Fertility Desires and the Course of Fertility Decline in sub‐Saharan Africa

Abstract

This research examines the demand for children in sub-Saharan Africa and how this demand bears on prospects for fertility decline. Fertility levels on average are high in the region. At issue is whether, and how, fertility decline will occur in the near or medium term. In sub-Saharan Africa fertility desires too are generally high, as we have documented elsewhere (Bongaarts and Casterline 2013; Casterline and Agyei-Mensah 2014). The backdrop for this research, then, is high fertility evidently sustained by high demand for children. We confine our analysis to the quantity of children, setting aside desires to space births. Caldwell et al. (Caldwell, Orubuloye, and Caldwell 1992; Caldwell and Caldwell 2002) stress the prominence of birth spacing in sub-Saharan African reproductive regimes, inferring from this that increased birth spacing will play a larger role in African fertility declines than has been the case in other regions. More recently, on the basis of analysis of Demographic and Health Survey (DHS) data Moultrie, Sayi, and Timæus (2012) conclude that “birth postponement” has made a large contribution to African fertility declines to date. Consistent with this argument, Casterline and Odden (2016) document that inter-birth intervals have lengthened during the early stages of fertility decline in Africa. By no means do we seek to dismiss the importance of birth spacing in African reproductive regimes. Rather, to set some reasonable bounds on the present research, our attention is limited to the desired quantity of children, as revealed by survey data on the ideal number of children and the parity-specific preference to have no more children. While it may be the case that intentional birth spacing reduces fertility to a greater extent in Africa societies than in other regions, it is highly unlikely that fertility in African societies will fall below four births per woman, not to mention falling to replacement level, unless most women have just a few children. And this in turn seems very unlikely in the absence of widespread deliberate effort to limit the number of children. That is, the postponement behavior posited by Timæus and Moultrie (2008) by itself seems an inadequate mechanism for reducing fertility to less than four births per woman, given that sexual activity begins at relatively young ages on average. Moultrie, Sayi, and Timæus (2012), while arguing that postponement can exercise a powerful fertility-reducing impact, concede that attainment of low fertility requires adoption of deliberate and effective family-limitation practices. For this reason it is crucial to assess the existing demand for children, and from this assessment to draw some inferences about the prospects for widespread desire to limit family size, which presumably could result sooner or later in lower fertility rates. The social science and demographic literature on family-limitation desires and behaviors in sub-Saharan Africa is not as rich or well developed as the corresponding literature on birth spacing. Instead, over the years many scholars, drawing on a wide range of quantitative and qualitative materials, have concluded that African women's planning horizons for fertility are short and that explicit quantity-of-children targets are for the most part not formulated. This assertion is found not only in scholarship on African reproduction in the past (van de Walle 1992) but also in recent scholarship based on fieldwork in multiple sub-regions as well as analysis of DHS (Agadjanian 2005; Bledsoe 2002; Johnson-Hanks 2005, 2007). Our analysis does not directly challenge this fundamental assertion, because the survey data we analyze do not provide sufficient basis for a rigorous challenge. We do, however, present evidence which is inconsistent with this assertion, evidence which suggests the emergence of a family-limitation mentality in African societies. Our research takes as a given that fertility desires affect fertility; were this relationship weak or non-existent, the rationale for this research would be weakened. More specifically, we assume that declines in desired fertility sooner or later translate into a decline in fertility. We do not interrogate this crucial assumption, but there is abundant empirical research validating it—investigations of the association between fertility desires and fertility outcomes at both the aggregate level (e.g. Lee 1980; Westoff 1990; Morgan and Rackin 2010) and the individual level (e.g. Westoff and Ryder 1977; Schoen et al. 1999). An important conclusion from this large literature is that the association is variable; that is, the correspondence between fertility desires and reproductive behavior is far closer in some settings than in others. No doubt in part this reflects variation across settings in the meaning of responses to the attitudinal items that have been standard in demographic surveys during the