the giving of a name to newborn children. The privilege of naming a child is usually requested by a close friend or relative of either the mother or father, or the parents may ask someone special to give a name to the child. Thereafter the 'namesake' sigoa has a special relationship to the child. He or she is like an additional grandparent, and treats the child with great indulgence. Children often visit their namesakes for periods of time and are sometimes adopted by them.
As a rule, adopted children are treated very well. Especially if the adoption was initiated by the receiving family, the children are treated with the same indulgence they would characteristically receive from their true grandparents. According to Rotuman ideology, people treat their adopted children particularly well 'so they will not think of their real mother and father'. A family would be shamed indeed if a child whom they had requested to adopt complained about his treatment and chose to return home. In some cases, however, the adoption is a matter of necessity and obligation, as when a family dissolves through death or divorce. Under such circumstances the adopted child may come to be regarded as a burden and is supported only begrudgingly.
This study is based upon demographic data collected from two sources-living informants, and the registers of births, deaths, and marriages for the Colony of Fiji. Data collection began in 1960 with a household census of Rotuma. In addition to household composition, an individual data sheet was made up for each adult, including information on family affiliations, residential history, and a variety of factors relevant to the person's status in the community. Early in 1961 similar information was obtained from five Rotuman communities in Fiji. During the same periods, birth records dating back to 1904 were copied and alphabetized by mothers' names. Death, marriage, and divorce records were also copied, and on the basis of this information the data collected in the field were corrected. By analyzing record linkages it was then possible to determine the demographic fate of the large majority of Rotumans born since 1904.
For the purposes of this study a child is operationally defined as adopted if he is listed as living in a household in which neither of his parents reside. This definition has the obvious disadvantage of not distinguishing temporary from permanent residence, but since that information was not systematically collected, we are provided with no better alternative. The term "child" refers to an individual born in the year 1945 or later, thereby restricting him to fifteen years old or less. At about age sixteen a child comes to be regarded as a 'youth' haharagi, and he enjoys a considerable amount of mobility, whereas prior to that age his movements are restricted by his dependence.
As can be seen in Table 24, 23 percent of all Rotuman households included in the census contained adopted children. Remarkably, the percentage is nearly identical in Rotuma and Fiji.
The term "transaction" which is used in Tables 25 and 26 is defined as the adoption of one or more children of a particular natural parent. A household which contains three adopted children, one from one sibling set and two from another, has therefore been involved in two adoption transactions. Table 25 shows that most adopting households are involved in only one transaction and that the maximum is three. Table 26 shows the number of children involved and the percentage which have been adopted. Again the figures are nearly identical for Rotuma and Fiji.
The relationships between adopters and adoptees are presented statistically in Table 27. The adopter was operationally defined as the adult in the household most closely related to the child. If a man and woman were equally close, the woman was considered the adopter; this again is an operational assumption. In the large majority of cases, this adopter was either the household head or spouse of the household head. Table 27 shows a contrast between the figures for Rotuma and Fiji. In Rotuma there is a stronger tendency for adopters to be actual grandparents or of the grandparent generational level. In Fiji adopters are more likely to be of the parental generation.  Also, in Fiji there are more cases of children living with their older siblings.
Adoption rates were calculated also from sixty-six life history interviews collected on the basis of accessibility rather than sampling technique, but there is no reason to think that they represent a skewed population with regard to adoption. Eighteen of these adults (27 percent) spontaneously reported having had extensive adoption experience as a child. This is more than double the rate found by the synchronic analysis of household composition for 1960-1961. The discrepancy is partially explainable on the basis of the fact that many children then living with their parents had already had fosterage experience but had returned home, while others were destined to be adopted at a later date. The 12 percent figure is therefore best interpreted as the proportion of children living in foster homes at one point in time, not the proportion of children who have been adopted out at one time or another in their lives. On the other hand, there are reasons to suspect that adoption rates in Rotuma have changed over the past thirty or forty years. These will become clear as we more closely inspect the relationship between demographic patterns and fosterage.
