Demographic Change in Sub-Saharan Africa (1993)

Barney Cohen

Fertility rates are higher in sub-Saharan Africa (Africa) than in any other major region of the world, and considerable controversy surrounds the likelihood of these rates declining in the near future. Although mortality and fertility rates fell substantially in Latin America and Asia between 1965 and 1985, only mortality declined in Africa; fertility remained relatively stable, well above a level required to replace the population. Consequently, the region experienced extremely rapid population growth, with rates for some populations considerably above 3 percent per year (United Nations, 1991; Freedman and Blanc, 1992). A few countries, most notably Kenya, Botswana, and Zimbabwe, have begun the transition toward lower fertility, but smaller declines in fertility have been observed recently in many other countries. Nevertheless, fertility rates generally remain above six children per woman, and the question of whether Africa is more resistant to fertility change than other regions of the world is a topic of considerable debate

(Boserup, 1985; World Bank, 1986; Caldwell and Caldwell, 1987, 1988, 1990; Lesthaeghe, 1989; van de Walle and Foster, 1990; Caldwell et al., 1992).

The level of fertility in sub-Saharan Africa, as measured by the total fertility rate (TFR),¹ is approximately 6.0–6.5 births per woman. This figure masks considerable variation between regions and between individual countries. For example, the most recent estimate of the total fertility rate in Rwanda (8.5 births per woman in 1983) is almost double the most recent estimate for the population of black South Africa (4.6 births per woman in 1987–1989). More generally, fertility rates in East and West Africa are greater than those in Central Africa, in part because of the historically high prevalence of sexually transmitted diseases (STDs) in certain areas of Central Africa (Frank, 1983; Tambashe, 1992). The prevalence of STDs is associated with unusually high rates of infecundability in the region especially prior to the 1970s. Fertility was probably higher in East Africa than in West Africa during the 1970s and 1980s, although the difference appears to have lessened in the more recent past. Reported fertility rates rose in certain parts of Africa in the late 1960s and 1970s; however, it is not clear what proportion of the increase was the result of improvements in data accuracy.

In addition to the regional and national variation in fertility rates, there is often considerable variation in fertility within countries. Repeatedly, fertility surveys have recorded substantial differences in rates among ethnic, geographic, and socioeconomic groups. For example, fertility rates are consistently lower among women who live in urban areas, women who have more than primary school education, and women who work in the formal labor market. In Africa, the number of women in each of these socioeconomic groups has, at least until recently, been small, and the groups overlap considerably. Consequently, lower fertility among these women has a minimal effect on national-level TFRs.

The objective of this chapter is to summarize existing knowledge on levels, trends, and differentials in achieved fertility in sub-Saharan Africa. Although there have been several comprehensive reviews of fertility levels in Africa in the past (see, for example, Brass et al., 1968; Page and Coale, 1972; United Nations, 1987), new sources of data make it possible to update

the analysis to the early 1990s. By employing a wide variety of data sources, including some that have not been readily accessible in the past, estimates of fertility rates are presented for virtually all countries in sub-Saharan Africa.

True understanding of fertility trends in Africa is clouded by the extremely variable quality of demographic data in the region. Close examination of much of the data reveals gross inconsistencies that are the result of misreporting of ages and omitting or systematically displacing vital events. In an attempt to correct for obvious data errors, a mixture of direct and indirect estimation techniques is used to determine fertility rates. The indirect techniques are based on the examination of inconsistences within the reported data or on comparisons of observed data to values expected from various demographic models.

The chapter is organized as follows: Issues of data availability and quality are discussed in the following section. In the third section, four methods for estimating TFRs are presented. Characteristics of African fertility are presented in the fourth section. Next, recent data from the Demographic and Health Surveys are used to examine the possible evidence for declining fertility levels in Africa. The penultimate section compares recent fertility trends in Africa to those in other developing areas of the world. Finally, there is a summary and some concluding observations.

The state of demographic data collection in Africa has recently been reviewed by de Graft-Johnson (1988). Despite dramatic improvements since the 1960s, our knowledge and understanding of fertility levels and trends in Africa are still surprisingly weak. Until 1960, virtually no sub-Saharan African country had conducted a complete census. Consequently, little was known about the size or structure of the region’s population. In the few countries where censuses were undertaken, they were often unreliable and of very limited content. A fundamental problem facing researchers in Africa was that a large percentage of the adult population was unable to report its age accurately. Further, many early censuses did not include questions related to the number of children ever born and childhood mortality. In addition, vital registration data were virtually nonexistent throughout the region and, when available, were of questionable quality.

Fortunately, demographic data collection in Africa has improved considerably over the last 30 years. Although vital registration is still rare, most countries have conducted one and in many cases several censuses, though quality has been uneven. In addition, many countries have supplemented efforts to collect reliable demographic data with various ad hoc

national and subnational household demographic surveys. Some of the most accurate information comes from these large-scale demographic surveys. In particular, the World Fertility Surveys (WFS), an international data collection effort undertaken from the mid-1970s to the early 1980s, and the ongoing Demographic and Health Surveys (DHS) begun in the mid-1980s, have generated a reasonably accurate data base for calculating fertility levels and differentials for countries in sub-Saharan Africa. The WFS carried out surveys in nine sub-Saharan African countries: To date, the DHS has published demographic reports for 13 sub-Saharan African countries and issued preliminary results for 4 others. Reports for 4 additional African countries are scheduled for release by the end of 1993. Special attention is given in this chapter to the DHS because it is the source of most of the recent demographic data on Africa.