past four decades (e.g. Demographic and Health Surveys). More fundamentally, fluctuating correspondence between fertility desires and fertility outcomes is due to variation in the extent to which desires are implemented in those behaviors that directly determine fertility: sexual activity, contraception, and induced abortion (Bongaarts 1978). That fertility desires are imperfectly implemented is explicitly recognized in the notion of “unmet need for contraception”—that is, lack of contraceptive practice among women who state a desire to avoid pregnancy (in the short term or indefinitely) (Casterline and Sinding 2000; Bradley and Casterline 2014). Implementation of preferences is a central concept in the second portion of the analysis presented here, although we do not analyze contraception per se (instead examining rates of unwanted fertility). All this argues for caution in drawing simple inferences about likely future trajectories of fertility in sub-Saharan Africa on the basis of fertility desires alone. The complications just noted, and others (including the dangers of inferring fertility desires from reproductive behavior), are cogently spelled out in Johnson-Hanks (2007). In short, this research has two main goals: (i) to examine fertility desires (number of children) in sub-Saharan Africa: levels as compared to other major regions, and recent trends; and (ii) to assess the extent to which fertility decline in sub-Saharan Africa is contingent on decline in fertility desires (singly, and in combination with other reproductive changes). The empirical analysis relies on information collected in national demographic surveys from 1975 to the present. These surveys were part of four international survey programs: World Fertility Surveys (WFS), Demographic and Health Surveys (DHS), Reproductive Health Surveys (RHS), and Pan-Arab Project for Family Health (PAPFAM).1 Recent data for sub-Saharan Africa are drawn entirely from DHS surveys, while African WFS surveys provide historical depth for some of the analysis. The non-DHS surveys greatly enlarge the coverage of other regions (especially in the pre-decline and early-decline stages), affording sounder comparison with sub-Saharan Africa. In the comparative analysis of historical trends, the sample is 84 countries and 303 surveys, of which 38 countries and 134 surveys are in sub-Saharan Africa.2 In the analysis of prospects for future fertility decline, the sample is limited to 32 sub-Saharan African countries with a survey conducted since 2000. Throughout, the analysis is attentive to differences between sub-regions and countries. In particular, we break Africa into three sub-regions: East Africa (12 countries), Southern Africa (5 countries), and Middle and West Africa (21 countries). This regional breakdown proves a meaningful distinction, as could be anticipated from Cleland, Ndugwa, and Zulu (2011). The remainder of the globe is classified into Latin America and Caribbean (19 countries), South and Southeast Asia (15 countries), and West Asia and North Africa (12 countries). This yields six regional/sub-regional groupings in total. For convenience we refer to these as regions, although strictly speaking the three sub-Saharan African categories are sub-regions. All calculations required for this analysis have been performed by the authors using the individual-level survey data files for the 303 surveys. These demographic surveys offer three measures of fertility desires: ideal number of children; preference for another child: whether another child is desired and, if so, the desired waiting time; and wantedness of recent births, as reported retrospectively. These measures permit construction of multiple indicators of fertility desires. We rely on two of the three items, namely the ideal number of children and the preference for another child. We depend more heavily on the latter because numerous methodological studies have demonstrated that the preference for another child is the most valid and reliable of the three measures listed above (Casterline and El-Zeini 2007). While the prospective preference item is not a direct question about the desired quantity of children, when the responses are tabulated on a parity-specific basis they are revealing of the desired number. For example, if a woman of parity three states that she does not want any further children, one may infer that three (or less) is her desired number. A qualification of this point is that parity and age are confounded: women of higher parity are older ceteris paribus, and to the extent that age independently affects the desire to terminate childbearing, parity-specific desires to stop may reflect considerations of age as well as desired number of children. The prospective preference item, in combination with the birth histories collected in all DHS surveys, provides a basis for calculating wanted and unwanted period fertility rates (TFR). The second portion of the analysis makes use of such rates. The