DECISION-MAKING MODELS OF FOSTERAGE AMONG THE ROTUMANS
We begin our analysis by making a working assumption that parents have primary rights of allocation in children and that the rights and obligations of others with respect to the children are based primarily on their kinship relation to one or both of the parents. From this assumption we derive the hypothesis that the decision to adopt out children is affected by their parents' ability to care for them. It follows therefore that adoption rates should vary with parental circumstances. We would predict a higher rate of adoption among children from disrupted or broken families than among children from intact families.
In order to test the validity of this hypothesis we approached the demographic data from another point of view, by checking the whereabouts of children born after 1945 from each sibling set for which information was available. In most cases it was possible to determine parental circumstances: whether or not the children were born out of wedlock, whether one of the parents had died, whether they were both alive but separated or divorced, or whether they were living together at the time of the census. The results appear in Table 28.
The rates in Table 28 are stated in terms of minimums and maximums as a consequence of failure to locate a certain proportion of the children. This is due to an incomplete census in Fiji and to the movement of some children from households that were not recorded until near the end of the census to households previously recorded. The figures support the hypothesis. By far the highest rate occurs when a family is dissolved by the death of the mother. Next come children from families dissolved by separation or divorce, followed by children of unmarried mothers and families dissolved by the death of the father. The lowest rate occurs among the complete families. These differences are sufficient to warrant separate treatment for each of the circumstances of adoption. In the analysis that follows only cases are considered in which the whereabouts of children are known.
Family dissolved by death of mother. There were eighteen determinate instances of families dissolved by maternal death, involving forty-two children. In five cases (eight children) fathers kept their immature offspring; in twelve cases (thirty-two children) they gave away all of them in adoption, and in one case a father kept one child and gave away one in adoption. Four of the fathers who retained their children, as well as the father who allowed one of his children to be adopted, were remarried at the time of the census and another was residing with his mother. In only one case was a widower living alone with his children; since his wife had only recently died there is reason to believe that the situation was temporary. Leaving aside this aberrant case, which accounts for four of the eight children, it is interesting to note that each father retained only one child, and that they were all over the age of thirteen. Also, four out of five were males.
How can this evidence be stated in terms of a decision-making model? It is first necessary to point out that the decision as to whether or not a man keeps his children under these circumstances is a strategic one. It is neither appropriate nor inappropriate for him to keep his children with him provided he can care for their needs satisfactorily. Having a mature woman in the house and retaining only one child, a male beyond the age of puberty, are ways of maximizing his ability to meet this requirement. Decisions ultimately depend on a variety of personal considerations, including willingness to be tied down, degree of affection, and personal feelings of adequacy. Nevertheless we can sum up the data as follows: widowers either give away all their children in adoption or retain only males past the age of thirteen. As a retrodictive model this accounts for sixteen out of eighteen decisions, or 89 percent of the cases.
On the basis of our assumption that children are valuable commodities and that therefore their allocation is governed by culturally explicit norms, it is presumed that the choice of where to send a child for adoption involves a normative decision. There are three relevant principles that combine to produce a decision-making model for this situation: women's rights over children take precedence over men's rights; the rights of genealogically close kinsmen take precedence over the rights of genealogically distant kinsmen; the rights of senior kinsmen take precedence over the rights of junior kinsmen. The model can be stated as follows:
When ego is a widowed male and he decides to adopt out one or more of his children, first choice should be given to his wife's mother;
A person's nuclear kindred is here defined as consisting of all his lineal relatives and their siblings. An individual is considered "available" if he or she is a core member of a household, as opposed to a marginal or peripheral member, and resides in the same locality (either on Rotuma or Fiji) as the child at the time of his mother's death. The emphasis on core membership in a household is related to the requirement that the adopter have sufficient control of resources to assure the child's subsistence. Operationally a person is considered a core member of a household if he or she is a member of the head's nuclear kindred or a member of the household head's spouse's nuclear kindred.
The actual distribution of the thirty-three children who were adopted out upon the death of their mothers, is indicated in Table 29.
Leaving aside three problematic cases for which insufficient information is available, the model accounts for all but two of eighteen transactions (87 percent). It also accounts for the allocation of 93 percent of the children.
