Social Organization and the Prevalence of Nonmaternal Residence
A Model
Esther Goody's classic work on fostering developed a model in which fostering is a function of kinship, marriage, and inheritance systems on the one hand, and of social and political complexity on the other. In essence, the first group of variables determines both the extent to which the members of the lineage(s) to which the child belongs, have an interest in exercising direct rights over the child and the extent to which a child has property or other rights elsewhere. (It can also, through the intermediary of residence patterns, affect the extent to which it is in the direct interest of the parents, particularly the mother, to send a child elsewhere to live with kin in order to maintain close contacts and/or support there later—for example, in those societies where marital residence is neither uxorilocal or matrilocal but where women return to live near their own kin in old age). The second set determines the extent to which circulation of children, not only to kin but also to nonkin, can be used to develop or strengthen patron-client or alliance relationships either for training or to enhance social mobility.
Goody summarized the general implications of these considerations with three broad propositions concerning traditional fostering patterns (E. Goody, 1982, p. 275):
1. In undifferentiated segmentary societies, parental roles are unitary and there is little delegation of childrearing; these societies are characterized by very little, if any, fostering other than as response to family "crises" such as divorce or widowhood. In undifferentiated societies with matrilineal or double-descent systems, parental roles are not unitary; generally, however, only jural status and reciprocities involve anyone other than the biological parents (e.g., the mother's brother), while childrearing is still vested primarily in the biological parents and there is little occasion for fostering outside crisis situations.
2. Differentiated states are characterized by fostering extending beyond response to family crises. Goody makes in addition a distinction between simple differentiated states and more complex hierarchical states:
3. Simple differentiated states are characterized by fostering primarily to kin.
4. Complex hierarchical states are characterized by a greater importance of fostering to nonkin as well as to kin and by fostering for social mobility and for forging alliance or patron-client relations.
As Goody also indicated, this model can readily be extended beyond traditional social structures to include the effects of increasing social and
economic differentiation associated with contact with other populations (particularly trading and colonial contacts) and with modernization. Indeed, her model implies increasing levels of child circulation, especially circulation to nonkin for education, training, and sponsorship purposes, with the increasing social differentiation associated with modernization. The model could also be extended to include changes in kinship, marriage, and inheritance systems, although this seems less urgent given the slower pace at which these appear to be changing.
In Goody's model, modernization was not treated as a separate dimension but rather as an extension of the complexity variable. Figure 9.5 shows a simple extension that incorporates not only modernization as distinct dimension but also the fact that our data relate to nonmaternal residence rather than to fostering. Two sets of primary variables are distinguished. In the first are found the intensity and patterns of child circulation. In the second are found a series of variables reflecting other aspects of social organization. These are grouped under three headings:
Kinship and Marriage Variables
1. The form and strength of the lineage organization and inheritance systems provide an indicator of the potential interest of other lineage members in childrearing.
Figure 9.5.
Nonmaternal Residence and Social Organization
2. Women's marriage and residence patterns affect both the possibility for her children to reside with her and the desirability of having at least one raised elsewhere. We distinguish three key variables here:
3. Coresidence of spouses may constrain the possibilities for coresidence of children and their mothers since, where spouses do not live in the same household, it is not uncommon for the older children, more particularly the boys, to live with their father rather than with their mother.
4. Frequency of marriage dissolution also constrains the possibilities for mother-child coresidence, since older children of both sexes are then unlikely to stay with their mother: in patrilineal societies they are most likely to follow the husband or to go to a member of his lineage; in matrilineal ones they are more likely to go to one of their mother's kin.
5. Marital residence rules in combination with the residence patterns of women in old age, after divorce, or following widowhood determine the desirability of sending at least one child to be raised in the place the mother will later live.
"Traditional" Forms and Levels of Societal Complexity
1. Complexity of the traditional political structure.
2. Socioeconomic stratification as reflected in the traditional occupational or class structure.
"Modern" Forms and Levels of Complexity
1. Educational levels and heterogeneity.
2. Occupational diversification related to modernization.
Finally, two additional sets of variables have been included in figure 9.5 to complete the overall model, namely a control variable (the level of female adult mortality) and fertility itself.