ideal number of children has proven a less reliable, and arguably less valid, indicator of fertility desires than prospective preferences (Casterline and El-Zeini 2007). Test–retest reliability is low. Validity is compromised by an aversion to offering an ideal that is less than the already attained number of children (so-called rationalization) and by non-numeric responses (“up to God”). But the stated ideal number of children has the virtue of being a concise expression of the overall demand for children. We use it here to obtain a first look at fertility desires. To minimize the validity threat presented by rationalization, we limit the sample for analysis of the ideal number of children to women whose first union commenced within ten years of the survey interview. These are mainly younger women with a small or modest number of children ever born, and hence there is less risk of rationalization.3 Women supplying non-numeric responses are excluded. While non-numeric responses have become relatively rare in recent demographic surveys in sub-Saharan Africa and elsewhere (Frye and Bachan forthcoming), in a minority of the surveys in our research such a response is relatively prevalent: in about 20 percent of the surveys the proportion of women providing a non-numeric response exceeds 10 percent (among women within ten years of start of first union). As noted, this is one reason for favoring prospective preferences as an indicator of fertility desires. There is some variation among the survey programs in the questionnaire item that asks for the respondent's ideal number of children: WFS has the deviant wording. The additional clause in the DHS has been demonstrated to reduce slightly the reported ideal number of children, presumably by reducing the extent of rationalization. Hence the WFS item yields a higher ideal ceteris paribus, producing a slight exaggeration of the decline in the ideal in those countries where the first survey is WFS. As is the case with the ideal number of children, the fertility preference items in these survey programs are similar but not identical: Again the outlier is the WFS, which offers only the affirmative side and therefore is subject to acquiescence bias. To the extent such bias occurs, it will cause upward bias in WFS estimates of the desire for another child, leading to an exaggeration of the decline in the proportion of women wanting another child (because in countries with a WFS it is historically the first survey). Note that trends in the ideal number of children and prospective preferences are subject to bias in the same direction when the WFS is the earliest survey. Because the prospective preference item was administered only to currently-in-union women in some surveys (as against both never-in-union and currently-in-union women in other surveys), we confine the analysis of prospective preferences to currently-in-union women, to ensure that estimates of variation across countries and over time are not affected by differences in sample selection on union status. For the descriptive analysis of levels and trends—the first step in this research—estimates are generated survey-by-survey. These estimates are then pooled within region, and the remainder of the analysis relies on these survey-level estimates. With these survey-level estimates, trends are calculated country-by-country, and for summary purposes a regression line is estimated region-by-region with fixed effects for country. This fitted regression line can be viewed as a convenient average of the within-country trends. The regression specification has linear splines—connected straight lines with disjuncture at “knots” —to accommodate trends that are non-linear (i.e. more rapid or slower pace of change at different stages of fertility decline). The second step in this research is to assess the role of fertility demand as an obstacle to fertility decline in sub-Saharan Africa and, conversely, to consider how declines in desired fertility might facilitate decline. We conduct this assessment through a simulation exercise that employs a reproductive model containing fertility desires as an explicit element, thereby allowing for an examination of the responsiveness of fertility to hypothetical changes in desires. We adopt this approach to generate a balanced and numerical assessment of the potential effect on fertility rates in Africa of changes in fertility desires. The reproductive model is as follows: If the product rnpn is regarded as the nuptiality component—the proportion never married and the fertility rate among the never married4—equation 1 can be viewed as an expression for fertility with four essential elements: This equation establishes terms for the debate about the relative weight in determining fertility decline of fertility demand vs. implementation of this demand that differ from those that have been common in the scholarly literature to date, most notably as articulated in the highly influential article by