One of the most striking aspects of the allocations is that of seven instances in which more than one child was part of a transaction, six cases involved the MoMo as adopter. This suggests that the maternal grandmother is the only direct mother substitute. In cases where the maternal grandmother is not available and more than one child is adopted out, they appear to be parcelled out one to a household, beginning with the most eligible adopter, as defined by the model, and working down.
It is perhaps necessary to speculate at this point on the relationship between "rights" and "obligations" as they affect the allocation of children. We are assuming it is a widower's obligation to grant first choice of children to the senior woman in his wife's nuclear kindred; that she has a right to be given that choice. According to Rotuman custom, however, it is probable that she is obliged to take at least one unless she has a valid reason for not doing so, such as inability to care adequately for him. The fact that maternal grandmothers frequently take several children in single transactions whereas others do not suggest that they are both more willing and more obligated to adopt children than are other relatives.
Family dissolved by death of father. There were twenty determinate instances of families dissolved by paternal death, involving forty-seven children. In fourteen cases (thirty children) the widowed women kept all of their children with them; in five cases (sixteen children) the mothers kept at least one child with them and allowed at least one to be adopted out; and in one case a mother allowed her only child to be adopted out.
It is apparent from this distribution that widows, in contrast to widowers, ordinarily keep their children with them. Furthermore, the characteristics of the children whom widows keep are different from those kept by widowers. Thus in four of the five cases in which mothers kept some children and gave up some in adoption, they kept the youngest ones with them. These data suggest that it is inappropriate for widows to seek adopters for their immature children, particularly younger ones, and that the adoptions that do occur are initiated by the adopters. This conclusion is supported by two lines of evidence: first, in none of the eight adopting households were there children in the same age range (within five years) as the adopted child; second, the distribution of adopting relatives does not follow the principles described above. Only one case conforms to the model. The actual distribution is indicated in Table 30.
It is also significant that namesakes were adopters in two of the eight transactions. Namesakes may request a child in adoption, but they are rarely asked to foster a child by a parent seeking an adopter.
Family dissolved by separation or divorce. Complications are introduced when a family unit is dissolved by separation or divorce by the fact that both parents are alive. According to Rotuman ideology the first child of a union belongs to the father's side of the family, the second to the mother's side, the third to the father's, the fourth to the mother's and so on. The selection of namesakes usually conforms to this rule. Rotumans say that when a couple separates, their children are divided according to this principle but when we examine the distribution of cases, it becomes clear that either the ideal rule does not operate or it is modified by other considerations. For example, when only one child was involved he went to the mother or one of her relatives in eleven out of thirteen cases, although the child supposedly belonged to the father's side. In the fifteen instances in which two children were involved, both went to the mother's side in five cases, both went to the father's side in two, and in eight cases the children were divided. Indications are that both children go to one side only in the case where one of the parents gives up his or her rights through desertion. In the eight instances in which two children were divided, the ideal rule was conformed to in seven, with the first child going to the father's side and the second to the mother's. When three children were involved, the children were divided according to this principle in only one out of six cases, and when families with more than three children were dissolved by separation or divorce the division of children did not conform to the rule in any of the thirteen instances. Apparently only when two children are being divided are all other things "equal" and the deal rule an effective predictor.
The number of considerations that go into dividing up children under these circumstances are sufficiently numerous that we cannot readily devise an effective decision-making model to account for it on the basis of the demographic data alone. No significant differences were found between the distribution of males and females or younger and older children. In general, though, the majority of considerations appear to favor allocation to the mother's side (69 of 106 children for whom information was available). Of the children who remained with their parents, forty-eight were with their mothers and twenty-five with their fathers. The residential fate of those who were adopted out is indicated in Table 3 1.
The location of children adopted out indicates that the same principles governing the allocation of widower's children apply to the allocation of children of divorced parents, with the qualification that fathers in this case confine their choice to consanguineal relatives and mothers do likewise. This decision-making model- that individuals first approach their mothers, followed by the senior female in their nuclear kindred, or if no females are available, the senior male, when seeking adopters for their children- accounts for twenty-two out of twenty-six transactions. It fails to account for three transactions and there was insufficient information for one case (MoBr).