The quality of DHS data was recently analyzed by DHS staff and found to be generally acceptable. But, in cases where data problems were identified, they were determined to be most severe in sub-Saharan Africa in comparison to other regions of the developing world (Institute for Resource Development, 1990:2). For example, Arnold (1990) identified errors in the coverage and timing of births, including (1) systematic displacement of children’s birth dates, (2) disproportionate numbers of women’s ages heaped on digits ending in 0 and 5, and (3) missing or incomplete information in some birth histories. These problems were determined to be most severe in Botswana, Burundi, Liberia, Mali, and Togo. Problems in the first category arose, in part, because some interviewers appear to have deliberately altered the ages of children under 5 to avoid asking an extensive series of questions on the health and well-being of young children. A second assessment of the quality of DHS data focused on women’s age at first birth and judged that response problems were most severe in African countries, especially Mali and Liberia (Blanc and Rutenberg, 1990). The African data suggest that some women omit information about early births or displace the dates of low-parity births forward in time, making children appear younger than they really are. Finally, Rutstein and Bicego (1990) report that less than 80 percent of women interviewed in Africa provide accurate birth dates for their children.

Fortunately, the effect of displacement problems on fertility levels is relatively minor. For example, Arnold and Blanc (1990) calculate that without any displacement, the total fertility rate in Liberia, the country with the most displacement, would have been 6.5 instead of 6.3 births per woman between 1983 and 1988. Nevertheless, it is important to acknowledge that there is always the danger of drawing incorrect conclusions from data collected in areas where vital events go unrecorded. Consequently, a single point estimate of fertility from Africa should be interpreted with some caution.

In this chapter, DHS and WFS data are supplemented by data collected in censuses and other national demographic surveys. Where no other information was available these data are augmented with findings from large-scale subnational studies. Data from small-scale studies conducted at the district or provincial level have not been used due to concerns regarding their generalizability.

Naturally, censuses and surveys are carried out in different countries at different times. But, unlike the estimates presented by several organizations (including the United Nations and the U.S. Bureau of the Census), the estimates presented here are not standardized on a specific year. Rather, the current goal is to present the reader with the original data from which standardized estimates are derived.

Four distinct strategies are used here to obtain independent estimates of fertility. The first strategy is to calculate fertility directly, without adjusting for any apparent inconsistencies in the data. This method requires information on the number of women of childbearing age, their ages, and the number of births to these women during a given time period, typically five years. Direct estimates of fertility are reported only when the quality of the data was thought to be adequate, for example, as in all the WFS and the DHS. In these cases, fertility estimates are derived by using retrospective birth histories.

Experience has shown, however, that response errors in census and survey data can often lead to biased or inaccurate estimates of the fertility rate. Response errors in birth history data arise mainly from age misreporting and the omission or systematic displacement of vital events. For example, many women incorrectly report their own age or the ages of their children. Similarly, in the absence of written records, women often forget births that occurred in the distant past and make systematic errors when estimating the timing and spacing of events (Potter, 1977). Older women, women with little education, women who were not in sanctioned unions at the time of their first birth, and women whose children have moved away or died are particularly likely to make these types of errors. Obvious errors, such as birth intervals of less than 6 months or first births to women under 10 years of age, can often be detected by the interviewer or the researcher and perhaps corrected by cross-referencing birth dates with well-known historical events. Errors resulting from omitted births are much harder to correct.

Demographers have developed alternative methods designed to improve their ability to make “indirect” inferences about fertility from poor or incomplete data (Brass et al., 1968; United Nations, 1983). Most of these methods involve the identification of internal inconsistencies in the reported

data or the comparison of observed data to model fertility schedules. In cases where the direct and indirect estimates of fertility are substantially different, the indirect estimates are usually preferred. Three of the four strategies used in this chapter employ indirect techniques.

One strategy is based on the principle of comparing reported births in a given time period with women’s responses to questions regarding the number of children ever born. A full description of this method (commonly called the method of P/F ratios) can be found in United Nations (1983). The data requirements for this method are identical to those for direct estimation except that they include information about the number of children ever born. Where the data allowed, this technique was used to check and, if necessary, to adjust the survey or census estimates for apparent misreporting. Unfortunately, because this method relies on equating current and past experiences, it has the potential for producing biased estimates of the total fertility rate when fertility has recently declined (United Nations, 1983:32). Nonetheless, at least for the earlier time periods, this method arguably produces the most accurate estimates possible.

Because early censuses often did not include specific questions on fertility, the age structure of the population may be the only information available to estimate the total fertility rate. In these cases, fertility estimates are inferred by using stable population theory, which is based on assumptions of constant fertility and mortality. The only data requirements for these estimates are the age structure of the population, the growth rate, and an estimate of the level and pattern of mortality. Because results from this estimation method are not particularly robust and are quite sensitive to different mortality assumptions, it is used only in the absence of other alternatives. In an attempt to check the robustness of these approximations, similar estimates are also derived by using a method developed by Coale (1981) and later extended by Venkatacharya (1990). This method, labeled the Coale method, also relies on stable population theory and requires an estimate of the population growth rate, the proportion of both sexes under the age of 15, and an estimate of mortality for children up to age 5. Assuming constant fertility rates for the population under consideration, Coale (1981) suggested that his method would yield reasonable estimates of the total fertility rate for 7.5 years prior to the census date, even if the census or survey was characterized by severe age misreporting.²

Table 2–1 provides estimates of the TFR for 38 African countries for which data were available at some point during 1960–1992. For ease of comparison, all estimates were converted into point estimates of the total fertility rate. In reality, a particular figure may be the midpoint of a range of plausible estimates. Table 2–1 highlights the paucity of demographic data in many African countries. Despite considerable improvements in the availability of data during the past 10 years, 12 countries still have fewer than four data points since 1960. In other cases, although data exist, they are of extremely variable quality. Consequently, fertility trends over time may appear more erratic than they truly are. For example, in Ethiopia, the data imply a substantial increase in fertility during the 1970s followed by a rapid decline in the 1980s. Both trends are almost certainly exaggerated.

Although the data are often sketchy, several important conclusions may still be drawn about fertility in Africa. Few countries in Africa have TFRs less than 6.0, and nowhere is fertility currently less than 4.0 births per women, a rate well above that required for replacement. Africans have a strong preference for large families. Children are prized not only as the means of preserving family lines, but as positive economic assets that provide labor, wealth, risk insurance, and old-age security to their parents.