Pritchett (1994). We develop this point in the final section. A distinctive feature of equation 1 is the two elements rw and ru. We term these conditional rates (wanted and unwanted, respectively), because the denominators for these two rates are restricted to woman-years during the reference period wanting or not wanting another child, respectively (among ever-in-union women). That is, these are rates conditional on fertility preferences. By contrast, conventional unconditional wanted and unwanted rates (e.g. as routinely calculated from DHS surveys) include all women in the denominator. The denominators for the conditional rates are obtained by assuming that parity-specific desires for another child at the time of the survey apply to the entire reference period. Because the conditional rates are restricted to ever-in-union women at risk of wanted or unwanted births, they are more appropriate than unconditional rates for examining the implementation of fertility preferences. They also provide more direct counterparts to demand satisfied, an increasingly popular refinement of unmet need for contraception (Bradley et al. 2012). Demand satisfied is contraceptive prevalence among women who want to postpone or stop childbearing; that is, its denominator is restricted to women who want to avoid pregnancy. Unmet need, by contrast, includes all women in the denominator (although typically only currently-in-union women) and therefore is an unconditional measure. nuptiality: more adult years spent never-in-union preference composition: more ever-in-union years spent not wanting another child (vs. wanting a child) wanted rates: decline in rate of childbearing among those wanting another child (conditional wanted rate) unwanted rates: decline in rate of childbearing among those not wanting another child (conditional unwanted rate) The second source reflects changes in the desired number of children. The third and fourth sources are matters of more effective birth control, which can be achieved via either contraception or induced abortion. (A decline in sexual activity would also depress these rates.) That is, the third and fourth sources connect this research to the central concepts “unmet need for contraception” and “satisfaction of demand.” On the basis of equation 1, one can perform simulations that generate hypothetical fertility change (change in TFR) when one or more of the elements of the equation are assigned different empirical values. Numerous simulations might be informative; the crucial decision is the choice of alternative values. We ask: based on this equation, how would present levels of fertility in sub-Saharan Africa change if there were an instantaneous shift to values of each element as observed in contemporary low-fertility populations in other regions? For the latter, we use data from seven national surveys that yield a TFR below 2.3 births per woman.5 Substitution of this set of alternative values element-by-element6 yields hypothetical age-specific fertility rates that are then summed to hypothetical TFRs.7 For this simulation exercise, the starting point is the most recent survey since 2000 in 32 sub-Saharan African countries in East, Middle, and West Africa, along with the average values from the seven low-fertility surveys (see endnote 5).8 We use a 36-month reference period for the fertility rates (months 1–36 preceding the survey). Wanted vs. unwanted fertility is calculated via the “aggregate prospective” method of Casterline and El-Zeini (2007). This method relies on the prospective fertility preference item, which as reported above is generally recognized as the most valid and reliable of the core fertility desires items used by the major survey programs. Typically the aggregate prospective method produces higher estimates of unwanted fertility than other available methods.9 We present the results in two sub-sections, corresponding to the two main steps in this research. This discussion lays out the empirical facts; we postpone most of our interpretation to the final section. To obtain an initial picture of levels of fertility desires and their association with levels of fertility, with an emphasis on cross-regional comparison, we consider the mean ideal number of children reported by women who are within ten years of the start of their first union. The level of fertility is represented by the period total fertility rate (TFR) for months 1–36 preceding the survey. The full data for this analysis are shown in Figures 1a and 1b: trends for 84 countries, as derived from 303 surveys, plotted separately for each of the six regions. The country-by-country trends are shown with dashed lines, and the fitted line for the region is a bold solid line. Note that the x-axis is ideal number of children reversed, so that left-to-right corresponds to fertility decline. As expected, realized fertility (TFR) declines as ideal fertility declines; there is no evidence of departure from this regularity. But the rate