The study of fertility in sub-Saharan Africa is fraught with the complexity of its family arrangements (Lesthaeghe 1989
). Depending upon the region and ethnic group, husbands and wives may spend little time actually living together, children may routinely be shared among families (fosterage), uncles may be more important than fathers in terms of childrearing, mothers may be more dependent upon their children than upon a husband for security, and women may find themselves in a polygynous relationship with an older man. These more traditional variations on family life are, of course, less common in cities, where Westernization has brought social change, not just economic change. Nonetheless, it is important to note that we have relatively few measures of these important aspects of life beyond ethnicity and religion, which serve as proxies for the type of family system in which a woman in likely to be enmeshed.
We have shown that there is a very clear spatial pattern in Accra of delayed marriage, and it is slightly better aligned with the vernacular neighborhoods than with the organic neighborhoods. This seems to be related to the fact that delayed marriage is influenced by ethnicity and religion, and the vernacular neighborhoods appear to capture variability in these characteristics, whereas we created the organic neighborhoods on the basis of a set of variables that were defined as status, since the enumeration areas were statistically more differentiated on that basis than on the basis of ethnicity and religion.
There is a less clear spatial pattern of childbearing after marriage, but it seems more closely aligned with the organic neighborhoods than with the vernacular neighborhoods. Thus, a woman’s cultural group seems to be more influential in the decision to delay marriage, whereas status seems to be the more important predictor of fertility within marriage. The slum neighborhood of Nima provides an interesting example of this. A majority of the adult population is Muslim and ethnically they are neither Ga nor Akan (the numerically largest ethnic groups in Accra). At the same time, it has one of the lowest status scores of any of Accra’s vernacular neighborhoods. The cultural mix is associated with one of the highest levels of delayed marriage, but the generally low status is associated with one of the highest percentages of ever-married women with at least three children. Nima’s overall level of fertility among women of reproductive age is lower than would be expected solely on the basis of status because the cultural pattern in the area is to delay marriage, even though fertility is high once a woman is married. Delayed marriage is seemingly not a route to lower fertility in this neighborhood, at least partly because it is not associated with women using that delay as an opportunity to improve their education. In Nima, 32 percent of women aged 15–24 had no schooling, and 76 percent had less than a secondary level of education, which are among the highest levels in Accra. Nor are they working in substantially higher proportions than in other neighborhoods. Forty percent of women 15–24 worked outside the family for even one day a week in Nima, which was only slightly higher than the average of 35 percent for all neighborhoods. It is probable that they are in charge of domestic duties within the household, perhaps including baby-sitting, but we could only know that from fieldwork, not from the census.
We measured the overall level of fertility in a neighborhood by creating an individual-level age-standardized measure of fertility that could then be directly aggregated to the neighborhood level. From the census we calculated the mean and standard deviation of the number of children ever born to women according to single year of age between the ages of 15 and 49. We then calculated each woman’s difference from the mean for her age in standard deviation units (z-scores). We label this as CEBz. Thus, a CEBz of zero means that a woman has borne exactly the same number of children as the average for all women her age in Accra. A positive value indicates that she has more children than average, and a negative value indicates that she has fewer children than the average for all women her age. We then aggregated the individual scores to the neighborhood level to generate an average fertility score for each neighborhood. This allowed us to decompose the overall level of fertility into the share attributable to delayed marriage and that attributable to childbearing after marriage.
Among the vernacular neighborhoods, the combination of the percent of women 15–24 who were never-married and the percent of women 15–49 who had given birth to fewer than three children accounted for 82 percent of the variability in CEBz, and the standardized beta coefficient for delayed marriage was almost twice the size of the coefficient for fertility after marriage. Thus, we can conclude that 64 percent of the variability in overall fertility was due to delayed marriage and the remainder to childbearing after marriage. The explained variance in overall fertility was also .82 among the organic neighborhoods, and delayed marriage was the most important component, but less than so than among the vernacular neighborhoods. Delayed marriage accounted for 55 percent of the overall level of fertility and childbearing after marriage accounted for 45 percent. These calculations underscore our results that suggest that the spatial dynamics and predictors of delayed marriage are different from those of reproduction after marriage. These findings also suggest that the context within which decisions about marriage are made are likely to be different from those in which decisions will be made about how many children to have, once at risk of having children.
At the individual level, we found a measurable contextual effect only for delayed marriage, and not for fertility after marriage. This is consistent with the literature, noted above, that neighborhood context may be more influential at the younger ages than at the older ages. However, the overall lack of explanatory power at the individual level is puzzling. It is not simply that the contextual effects do not emerge strongly—none of the fixed effects were powerful predictors. To be sure, all relationships were in the expected direction, especially the relationship with education, but the overall size of the effects was relatively small. We tried every possible combination of variables (not shown) without any improvement in the results. The census data have the strong advantage that they permit a detailed spatial analysis, but they have the strong disadvantage that the relatively limited number of questions asked on the census appear not to be capturing those things that are most influencing reproductive decisions in Accra.
The relatively small contextual effects, regardless of how spatially bounded the context (e.g, vernacular or organic in nature), fits into the larger discussion in the literature about the nature of these effects (see, for example, Subramanian et al. 2009
). Entwisle (2007)
and Matthews (2008)
are among those who have noted that (1) neighborhoods of residence (at which location we collect most data in the social sciences) may not be as salient as we think in determining human behavior; and (2) even if they are, the multilevel statistical techniques currently employed, as in this research, may not be as appropriate as we would like them to be for this task. We agree with both Entwisle and Matthews that the next step must involve more intensive field research that includes a clear spatial component. This may also require that we turn the tables on the usual approach to defining neighborhoods which is, as we did in this research, to define them on the basis of the expected “predictor variables,” and then we see whether the observed data, in this case fertility levels, match our expectations. Most research findings, including our own in this research, produce very modest levels of explanation of behavior at the individual level and so an alternative research strategy may be to create “contexts” of high and low fertility and then investigate, through data mining and field work, the factors that explain the individual variability in behavior among people living in those differing contexts.
In this paper, we have been able to take only a cross-sectional slice of fertility behavior in Accra, during a transitional time when fertility had stalled in Ghana and in Accra more specifically, and so we are not in a position from these data to fully understand these potentially dynamic factors underlying fertility change. Nonetheless, we have shown that an important component of the variability in fertility levels in Accra is the age at which young women marry. Over time, increasingly delayed marriage would have the effect of lowering overall levels of fertility through its tempo effect. In order for this to be associated with a stall in fertility, however, it would had to have meant that reproduction was actually rising within marriage. And, in fact, data from the GDHS do show that in 2003 the number of children ever born to ever-married women aged 25 through 39 was higher than it had been among ever-married women of the same age in 1998. From this, we can infer that the stall in fertility in Accra was largely a tempo effect from delayed marriage and, as we noted above, delayed marriage is also a major contributor to the intra-urban variability in fertility within Accra. On the other hand, data from the 2008 GDHS show that the delay in marriage leveled off between 2003 and 2008 after reaching near universality among women 15–19 and more than two-thirds of women aged 25–29. At the same time, the number of children born to ever-married women was lower at every age in 2008 than it had been in 2003, leading to an overall drop in fertility among women in Accra. We will be able to test the expected spatial patterns of these changes within the city after completion of the 2010 Census of Population and Housing.