In the past, high fertility in Africa resulted from early and near universal marriage,³ and extremely low rates of efficient contraception. Fertility has been controlled (outside geographic areas of pathological sterility) by social pressures against premarital sex, the practice of postpartum sexual abstinence, and long breastfeeding periods that lead to lengthy lactational amenorrhea (see Chapter 3; also Caldwell and Caldwell, 1977, 1987; Page and Lesthaeghe, 1981). Bongaarts et al. (1990) recently estimated that fertility in Africa would increase by 72 percent if the fertility-inhibiting effects of breastfeeding and postpartum abstinence were removed. These fertility-reducing practices have probably evolved principally to ensure exceptionally long birth intervals in an effort to combat high rates of infant mortality. Recently there are signs that some of these cornerstones of African fertility may be weakening (see Chapter 3; also Schoenmaeckers et al., 1981; Caldwell et al., 1992; Westoff, 1992).

Most recent estimates of the TFR in each country are shown in the map (the data for the map are from Table 2–1). African fertility rates are not homogeneous across the continent, and close investigation of demographic data reveals considerable diversity. In the 13 West African countries in Table 2–1, the estimated TFRs lie within a narrow range between 6.2 and 7.4 births per woman. Fertility in West Africa is highest in Niger, where the TFR was estimated to be 7.4 children per woman in 1992, and lowest in Nigeria, where the TFR was estimated to be 6.2 children per woman in 1990. For the 12 East African countries discussed in this chapter, fertility ranges from 5.5 children per woman in Zimbabwe to 8.5 children per woman

Total fertility rates in sub-Saharan Africa. SOURCE: See Table 2–1; estimates are the most recent available.

in Rwanda. In the majority of East African countries the TFR is between 6.4 and 7.0 children per woman.

Average fertility in Central Africa is somewhat lower than it is in either East or West Africa. Fertility in Central Africa was lower during the 1960s and 1970s than elsewhere in Africa because of high levels of infertility and subfecundity. In all probability, the most common cause of infertility in sub-Saharan Africa was a high prevalence of gonorrhea (Frank, 1983). Childlessness is a good indicator of overall infertility, and a rate of childlessness of 3 percent is what one might expect to see in a developing country (Frank, 1983). In Central Africa, the proportion of women aged 45 and over who have not had a live birth ranges from 11 percent in Chad and Angola to more than 20 percent in Congo, Gabon, and Zaire (Frank, 1983).

Fertility estimates are included for five countries in southern Africa: Botswana, Lesotho, Namibia, South Africa, and Swaziland. Most recent estimates for Botswana, South Africa, and Swaziland indicate that the TFR in these three countries lies between 4.5 and 5.0 children per woman. Fertility is slightly higher in Namibia and Lesotho, although the latest estimate of the TFR in Lesotho (5.2 children per woman in 1986) is now quite dated.

The black population of South Africa experienced a decline in fertility before any country of sub-Saharan Africa (Caldwell and Caldwell, 1993). This fact has been poorly documented for two reasons. First, a long period of international political isolation has meant that little has been written about the demographic situation in South Africa over the last 30 or so years. Second, the South African government has been reluctant to provide open access to demographic data. In addition, the primary means of collecting demographic data—the registration of births and deaths—has not been complete because black South Africans have not felt sufficiently vested in ensuring its accuracy (Caldwell and Caldwell, 1993).

Fortunately, the situation has improved recently. The country has entered a period of political, social, and economic reform, and new data have emerged from a DHS-type household survey conducted in 1987–1989 under the auspices of the Human Sciences Research Council of South Africa. There were some initial concerns about the general lack of documentation of the sampling and methodological strategies that were employed to collect these data (see, for example, Freedman, 1992; Caldwell and Caldwell, 1993), but these issues have now been rectified (R.Freedman, personal communication, 1993). These data reveal that the black population of South Africa currently has lower rates of childbearing and higher rates of modern contraceptive use than any country in sub-Saharan Africa. However, there will probably be a great deal of debate about the extent to which comparisons should be drawn between the South African experience and the experiences of other countries in the region.

Undoubtedly, African fertility has varied over time. The longstanding

belief that African fertility levels have remained the same for long periods is probably a result of the absence of reliable time-series data (Page, 1988). The seemingly regular occurrence of catastrophic events certainly creates substantial short-run variations in demographic rates that often go unrecorded. However, the subject of short-term fluctuations in fertility lies beyond the scope of this chapter.⁴ The only changes in fertility rates discussed below are the more permanent ones that have occurred over longer periods of time.

There is some evidence to suggest that fertility rates rose in several African countries during the 1960s and 1970s. As stated earlier, it is unclear what proportion of the change is genuine and what proportion is attributable to improvements in data collection. A popular example of a country in which fertility rates may have increased is Kenya. Historical fertility estimates for Kenya are available from the 1962 Post-Enumeration Sample Census, the 1969 census, and the 1977 National Demographic Survey. At face value, the data from these sources indicate that fertility rose dramatically from 5.3 births per woman in 1962 to 6.6 in 1969 and to 8.0 in 1977. Extensive manipulation resulted in official estimates being revised to 6.8 for 1962 and 7.6 for 1969. However, it is now apparent that the shapes of the age-specific fertility distribution derived from both the 1962 and the 1969 censuses were almost certainly biased (Blacker et al., 1979). Despite the extensive official data manipulation, TFR estimates for 1962 and 1969 are probably still too low. Fertility probably increased in Kenya between 1962 and 1977, but the true extent of the increase is unknown.