Obviously regional and rural–urban breakdowns are as inadequate to handle this type of model as they are methodologically limited. The use of ethnic group as a unit of analysis, however, provides a solution to both substantive and methodological problems. Our further analysis proceeds, therefore, on the basis of results for ethnic groups.
Ethnic Group As Units of Analysis
Using ethnic groups with the WFS Household Questionnaires is not straightforward. We need, therefore, to describe our procedures in some detail.
Direct information on ethnic groups is available for all children only in Cameroon, although we can assume that practically all the children in the Lesotho survey are Sotho. In the Ivory Coast, ethnic group was asked only
for persons over age 15; we have linked the household head's data to each child and made the simplifying assumption that children were of the same ethnic group as their head of household.[6] In Sudan, too, ethnic group was not ascertained for individuals but at the level of the household; here, unfortunately, the meaning of the codes used for ethnic group is not available, so Sudan must be dropped from the analysis.[7]
In countries where ethnic group was not ascertained at all in the Household Questionnaire, we must use instead information from the Individual Questionnaire administered to women of reproductive age. In other words, we must link the data from two files. One possible procedure would be to identify for each child in the Household Files the data for his or her mother (if present) in the Standard Recode Files and then to assign to each child the ethnic group of the mother. Where the mother was not present one might assign the ethnic group of, say, the oldest woman interviewed in the household. Unfortunately, quite apart from any errors that might be introduced by interethnic marriages, this procedure would have the effect of excluding all children enumerated in households where there was no woman of eligible age interviewed. Since we suspect that a not insignificant portion of child circulation is movement of children to elderly persons, to help them in their household tasks and to provide companionship, this would be a potentially very serious loss. We have, therefore, opted for an alternative approach. We first split the Standard Recode Files into the smallest sampling areas used (there are between 150 and 250 areas per country in the countries concerned) and examined the ethnic distribution of women of reproductive age in each area. We then assigned to each child in the Household Files a probability of belonging to each ethnic group equal to the proportion of women in that ethnic group in his or her sample area. In other words, when making estimates for ethnic group A, each child has received a weight equal to the proportion of women in his or her sampling area who were from group A. Since the ultimate sampling units were usually small and rather homogeneous ethnically, this works rather well. In terms of the ethnic groupings used here, on average over 80 percent of the women in an area belonged to the same group; just over one-quarter of the areas were fully 100 percent homogeneous, and in over half, 90 percent or more of the women interviewed belonged to a single group.[8]
Finally, we should note that data on ethnic group can be quite hard to collect. The amount of assistance given the interviewers and coders (e.g., lists of the various names used for the different ethnic groups and subgroups, and detailed instructions concerning the way in which subgroups were to be recorded) varied considerably between countries. Moreover, ethnic group data can be extremely sensitive. Given the difficulties involved, it is perhaps not surprising that we cannot present results by ethnic groups for all the countries. Not only are ethnic group codes not available for Sudan, as we have
already mentioned, but, in addition, permission to use the Nigerian data does not extend to publication of any data by ethnic group.
For this chapter we have used for each country the broadest ethnic groupings given in the Standard Recode Files (apart from a few exceptions related mainly to sample-size considerations). The disadvantages of using these particular groups stem from the fact that many are broad language groups rather than ethnic groups per se; some are, therefore, highly heterogeneous internally (e.g., the Northern Ghanaian groups). For these cases our data are rather bland averages over quite highly contrasting groups. The main advantage of using these groups is ready comparability with other analyses based on the WFS data, including analyses found elsewhere in this volume.
Our estimates of proportions of children not living with their mother, specific by the children's age and sex, are documented for the five countries where documentation is possible in table 9.3. Our subsequent analysis attempts to relate the age- and sex-specific levels of nonmaternal residence to selected social organization variables. More specifically we relate them here to the following:
1. Among the kinship and marriage variables we have used lineage organization, coresidence of spouses, and the frequency of marriage dissolution.