of decline varies across regions, and, of possibly more significance, conditional on ideal number of children the level of fertility varies considerably across regions. Equivalently, conditional on level of fertility, the ideal number varies considerably across regions. These conclusions are more transparent in Figure 2 and Table 1a. Figure 2 brings together in one graph the six regional fitted lines. What is most striking in this figure is the higher mean ideal number of children in the three sub-Saharan sub-regions as compared to the three other major regions. The mean ideal is especially high in Middle and West Africa. For more precise illustration, Table 1a presents fitted TFRs at a selection of mean ideal number of children (5.5, 3.5, 2.5). Where the data permit cross-regional comparisons—that is, when the mean ideal ranges between 3.0 and 5.0—the TFR exceeds the mean ideal by a smaller amount in sub-Saharan Africa than in Latin America and the Caribbean and in South and Southeast Asia. That is, there is closer correspondence between ideal and realized fertility in Africa; from a comparative perspective, the closer correspondence between ideal and realized fertility is a distinctive feature of the reproductive regimes in sub-Saharan Africa.10 Or, put otherwise, at the aggregate level ideal and realized fertility are more in equilibrium in Africa than in other regions. This finding suggests less excess fertility in sub-Saharan Africa, a conclusion verified in more direct analyses of unwanted fertility rates (Sedgh, Singh, and Hussain 2014). Of course, avoidance of unwanted births is easier to achieve, with less recourse to deliberate birth control, when the ideal number of children is high. A further finding—evident from Figure 2 and documented in Table 1b—is the closer correspondence between the rates of decline in ideal and realized fertility in the three sub-Saharan African regions as compared to the other three major regions. In these latter three regions, the slopes are well in excess of 1.0, indicating more than a one-birth decline in TFR for each one-child decline in the ideal number of children. That is, realized fertility falls more rapidly than ideal fertility. This is not surprising in view of the higher excess fertility in the non-African regions at the onset of their fertility decline; in these regions, as compared to Africa, there was more opportunity for fertility decline to consist of elimination of excess fertility. But this pattern of slopes in Table 1b does not follow by necessity: had the ideal number of children declined rapidly in the non-African regions, substantial fertility declines could have occurred without reduction in the amount of excess fertility—that is, slopes near 1.0. By the same reasoning, the slopes less than 1.0 in East Africa and Middle and West Africa when the mean ideal is greater than 5.0 indicate that excess fertility increases in the early stages of fertility decline in these African regions. We suggested earlier, drawing on past methodological research, that the ideal number of children is an inferior measure as compared to the prospective preference for another child, and furthermore the latter affords a more detailed picture of the evolution of fertility desires as the reproductive career unfolds. In Figures 3a and 3b, we examine historical trends in the parity-specific desire to stop childbearing in 18 African countries (nine in East Africa and nine in Middle and West Africa). The 18 countries are purposively selected on the basis of population size and the availability of a time-series of surveys. The mean percentage of women expressing a preference not to have another child is shown for women of parity 2, 4, and 6. These figures allow for consideration of both levels and historical trends. This is the only portion of our analysis that explicitly indexes by historical time. First, consider levels. Vertical distance in Figures 3a and 3b reflects an increase with parity in the desire to stop. An increase with parity is consistent with a “family-limitation” mentality—that is, a desire to terminate childbearing once a certain number of births have occurred (Knodel 1977). No increase with parity, in contrast, would indicate an absence of the family-limitation mentality, as is suggested in research by Johnson-Hanks (2007) and Moultrie, Sayi, and Timæus (2012). In the event, the survey data show that women are prepared to indicate a desire to end their childbearing and, further, that this desire increases markedly with parity, especially in the period since 2000. In East Africa (Figure 3a), the difference between women of parity 2 and women of parity 6 is on the order of 30–40 percentage points in the recent period. The differential is of smaller magnitude in Middle and West Africa (Figure 3b) but unmistakable nevertheless, on the order of 20–30 percentage points in most countries. This is the case even in the Sahelian countries, with the clear exception of Niger. In