Unmarried mothers. The allocation of children of unmarried mothers also involves a wide range of considerations, including the degree of shame felt by the mother, her ability to care for the child, and the degree to which she sees the child as a hindrance to her eventual marriage. Therefore her decision whether or not to keep a child is a strategic one which is not easily accounted for, although once again the majority of children who were adopted out can be located on the basis of the previous model. Seventy-four of one hundred "illegitimate" children for whom information was available were living with their mothers, two were with their fathers, and the rest were adopted out. The location of those who were adopted out is shown in Table 32.
Of twenty transactions, thirteen conform to the model, two do not, and for five there is insufficient information.
Complete families. Children from complete families have the lowest adoption rate, and there is good reason to believe that they are rarely adopted out on the initiative of the parents. Couples who are suspected of not wanting to keep their children are likely to be the victims of gossip and joking, with the accusations made that they either did not love their children or were incapable of taking care of them adequately. As a result, the initiative to adopt children from complete families usually comes from the adopter. Prospective adopters must meet several requirements. First, they must have a convincing reason for wanting a child. The most common reason, and a thoroughly convincing one to most Rotumans, is lack of children in their household. Thirty-seven out of forty-seven adopting households had no children other than adopted children within the same age range (five years) as the children they adopted. They must also be in a position to take care of the child both economically and emotionally, at least as adequately as the natural parents and preferably better. This is the primary justification for both parties to the transaction. When asked why they give up children in adoption, Rotuman parents invariably responded, "They are able to do more for our child than we can." Another consideration is the relationship between the prospective adopter and the parents. The request of a namesake, for example, is difficult to refuse because of the special relationship to the child. If the adopter is senior in kinship rank it is also difficult to refuse, but if he is junior the other considerations must be strong enough to be compelling. Although kinship rank is a significant factor in selecting a child for adoption, the other considerations are sufficient to render of low predictive value any model based upon kinship principles alone.
The actual distribution of adopted children from complete families in relation to their adopters is shown in Table 33.
The adoption of children from complete families differs from those transacted under other circumstances insofar as both parents are present and have rights over the child. The parents must therefore have some way of deciding between them which child should be adopted out when one of their consanguineal relatives makes a request. In this case the ideal rule, that the first child belongs to the father, the second to the mother, etc., appears to be a major determining factor. The distribution of children adopted out to the father's and mother's side by birth order is as follows:
Thus eighteen out of twenty-three children on the father's side (78 percent) and thirteen out of twenty-two children on the mother's side (59 percent) come from their expected positions in the birth order.
The data presented in this chapter suggest some generalizations that seem to hold for Rotuma and may have widespread applicability throughout Oceania and perhaps in other culture areas as well. First, it appears necessary to distinguish, for analytical purposes, between adoption that is initiated by an adopting agent and adoption initiated by a parent of the child. They are likely to differ in many ways and correspondingly are likely to require somewhat different models to account for them. In Rotuma the model needed to account for adopter-initiated transactions is complicated by motivational or strategic considerations so that it is impossible, on the basis of the evidence available to us, to specify from whom a person is likely to request a child for adoption. In contrast, for parent-initiated adoption the norms are explicitly based upon kinship rights and obligations, and it is possible to specify with a high degree of probability to which relatives children are sent. A second generalization that is suggested by the data is that the role of women in adoption transactions is more central than that of men, particularly in regard to adoption of younger children. This stems from the fact that a mother or mother surrogate is crucial to a child's well-being. Her presence is necessary if she is to fulfill her role obligations, while a father's obligations are essentially economic and can be fulfilled at a distance. It has been pointed out to me by Ward Goodenough that in Truk there is a strong feeling against allowing a woman who is not a consanguineal relative of a child to care for it; it is felt that a stepmother will discriminate against her stepchild.  This corresponds to Rotuman feelings on the subject and helps to account for the tendency for children to stay with the mother in cases of divorce and for them to be taken over by a grandmother or aunt rather than a stepmother in the event of their mother's death. One further generalization is suggested: children create a bind between the families of spouses that is more readily ruptured by divorce or separation than by the death of one of them. Thus the apparent willingness of widowers to place their children with their wives' relatives is in marked contrast to the claims they appear to exercise following a separation. Giving children up in adoption is an obvious way to keep affinal ties alive; removing children is an important part of severing relations. 