An increase in fertility in recent decades appears more certain in Cameroon and certain other Central African countries. Of the eight Central African nations included here, only in Cameroon and Angola was fertility higher than 5.5 births per women before 1975. Currently, fertility is estimated to be above this level in all six countries for which data are available after 1975. This increase in fertility has been attributed largely to a reduction in the historically high incidence of pathological sterility in Central Africa resulting from widespread STDs (Frank, 1983; Tambashe, 1992).

Recent indications exist that fertility may be falling in several sub-Saharan African countries, a topic that is taken up in more detail below. Table 2–1 shows fertility falling in 26 countries. In most cases, however, the declines in fertility are quite small, less than one birth per woman. The observed declines could be a function of using alternative strategies to estimate fertility at different points in time or the result of using unreliable

data. Evidence of declining fertility is strongest in Kenya, Botswana, and Zimbabwe. Apart from Ethiopia and Malawi, where the data are too unreliable to detect the magnitude of a decline with any measure of certainty, these are the only countries that show a recent fertility decline of more than 1.0 birth per woman. Recent data from various DHS surveys indicate that fertility may be falling in other countries as well, including Senegal, Zambia, and parts of Nigeria. However, in these and other cases, the changes in fertility are smaller and less definite. Further observation is required to confirm these trends.

Detecting the start of a decline in fertility is not always easy. The number of children a woman bears is the outcome of a series of complex interactions among biological, social, economic, and other factors. These factors can be conveniently divided into the proximate (biological and behavioral) determinants that directly influence fertility, and all other social and economic factors that affect fertility only indirectly through the proximate determinants (Bongaarts and Potter, 1983). Early detection of fertility decline through behavioral changes affecting the proximate determinants is made difficult because certain aspects of socioeconomic development can have competing effects on the proximate determinants that cancel each other at low levels of development (Lesthaeghe et al., 1981). Hence, early detection of fertility decline would be considerably easier if we had more precise measures of each of the proximate determinants and a clearer understanding of how these variables relate to one another. (See Chapter 3 for a more indepth discussion of proximate determinants in sub-Saharan Africa.) Finally, fertility and nuptiality patterns may be affected by the recent implementation of structural adjustment programs by several African governments in response to worsening economic conditions. It is unclear whether any crisis-induced reductions in fertility would be sustainable for long periods of time. It is worth noting, however, that economic conditions in Kenya, Botswana, and Zimbabwe have been better than average for the region during the past 10 years (van de Walle and Foster, 1990).

A great deal can be learned about the timing and intensity of childbearing from a simple examination of the shape of the fertility schedule by age. However, construction of fertility schedules requires reasonably accurate data because the omission or misplacement of vital events can lead to serious errors. Recent estimates of age-specific fertility rates (ASFRs) are presented in Figures 2–1 to 2–4. The data for these figures are provided in Table 2–2. The data reveal a remarkable degree of similarity among estimates across countries, particularly given the many difficulties associated with collecting accurate demographic data in an area where significant per-

centages of the population are unable to recall either their own or their children’s dates of birth. Below, first births and the timing of subsequent births are discussed.

The age at which a woman gives birth for the first time has important implications for the health and well-being of the mother and her child (Haaga, 1989). Compared to older women, teenage mothers face greater risks of pregnancy- and delivery- related complications, maternal morbidity and mortality, and having premature or low-birthweight babies. Significant educational and economic consequences also result from having children at a very young age. The most publicized examples of these consequences relate to lost educational opportunities (Working Group on the Social Dynamics of Adolescent Fertility, 1993). Under a natural fertility regime, with little or no use of modern contraception, the mother’s age at first birth is also an important determinant of completed family size.

In general, African countries have relatively high rates of adolescent fertility, and the median age of women at first birth in sub-Saharan Africa is approximately two years younger than it is in North Africa, Asia, or Latin America (Arnold and Blanc, 1990). Table 2–2 reveals that the largest differences between age-specific fertility schedules for a subset of countries in sub-Saharan Africa occur among women aged 15–19. This finding is largely attributable to differences in the average age of marriage in these countries. For example, teenage fertility is highest in Mali, where the mean age of 15.7 at first marriage is among the lowest in Africa (Traore et al., 1989). Similarly, adolescent fertility is low in Burundi where the mean age at first marriage is relatively high, 19.5 years (Segamba et al., 1988).

In the past, the vast majority of childbearing in Africa probably took place within the institution of marriage. However, recent survey data have identified a weakening of the link between the age at which a woman first marries and the age at which a woman first gives birth. What appears to be occurring in several countries is that the age at first marriage is increasing while the age at first birth is remaining constant. Consequently, several countries have recorded an increase in births among unmarried adolescents. In Kenya and Botswana, more than 70 percent of teenagers who give birth either are unmarried or become pregnant before they marry (Population Reference Bureau, 1992; see Chapter 4, for a discussion of marriage trends and the relation of marriage and fertility.)

Figures 2–1 to 2–4 illustrate another essential feature of fertility in Africa: Unlike Western populations, childbearing continues throughout a woman’s reproductive years with no obvious “stopping” behavior. The peak of childbearing occurs between 20 and 29 and falls slowly, indicating little parity-specific limitation. In societies that practice family limitation, fertility rates depart from a natural fertility schedule as women age, because women use efficient methods of contraception to prevent pregnancy once they have achieved their desired family size. There is little evidence of a stopping pattern in any of the fertility schedules for sub-Saharan Africa, despite the reported practice of terminal abstinence in some societies.

It is important to note, however, that there is some debate about the path that fertility decline in Africa is likely to take and, consequently, the effect of a decline in fertility on the shape of the age-specific fertility distribution. Caldwell et al. (1992) argue that the nature of fertility decline in Africa is likely to be very different from that observed in Asia or Latin America. The reasons are related to differences in constraints on premarital and extramarital sexuality, and different emphases on the need and reasons for birth spacing. In Africa, efficient contraception may be used to space children more efficiently rather than as a means to lower completed family

FIGURE 2–1 Age-specific fertility rates: western Africa.