short, throughout Africa the expressed desire to stop childbearing increases with parity, a pattern revealing of a consciousness about the reproductive career that is consistent with family-limitation behavior.11 From these data alone, we cannot say whether women have long time horizons nor whether they adhere to a relatively fixed target number of children. But on their face, these estimates refute the notion that parity-dependent birth control is an alien concept in recent decades in the diverse societies of sub-Saharan Africa. Second, consider trends. The picture in Figures 3a and 3b is mixed. In most countries there is little increase at parity 2; this indicates that small-family desires have not yet taken hold. At higher parities, especially parity 6, there is an upward trend in most countries. In general an upward trend at parity 4 and 6 is less evident in the countries of Middle and West Africa; indeed the majority of countries there show little trend at parity 4, let alone parity 6. Lack of trend especially characterizes the Sahelian countries. Evidently family limitation, in a subjective sense at least, is establishing itself more rapidly in East and Southern Africa than in Middle and West Africa. The assessment of trends in Figures 3a and 3b is complemented by investigation of the association between the desire to stop childbearing and the total fertility rate in Figures 4a, 4b, and 5 and Tables 2a and 2b. These figures and tables follow the same methodology as the analysis of the ideal number of children in Figures 1a, 1b, and 2 and Tables 1a and 1b. The figures and tables are confined to parity 3, which captures the important distinction between women prepared to stop at a rather small number of children and those who wish to proceed to a moderate or large number. It is clear from Figures 4a and 4b that the TFR declines as the desire to stop at parity 3 increases. The country-by-country trends vary around the fitted regional slope, and some noise is apparent (note East Africa and Latin America and Caribbean), but on the whole the pattern of association is consistent. The fitted regional slopes are extracted and presented in Figure 5, and visual assessment suggests that the slopes are steeper in the three African regions. A steeper slope means a larger decline in TFR for each percentage point increase in the desire to stop. The slopes of the fitted lines presented in Table 2b only weakly support this visual assessment, however. A more valid generalization is that the slopes can be regarded as roughly equivalent among regions—Figure 5 and Table 2b both support this conclusion. The noteworthy regional differences evident in Figure 5 and Table 2a concern levels, not trends: conditional on fertility preferences (i.e. percentage desiring to stop childbearing at parity 3), the TFR tends to be lower in sub-Saharan Africa than in the other regions. Or, equivalently, at high levels of fertility (e.g. TFR>5.0) the parity-specific desire to stop tends to be lower in sub-Saharan Africa and especially in Middle and West Africa. Figure 5 contains exceptions to this generalization, in particular the somewhat low TFR in South and Southeast Asia when the desire to stop is low (i.e. onset of fertility decline). Still, the overall pattern is for the TFR to be lower in Africa when one conditions on the parity-specific desire to stop childbearing. We suspect this regional differential mainly reflects behaviors, especially postpartum practices, that produce long average inter-birth intervals in many African societies (Casterline and Odden 2016)—that is, lower rates of childbearing among women who want more children in sub-Saharan Africa as compared to other regions. A close look at Figure 5 also yields differences between the African regions that deserve mention, specifically lower fertility conditional on preferences in Middle and West Africa as compared to East Africa (or, equivalently, lower desire to stop in Middle and West Africa than in East Africa at any given level of fertility). This intra-Africa differential is consistent with all the results discussed to this point, including results from our analysis of the ideal number of children: a stated desire to limit family size is far less prevalent in Middle and West Africa. Finally, Figure 5 reinforces the conclusion derived from Figures 3a and 3b that the desire to stop at a moderate number of children has increased rather rapidly during the past 10–20 years in some African countries. Judging from the fitted lines in Figure 5, if recent trends were to persist, within a decade or so the majority of women in East Africa will resemble their counterparts in Latin America, Asia, and North Africa in preferring to limit their childbearing to a small number of children (less than four). This is already the case in Southern Africa. The same cannot be said of Middle and West Africa, which has far more distance to go to close the gap with non-African regions in the desire to limi