2. For complexity of the "traditional" society we have used an indicator of political complexity and a measure of caste and class stratification.
3. For "modern" forms of societal complexity we first examined four indicators of educational and occupational heterogeneity—the proportion literate among women aged 15–49, their average years of schooling, the proportion of their husbands working in high-level jobs or as employers themselves, and the proportion of husbands in the traditional self-employment sector. A factor analysis showed that the proportion literate loads much more heavily on the first factor than do any of the other three variables. For simplicity of interpretation, therefore, we have simply used the proportion literate. Proportion urban is also included as a separate variable in some analyses, since it is not strongly correlated with educational or occupational differentiation between ethnic groups.
Our covariates—literacy, urbanization, marital residence, and coresidence of spouses —were all derived from the WFS files. They are thus derived from the same sample of households as our child-residence variables. The first three can be derived immediately from the Standard Recode Files, which refer to women of reproductive age located in the households covered by the Household Questionnaire.[9] They were operationalized as follows:
1. Literacy: percentage reporting themselves as literate, women aged 15–49
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
2. Urbanization: percentage residing in urban areas, women aged 15–49
3. Marriage dissolution: percentage of all ever-married women aged 15–49 who have ever been divorced or widowed.
Our fourth covariate is defined analogously:
4. Coresidence of Spouses: percentage of all currently married women aged 15–49 recorded as living in the same household as their husband.
Unlike the first three, however, this fourth covariate was not derived directly from the Standard Recode Files. These files do include data from a question asked each currently married woman as to whether her husband usually lived in the same household as she did. However, the question used varied in content between countries. Moreover, the data obtained were not included in the Standard Recode Files for all the countries. We have, therefore, used an alternative source of information located in the Household Files. On each Household Questionnaire, a link was made between each husband and his wife or wives, and a code identifying him and his wife/wives (the "couple code") was included routinely in the Household Files. Women with no husband present received a special code. In some countries, an explicit distinction was made between those with no husband present because they did not have a husband at the time of the survey (single, widowed, and divorced women) and those whose husband was elsewhere; this distinction can also easily be made by drawing upon data for a separate variable on marital status. In these countries it is a simple matter to estimate the percentage of currently married women whose husbands were enumerated in the same household, using just the Household Files. Not all countries included information on marital status in their Household File, however. For Kenya and Ghana, for example, there is no separate variable for marital status, and for the couple code no distinction is made between women who have no husband and those whose husband is elsewhere. For these countries we have used the proportion of women not currently married obtained from the Standard Recode File to adjust the proportion of women with no husband present recorded in the Household File. The estimated proportion of currently married women whose husbands are elsewhere is given simply by:
The data for our factors—lineage organization, traditional political complexity, and stratification —were extracted from Murdoch's Ethnographic Atlas (1962–1967, with updates through 1983), after identification of the ethnic group(s) in the Atlas corresponding most closely with the categories used in the WFS surveys. Since the number of observations incorporated in this analysis is rather small (sixty broad ethnic groups), we have reduced all the factors to simple dichotomies:
1. For our variable lineage, the small number of groups exhibiting bilateral or duolateral traits have been combined with the matrilineal groups:[10] our variable thus contrasts patrilineal societies (75 percent), with all the others (25 percent).
2. Political complexity contrasts societies with no chiefs or only petty chiefs on the one hand (65 percent), with states and with societies with paramount chiefs on the other hand (35 percent).[11]
3. Stratification contrasts societies with no stratification (or only despised, usually small, "caste" groups) (65 percent) with those exhibiting stratification by wealth or other more complex stratification (35 percent).[12]
Results
By way of exploration we pursued two lines. On the one hand, we conducted a factor analysis on the proportions of children not living with their mother in order to identify common patterns by age, sex, and rural-urban residence. On the other hand, we examined the effect of our social organization variables on nonmaternal residence levels using multiple classification analysis (MCA). First we did this on a very exploratory basis using the proportions by age; MCA is not the most appropriate technique in this regard, but it should suffice for the present exploratory purposes. Then we applied MCA to the factors extracted from the factor analysis. The results are presented in tables 9.4 through 9.6.