FIGURE 2–4 Age-specific fertility rates: southern Africa.

size. There may also be a growing demand for efficient contraceptives among unmarried women who want to delay the onset of marriage and childbearing. Consequently, Caldwell et al. predict that fertility decline in Africa is likely to involve a simultaneous uptake in contraceptive use at all ages.

The relationship between various indicators of socioeconomic development and family size is an important topic that is of direct relevance to planners and policymakers attempting to integrate population variables into development planning. Place of residence and education are examined here because they are usually two of the most efficient predictors of fertility decline.

Previous studies have consistently observed that women living in urban areas have fewer children than their rural counterparts. The explanation for this difference is often that women in urban areas tend to have more education and are more likely to participate in the formal labor market. Conse-

quently, these women are more likely to appreciate the advantages of having a smaller family. At the same time, urban women are assumed to have better knowledge of, and access to, modern contraception than women in rural areas.

Age-specific fertility rates by place of residence are shown in Table 2–3 for selected countries that participated in the DHS. The data confirm our a priori expectations: Rural fertility is substantially higher than urban fertility in every country even in those countries where national-level fertility estimates do not indicate a recent decline in childbearing (for example, Mali, Togo, and Uganda). The average difference in total fertility is 1.8 births per woman; however, the difference ranges from less than 1.1 births per woman in Cameroon and Liberia to more than 2.5 births per woman in Ethiopia, Tanzania, and Togo. The urban-rural differential is usually highest among those aged 15–19 and 45–49, reflecting differences in age at marriage and limitation of family size, respectively.

Fertility has also been closely associated with female educational levels, although identifying the direction of any causal relationship between fertility and education is complex (Cochrane, 1979). Lower levels of fertility are associated usually with higher levels of education. Typically, the explanation for this association revolves around the fact that more educated women are more likely to delay marriage and to work for paid employment in the formal labor market after leaving school. Consequently, the demand for children may be inversely related to educational level. Literacy skills may improve women’s ability to practice efficient contraception and may empower them with more decision-making authority in the household. However, it could be that the initiation of childbearing is a factor in the termination of education.

Several studies have found that low levels of education as opposed to no education may actually be associated with relatively higher fertility. Small amounts of education may break down birth-spacing practices, including long breastfeeding intervals and postpartum abstinence, without lowering fertility desires or increasing age at marriage. On the other hand, higher levels of education are almost always associated with the lowest fertility.

Age-specific fertility rates by level of education for countries participating in the DHS are presented in Table 2–4. The essential point to take from this table is that fertility does not appear very responsive to small amounts of education. The average TFR for women with no education, 7.0 (as shown at the bottom of the table), is identical to that for women with 1 to 4 years of education. In five countries, Burundi, Kenya, Liberia, Mali, and Nigeria, fertility rises with a small amount of education. (Fertility may

rise with small amounts of education in other countries as well, but the increase may be masked by the way the education categories were formed.) Even with 5–7 years of education, the average TFR remains high at 6.1 births per women. This figure represents a modest reduction in fertility of less than one child per woman, compared to women who have never attended school.

Fertility is considerably more responsive at higher levels of education. Women with eight or more years of education have many fewer children than women with no education. The difference ranges from 1.3 children per woman in Ghana to 3.5 children per woman in Zimbabwe. One explanation is that secondary education has a large positive and significant effect on the average age at first marriage and the average age at first birth (Westoff, 1992).

Sub-Saharan Africa is the only major region of the developing world that has not yet undergone a general decline in fertility. So the question of when fertility is likely to decline is a pressing one. The possibility that fertility may already be declining in several African countries was mentioned in an earlier discussion of TFRs. The evidence presented in Table 2–1 indicates that many sub-Saharan African countries are experiencing minor declines in fertility. However, small declines in fertility could be artificial—the product of using different strategies to estimate fertility at different points in time, or the product of unreliable data. Consequently, more detailed analysis is required to confirm whether small observed declines in fertility are genuine.

This section provides corroborating evidence for fertility declines in the sub-Saharan African countries that are participating in the DHS, when it exists. Two strategies are used to judge the veracity of recent trends in TFRs for DHS countries. The first strategy is based on a simple fact: It is easier to accept a change in fertility as genuine if it is accompanied by a change in one of the proximate determinants of fertility that would account for it (e.g., an increase in age at marriage or in the use of contraceptives). Consequently, changes in fertility are compared with changes in marriage patterns, changes in acceptance and mix of contraceptive methods, and changes in family size preferences. The second strategy is to conduct an independent, in-depth analysis of birth histories by using life table techniques.

Table 2–5 presents direct estimates of the total fertility rate for each of the 11 African countries for which detailed DHS information were available. The surveys were conducted over a 5-year period, 1986–1990. The fertility estimates do not all refer to the same date. They have been calculated by using recall periods of 0–3 years and 4–7 years prior to the date each country was surveyed.

Fertility appears to have fallen in 10 of the 11 countries included in Table 2–5. The average decline in fertility is slightly less than 1 child per woman, with a range of 1.2 children per woman in Zimbabwe to 0.4 in Ghana. In eight countries, the declines in fertility are statistically significant at the 5 percent level. Uganda is the only country that shows no evidence of a decline.

In reality, the eight cases with statistically significant declines can be split into two groups. Group A includes Kenya, Botswana, and Zimbabwe, where declines in fertility appear to be reasonably certain. The decline in fertility rates in all three countries has been accompanied by an increase in the proportion of women using modern contraception. (See Table 2–6 for current levels of modern contraceptive use.) DHS data in Table 2–5 show

that the group B countries, Burundi, Mali, Nigeria, Senegal, and Togo, also experienced statistically significant declines in fertility; however, these declines have not been accompanied by substantial increases in the median age at marriage, the percentage of women using modern methods of contraception, or a substantial decrease in ideal family size, which might explain the decline in fertility. (See Table 2–6 for most recent estimates.)