Finally let us return to consider the discrepancy in adoption rates for the population consisting of those adults whose life histories were taken (28 percent) and those cases occurring in the census material (12 percent). It was asserted that this probably represented, in part at least, a real decrease in fosterage rates through the years. If this is so, there are at least two logical possibilities to account for it. One is that Rotuman ideas about adoption, i.e., their decision-making procedures, have changed so that they adopt less frequently than before. The other is that their decision-making procedures have remained the same, but that the frequency of events leading to fosterage has altered. The former I would call culture change because it represents a different way of doing things. The latter can be called material change. Culture change includes the incorporation of new principles into decision-making models, the deletion of established principles, or a reordering of priorities. In any case, when culture change has occurred, the frequencies of specific choices alter, even though the circumstances under which they are made do not.
There is no evidence at our disposal to suggest that culture change has occurred with regard to adoption over the past three or four decades in Rotuma. The changes that may have taken place with respect to political and property functions had already occurred by the beginning of the twentieth century. On the other hand, there is definite evidence that material change has occurred. Specifically, a dramatic decrease in death rates during the past thirty years has meant that fewer families are dissolved by the death of a parent, and, as we have seen, the death of a parent, particularly the mother, significantly increases the probability of adoption.
The goal of this chapter has been to clarify, as well as possible on the basis of the information available, the nature of the decision-making processes through which Rotuman children are transferred from the households of their parents to adopting households. We have had to rely mainly upon demographic data and a general knowledge of Rotuman culture. Adoption was not systematically studied while the field investigation was being carried out; hence we have had to make inferences from an examination of the distribution of cases. Our procedure has been to try out in varying combinations the known repertoire of considerations that might be applicable to the activity of adopting children. This in my opinion is the central idea behind decision-making analysis. It is an attempt to register and understand the full range of considerations used by a group of people when making decisions, and then to show how these combine to generate predictable behavioral performances.
 In this chapter the term "adoption" is used to include what others have called "fosterage" as well as adoption in its strictest sense. The Rotumans do not make any such distinction lexically, nor was this distinction made in the collection of my field data. A number of people read a first draft of this chapter and made valuable criticisms. They include: Stephen Boggs, Vern Carroll, George Grace, Ward Goodenough, Robert Heighton, Irwin Howard, Chris Hulse, Sharif Kanaana, Bernd Lambert, and David Schneider. I am deeply indebted to all of them. [return to text]
 For a discussion of "decision-making models" as the phrase is used here, see Howard 1963b:409-412. [return to text]
 The five communities in Fiji were Suva, Vatukoula, Levuka, Tavua, and Lautoka. [return to text]
 For the purposes of this analysis all Rotuman households are considered as constituting one interaction system. This I feel is justified by the great deal of movement and communication back and forth between Rotuma and Fiji and between localities within these two areas (Howard 1961:277). Until 1961 most Rotumans maintained at least sporadic contact with their close kinsmen in every community. As a working assumption, this leads us to lump cases of adoption together without regard to locality, although it is presumed that proximity is one of the factors taken into consideration when a child is adopted out. [return to text]
 Cession took place in 1881. [return to text]
3. It shall not be lawful for any adopted child to be considered as a member of the family adopting it so far as inheriting by will shares in the family lands is concerned unless the person adopting it shall be the test of the family or sole owner of such land. [return to text]
 Rotuman consanguineal kinship terminology is strictly generational for all but ego's own generation, within which the cross sex versus same sex distinction operates. [return to text]
 This finding is the opposite of that found by Kay (1963b) and Finney (1965:293) for rural and urban areas in Tahiti. Among the Rotumans this difference is probably related to the demographic situation. Rotuman immigration to Fiji has been relatively recent (Howard 1961: 277) and the population is skewed toward younger age groups. [return to text]
 Personal communication. [return to text]
Churchward, C. Maxwell
Goodenough, Ward H.