Turning first to the most exploratory MCA statistics on the proportions of children not residing with their mother, by age (table 9.4), we can note the following:
1. As expected, the explanatory power of our social-organization variables increases with the age of the child. The proportion of the variance in nonmaternal residence that they explain increases from 47 percent (39 percent, after adjustment for the large number of variables relative to the number of observations) for children under 5 years to 70 percent (66 percent after adjustment) for those aged 10–14. The effect of differences in social organization is clearly stronger for older than for younger children, who still tend to live with their mothers even in those ethnic groups that have social organization characteristics associated with high levels of nonmaternal residence in general.
2. Among the factors, lineage type is usually the most important factor in terms of zero-order associations, and it remains so when all variables are introduced simultaneously. Societies with matrilineal characteristics and those exhibiting any bilateral or duolateral traits tend to have slightly higher proportions of children not living with their mother. When the other variables are introduced, however, the effect generally becomes less marked (especially for the older children), although the beta values are statistically significant.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
3. Somewhat surprisingly, political complexity and class/caste stratification—two variables one might expect to be of considerable importance in the light of Esther Goody's model of traditional child fostering—do not exhibit significant systematic effects in the expected direction. The deviations are often small (especially for the central age group of children), and even where they are larger they do not reach the 5 percent significance level. The irregular pattern may, of course, be the result of relatively weak operationalization of these variables. It is also possible, however, that "modern" forms of social differentiation may be taking over from "traditional" ones.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
4. Among the covariates measuring forms of heterogeneity associated largely with modernization, however, education does not appear to have a very strong effect. The beta coefficients for literacy are generally small and cannot be said to exhibit the pattern we had hypothesized, namely that higher levels of literacy would be associated, through greater educational heterogeneity, with more child circulation, especially for older children. The beta values for the two older age groups are negligible. The coefficient for the youngest age group, however, is not entirely negligible (and reaches the 1 percent significance level). Although higher levels of literacy do not appear to lead to more child circulation among older children, they do appear to lead to slightly higher proportions of very young children living away from their mothers. Presumably this is a reflection of the increasing difficulty in caring for very young children already discussed in the context of rural-urban differentials. It could also be related to the sending of children to their grandmothers when they are to be weaned—especially if they are to be weaned relatively young—as referred to by Isiugo-Abanihe (1985).
5. The estimated effects for degree of urbanization are however somewhat larger, particularly for older children, as was expected. For young children, higher levels of urbanization appear to have little impact on the proportion of children not living with their mother; perhaps any effect they have has already been taken up in the effect of literacy. For the oldest children the levels have a noticeable impact, as hypothesized: more urbanized ethnic groups have higher proportions not living with their mothers.
6. Our nuptiality variables play a more important role. As expected, coresidence of husband and wife has no effect for younger children, but it has a clear (and highly significant) effect for children above age 5. The magnitude of its effect is comparable to that of urbanization.
7. By far the largest effects of all are recorded for the variable measuring frequency of marriage dissolution. This variable is highly significant for all age groups and, as expected, increases markedly with age.
Overall, the marriage and kinship variables dominate over our social-complexity variables. Before jumping to conclusions, however, we should look at tables 9.5 and 9.6.
Table 9.5 shows the results of the factor analysis. When no distinction between rural and urban areas was made within each ethnic group (top panel), only one factor emerged: "all age and sex groups" loads relatively heavily on this factor (although the factor loadings are perceptibly lower for children under 5). When a distinction was made between rural and urban areas (bottom panel), three main factors emerged. The first is dominated by variation in rural values, the second by variations in values for older children
(especially older urban children), whereas the third factor predominantly reflects variation among very young children in urban areas. Rural, presumably more traditional, variations in child circulation levels thus emerge as the most important component of interethnic variability. Next comes a dimension capturing differences in child-training and child-labor patterns, especially although not exclusively in urban areas. And this is followed by a dimension reflecting interethnic differences in child-nurturing patterns in urban areas.