Fertility decline has been most dramatic in Kenya, Botswana, and Zimbabwe. In Kenya, fertility rates fell by slightly more than 0.5 child per woman between 1987 and 1989 and by 1.2 children per woman between 1979 and 1989. Overall, the Kenyan DHS results not only indicate that the transition to lower fertility is already under way, but also suggest that it

may accelerate in the future (Cross et al., 1991). Fertility decline has been even more dramatic in Zimbabwe where the DHS-based estimate implies that fertility fell by 1.2 children per woman between 1982–1984 and 1985– 1988. The rate of decline in Zimbabwe is even greater if one accepts the results from the Intercensal Demographic Survey, which yield an estimated TFR of 5.1 for 1987, instead of the DHS estimate of 5.5 for 1985–1988. In Botswana, the DHS-based fertility estimates are 5.9 in 1982–1984 and 4.9 children per woman in 1985–1988. Other sources have yielded considerably higher fertility estimates in Botswana in the recent past: 6.5 in the 1984 Family Health Survey (FHS) and 7.1 in the 1981 census. Comparing the results from the various surveys is difficult, however, because different questions and methods of estimation were used. Furthermore, close comparisons of the DHS and the FHS revealed that the two samples differed appreciably in composition (Thomas and Muvandi, 1992) and that the number of children ever born was inconsistent between surveys, at least for older women (United Nations, 1992). Finally, the estimate of the TFR based on the 1981 census was probably too high, tending to exaggerate the extent of the decline in fertility over the last 10 years (Rutenberg and Diamond, 1993). Nevertheless, the difference in fertility rates between the various sources is too great to be attributable solely to problems of sampling or estimation.

Countries in group A are distinguished from countries in group B by the fact that recorded decreases in fertility have been accompanied by changes in other fertility-related variables. As discussed above, collecting accurate demographic data in sub-Saharan Africa is difficult. Hence, for countries that have a low TFR, it is reassuring to find a low average desire for children, a high use of efficient contraception, and a high median age at first marriage. Table 2–6 summarizes the data on these key indicators. Two crude measures of fertility preference are presented: the mean ideal family size and the proportion of all women who want no more children. Of the women in the 11 populations, only women in Kenya, Botswana, and Zimbabwe considered the mean ideal family size to be fewer than 5 children.⁵ Among all women in Kenya, the mean ideal number of children fell dramatically from 5.8 in 1984 to 4.4 in 1989 (Kenya, 1989). Also of note is the age pattern of preference in 1989, with women aged 40–44 preferring 5.5 children, women aged 25–29 preferring 4.4, and women aged 15–19

preferring 3.7. Furthermore, compared to women in other countries, women in these three countries were more likely to indicate at the time of the DHS that they did not want additional children. In Kenya, nearly half the married women in the DHS sample reported that they wanted no more children. In Zimbabwe and Botswana, the proportion was closer to one-third.

Another important indicator of women’s fertility intentions is use of contraception. In group A countries, more than 14 percent of married women used some type of modern contraceptive at the time of the DHS. In Botswana, the increase in the number of women using modern contraceptives since 1984 is large enough to account for the observed decline in fertility (van de Walle and Foster, 1990). The level of contraceptive use among married women is even higher in Zimbabwe: Almost all women who were interviewed knew of at least one method of contraception; 93 percent of married women had used contraception at one time; and 36 percent of married women were currently using a modern method. Contraceptive prevalence in Kenya has also increased sharply: 27 percent of married women reported using some method of contraception in 1989, a fourfold increase in use from 1977–1978 levels (Kenya, 1989). Part of the increase in the demand for contraceptives may be the result of women wanting to space children more efficiently rather than wanting to limit family size (van de Walle and Foster, 1990; Caldwell et al., 1992).

Additional information on the nature of the fertility decline in Kenya, Botswana, and Zimbabwe can be obtained from an analysis of birth histories. These histories allow us to describe the family formation process in greater detail. For example, rather than just observing that a woman has had a birth in the last 12 months, birth histories can tell us birth order and the timing between pregnancies. Consequently birth histories enable the researcher to identify changes in fertility behavior that are specific to parity. Because the vast majority of women in Africa marry and have children, changes in fertility behavior between cohorts are most likely to occur after women have had several children. Consequently, a parity-specific investigation of fertility behavior can be a useful way to detect small changes in fertility for a country at the beginning of a transition toward lower fertility (Brass and Juarez, 1983).

Life table techniques provide the best way to analyze birth history data. This approach views the family formation process as a series of steps in which women pass successively through marriage to first births to second births and so on, until they achieve their completed family size. When data are available for a cohort of women who have completed their fertility, parity progression ratios (PPRs) can be calculated directly. For each age cohort, these ratios measure the proportion of women who have had n children and proceed to have an additional child. Because the majority of women included in cross-sectional data sets have not completed their fertil-

ity, it is inappropriate to calculate PPRs. Another measure, analogous to the PPR, is suggested by Rodriguez and Hobcraft (1980). It measures the proportion of women in an age cohort who, having attained an nth birth, go on to an (n+1)th birth within 60 months. This measure is usually called a censored parity progression ratio (CPPR). One advantage of CPPRs over TFRs is that they are robust with respect to the problem of misreported dates of births. As long as the index birth is reported as having occurred within 60 months of the preceding one, the CPPR is unaffected.

Unfortunately, Rodriguez and Hobcraft’s measure is biased because it systematically excludes women with long birth intervals (Brass and Juarez, 1983). A simple adjustment to correct this limitation has been proposed by Brass and Juarez (1983). This variant of the CPPR is reported below and denoted B₆₀. Because no formula exists to calculate the standard errors associated with the B₆₀s, inferences about changes in fertility trends may be drawn only by examining general patterns in the data.

Table 2–7 presents the B₆₀s for Botswana, Kenya, and Zimbabwe. To simplify the table, neighboring parity progression ratios for the same cohort have been combined by multiplying consecutive indices together. The summary measures represent the combined probability of having two additional births, that is, the probability of moving from the nth to the (n+1)th parity within 60 months multiplied by the probability of moving from the (n+1)th to the (n+2)th parity within 60 months of the previous birth.

The B₆₀s reveal the pattern of fertility decline that has occurred in these countries. The most important finding is that there appears to have been a general reduction in fertility across cohorts of women at all parities in the probability of having additional births. In Kenya, the largest decline across cohorts appears to have been among women with seven children. Fewer women who have given birth to seven children now continue to have an eighth and a ninth child. In Botswana, the B₆₀s imply that important reductions in fertility have also occurred at low-order parities (first to third and third to fifth births). In Zimbabwe, the greatest reductions have occurred in the middle-order parities (third to fifth and fifth to seventh parities).

Evidence in favor of a fertility decline is not as strong in the other countries exhibiting a decline in Table 2–5, namely, Burundi, Mali, Nigeria, Senegal, and Togo. As shown in Table 2–6, in all five countries the demand for children, as measured by mean ideal family size, remains high; and few married women are currently using modern methods of contraception.

Of the countries in group B, Senegal is the country likeliest to be on the verge of undergoing a decline in fertility. The total fertility rate in Senegal was 7.1 births per woman in 1978, according to WFS data. According to

the DHS, fertility fell to 6.6 by 1986, and provisional analysis of the recent census suggests it fell further to 6.3 in 1988 (Senegal, 1992). It is important to note, however, that the census and the DHS estimates are based on different kinds of data and different methodologies. The DHS collected detailed birth histories from all women. The census only asked women about the number of births they had in the last 12 months. Unfortunately, the latter approach is particularly susceptible to reporting bias because women tend to omit systematically children that have died or moved away. It has been established that the number of births in the 12 months preceding the census was underestimated (Senegal, 1992). Consequently, part of the difference between the fertility estimates from the DHS and those from the census may be attributable to flaws in the census data. Nevertheless, both the DHS and the census estimates are at least half a child per woman lower than the WFS estimates.

Some support for a decline in fertility in Senegal comes from apparent changes in contraceptive use and marriage patterns. The percentage of women currently using any method of contraception rose from 3.9 in 1978 to 11.4 in 1986. At the same time, there appears to have been an increase in the average age at first marriage among Senegalese women. DHS data indicate that the median age at marriage for women aged 40–44 at the time of the survey was 16.1 years in comparison to 17.2 years for women 20–24, and provisional results from the 1988 census indicate that the average age at first marriage is continuing to rise (Senegal, 1992). Westoff (1992) argues that this change in marriage patterns is sufficient to account for the slight drop in fertility. However, if the decrease in fertility is the result of an increase in the age at first marriage and first birth, then the greatest declines in fertility should have occurred among women in the youngest age groups. A simple comparison of age-specific fertility rates between data from the WFS and the DHS would suggest that changes in fertility have occurred in the youngest ages. The age-specific fertility rates for the two youngest age groups are between 8 and 16 percent lower in the DHS than they were in the WFS, but are virtually identical for the older four age groups. However, this pattern is not confirmed by a detailed internal analysis of the DHS data. These data suggest that fertility levels fell for all cohorts. A comparison of ASFRs for 0–3 and 4–7 years prior to the survey shows that the largest reductions in fertility occurred among women in the older age groups (Arnold and Blanc, 1990). Moreover, inspection of the B₆₀s for Senegal in Table 2– 8 indicates that fertility has fallen most consistently for parities three and higher. These discrepancies prevent Senegal from being placed in group A among countries in which declining fertility is confirmed. Even so, it is probable that at least some part of the observed fertility decline in Senegal is genuine, particularly the portion that results from an increasing age at first marriage.

In Nigeria, the DHS data indicate a decline in fertility of 1.3 children per woman, the largest absolute decline reported for any African country for which DHS data are available. Although some proportion of the decline is certainly genuine, other evidence does not support a decline of this magnitude. First, the use of contraception is extremely limited among Nigerian women. As shown in Table 2–6, only 6.0 percent of married women used any method of contraception in 1990, and only 3.5 percent used modern methods. Furthermore, the proportions of Nigerian women that are married by age 17 and that become mothers by age 20 appear to be stable across cohorts (Nigeria, 1992). On the other hand, the pattern in B₆₀s in the lower parities is consistent with a fertility decline. However, there is evidence to suggest that the number of births in the 5 years preceding the survey was underestimated (Nigeria, 1992).

More detailed analysis of data from Nigeria shows that the national

averages mask very important regional differences in fertility patterns. Table 2–9 highlights the extent of these differences. In southwest and southeast Nigeria, the TFR is less than it is in the northeast and northwest regions, by one child per woman. Furthermore, there is enormous variation in the average age at first marriage and the level of contraceptive use among the regions. In the northeast and the northwest regions, almost half of all women are married by age 15. Marriage occurs later in southeast and southwest Nigeria, where the median age at first marriage is closer to 18 and 20, respectively. Current use of family planning is also much higher in southwest Nigeria, where 15 percent of currently married women were using some method of contraception at the time of the survey, compared with 9 percent in southeast Nigeria and 2 percent or less elsewhere. The contrast in rates of modern contraceptive use between regions is even greater. Therefore, upon closer examination of fertility in Nigeria, it appears that a fertility transition is limited to the southwest (and possibly the southeast) region(s) of the country and is not a country-wide phenomenon. Nonetheless, this transition is significant because this part of Nigeria has a larger total population than most African countries.

The small declines in fertility recorded in Mali, Burundi, and Togo probably should not be interpreted as heralding the onset of a major fertility transition in those countries. A comparison of the total fertility rates in the two reference periods shows fertility falling by less than 1.0 birth per woman in each of these three countries. At the same time, there were considerable problems in the DHS for all three countries with the displacement of births (Arnold, 1990). The most likely explanation for the observed decline in fertility is that the data are flawed in some way.