Table 9.6 takes the three components derived from the factor analysis and submits them to an MCA. Since an explicit distinction was made between rural and urban areas in creating the components, the variable urbanization can be dropped. Before discussing the results we should note that statistical significance is slightly harder to reach here than in table 9.4, because of a slightly smaller sample size: ten ethnic groups with a very small number of observations for a particular age–sex–rural/urban residence group were excluded. Not surprisingly, the results for the first component—the general, predominantly rural, prevalence of nonmaternal residence—are quite similar to those shown in table 9.4. The proportion of the variance explained is about 60 percent (54 percent after adjustment). Frequency of marriage dissolution is the only variable to be significant at the 1 percent level, although both the other covariates—coresidence of spouse, and literacy —and the lineage factor are significant at the 5 percent level. For the second and third factors, which load heavily on particular age groups and on urban areas, the model performs less well overall here, with proportions explained around 30 to 40 percent. The beta values are, in general, smaller than in table 9.4, and the 5 percent significance level is rarely reached. As in table 9.4, however, systematic deviations in the expected direction are observed for lineage organization and for frequency of marriage dissolution (although, as expected, the latter has no impact on the "urban nurturing" factor). Comparing table 9.6 with table 9.4 shows that the main features of table 9.4 are confirmed at least in the rural patterns where the model performs best.
These results draw attention to two issues. First, they confirm the prime importance of kinship and marriage patterns in general, and of the prevalence of marriage dissolution in particular, for the prevalence of child circulation. Their importance should come as no surprise. Esther Goody has already identified the importance of lineage organization in general, and there are ample grounds for expecting marriage dissolution to play a major role: (1) To begin with, there is the direct effect of divorce (and widowhood) already referred to—what Esther Goody refers to as crisis fostering. The anthropological literature contains numerous references to children, especially older children, living with their father or with other relatives after a marriage is dissolved. (2) There may be indirect effects leading to nonmaternal residence even when the parents' marriage is still intact. For example there may be
"insurance" against marital dissolution in societies marked by high instability of marriage (Frank, 1985). Where a married woman returns to her own kin at widowhood or divorce, there may be a greater tendency to ensure that at least one child is raised there in order to make certain that she always has a child to return to who will help support her; the child is also more likely to have rights there. Or, as Lallemand (1976) described for the Mossi, older women in the lineage may use a high risk of divorce as an argument for taking a child away from a woman who has not been married long in the compound, on the grounds that the child should be reared by women who have proven to be loyal and stable members of it. (3) Finally and more generally, as the Goodys suggested nearly 20 years ago in the context of child-fostering (J. Goody and E. Goody, 1967), circulation of children and circulation of women may be two intertwined elements of a single system. Where either marriage itself and/or the transfer of rights in a woman's reproductive capacity to her husband's lineage on marriage are essentially irreversable, neither children nor women circulate; where marriage and rights in a woman's reproduction are more flexible, both circulate.
The second finding is the rather surprising absence of systematic marked effects related to educational levels (at least as measured by literacy) and the only moderate effect of urbanization. These are found to play only a relatively minor role in determining differences between ethnic groups—although urbanization is significant for older children. However, this does not rule out the possibility that these variables exercise their influence within ethnic groups, affecting the pattern rather than the prevalence of child circulation. As figure 9.6 illustrates for literacy—which had almost no effect on between-group differences—within ethnic groups these variables indeed play a strong role. Within ethnic groups there are marked differentials in the proportion of children not living with their mother, depending on the literacy level of the area where the child lives. The prevalence of nonmaternal residence is nearly always higher for areas that have intermediate or high levels of literacy, indicating either more circulation within, or movement of children to, the more privileged areas. These variables can, therefore, be seen as constituting important determinants of internal redistribution patterns, even if they have only a small impact on the overall level of redistribution.