Mali has an extremely low mean age at first marriage, and more than half of all women report that they were married by age 16, as shown in Table 2–6. The desire for children is greater in Mali than in any of the other countries in sub-Saharan Africa covered by the DHS, and relatively few women want no more children. Only 1.2 percent of currently married women are using efficient contraception. Moreover, there is little evidence of parity-specific reductions in fertility in the B₆₀s. These conditions are inconsistent with the reported decline in fertility. It is not surprising, therefore, to find evidence of a serious displacement of births in Mali. The problem appears particularly acute for children that died and for children whose exact month of birth was not known (Arnold and Blanc, 1990).

Fertility does appear to have fallen in Bamako, the capital of Mali. More than 16 percent of currently married women in Bamako are using some means of contraception, and 6 percent are using a modern method. Elsewhere in Mali, only 4 percent of married women are using any method of contraception, with less than 1 percent using a modern method. It is probably significant that in Bamako, the total fertility rate—an indication of

fertility at the time of the survey—was considerably lower than the number of children ever born to women aged 40 to 49—an indicator of cumulative fertility in the past (Traoré et al., 1989). However, because of migration, the lifetime fertility of women aged 40–49 does not necessarily reflect cumulative fertility in Bamako.

In Burundi, age at first birth is relatively stable across age cohorts, though quite high by sub-Saharan African standards (Segamba et al., 1988). The DHS recorded very low use of modern contraceptives with only 1 percent of currently married women using a modern method of contraception. The B₆₀s for Burundi provide a final check for any signs of a reduction in fertility that is parity specific, but provide no clear indication of parity-specific reductions in fertility. Hence, the recorded decline in fertility is suspect.

In Togo, the decline in fertility is likely to have been exaggerated by poor data, although a number of indications are consistent with fertility decline. Specifically, contraceptive use, primarily abstinence, is very high; contraceptive awareness is virtually universal (93.5 percent); family planning receives widespread approval (68.9 percent); and recently, there has been a 10 percent decline in the number of women marrying by age 20. However, there is little evidence of a parity-specific reduction in fertility in the B₆₀s for Togo.

Finally, Group B should also include the most recently surveyed DHS countries for which data are not yet completely available. These include Cameroon and Zambia, for which first country reports have been published, and Malawi, Namibia, Niger, and Tanzania, for which only preliminary reports have been released. Because standard recode files are currently unavailable, the interpretation of recent fertility trends for these countries are based solely on the relevant DHS publications. None of these countries is included in Table 2–5 because the relevant DHS publications do not report age-specific fertility rates for 0–3 years and 4–7 years prior to the survey, the intervals indicated for the other populations. Neither do any of the reports discuss the reliability of birth history data. For example, none discusses the extent of misreporting of childrens’ dates of births that in earlier surveys resulted in the appearance of declines in fertility.

The Cameroon report indicates a modest decline in fertility, from 6.4 births per woman in 1978 to 5.8 births per woman in 1991. This latter estimate is very close to the 1987 census estimate of 5.7 births per woman derived using P/F ratios. Nationally, the use of modern contraceptives is very limited (around 4.3 percent among currently married women) and not supportive of a sustained decline in fertility. However, in Douala and Yaoundé, the two largest cities in Cameroon, use of modern contraceptives is much higher (12.1 percent of currently married woman), and the total fertility rate for these two cities is 4.4 births per woman.

In Zambia, the DHS reports that the TFR of 6.5 children per woman is down approximately 10 percent from the (heavily adjusted) figure obtained from the 1980 census. Contraceptive prevalence stands at approximately 9 percent nationally, 15 percent in urban areas, and 3 percent in rural areas. At least part of the observed decline in fertility may be the result of a trend towards later age at first marriage among younger women (Zambia, 1993:60), but comparisons of age-specific fertility rates between the DHS and the census figures show that most of the decline in fertility appears to have occurred for women aged 25–39 and not in the age group 15–19 as might be expected. Hence, the conclusion that fertility is falling in Zambia is tentative and further assessment of these newest data is warranted when they become available.

Recent data from the Malawi DHS proved the most difficult to interpret. Until 1984, estimates of the TFR in Malawi were thought to be fairly reliable. Fertility rates appear to have changed little in the late 1970s and early 1980s in Malawi, and three independent sources—the 1977 census, the 1982 Demographic Survey, and the 1984 Family Formation Survey— were all in close agreement that the TFR stood between 7.6 and 7.7 births per woman (Malawi, 1980, 1987a,b).

Data from the 1987 census, however, showed an unadjusted TFR of 5.7 births per woman (National Statistical Office, Zomba, personal communication, 1992). This is almost certainly a gross underestimate, perhaps because the information was not necessarily supplied by the woman who bore the children. By applying a P/F adjustment, the TFR was revised to 8.0 births per woman (Table 2–1), a considerable difference from the unadjusted figure but not out of line with some expectations (see for example, House and Zimalirana, 1992:144). Consequently, the DHS preliminary estimate of the TFR, 6.7 children per woman, is a surprise, being substantially lower than the adjusted census figure and even below the mean number of children ever born to women aged 40–49 in both the 1987 census and the 1992 DHS. (When fertility is not changing, the number of children ever born is generally lower than the unadjusted total fertility rate because older women tend to systematically omit certain births.) However, 1990 and 1992 were drought years, which may have affected behavior over the short term. And the recent massive AIDS-prevention campaign may have achieved lower fertility through increased use of condoms, especially outside of sanctioned unions.

If all these estimates are correct, then the conclusion would be that fertility has fallen only in the last 3 or 4 years. Two qualifications apply. First, as stated above, adjusting the period fertility rates from the 1987 census on the basis of a comparison of P/F ratios is only valid if fertility has been approximately constant over the recent past. If fertility declined before 1987, then cumulated period rates cannot be expected to equal lifetime